University of Southern California Dissertations and Theses

Neuropsychological functioning in older adult long-term cancer survivors

(USC Thesis Other)

Neuropsychological functioning in older adult long-term cancer survivors

PDF

Download a page range

Download transcript

Copy asset link

Request this asset

Transcript (if available)

Content

NEUROPSYCHOLOGICAL FUNCTIONING IN OLDER ADULT
LONG-TERM CANCER SURVIVORS

by
Lara Heather Heflin

A Dissertation Presented to the
FACULTY OF THE GRADUATE SCHOOL
In Partial Fulfillment of the
Requirements for the Degree
DOCTOR OF PHILOSOPHY
(PSYCHOLOGY)

December 2007

Copyright 2007 Lara Heather Heflin

ii
Acknowledgements

I would like to thank the National Institute of Health, whose funding helped
make this project possible through the support of the following grants: T32
AG00037, R03 CA119697, R01 AG08724, and P50 AG05142.
My deepest thanks to the people who gave support and assistance on this
project. Thanks first to my mentor and advisor, Beth Meyerowitz, who had the
patience, knowledge, and skill to train me as a doctoral student and to guide the
development of this research. Thanks also to Margy Gatz, who has been like a
second mentor and advisor, helping train me and providing hours of methodological
and SAS help.
Thanks to my colleagues at the Karolinska Institute in Sweden, who, along
with Margy Gatz, designed the Swedish Twin studies, painstakingly collected data,
kept meticulous documentation, and shared their data. Particular thanks to Paul
Lichtenstein, Nancy Pedersen, and Per Hall for their feedback. Thanks to my
colleagues at the ADRC and the CSP, and in particular, Wendy Mack, who assisted
with CSP data, provided comprehensive statistical guidance, and served on my
committee; Helena Chui, who is PI on the ADRC study; Peg Balcius, who completed
the database match between the ADRC and CSP; and Linda Clark, who assisted with
a number of ADRC data-related issues.
Many thanks to my committee members, including Kathleen Chambers and
David Schwartz, who helped shape the project, and provided ongoing feedback and
guidance.

iii
Thanks to Carol McCleary, whose expertise and training not only informed
this research, but also significantly influenced my professional development and
goals.
I’d also like to acknowledge the friends and family who helped sustain me
throughout this process. Thanks to my parents, Dawn and Mike Allard, and Harry
and Connie Heflin, who have given far more than I can sum up here. Thanks to my
fiancée, Ed Stephens, who has supported my endeavors with endless patience,
understanding, and confidence. Thanks to my sister, Tanya Rice, and my dearest
friends, Sindy Oh, Ashley Borders, Jessica Brommelhoff, and Melanie Friedlander,
all of whom have provided innumerable hours of social and instrumental support
along with their friendship. Thanks to my wonderful classmates, Maggi Mackintosh,
Alisha Alleyne, Steven David, Robin Toblin, and Tania Abou-Ezzadine, and to
Poorni Otilingam, Kysa Christie, and Kathy Stephens, for warm friendship and
support. Thanks to Alan Burke, for his social support and computer expertise.
Thanks, finally, to the countless friends, colleagues, and professors who
assisted me along the way; you know who you are, and you have my gratitude.

iv
Table of Contents
Acknowledgements ii
List of Tables v
List of Figures x
Abstract xi
Chapter 1: Introduction 1
Chapter 2: Study 1 52
Method 53
Results 69
Discussion 85
Chapter 3: Study 2 96
Method 97
Results 112
Discussion 166
Chapter 4: General Discussion 176
References 182

v
List of Tables
Table 1. Brain regions and cognitive functions measured by 21
neuropsychological tests

Table 2. Articles examining neuropsychological functioning in cancer 25
survivors

Table 3. Comparing individuals with complete data (for significant 72
(p<.10) Hypothesis 1 predictors) to individuals with incomplete
data: continuous variables

Table 4. Comparing individuals with complete data (for significant 72
(p<.10) Hypothesis 1 predictors) to individuals with incomplete
data: categorical variables

Table 5. Comparing individuals with complete data (for all Hypothesis 73
1 predictors) to individuals with incomplete data: continuous
variables

Table 6. Comparing individuals with complete data (for significant and 73
nearly significant Hypothesis 1 predictors) to individuals with
incomplete data: categorical variables

Table 7. Spearman’s correlations between Study 1 variables for 74
Hypothesis 1 sample (N=357 twin pairs).

Table 8. Pearson’s correlations between cognitive battery tests (N=43 75
matched pairs)

Table 9. Demographic and covariate descriptive statistics for Hypothesis 77
1 – twin pairs discordant for cognitive dysfunction

Table 10. Cancer Diagnoses in Long-term Cancer Survivors for 78
Hypothesis 1 - twin pairs discordant for cognitive dysfunction

Table 11. Demographic and covariate statistics to check match precision 79
for Hypothesis 2 - matched pairs

Table 12. Cancer diagnoses in long-term survivors for Hypothesis 2 – 80
matched pairs (n=43 pairs)

Table 13. Conditional logistic regressions predicting cognitive dysfunction 80
with long-term cancer history and each potential covariate

vi
Table 14. Conditional logistic regression predicting cognitive dysfunction 81
with long-term cancer history and covariates

Table 15. Conditional logistic regressions predicting cognitive 82
dysfunction with long-term cancer history and each potential
covariate (n=189 pairs)

Table 16. Comparing cognitive test performance of Swedish Twin long- 84
term cancer survivors to matched cancer history-free controls

Table 17. Date range tests were included in the ADRC battery, and % 117
missing and not administered data for long-term cancer survivors

Table 18. Check of match precision for ADRC long-term cancer survivors 121
and cancer history-free comparison group

Table 19. Descriptive statistics for individual growth parameters for 123
separate within-person regressions as a function of years since 1st
test - long-term cancer survivors and comparison group

Table 20. Check of match precision for cancer survivors diagnosed with 124
cancer during their participation in the ADRC and cancer
history-free comparison group

Table 21. Descriptive statistics for individual growth parameters from 126
separate within-person regressions as a function of years since 1st
test - people diagnosed with cancer during study and comparison
group)

Table 22. Estimates of fixed effects, variance components, and fit indices 129
for individual growth models predicting Mini Mental State
Examination (MMSE) scores for long-term cancer survivors and
comparison group

Table 23. Estimates of fixed effects, variance components, and fit indices 130
for individual growth models predicting Animals scores for
long-term cancer survivors and comparison group

Table 24. Estimates of fixed effects, variance components, and fit indices 131
for individual growth models predicting FAS scores for long-term
cancer survivors and comparison group

vii
Table 25. Estimates of fixed effects, variance components, and fit indices 132
for individual growth models predicting Boston Naming Test
scores long-term cancer survivors and comparison group

Table 26. Estimates of fixed effects, variance components, and fit indices 133
for individual growth models predicting Word List Learning scores
for long-term cancer survivors and comparison group

Table 27. Estimates of fixed effects, variance components, and fit indices 134
for individual growth models predicting Word List Delayed Recall
scores for long-term cancer survivors and comparison group

Table 28. Estimates of fixed effects, variance components, and fit indices 135
for individual growth models predicting Word List Recognition
scores for long-term cancer survivors and comparison group

Table 29. Estimates of fixed effects, variance components, and fit indices 136
for individual growth models predicting Figure Copying scores for
long-term cancer survivors and comparison group

Table 30. Estimates of fixed effects, variance components, and fit indices 137
for individual growth models predicting Digit Span Forward scores
for long-term cancer survivors and comparison group

Table 31. Estimates of fixed effects, variance components, and fit indices 138
for individual growth models predicting Digit Span Backward
scores for long-term cancer survivors and comparison group

Table 32. Estimates of fixed effects, variance components, and fit indices 139
for individual growth models predicting Geriatric Depression Scale
scores for long-term cancer survivors and comparison group

Table 33: Repeated measures ANOVA comparing neuropsychological 140
functioning of long-term cancer survivors to cancer-free individuals

Table 34. Mixed models predicting MMSE with time since long-term 142
cancer diagnosis

Table 35. Mixed models predicting Animals with time since long-term 143
cancer diagnosis

Table 36. Mixed models predicting FAS with time since long-term cancer 144
diagnosis

viii
Table 37. Mixed models predicting Boston Naming Test (BNT) with time 145
since long-term cancer diagnosis

Table 38. Mixed models predicting Word List (WL) Learning with time 146
since long-term cancer diagnosis

Table 39. Mixed models predicting WL Delayed Recall with time since 147
long-term cancer diagnosis

Table 40. Mixed models predicting WL Recognition with time since 148
long-term cancer diagnosis

Table 41. Mixed models predicting Figure Copying with time since 149
long-term cancer diagnosis

Table 42. Mixed models predicting Digit Span Forward with time since 150
long-term cancer diagnosis

Table 43. Mixed models predicting Digit Span Backward with time since 151
long-term cancer diagnosis

Table 44. Mixed models predicting Geriatric Depression Scale with time 152
since long-term cancer diagnosis

Table 45. Estimates of fixed effects, variance components, and fit indices 154
for individual growth models predicting Mini Mental State
Examination (MMSE) scores for individuals diagnosed with
cancer during the ADRC study, and matched comparison group

Table 46. Estimates of fixed effects, variance components, and fit indices 155
for individual growth models predicting Animals scores for
individuals diagnosed with cancer during the ADRC study, and
matched comparison group

Table 47. Estimates of fixed effects, variance components, and fit indices 156
for individual growth models predicting FAS scores for individuals
diagnosed with cancer during the ADRC study, and matched
comparison group

Table 48. Estimates of fixed effects, variance components, and fit indices 157
for individual growth models predicting Word List Learning
scores for individuals diagnosed with cancer during the ADRC
study, and matched comparison group

ix

Table 49. Estimates of fixed effects, variance components, and fit indices 159
for individual growth models predicting Word List Delayed Recall
scores for individuals diagnosed with cancer during the ADRC
study, matched comparison group

Table 50. Estimates of fixed effects, variance components, and fit indices 160
for individual growth models predicting Geriatric Depression Scale
scores for individuals diagnosed with cancer during the ADRC
study, matched comparison group

Table 51. Estimates of fixed effects, variance components, and fit indices 164
for individual growth models predicting Animals scores and
controlling for depression for individuals diagnosed with cancer
during the ADRC study, and matched comparison group

Table 52: Repeated measures ANOVA comparing neuropsychological 165
functioning of cancer survivors diagnosed with cancer during the
ADRC study and matched cancer-free individuals

Table 53. Comparing cognitive test performance of Swedish Twin 178
long-term cancer survivors to ADRC long-term cancer survivors
at Time 1

x
List of Figures
Figure 1. Number of twin pairs with and without cognitive dysfunction 76
who were cancer survivors

Figure 2. Sample selection for ADRC long-term cancer survivors 114

Figure 3. Prototypical performance trajectories on Animals for 160
individuals diagnosed with cancer during ADRC study and
matched cancer history-free comparison group

Figure 4. Prototypical performance trajectories on Word List Learning 161
for individuals diagnosed with cancer during ADRC study and
matched cancer history-free comparison group

Figure 5. Prototypical depression trajectories on Geriatric Depression 162
Scale for individuals diagnosed with cancer during ADRC study
and matched cancer history-free comparison group

xi
Abstract
Cancer and its treatment have been associated with a range of
neuropsychological deficits, but the presence and temporal pattern of such problems
in long-term cancer survivors are poorly understood. Moreover, although cognitive
problems are likely to become most pronounced in older age, few studies have
focused upon older cancer survivors or controlled for potential confounding
variables. This research investigated cognitive functioning in older adult long-term
cancer survivors in two studies. Using a co-twin control design, the first study found
long-term cancer survival remained a significant risk for cognitive dysfunction,
controlling for exercise, education, and depression. The first study also found long-
term cancer survivors performed significantly worse than matched controls on verbal
and visual recognition, but not on recall, verbal functioning, visuoconstruction, or
information processing speed. The second study failed to replicate this pattern of
deficits, finding no differences on any neuropsychological tests between long-term
cancer survivors and a matched comparison group of individuals without a cancer
history. However, individuals for whom pre- and post-cancer neuropsychological
assessments were available showed significant declines on word list learning,
compared to matched individuals without a cancer history. Implications of these
findings are discussed.

1
Chapter 1: Introduction
Improvements in cancer screening and treatments have contributed to longer
survival for many cancer patients (Howe et al., 2006; Jemal et al., 2004), increasing
the importance of identifying long-term side effects of treatments. Although
cognitive functioning has not received as much attention in the research literature as
have other side effects of cancer treatments such as fatigue, adequate cognitive
functioning is necessary for people to continue their careers, leisure activities, and
daily life (Bradley et al., 2005; Meyers, 2000). The relative dearth of research on
long-term cognitive deficits following cancer treatment provides cancer patients and
medical professionals few answers about the duration and severity of cognitive
problems to expect after cancer.
In addition, the focus within the existing research on relatively young cancer
patients has prevented an examination of possible interactions between cancer-
related cognitive deficits and normal age-related cognitive decline that other
researchers have proposed should be examined (Ahles & Saykin, 2007; Ahles et al.,
2002). Just as cognitive declines can occur in older age as part of a progressive
dementia process (Reischies, 1998), as long-term cancer survivors age, cognitive
changes that appeared relatively subtle and innocuous at younger ages could progress
to more serious cognitive deficits. In contrast, younger individuals might be able to
more easily compensate for any cognitive difficulties (Stern, 2003) experienced due
to cancer and thus mask some of the deficits they might experience. As an indication
of the relative scarcity of information on older adult cancer survivors, only one study

2
(van Oosterhout et al., 1996) out of twenty-eight in a recent meta-analysis of
neuropsychological effects of cancer treatments assessed participants at a mean age
of 65 years or older (Anderson-Hanley et al., 2003).
Only recently have investigators begun focusing attention on cognitive
functioning in older adult cancer survivors (Heflin et al., 2005; Buckwalter, Crooks,
& Petitti, 2005; Hurria et al., 2006; Eberhardt et al., 2006). The lack of research on
cognitive functioning in older adult cancer survivors is particularly distressing in
light of our recent research showing greater serious cognitive dysfunction in older
adult cancer survivors compared to co-twin controls (Heflin et al., 2005), and
prospective longitudinal research that suggests subgroups of older adults experience
cognitive declines post-treatment (Hurria et al., 2006). This study focuses primarily
upon older adult cancer survivors in the hopes of better understanding the association
between cancer survival and cognitive functioning in older age.
Presence of cognitive deficits in long-term cancer survivors
Among the most fundamental questions related to this area of research is
whether cognitive problems that are experienced shortly after cancer treatment
portend long-term cognitive deficits. In theory, if cognitive problems related to
treatment are transient, they are likely to be perceived as more benign and acceptable
side-effects of cancer treatment. In contrast, if cognitive problems related to
treatment are lasting, patients would likely want to have that information in order to
make treatment decisions consonant with their values and to plan for future
activities, such as returning to work.

3
Previous research does not unanimously support the existence of long-term
cognitive problems in cancer survivors. There are a number of studies that have
found short-term deficits, however. During and immediately after treatment, cancer
patients have been found to have cognitive deficits in attention, executive
functioning—defined here as the higher-order skills, such as planning, organization,
mental multi-taking, self-monitoring, and flexible thinking, that influence how
individuals problem-solve and perform other cognitive tasks (Zelazo et al., 1997;
Lezak et al., 2004)—information processing speed, motor functioning, memory, and
language (Silberfarb et al., 1980; Herzig et al., 1983; Brezden et al., 2000; Wieneke
& Dienst, 1995; Tchen et al., 2003). A meta-analysis that primarily reviewed studies
of cancer survivors within 5 years following adjuvant treatment found consistent
deficits in executive functioning, verbal memory, and motor functioning (Anderson-
Hanley et al., 2003). These studies suggest that findings of measurable cognitive
deficits within the first five years following systemic cancer treatment are quite
robust.
In contrast, research findings in long-term cancer survivors—those survivors
5 years or more post-diagnosis—have been more equivocal. Some research has
suggested that long-term cancer survivors experience cognitive deficits (Heflin et al.,
2005; Ahles et al., 2002; van Oosterhout et al., 1996; Cull et al., 1994). Our previous
research found that older adult long-term survivors of heterogeneous cancers had
increased risk for serious cognitive dysfunction compared to co-twin controls (Heflin
et al., 2005). Other cross-sectional studies with more homogenous samples and

4
sensitive neuropsychological tests have similarly found cancer survivors, particularly
those treated with chemotherapy (Ahles et al., 2002) or chemotherapy and
prophylactic cranial irradiation (van Oosterhout et al., 1996), to be at increased risk
for cognitive dysfunction. Ahles and colleagues (2002) found poorer verbal memory
and information processing speed in cancer survivors approximately 9 years post-
chemotherapy compared to cancer patients who received only local treatment.
Moreover, poorer cognitive performance was correlated with having received more
cycles of chemotherapy, lending support to the idea that cancer treatment may be
responsible for long-term cognitive deficits.
Other research has cast doubt on the idea that cancer survivors are likely to
experience long-term cognitive problems. Evidence for this perspective comes from
a cross-sectional epidemiological study that failed to find significant differences in
gross cognitive functioning among older adult female cancer survivors compared to
older females who had never been hospitalized for cancer (Buckwalter, Crooks, &
Petitti, 2005), and studies finding greater evidence for cognitive improvement than
declines following cancer treatment (Mar Fan et al., 2005; Wefel et al., 2004b;
Fliessbach et al., 2003; Schagen et al., 2002; Kaasa, Olsnes, and Mastekaasa, 1988).
It is possible that Buckwalter and colleagues (2005) failed to find significant risk for
cognitive dysfunction among cancer survivors due to their reliance on self-report of
cancer history, which the authors acknowledge has been found to be inaccurate and
could be influenced by poor cognitive status, and their use of a fairly insensitive
cognitive measure. More interesting is the second group of studies, which found that

5
only small subgroups appear to demonstrate cognitive decline following cancer
treatment, with little evidence of cognitive declines for most cancer survivors
following cancer treatment. These studies arguably provide the strongest evidence
that most long-term cancer survivors may not experience significant cognitive
deficits.
If most cancer patients’ cognitive functioning is observed to improve or
remain stable rather than decline after treatment, as these studies suggest, it would
appear unlikely that cancer treatment causes either early or long-term cognitive
problems in most cancer survivors. Mar Fan and colleagues (2005) recently found
that the percentage of cancer survivors who demonstrated moderate-severe cognitive
dysfunction declined from 16% pre-treatment to only 3.8% after two years.
Similarly, Wefel and colleagues (2004b) found that although 61% of breast cancer
patients had declined cognitively 3 weeks after treatment compared to pre-treatment,
45% of those patients had improved at 1 year post-treatment and an additional 45%
had stabilized, presumably at their lower level of functioning, with 10% showing
some improvement and some declines. Even cancer patients who do not show
improvements initially may improve over time. Schagen and colleagues (2002)
found that, compared to their performance on a previous assessment at 2 years post-
treatment, cancer patients treated with chemotherapy demonstrated improvement on
a range of neuropsychological tests, even while a comparison group of patients
treated locally demonstrated declines. Interestingly, the improvements occurred

6
between 2 and 3.5 years post-chemotherapy, suggesting that cognitive improvement
might continue for years following cancer treatment.
Additional evidence for cognitive improvement post-treatment was found by
Fliessbach and colleagues (2003), who assessed the neuropsychological functioning
of 10 central nervous system cancer patients prior to treatment with chemotherapy, 4
months after treatment, 12 months after treatment, and 21-69 months after treatment.
None of the cancer patients showed significant declines over time compared to their
pre-treatment performance, and the cancer patients who demonstrated deficits prior
to cancer treatment improved over time. Kaasa and colleagues (1988) also found
cognitive improvement or no cognitive changes after a 14-week follow-up of cancer
patients treated with chemotherapy or radiation. However, given the brief period of
time between neuropsychological test administrations, and the lack of a healthy
comparison group that may also show improvement over time, it is possible that
practice effects artificially inflated the post-treatment test results of both these
studies.
Of note, the studies that have found cognitive deficits in long-term cancer
survivors have been retrospective cross-sectional studies (Heflin et al., 2005; Ahles
et al., 2002; van Oosterhout et al., 1996; Cull et al., 1994), whereas studies that have
collected cognitive functioning data at more than one time point have found
cognitive improvement from pre-treatment to post-treatment or at two time points
post-treatment (Fliessbach et al., 2003; Schagen et al., 2002). Because longitudinal
studies can measure changes in cognitive functioning over time and can therefore

7
determine whether cancer survivors experience decline rather than simply deficits, it
is tempting to put greater faith in the results of those longitudinal studies than the
cross-sectional ones and conclude that any cognitive deficits that cancer survivors
experience shortly after treatment will likely lessen over time. However, differential
attrition rates, where cancer survivors with cognitive deficits were less likely to
participate in follow-up assessments (Schagen et al., 2002), small sample sizes
(Wefel et al., 2004b; Fliessbach et al., 2003; Schagen et al., 2002), a focus on brain
tumor patients (Fliessbach et al., 2003), and a failure to use control groups to account
for practice effects (Wefel et al., 2004b; Fliessbach et al., 2003; Kaasa, Olsnes, and
Mastekaasa, 1988) diminish the credibility and generalizability of the existing
longitudinal studies’ findings. Despite these limitations in the longitudinal studies,
given the absence of clear indications from longitudinal studies that cancer survivors
experience post-treatment declines, it is unclear why retrospective studies have
repeatedly found deficits in cancer survivors (Anderson-Hanley et al., 2003).
Several possibilities exist that could explain the discrepancy between cross-
sectional and longitudinal studies. It is possible that the presence of cognitive deficits
in cross-sectional studies of cancer survivors is due largely to cognitive deficits that
existed prior to cancer treatment, perhaps due to third variable risk factors that
increase risk for both cancer and cognitive problems. This possibility will be
explored in greater depth later in this paper. In addition, the cross-sectional studies
focused upon individuals who had survived cancer for longer periods of time, with
average cancer survival lengths of 3.9 to 14 years (Heflin et al., 2005; Ahles et al.,

8
2002; van Oosterhout et al., 1996; Cull et al., 1994), compared to longitudinal
studies, which tested individuals 2-3.5 years post-treatment (Fliessbach et al., 2003;
Schagen et al., 2002). It is therefore possible that some cancer treatments cause late
cognitive effects or damage that results in a slow, progressive decline most clearly
observed after some time, and as increasing age puts individuals at risk for cognitive
deficits more generally. A study with multiple time points of cognitive functioning
later in cancer survival could address this last possibility by examining whether long-
term cancer survivors experience more rapid rates of cognitive decline than
individuals without a cancer history. The existence of a more rapid rate of cognitive
decline would support the idea of damage that causes progressive cognitive deficits
in cancer survivors, and help explain why later cross-sectional studies have found
deficits in long-term cancer survivors.
Trajectories of cognitive functioning in older adult long-term cancer survivors
Research thus far has provided little information about the progression of
cognitive functioning over time in cancer survivors, relative to the general
population. Because most of the studies that have suggested the presence of long-
term effects have been retrospective cross-sectional studies (Heflin et al., 2005;
Ahles et al., 2002; van Oosterhout et al., 1996; Cull et al., 1994), we cannot know
whether the cognitive functioning of cancer patients is declining at a faster rate than
other individuals, or whether they experienced an initial decline in cognitive
functioning from which they never fully recovered. This distinction is important for
two reasons: 1) the presence of lasting effects from a discrete insult in cancer

9
survivors may implicate different mechanisms than a more rapid rate of decline, and
2) it is important for cancer survivors to be aware of possible courses of their
cognitive functioning, so that they can be vigilant for declines later in life, or learn
how to cope permanently with stable long-term effects if needed.
The theory of cognitive reserve (Satz, 1993; Katzman, 1993; Stern, 2002)
offers one way to understand the cognitive deficits found in long-term cancer
survivors. The cognitive reserve theory posits that individuals have a certain amount
of “brain reserve capacity” that buffers them against showing cognitive problems
(Stern, 2003). Factors such as higher levels of education (Stern, 2003), genetic
predisposition to high intelligence (Lee, 2003), or lifestyle factors (Scarmeas &
Stern, 2003) might increase cognitive reserve, allowing individuals to sustain more
neurological injury or greater progression of neurological disease before
demonstrating cognitive problems (Katzman, 1993). Discrete events causing
neurological damage might decrease cognitive reserve, and place individuals closer
to the threshold at which cognitive problems are observed (Ropacki & Elias, 2003).
In line with this theory, cognitive dysfunction among cancer survivors may be the
result of a relatively circumscribed cognitive insult that places them closer to the
threshold at which an individual meets criteria for having cognitive dysfunction. The
current body of research on long-term cancer survivors, which has found cognitive
deficits in long-term cancer survivors but has failed to find consistent evidence of
more rapid decline in cancer survivors than in healthy comparison groups (Mar Fan,
2005; Wefel et al., 2004b; Fliessbach et al., 2003; Schagen et al., 2002; Kaasa,

10
Olsnes, and Mastekaasa, 1988), is consistent with the idea of a cancer-related
cognitive insult that decreased cognitive reserve.
Alternatively, cognitive dysfunction among long-term cancer survivors may
be the result of accelerated cognitive decline over time caused by an interaction
between cancer-related cognitive deficits and the normal aging process (Ahles &
Saykin, 2007; Ahles et al., 2002). This perspective suggests that cancer treatments
might interact with biological factors to increase the rate at which cognitive decline
occurs (Ahles & Saykin, 2007). Biological factors that researchers have suggested
might contribute to such interactions include DNA-repair, neural plasticity, and
neural repair genes that could influence the ease with which the brain could recover
from neurological damage caused by cancer treatment (Ahles & Saykin, 2007). If
such interactions occur, the effects would likely be seen in older age, and manifest in
a more rapid rate of cognitive decline in cancer survivors than in similar individuals
without a history of cancer. As mentioned earlier, however, few studies have focused
upon older adult cancer survivors, leaving both the course and type of cognitive
functioning in older adult cancer survivors very poorly understood.
Pattern of cognitive deficits in long-term cancer survivors
Localization of deficits
Examining the pattern of neuropsychological deficits within the context of
systems or regions of the brain that appear to be producing the deficits can provide
an organizational framework for understanding the deficits. This process is referred
to as localization of deficits, or localization of damage (Lezak et al., 2004; Damasio

11
& Damasio, 1989). Unlike early theories of localization that postulated “centers” of
functioning, more current theories of localization postulate that “parallel, serial,
recursive, and nonlinear” (p.12) neural systems are involved in producing cognitive
abilities (Damasio & Damasio, 1989). Although neuropsychologists rely upon this
process of localization of deficits to aid in differential diagnosis, there exists some
debate about whether neurological localization of deficits is a reliable process.
One reason for the skepticism about localization is simply that the brain is
such a complex organ, with numerous circuits connecting disparate regions and
working together to perform even relatively simple tasks (Lezak et al., 2004; Nolte,
2002; Zakzanis et al., 1999). Because of neurological interconnections and the
multiple systems upon which cognitive abilities depend, a specific cognitive problem
can result from damage to any one of multiple regions of the brain (Lezak et al.,
2004; Damasio & Damasio, 1989). Exacerbating the problem, most
neuropsychological tests do not measure abilities in isolation from one another, but
require multiple cognitive abilities in order for a person to successfully complete a
test (Lezak et al., 2004). For instance, a neuropsychological test that asks individuals
to name the objects represented in pictures (e.g., the Boston Naming Test) seems to
require at least the following abilities: accurate perception of the stimuli, recognition
of the objects pictured as something familiar, knowledge of and access to what the
object depicted is, knowledge of and access to the name of the objects pictured (the
phonological representation of the objects), and the motor abilities to articulate the
name of the object (DeLeon et al, 2007). Although such a test is believed to be a

12
good measure of verbal abilities, identifying its constituent cognitive components
suggests that diverse areas of the brain are likely to be involved in completing the
task correctly. In short, the wide distribution of neurological circuits required to
perform cognitive tasks, and the difficulty in assessing specific cognitive functions
with current neuropsychological tests, would thus seem to make localization of
deficits a highly inaccurate task. In addition, evidence from imaging research shows
great diversity in the structure of neuroanatomical regions, and the neurological areas
that appear active when individuals are engaged in cognitive tasks (Brazis et al.,
2007; Amunts & Zilles, 2006). This diversity undermines the basic premise of
consistency of neuroanatomical structure and function across individuals that is
important for localization of deficits.
Despite these factors, neuropsychological deficits or syndromes have been
found to reliably indicate specific neurological areas of damage or diagnoses,
supporting the idea that it is possible to examine patterns of deficits and preserved
abilities to infer areas or systems that are likely to be damaged (Lezak et al., 2004;
Waxman, 2000; Damasio & Damasio, 1989). Animal studies that demonstrate
specific functions are associated with discrete regions or systems also support
localization of cognitive function (e.g., Treit et al., 1993), as does human research
that demonstrates dissociation between systems in producing abilities (e.g.,
Rascovsky et al., 2002; Perry et al., 2000; Winocur et al., 1996). In addition, an
exhaustive meta-analysis found patterns of neuropsychological deficits can
differentiate several common neuropsychiatric syndromes, and that these syndromes

13
are typically characterized by deficits most pronounced in the neuropsychological
tests which are used to assess the neurological region that have been found to be
most damaged in the syndrome (Zakzanis et al., 1999). These bodies of research
suggest that, despite the difficulties with localization described earlier, there remains
good evidence that disruption in specific cognitive abilities can reliably implicate
damage or dysfunction in distinct neurological areas.
Brain regions and functions
A basic understanding of the neurological substrates of brain systems and the
neuropsychological sequelae of dysfunction provides a foundation for interpreting
the pattern of results from studies of long-term cancer survivors. The focus in this
section is constrained to the neurological structure of the two systems for which the
literature on neuropsychological deficits in adult cancer survivors (discussed
immediately following the overview of functional neuroanatomy) most frequently
suggests deficits: the frontal-subcortical system, and the temporal lobe system.
The frontal lobes, within the frontal-subcortical system, are comprised of
three main areas: the motor cortex, the premotor cortex, and the prefrontal cortex
(Snell, 2001). The motor cortex contains descending projections to subcortical
structures involved in producing goal-directed movement, including the spinal motor
neurons, basal ganglia, and red nucleus (Snell, 2001; Waxman, 2001). Through these
connections, the motor cortex produces the body’s movements (Snell, 2001;
Waxman, 2001). The motor activity that the motor cortex produces is directed by
information received from the premotor cortex, which “store(s) programs of motor

14
activity assembled as the result of past experience,” (Snell, 2001, p. 287). The
premotor cortex also contains descending corticospinal projections, which are
believed to contribute “speed and agility” to volitional movements (Afifi &
Bergman, 1998, p. 347). Consequently, slowed motor functioning or difficulty with
fine motor functioning can result from damage to the pre-motor cortex, or from
damage to the subcortical motor circuits (Afifi & Bergman, 1998; Jason, 1990).
Verbal fluency can also be affected by damage to the pre-motor strip or motor
circuits, because rapid word generation appears to involve the left pre-motor cortex
(Condon et al., 1997).
The prefrontal cortex contains the dorsolateral, orbitofrontal, and medial
cortices. It is connected to the basal ganglia, thalamus, hypothalamus, and several
cortical areas by a series of circuits that are involved in cognitive, emotional,
behavioral, and motor functioning (Snell, 2001; Troster, 1998). The prefrontal cortex
uses these circuits to plan and initiate purposeful movements and behavior (including
emotional expression) in response to internal information, goals, previous
experiences, and external factors (Waxman, 2000).
The different regions of the prefrontal cortex contain projections to different
areas of the brain, and can result in distinct types of deficits. The dorsolateral cortex
has connections to the parietal cortex, which in turn projects to the cingulate cortex,
basal ganglia, and superior colliculus (Waxman, 2000, Nolte, 2002). Damage to this
system can result in working memory deficits, and in personality changes such as
indifference (Waxman, 2000). The orbitofrontal cortex is more closely associated

15
with the temporal lobe system, with connections to the amygdala and temporal
cortex (Nolte, 2002). Damage to the orbitofrontal system is associated with
disinhibition, attentional deficits, poor social behavior, and poor judgment (Waxman,
2000). This portion of the prefrontal cortex also contains Broca’s area, which is
important for the motor movements necessary to produce speech (Snell, 2001).
Damage to the medial prefrontal cortex can result in difficulty initiating activities
(Brazis et al., 2007; Waxman, 2000).
Furthermore, because frontal lobe systems are involved in organization
(Fletcher et al., 1998) and memory search strategies (Janowsky et al., 1989), damage
to frontal-subcortical circuits can affect performance on a variety of tests that are not
clearly measures of frontal lobe functioning. On tasks that require organizing
complex or disorganized information, individuals with frontal-subcortical damage
are likely to perform more poorly than peers (Lezak et al., 2004). For example, poor
performance on the copying component of the Rey-Osterreith Complex Figure Test
can be the result of poor organization of the visual information and poor planning
when the task is approached. Recall of disorganized or highly complex information
also suffers in individuals with frontal-subcortical damage because disorganized
information is more difficult than organized information to recall (Tremont et al.,
2000; Lezak et al., 2004). Learning is frequently diminished due to slowed
information processing speed (Kieseppa et al., 2004); individuals with subcortical
damage appear unable to process information rapidly enough to learn it while
attending to the next item they are to learn. Additionally, difficulty with memory

16
search strategies can result in poor recall of learned material, particularly for visual
material (Busch et al., 2005; DeCarli et al., 1995).
Certain abilities are typically relatively well-preserved in the context of
frontal-subcortical system damage, however. Although difficulty retrieving
information due to poor memory search strategies can result in poor performance on
recall tasks, individuals with frontal lobe damage can typically perform well on
recognition tasks (Jetter et al., 1986). And while learning new information may be
slow due to difficulty organizing and attending to information, that which is learned
is able to be recognized even after long delays (Lezak et al., 2004). This difficulty
with recall in the presence of preserved recognition occurs because the temporal
lobes, which are responsible for memory consolidation, are intact (Du et al., 2005).
The intact temporal lobes enable individuals to learn, albeit at a pace determined by
how quickly they are able to process the information (Kieseppa et al., 2005). The
information is then learned and available within the brain, but unable to be accessed
easily (Tullberg et al., 2004).
The temporal lobe system is comprised of two main areas: the temporal
cortex, and medial temporal structures. The medial temporal structures include the
amygdala, hippocampus, cingulate cortex, parahippocampal (or entorhinal) cortex,
and perirhinal cortex (Squire, Stark, & Clark, 2004; Nolte, 2002). Several of the
temporal lobe structures and systems are essential for language functions (Lezak et
al., 2004; Nolte, 2002; Hamberger et al., 2001; Sawrie et al., 2000). Projections
from the thalamus to primary auditory cortex, middle temporal gyrus, and auditory

17
association cortex appear to be involved in hierarchical auditory processing, or the
processing of basic stimuli such as phonemes into recognizable words (Liebenthal et
al., 2005; Wessinger et al., 2001). The auditory projection system connects to the
posteriorally-located Wernicke’s area and the surrounding cortex, and is necessary
for language comprehension (Lezak et al., 2004). A system involving the primary
auditory cortex, the superior temporal gyrus, the supramarginal gyrus, and portions
of the frontal lobe is active in localizing sounds (Brunetti et al., 2005). Within the
non-language dominant hemisphere (typically the right hemisphere), temporal lobe
connections to the occipital lobe are important for visuospatial tasks, including
recognizing and understanding visual material, processing the ‘gestalt’ of visual
stimuli, learning and remembering visual material, and visuoconstruction (Lezak et
al., 2004; Doyon & Milner, 1991). Systems within the right posterior portion of the
temporal lobe similar to those in the left hemisphere are involved in understanding
the melodic components of speech, or prosody (Kandel et al., 1995).
As alluded to above, the structures and systems of the temporal lobes, and
particularly the medial temporal lobes, are essential for verbal and visual learning,
memory consolidation, and verbal fluency (Du et al., 2003; Nolte, 2002; Pihlajamaki
et al., 2001; Squire & Zola-Morgan, 1991). Sensory and emotional input is sent from
a number of regions, including the thalamus and amygdala, to the entorhinal cortex
and then on to the hippocampus, providing the hippocampus with internal and
external information (Nolte, 2002). These projections are highly involved in learning
and long-term memory (Olsen et al., 2007; Fletcher et al., 1995). Within this system,

18
the hippocampus is believed to be one of the most central structures for learning and
consolidating memories for this new information (Squire, Stark, & Clark, 2004; Du
et al., 2003); when there is damage to the hippocampus, learning and memory of new
information can be profoundly affected. Individuals with hippocampal damage are
likely to retain the ability to repeat strings of numbers or other information, a task
that is more dependent upon attention than learning, but have difficulty retaining that
information over time (Nolte, 2002).
There are substantial interconnections between the frontal-subcortical system
and the temporal lobe system, and both systems are involved in a number of related
cognitive and emotional functions. For instance, semantic memory or knowledge, or
knowledge of factual information, requires the left temporal cortex, cingulate cortex,
and frontal association areas (Afifi & Bergman, 2005). In addition, both systems play
roles in emotional functioning, with the frontal-subcortical (and particularly the
orbitofrontal cortex system) system contributing to emotion regulation (Bechara et
al., 2000), and the medial temporal system (and particularly the amygdala and its
connections) involved with autonomic arousal and emotional behavior (Afifi &
Bergman, 2005). Indeed, the medial frontal area is closely connected to the limbic
system within the medial temporal lobe, and the overlap between these systems is
involved in initiation of activity and behavioral responses to external stimuli based
upon past experiences (Brazis et al., 2007). Such interconnections make it difficult to
localize deficits in cognitive abilities that require both systems.

19
Fortunately, researchers have demonstrated dissociations between the
systems, particularly when comparing groups of patients who suffer damage
primarily to the temporal systems versus to the frontal systems. One such study
compared patients with Alzheimer’s disease to those with frontotemporal dementia
on “theory of mind” tests, which require people to make judgments about the
thoughts and emotions of others, verbal fluency, mental flexibility/problem solving,
visuospatial functioning, and memory (Gregory et al., 2002). Patients with
frontotemporal dementia had difficulty with “theory of mind” tests, but no difficulty
with memory tests, while patients with Alzheimer’s disease showed the opposite
pattern (Gregory et al., 2002). In addition, the “theory of mind” tests were highly
correlated with MRI-confirmed atrophy in the ventromedial frontal cortex (Gregory
et al., 2002). A retrospective study that compared previously completed
neuropsychological assessments of individuals with autopsy-confirmed Alzheimer’s
disease and frontotemporal dementia found a double dissociation between temporal
system and frontal-subcortical system functions (Rascovsky et al., 2002). These
authors found that the frontotemporal dementia patients performed more poorly than
the Alzheimer’s disease patients on verbal fluency tasks, while the Alzheimer’s
disease patients performed more poorly than the frontotemporal dementia on
memory and visuospatial tasks (Rascovsky et al., 2002). Dissociations have even
been found between two groups of frontotemporal dementia patients— frontal-
variant frontotemporal dementia (patients whose imaging indicated pronounced
frontal degeneration), and temporal-variant frontotemporal dementia (patients whose

20
imaging indicated pronounced temporal degeneration) (Perry & Hodges, 2000). The
temporal-variant frontotemporal dementia patients performed more poorly than the
other groups on tests of semantic memory, but performed well on attention and
executive function tests, whereas the frontal-variant frontotemporal patients
performed more poorly then the other groups on tests of attention and executive
functioning, with preserved semantic memory (Perry & Hodges, 2000). These
findings suggest that memory and semantic memory tasks are differentially sensitive
to temporal lobe system deficits, whereas “theory of mind,” attention, executive
functioning, and possibly verbal fluency tests, are likely differentially sensitive to
frontal lobe deficits.
Table 1 shows the cognitive abilities, and neurological systems understood to
be instrumental in producing these abilities, measured by the tests used in this study.
The substantial overlap of systems in producing cognitive abilities is not represented
in the table; the table is an over-simplification for illustrative purposes. As the table
shows, most tests are believed to require certain basic processes such as attention and
verbal comprehension (which is typically necessary at least for understanding the test
instructions), and therefore all tests could be construed as measuring attention.
However, the best measures of such basic processes are tests that do not also rely
upon many additional processes. When tests of basic processes suggest an individual
is intact, they are unlikely to be responsible for poor performance on higher-order
cognitive tests.

21

Table 1. Brain regions and cognitive functions measured by neuropsychological tests
Neuropsychological Tests: Judgment Information Animals FAS
Boston
Naming
Test
Brain Regions and Functions:
Frontal-Subcortical System:
Attention x x x x x
Working memory
Organization / Planning
Reasoning x
Set shifting/ cognitive flexibility x
Self-monitoring x x
Information retrieval x x x x
Social knowledge x
Motor functioning
Speech/Language production x x x x x
Subcortical System:
Information processing speed x x x
Temporal System
Language comprehension x x x x x
Semantic knowledge x x x
Verbal fluency x x
Verbal memory
Visuospatial functioning x
Visuoconstructional functioning
Visual memory

22

Table 1, Continued
Neuropsychological Tests:
Word List
Learning
Word List
Recall
Word List
Recognition
Prospective
Memory
Brain Regions and Functions:
Frontal-Subcortical System:
Attention x x x x
Working memory
Organization / Planning x x x
Reasoning
Set shifting/ cognitive flexibility
Self-monitoring x
Information retrieval x
Social knowledge
Motor functioning
Speech/Language production x x x
Subcortical System:
Information processing speed x
Temporal System
Language comprehension x x x x
Semantic knowledge x x x
Verbal fluency
Verbal memory x x x
Visuospatial functioning
Visuoconstructional functioning
Visual memory

23

Table 1, Continued
Neuropsychological Tests:
Figure
Copying
Digit
Span
Forward
Digit Span
Backward
Symbol
Digit
Block
Design
Brain Regions and Functions:
Frontal-Subcortical System:
Attention x x x x x
Working memory x x
Organization / Planning x
Reasoning
Set shifting/ cognitive flexibility x
Self-monitoring x x
Information retrieval x x x x
Social knowledge
Motor functioning x x
Speech/Language production x x x
Subcortical System:
Information processing speed x x
Temporal System
Language comprehension x x x x x
Semantic knowledge
Verbal fluency
Verbal memory
Visuospatial functioning x x
Visuoconstructional functioning x x
Visual memory

Note: Due to the high number of neuropsychological tests that exist, only tests used in this study were
included in this table. A bold-type x (x) indicates the cognitive function(s) that the test is believed to
measure best.
References for table: Lezak et al., 2004; Kaufman & Lichtenberger, 1999; Morris et al., 1989;
Schmolck et al., 2002; Kramer et al., 2005; Hamberger et al., 2001; Kliegel et al., 2000.

24
Brain regions affected in long-term cancer survivors
When the literature on cognitive problems in cancer survivors is examined
with the goal of localizing deficits, two broad neurological systems are implicated:
the frontal-subcortical system, and medial temporal lobe system. As can be seen in
Table 2, cognitive domains frequently found to be affected in long-term cancer
survivors include verbal memory, visual memory, information processing speed,
attention and executive functioning (Ahles et al., 2002; van Oosterhout et al., 1996;
Cull et al., 1994), while domains of verbal functioning, motor functioning and
visuospatial functioning have not been found to be worse in long-term cancer
survivors (Ahles et al., 2002; Cull et al., 1994).
The common neurological thread that unites the deficits most frequently
found in long-term cancer survivors is that all depend upon well-functioning frontal-
subcortical circuits, whereas the cognitive areas that have been found to be spared in
long-term cancer survivors are less dependent upon this circuitry (Jetter et al., 1986;
Lezak et al., 2004). When Schagen and colleagues (2002) wrote a review of their
own 4 studies in this area, they concluded that the diffuse pattern of chemotherapy-
related cognitive deficits found in their studies was similar to that found in traumatic
brain injury. It is unlikely, however, that cancer survivors have a higher than average
rate of traumatic brain injury. Because the blunt force of traumatic brain injuries
results in contusions, edema, and shearing as the brain moves through the skull, and
because cancer or its treatment would be unlikely to cause similar damage (with the
possible exception of edema, which as of yet has not been reported in the literature)

25

Table 2. Articles examining neuropsychological functioning in cancer survivors

1st Author Year N Sample
Comparison
Group Study design Treatment
Time
since
Treatment Neuropsychological Findings
Silberfarb 1980 177
Mixed
cancer pts
Psychiatric
pts and
healthy
normal
comparison
group
Cross-
sectional
Pts on
chemo, &
pts on
chemo +
radiation
Current
(for those
on chemo)
Cognitive Capacity Screening Test:
Chemo pts worse than psych pts
Trails B: Chemo pts worse than psych
Digit Symbol: All cancer pts worse
than pts & normals
Zung: No difference
BMAACL anxiety: No difference
BMAACL hostility: No difference
BMAACL depression: No difference
BMAACL distress: Cancer pts less
distressed
Oxman 1980 10
Mixed
cancer pts
Each pt
served as
own control Longitudinal Chemo
Before tx,
24 hours
post-tx, &
1 month
post-tx
Digit symbol: No differences after
chemo
Digit span forward: No differences
after chemo
Digit span backwards: No differences
after chemo
Trails B: No differences after chemo

Johnson 1985 20
Small cell
lung
cancer pts
Compared
pts tx with
chemo +
PCI to pts
tx with
chemo
alone
Cross-
sectional,
comparison
between
non-
randomized
tx grps
PCI +
chemo
2.4-10.6
yrs post-tx
(median=
6.2)
WAIS: VIQ 1 SD > PIQ
Trails: No significant findings
Category test: No significant findings
Aphasia screening battery: No
significant findings
WMS logical memory: Correlated with
memory complaints
WMS visual memory: Correlated with
memory complaints
WMS associate learning: Correlated
with memory complaints
25

26

Table 2, Continued.
1st Author Year N Sample
Comparison
Group Study design Treatment
Time
since
Treatment Neuropsychological Findings
Kaasa 1988 95
Non-small
cell lung
cancer pts
< 70 yrs
old
Chemo grp
compared to
radiation grp
before and
after tx
Longitudinal
randomized
trial
Chemo or
radiation
Before tx
and 14
weeks
post-tx
Benton Visual Retention Test: More
chemo pts performed poorly than
radiation pts
Verbal Learning Test: Improvements
in both grps
Trails A & B: Improvements in both
Tucker 1989 24
Leukemia
or non-
Hodgkin's
lymphoma Norms
Cross-
sectional
Chemo +
cranial
irradiation
(PCI)
Mean =
11.5 years
(range= 4-
13)
WAIS-R: Average VIQ, PIQ, and FSIQ
Verbal memory: Average
Visual memory: Average
Meyers 1991 14
Mixed
cancer pts None
Retro-
spective
Interferon
-alpha
(immuno-
therapy)
Average=
28 weeks
WAIS-R, Dementia Rating Scale,
Logical Memory, Benton Visual
Retention, Wisconsin Card, Trails A &
B, Multilingual Aphasia Exam, Line
Bisection, Finger tapping, Grooved
Pegboard, MMPI:
10 of 14 pts had “findings suggestive
of frontal-subcortical dysfunction”
Meyers 1992 47
Metastatic
solid
tumor pts
Test norms,
& pts not tx
with
biologics
(immuno-
therapy)
Cross-
sectional
Chemo
and some
biologics
At least 3
weeks
Digit Span, Arithmetic, Digit Symbol,
Similarities, Block Design, Logical
Memory, Revised Benton Visual
Retention Test, Trails A & B, Grip
Strength, Finger tapping, FAS: 32%
had > 1 abnormal score (>1.5 SD from
norms); 53% memory problems, 47%
frontal lobe, 40% visual-motor
scanning, and 33% attention. Poor
scores related to biologics (p=.007).
26

27

Table 2, Continued.
1st Author Year N Sample
Comparison
Group
Study
design Treatment
Time
since
Treatment Neuropsychological Findings
Cull 1994 64
Small cell
lung
cancer pts
in
remission
Neuropsych-
ological test
norms
Cross-
sectional
PCI, PCI
+ chemo,
or
radiation
2 yrs or >
post-tx
(mean=3.9
yrs)
National Adult Reading Test: No sig.
findings
Williams delayed recall test: 16%
impaired
Digit symbol: 45% impaired
Complex figure test: 68% impaired-recall
Trails A & B: No sig. findings
Rotterdam Symptom Checklist: No sig.
findings
Hospital Anxiety and Depression scale:
No sig. findings
Wieneke 1995 28
Stage I
and II
breast
cancer pts
Pts own
estimated
pre-morbid
IQ
Cross-
sectional Chemo
1/2-12
months
WAIS-R FSIQ: No sig. findings
Digit Span: 1.5 SD < expected
performance
Digit Symbol: 1.5 SD < expected
performance
PASAT: 1.5 SD < expected performance
Trails A & B: 1.5 SD < expected
performance
CVLT: 1.5 SD < expected performance
Rey-O: 1.5 SD < expected performance
Categories: No sig. findings
Similarities: No sig. findings
COWA: No sig. findings
Block Design: 1.5 SD < expected
performance
Grooved Pegboard: 1.5 SD < expected
BDI: Uncorrelated with dysfunction
27

28

Table 2, Continued

1st Author Year N Sample
Comparison
Group Study design Treatment
Time since
Treatment Neuropsychological Findings
van Oosterhout 1995 14
Small cell
lung
cancer pts-
no brain
metastases
Matched
control
group
Prospective
longitudinal
Chemo &
PCI
At diagnosis,
during
chemo, after
chemo, 1 &
5 months
post-PCI
Groninger Intelligence Test:
No sig. findings
Stroop: Pre-tx pts worse than
control grp
Trails A & B: Pre-tx pts worse
than control grp
15-word test: No sig. findings.
van Oosterhout 1996 78
Small cell
lung
cancer pts
Healthy
controls
Cross-
sectional
Chemo
and PCI
for some
pts.
At least 2
years
(mean=5.4)
AVLT: Pts worse than controls
Digit Span Forwards: Pts
worse than controls
Memory Scanning: Pts worse
than controls
Stroop: Pts worse than
controls
Concept Shifting Test: Pts
worse than controls
Ahles 1998 177
Limited-
stage
small-cell
lung pts
Each pt
served as
own control
Substudy of
randomized
longitudinal
study Chemo
During &
post-chemo,
& post-
radiation
POMS: Chemo pts better
during chemo
Trails B: Improved between
chemo and T2, but worsened
between T2 and T3
28

29

Table 2, Continued

1st Author Year N Sample
Comparison
Group Study design Treatment
Time
since
Treatment Neuropsychological Findings
van Dam 1998 104
Breast
cancer
pts
Compared
high dose
chemo pts,
standard
dose chemo
pts, and no
chemo pts
(matched on
age & time
since tx)
Randomized
cross-
sectional
Chemo,
tamoxifen,
&
radiation
Average =
2 yrs (at
least 6
months)
since
chemo
AVLT: No sig. findings
Rey-O: High dose chemo pts worse
than locally tx on immediate recall
Digit Span: High dose chemo pts
worse than locally tx pts on
Backward
Digit Symbol: High dose chemo pts
worse than locally tx pts
Trails A & B: No sig. findings
D2 test: No sig. findings
Stroop: No sig. findings
Word Fluency: No sig. findings
Fepsy Finger Tapping Task: High
dose chemo pts worse than locally
tx pts
Fepsy Visual Reaction test: High
dose chemo pts worse than locally
tx pts
Fepsy Binary Choice test: No sig.
findings
Fepsy Visual Searching test: No
sig. findings
Cognitive Problems in Daily Life
checklist: Self-reported cog probs
unrelated to neuropsych,, related to
distress
HSCL-25: High dose chemo pts
more depressed than controls
29

30

Table 2, Continued.
1st
Author Year N Sample
Comparison
Group
Study
design Treatment
Time
since
Treatment Neuropsychological Findings
Schagen 1999 71
Breast
cancer
pts
Age-
matched
breast
cancer pts
without
adjuvant tx
Cross-
sectional
Chemo
with or
without
tamoxifen
& local
radiation
1.9 yrs
after
chemo
AVLT: Chemo grp worse on long delay
recall
Rey Complex Figure test: Chemo grp
worse
WMS Visual Reproduction: Chemo grp
worse
Digit Span: Chemo grp worse on Digits
Backward
Digit Symbol: Chemo grp worse
Trails A & B: Chemo grp worse on Trails
B
D2 test: No sig. differences
Stroop: : No sig. differences
Animals: Chemo grp worse
Fepsy Finger Tapping Task: Chemo grp
worse
Fepsy Visual Reaction test: Chemo grp
worse
Fepsy visual searching: : No sig.
differences
Dutch NART: : No sig. differences
Hopkins Symptom Checklist-25: No
correlation with measured cognitive
problems; sig correlation with reported
cognitive problems
30

31

Table 2, Continued.
1st
Author Year N Sample
Comparison
Group
Study
design Treatment
Time
since
Treatment Neuropsychological Findings
Brezden 2000 107
Breast
cancer
pts
Compared
current
chemo
patients to
those >1 yrs
post-chemo,
to healthy
controls
Cross-
sectional
Adjuvant
chemo
Current,
or mean =
2 yrs
HSCS memory: Chemo pts worse than
controls
HSCS language: Chemo pts & post-chemo
pts worse than controlsHSCS visual-motor:
Post-chemo pts worse than controls
HSCS spatial: No differences
HSCS attention: No differences
HSCS concentration: No differences HSCS
self-regulation and planning: No
differences between groups
POMS: No differences between groups
Ahles 2002 128
Breast
cancer
and
lymph-
oma pts
Pts at least 5
yrs post-
local tx
Cross-
sectional Chemo
Average =
approx. 9
years
post-tx
WAIS-III Vocabulary: No sig differences
WRAT reading: No sig differences
BNTL: No sig differences
COWA: No sig differences
Block design: No sig differences
CVLT: No sig differences
Logical Memory: Chemo pts worse
WMS visual reproduction: No sig
differences
Digit symbol: Chemo pts worse
Trails A & B: Chemo pts worse
Finger tapping: No sig differences
Thumb-finger sequencing: No sig
differences
Continuous Performance Test: No sig
differences
31

32

Table 2, Continued
1st
Author Year N Sample
Comparison
Group
Study
design Treatment
Time
since
Treatment Neuropsychological Findings
Schagen 2002 148
High-
risk
breast
cancer
pts
Breast
cancer pts tx
locally
Follow-up
from
previous
studies
Chemo,
radiation,
tamoxifen
Time 1=2
yrs post-
tx;
Time 2=
3.5 yrs
post-tx
AVLT, Complex Figure test, Digit Span,
Digit Symbol,Trails A & B, D2 test, Stroop,
Word Fluency, Fepsy Finger Tapping Task,
Fepsy Visual Reaction test, Fepsy Binary
Choice Test, Fepsy Visual Searching Test,
Dutch Adult Reading Test, & EORTC QLQ-
C30, HSCL-25:
No differences in cognitive deficits between
chemo grps and control grp at T2. Trend
towards improvement (1 SD change) on cog
tests for all chemo grps, and decline for
control. 33-45% of cog impaired high dose
chemo pts at T1 were lost to follow-up due
to relapse or death. Low correlation
between subjective complaints of cog.
problems and neuropsych function.
Neuropsych unrelated to tamoxifen use.
Tchen 2003 200
Breast
cancer
pts
Matched
cancer-free
controls
Cross-
sectional Chemo
2-6 weeks
post-
chemo
HSCS: Cancer pts worse than controls
CCPT:: No sig difference
Trails A and B: No sig difference
Fliess-
bach 2003 10
Primary
CNS
lymph-
oma pts Norms
Long-
itudinal Chemo
Pre-tx or
immediate
ly after
initiation,
4, 12, &
36 months
(median)
Number Connection Test (like Trails A): Pts
improved from baseline (n=3)
COWA: No sig findings
Verbal Learning and Memory Test: Pt
improved from baseline on recall (n=1)
Benton Visual Retention Test: No sig
findings
Block Design Test: No sig findings
32

33

Table 2, Continued.
1st Author Year N Sample
Comparison
Group
Study
design Treatment
Time
since
Treatment Neuropsychological Findings
Wefel 2004a 84
Breast
cancer
pts Norms
Cross-
sectional None
Pre-
chemo
Digit Span: No sig findings
Digit Symbol: No sig findings
Arithmetic: No sig findings
Letter-Number Sequencing: No sig findings
WMS Mental Control: No sig findings
Trails A & B: Pts worse than norms on A
HVLT: No sig findings
VSRT: Pts worse than norms on recall
NVSRT: No sig findings
Rey-O: No sig findings
COWA: No sig findings
Boston Naming: No sig findings
Sequential Commands: No sig findings
Category Test: No sig findings
Similarities: No sig findings
Block Design: No sig findings
JLO: No sig findings
Grip strength: No sig findings
Grooved pegboard: No sig findings
BDI: No sig findings
MMPI: No sig findings
STAI & BAI: No sig findings
Iconomou 2004 80
Solid
tumor
cancer
pts None
Pro-
spective
long-
itudinal Chemo
Pre-
chemo
and
immediate
post-
chemo
EORTC QLQ-C30 (QoL): MMSE
uncorrelated with self-report.
MMSE: 15% impaired at baseline. No
changes pre- to post-chemo.
Hospital Anxiety and Depression Scale:
Correlated with QLQ-C30 cognitive scale
33

34

Table 2, Continued.
1st Author Year N Sample
Comparison
Group
Study
design Treatment
Time
since
Treatment Neuropsychological Findings
Wefel 2004b 18
Breast
cancer
pts
(Stage
1-3) Norms
Long-
itudinal,
pro-
spective,
random-
ized Chemo
3 weeks
post-tx (6
mo post-
baseline)
& 1 year
post-tx
(18 mo
post-
baseline)
Digit Span: 22% declined T1 – T2
Digit Symbol: Mean improvement
Arithmetic: 12% declined T2 – T2
Letter-Number Sequencing: No sig
findings
WMS Mental Control: No sig findings
Trails A: 11% declined T1 – T2
HVLT: No sig findings
VSRT: 12% declined T1 – T2
NVSRT: Mean improvement from
baseline (non-sig after Bonferroni)
Rey-O: No sig findings
COWA: No sig findings
Boston Naming: No sig findings
Sequential Commands: No sig findings
Trails B: No sig findings
Category Test: Mean improvement
Similarities: No sig findings
Block Design: Mean improvement from
baseline (non-sig after Bonferroni)
Rey-O copy: No sig findings
LJO: No sig findings
Grip strength: No sig findings
Grooved pegboard: No sig findings
FACT: Uncorrelated with cog
dysfunction
34

35

Table 2, Continued.
1st Author Year N Sample
Comparison
Group
Study
design Treatment
Time
since
Treatment Neuropsychological Findings
Castellon 2004 72
Breast
cancer
survivors
Matched
non-cancer
grp & pts
not tx with
chemo
Cross-
sectional
Chemo &
some
tamoxifen
2-5 yrs
post-
diagnosis
Verbal Fluency (FAS & Animals):
Chemo pts worse than non-cancer grp
and no-chemo pts.
Verbal Memory (Logical Memory): No
sig findings.
Verbal Learning (CVLT): No-chemo
pts better than non-cancer grp. Chemo
+ tamoxifen pts worse than pts without
adjuvant tx.
Visual Memory (WMS Visual
Reproduction & Rey- O Long Delay):
Chemo pts worse than no-chemo pts.
Chemo + tamoxifen pts worse than pts
without adjuvant tx.
Visuospatial (Block Design, Rey-O
Copy): Chemo pts worse than no-
chemo pts. Chemo + tamoxifen pts
worse than pts without adjuvant tx.
Psychomotor Speed (Digit Symbol &
Trails A & B): No sig findings.
Reaction Time: No sig findings.
Executive Attention (PASAT & Stroop):
No sig findings.
Cognitive Failures Questionnaire:
Correlated with BDI, STAI, & fatigue.
BDI: No relation to neuropsych.
STAI: No relation to neuropsych.
SF-36: No relation to neuropsych.
35

36

Table 2, Continued.
1st Author Year N Sample
Comparison
Group
Study
design Treatment
Time
since
Treatment Neuropsychological Findings
Buck-
walter 2005 3664
Older
female
mixed
survivors
3123 women
who
reported no
history of
cancer
Cross-
sectional
Unknown
(probably
mixed)
Mean =
17.3 years
after 1st
hospitaliz
ed for
cancer
Telephone Interview of Cognitive
Status - modified (gross cognitive
function measure highly correlated
with MMSE): No significant between
grp differences.
Cimprich 2005 184
Early
stage
breast
cancer Test norms
Cross-
sectional None
23 days
after
diagnosis
(18 days
before
surgery)
Digit Span: Mean scores average
range
Trails A & B: Mean scores average
range
Three Shapes & Three Word Memory
Test: Mean scores average range.
Symptom Distress Scale: Predicted
subjective cognitive problems
POMS-Short Form: More anxiety
related to better Trails B and Digit
Span B
Mar Fan 2005 172
Young
breast
cancer
survivors
81 patient-
nominated
age-matched
healthy
controls
Long-
itudinal
Adjuvant
chemo
During, 1
and 2 yrs
post-
chemo
High Sensitivity Cognitive Screen:
During chemo, "more pts than
controls had moderate-severe cog
dysfunction." Pts with mod-severe
impairment improved at follow-ups.
3.8% worsened from mild -
moderate/ severe at later follow-ups.
No diff between pts who received
hormone tx vs those who didn't.
Trails B: Pts worse at 1 and 2 years
than controls.
36

37

Table 2, Continued.
1st
Author Year N Sample
Comparison
Group
Study
design Treatment
Time
since
Treatment Neuropsychological Findings
Downie 2006 21
Breast
cancer pts
(< 60
years) Norms
Cross-
sectional
Adjuvant
chemo
During
chemo
(between
3-6
cycles)
HSCS memory: 48% had dysfunction
HSCS language: 61% had
dysfunction
HSCS attention & concentration:
10% problems
FACT: No correlation with HSCS or
with self-reports of cognitive
problems.
Schagen 2006 184
High-risk
breast
cancer pts
Stage 1
breast, no
chemo
(n=57) &
healthy
controls
(n=60)
Longitudi
nal
Adjuvant
high or
standard
dose
chemo
Pre-
chemo &
6 months
post-
chemo
Trails A & B, & Stroop Color-Word
Test:
No difference between grps at 1st
timepoint on any neuropsych tests, or
in % of pts impaired. When
controlling for practice effect,
significantly more high-dose pts
declined than healthy controls (25%
vs 6.7%).
37

38

Table 2, Continued.
1st
Author Year N Sample
Comparison
Group
Study
design Treatment
Time
since
Treatment Neuropsychological Findings
Hurria 2006 28
Older
breast
cancer
pts (65
years or
older) Norms
Long-
itudinal
Adjuvant
chemo &
some
adjuvant
hormone
therapy
Pre-
chemo &
6 months
post-
chemo
WRAT-3 Reading: No sig findings.
Boston Naming Test: No sig findings.
COWA: No sig findings.
Hopkin's Verbal Learning Test-Revised:
Rey-O: Pts improved on immediate recall.
20% declined 2 SD on copy, 4% declined 2
SD on immediate recall, & 4% declined 2 SD
on delayed recall.
WAIS-III Block Design: No sig findings.
WAIS-III Digit Span: Pts declined.
Trails A & B: 7% declined 2 SD ob Trails A.
Stroop: No sig findings.
GDS: No sig findings.
FACT-B: No sig findings.

Abbreviations: Chemo= chemotherapy; cog=cognitive; grp=group; pts= patients, PCI=prophylactic cranial irradiation; sig=significant;
tx=treatment; T1=Time 1; T2=Time 2; T3=Time 3.
Abbreviations for Neuropsychological Measures: BMAACL= Brief Multiple Affective Adjective Checklist; WAIS = Wechsler Adult
Intelligence Scale; VIQ = Verbal Intelligence Quotient; PIQ = Performance Intelligence Quotient; FSIQ = Full Scale Intelligence Quotient;
Trails = Trail Making Test; WMS = Wechsler Memory Scale; NART = National Adult Reading Test; MMPI = Minnesota Multiphasic
Personality Inventory; COWA = Controlled Oral Word Association; PASAT = Paced Auditory Serial Addition Task; CVLT = California
Verbal Learning Test; AVLT = Auditory Verbal Learning Test; Rey-O = Rey-Osterreith Complex Figure; CCPT = Conner’s Continuous
Performance Test; BDI = Beck Depression Inventory; EORTC QLQ = European Organization for Research and Treatment of Cancer
Quality of Life Questionnaire; HSCL = Hopkin’s Symptom Checklist; POMS = Profile of Mood States; HSCS = High Sensitivity Cognitive
Screen; WRAT = Wide Range Achievement Test; BNT = Boston Naming Test; MMSE = Mini Mental State Examination; HVLT =
Hopkin’s Verbal Learning Test; VSRT = Verbal Selective Reminding Test; NVRST = Nonverbal Selective Reminding Test; JLO =
Judgment of Line Orientation; STAI = State Trait Anxiety Inventory; BAI = Beck Anxiety Inventory; FACT = Functional Assessment of
Cancer Therapy; GDS = Geriatric Depression Scale.
38

39
it is also unlikely that cancer or its treatment would result in these primary effects of
traumatic brain injury (Adams, 1984). These factors make traumatic brain injury an
unsatisfying analogy. However, the diffuse deficits the authors describe, and which
are frequently found in the broader literature of cognitive functioning in long-term
cancer survivors (Ahles et al., 2002; van Dam et al., 1998; van Oosterhout et al.,
1996), are also consistent with frontal-subcortical dysfunction (Malloy et al., 1998;
Roman et al., 2002). Perhaps the clearest pattern of such frontal-subcortical deficits
in long-term cancer survivors was found in van Oosterhout and colleagues’ (1996)
study of small cell lung cancer survivors who were 4-7 years post-treatment. These
researchers found deficits in selective inhibition, mental tracking, and information
processing speed, all of which are highly dependent upon good frontal-subcortical
functioning. Deficits were also found in verbal learning and verbal recall, but not in
verbal recognition, a pattern also consistent with frontal-subcortical dysfunction.
Although studies of neuropsychological functioning in cancer survivors often
demonstrate a frontal-subcortical pattern of neuropsychological deficits, the
consistency with which cancer survivors experience memory deficits also implicates
dysfunction in the temporal region of the brain. Several studies have found evidence
of difficulty with memory (van Oosterhout et al., 1996; Ahles et al., 2002; Wieneke
& Dienst, 1995; Schagen et al., 1999; van Dam et al., 1998). Perhaps the most
compelling example of these studies comes from Castellon and colleagues (2004),
who compared the neuropsychological test results of 36 breast cancer survivors 2-5
years post-diagnosis who had been treated with adjuvant chemotherapy to those of

40
17 locally-treated breast cancer survivors and 19 healthy controls. When comparing
the chemotherapy-treated breast cancer survivors to test norms, mild deficits were
found in verbal memory, verbal fluency, visual memory, visuospatial functioning,
reaction time, and executive functioning. The verbal and nonverbal memory deficits,
visuospatial deficits, and verbal fluency deficits are all consistent with temporal lobe
damage (Hamberger et al., 2001; Lezak et al., 2004). However, even in this study,
the deficits in reaction time and executive functioning provide evidence of frontal-
subcortical dysfunction (Lezak et al., 2004).
The physiological evidence of long-term white matter changes following
cancer treatment is consistent with subcortical damage (Brown et al., 1998). Brown
and colleagues (1998) studied a small sample of women with advanced breast cancer
who were treated with high dose chemotherapy, and found white matter changes in
75% of the patients at 3 months post-chemotherapy. These white matter changes had
increased at 6 months post-chemotherapy, and then remained stable, with no further
changes, at 1 year post-chemotherapy. In a separate study, altered activity within the
inferior frontal gyrus, cerebellum, and basal ganglia were found in breast cancer
survivors 5-10 years post-treatment with chemotherapy, which the authors attributed
to chemotherapy-related neurological damage and compensatory activation
(Silverman et al., 2007).
There is also physiological evidence for medial temporal lobe damage
following cancer treatment. However, changes in temporal regions of the brain have
typically been found to be short-lived (Yoshikawa et al., 2005; Inagaki et al., 2007).

41
Comparing breast cancer survivors who had been treated with chemotherapy to
breast cancer survivors not treated with chemotherapy, Inagaki and colleagues
(2007) found reduced volumes within the parahippocampal, cingulate, and prefrontal
gyrus and precuneus a year after surgery. Brain volume in these regions was
significantly correlated with measures of attention and visual memory. However, 3
years after surgery, no differences in brain volume between the two groups were
found. In a separate study, the same group of researchers found that, approximately 3
years after chemotherapy, hippocampal volumes did not differ significantly between
breast cancer survivors treated with chemotherapy versus those who did not receive
chemotherapy (Yoshikawa et al., 2005). These findings suggest that temporal lobe
changes, which could cause problems with learning and memory, would be likely to
be more evident in short-term cancer survivors than in long-term cancer survivors. In
sum, although there is evidence for both frontal-subcortical and limbic/temporal lobe
problems in long-term cancer survivors, more evidence suggests that subcortical
damage may be primarily responsible for the observed deficits in long-term cancer
survivors.
Potential confounding factors
To identify the factors most likely to be responsible for cognitive dysfunction
in cancer survivors, researchers have primarily focused on investigating the role of
treatment factors such as type and dose of treatment (e.g., Baile, 1996; Silberfarb &
Oxman, 1988). However, one of the problems with assuming that cognitive deficits
experienced by long-term cancer survivors are caused by cancer treatments is that

42
there remains the possibility that any differences in cognitive functioning found
between long-term cancer survivors and individuals without a cancer history existed
prior to cancer diagnosis and treatment. Indeed, some research has suggested that an
unusually high number of cancer patients do exhibit cognitive deficits prior to cancer
treatment, and exhibit similar patterns of cognitive deficits as those seen after cancer
treatment (Wefel et al., 2004; van Oosterhout et al., 1995; Komaki et al., 1995). One
explanation for such pre-treatment cognitive deficits, and the deficits found in cross-
sectional studies of long-term cancer survivors, is that confounding variables may
put some individuals at higher risk for developing both cancer and cognitive deficits.
These factors could be largely responsible for cognitive deficits in cancer survivors,
and may thus inflate the rate of cognitive problems in cancer survivors irrespective
of cancer treatment.
A sedentary lifestyle is one such factor that contributes to increased risk of
several types of cancers, including breast cancer (Leris & Mokbel, 2001) and colon
cancer (Martinez, 2005). A sedentary lifestyle also increases risk for cognitive
problems associated with cerebrovascular dementia (Butler et al., 1993). In contrast,
physical activity or exercise can reduce risk for these cancers (Leris & Mokbel,
2001; Martinez, 2005) and improve cognitive functioning (Colcombe & Kramer,
2003; Kramer et al., 2004).
Although the mechanisms through which exercise might improve cognitive
functioning or protect against cognitive impairment remain theoretical, animal
research suggests that exercise could be beneficial for the brain by increasing

43
capillary density (Black et al, 1990), promoting new cell growth in the hippocampus
(van Praag et al., 1999), and increasing production of brain-derived neurotrophic
factor, which enhances synaptic plasticity and may improve learning and memory
(Dishman et al., 2006). These theoretical mechanisms are interesting because they
suggest certain areas of cognitive functioning that would be particularly well
protected against decline—executive functions (Dishman et al., 2006; Kramer et al.,
1999) and memory (Dishman et al., 2006). These are the same cognitive areas that
are frequently found to be impaired in cancer survivors. Indeed, a meta-analysis that
examined the effect of exercise interventions on cognitive functioning in older adults
found that cognitive benefits of exercise were most pronounced in executive
functioning, relative to motor speed, visuospatial functioning, and controlled
processing (Colcombe & Kramer, 2003). Memory, unfortunately, was not included
as an outcome measure in this meta-analysis.
Heavy alcohol consumption is another behavioral factor that increases risk
for several cancers (Bagnardi et al., 2001), including breast cancer (Leris & Mokbel,
2001; Kropp et al., 2001; Bagnardi et al., 2001), colon cancer (Martinez, 2005; Stein
& Colditz, 2004; Bagnardi et al., 2001), and oral cancer (Stein & Colditz, 2004;
Bagnardi et al., 2001)., whereas low to moderate alcohol consumption can reduce
risk for some cancers (Kropp et al., 2001). Heavy alcohol consumption also increases
risk for dementia (Mukamal et al., 2003) and other cognitive problems (Hildebrandt
et al., 2004; Bruce & Jones, 2004).

44
Of note, heavy alcohol consumption has been found to be particularly
detrimental to individuals’ ability to perform executive functioning tasks such as
mental set switching and response inhibition (Hildebrandt et al., 2004), visuospatial
functioning and memory (Sullivan et al., 2002) and attentional tasks (Bruce & Jones,
2004). These deficits may be the result of disrupted networks connecting frontal lobe
regions, the cerebellum, and the thalamus (Sullivan, 2003). Indeed, chronic
alcoholism has been found to result in white matter changes (de la Monte, 1988),
which provides further evidence for subcortical damage associated with high alcohol
use. As noted when discussing the protective effects of exercise, executive
functioning abilities are commonly affected in cancer survivors, as are attentional
abilities. In addition, older adults with alcoholism have been found to have reduced
hippocampal volume relative to individuals without alcoholism, suggesting an
alcohol-related neurological substrate for memory problems (Sullivan et al., 1995).
However, reduced hippocampal volumes were not significantly associated with
memory problems or scores on memory tests (Sullivan et al., 1995). The well-
known Korsakoff syndrome that results from an alcoholism-related thiamine
deficiency also causes severe amnestic memory impairment (Lezak et al., 2004).
Although the effect size of memory problems in cancer survivors is large (Anderson-
Hanley et al., 2003), however, it seems unlikely that memory problems in cancer
survivors are caused by alcohol-related Korsakoff syndrome because memory
problems in cancer survivors have not been noted to be amnestic.

45
Education level is an environmental factor that, although it does not appear to
be a direct risk or protective factor for cancer, can increase or decrease risk for
cancer due to its correlation with other important behavioral factors and exposures
that can be risk or protective factors for cancer. For instance, women with higher
levels of education are commonly found to be at higher risk for breast cancer, but
this effect may be due largely to well-educated women having fewer children and
being older when they begin having children, which exposes them to more menstrual
cycles and thus higher amounts of hormones (Braaten et al., 2005). Similarly,
individuals with lower levels of education are more likely to engage in some
important risk behaviors, such as smoking, that increase risk for certain types of
cancer, such as lung and kidney cancers (Braaten et al., 2005). When viewed as a
proxy for such behavioral tendencies, education level is a risk or protective factor for
a number of cancers (Hemminki & Li, 2003).
The influence of education upon cognitive test performance is probably more
direct, and has been described as “potent and pervasive” (Lezak et al., 2004, p. 315).
Education has been found to affect nearly every cognitive ability, including verbal
abilities, visuospatial abilities, memory, and even simple copying tasks (see Lezak et
al., 2004, for an overview). Education has also been theorized to increase cognitive
reserve, the previously described cognitive cushion that buffers individuals from
reaching the clinical threshold at which cognitive impairment or dementia is
recognized (Katzman, 1993). Research examining a large sample of older adults in
France showed that the beneficial effects of education on cognitive reserve may be

46
mediated by improved conceptualization and controlled processing abilities, as well
as more cognitively demanding occupations (Le Carret et al., 2003). Despite the
influence of education on cognitive test performance, many norms do not account for
education level. This can result in poorly educated individuals who are cognitively
intact performing more poorly than well educated individuals with mild cognitive
impairment (Lezak et al., 2004). Education is thus an important factor to consider
when cognitive functioning is being used as an outcome variable.
Although not necessarily a risk factor for developing cancer, elevated
depression levels among cancer survivors relative to the general population (Raison
& Miller, 2003) may also contribute to high levels of pre- and post-treatment
cognitive problems, since depression can affect neuropsychological performance
(Kalska et al., 1999; Rosenstein, 1998). Estimates of the prevalence of depression in
cancer survivors vary depending upon type of cancer, measurement of depression,
and other study characteristics, but range between 0 - 58% (see, e.g., Massie, 2004
for review).
When depressed older adults are given cognitive tests, they tend to perform
more poorly upon tests of attention, motor speed, learning, memory, and executive
functioning (Baune et al., 2006; Elderkin-Thompson et al., 2007) compared to
individuals without depression. There is some evidence that learning and memory
problems in depressed individuals are mediated by deficits in attention and executive
functioning, which prevent individuals from attending to tasks and encoding and
organizing information sufficiently well to learn and remember it (Elderkin-

47
Thompson et al., 2007). Other research suggests that the poor cognitive performance
of depressed individuals is exacerbated by beliefs that they are doing poorly on
testing or feedback about errors, whereas non-depressed individuals use error
information to increase effort on cognitive tests (Holmes & Pizzagalli, 2007). In
addition, late-onset depression appears to be prodromal for dementia (Jorm, 2001),
making it difficult to know whether cognitive problems seen in the presence of
depression are a consequence of, or part of the same pathological process as,
depression.
The relationship between depression and cognitive functioning in cancer
survivors is not a simple one. For the most part, when studies of post-treatment
cancer survivors have examined depression, measurable cognitive dysfunction has
been unrelated to depression (Castellon et al., 2004; Brezden et al., 2000; Ahles et
al., 1996; Wieneke & Dienst, 1995; Silberfarb, Philibert, & Levine, 1980), although
it has been found to be associated with pre-treatment cognitive problems (Wefel et
al., 2004a). Depression is, however, fairly consistently associated with subjective
memory complaints in cancer survivors post-treatment (Iconomou et al., 2004;
Castellon et al., 2004; Schagen et al., 1999; van Dam et al., 1998). This effect is not
unique to cancer survivors; research examining older adults more broadly has also
found depression to be more highly associated with subjective memory complaints
than with current measurable cognitive problems (Reid & MacLullich, 2006). These
findings in cancer survivors have cast doubt upon the idea that cancer survivors have
insight into their own cognitive problems, suggesting that depression, rather than

48
objective cognitive problems, is responsible for their subjective experience of
cognitive problems. However, the same researchers found that subjective memory
complaints were associated with later cognitive deficits (Reid & MacLullich, 2006),
so it remains possible that individuals can detect more subtle cognitive changes than
can objective tests, or that individuals have a better awareness of changes from their
own baseline than do standard cognitive assessments.
Despite the associations between these factors, cancer, and cognitive
functioning, the extent to which behavioral factors may contribute to cognitive
problems in cancer survivors has not, to my knowledge, been previously
investigated. Similarly, many studies of cognitive functioning in long-term cancer
survivors (e.g., Heflin et al., 2005; van Oosterhout et al., 1996; Tucker et al., 1989)
fail to control for depression. Studies finding neuropsychological improvement from
pre- to post-treatment either did not control for depression (Fliessbach et al., 2003),
or found depression to be significantly related to cognitive impairment among cancer
survivors (Schagen et al., 2002), leaving open the possibility that cancer survivors’
cognitive problems remit along with any depression they may experience.
The frequent use of test norms (e.g., Fliessback et al., 2003; Wieneke &
Dienst, 1995; Cull et al., 1994; Tucker et al. 1989), rather than a separate
comparison group, also makes studies vulnerable to confounding factors, since
potential confounds beyond age, sex and education cannot be controlled for when
using norms. Indeed, a meta-analysis examining neuropsychological functioning
among cancer patients found that a different pattern of results was obtained when

49
cancer patients are compared to norms than to other comparison groups or to a
baseline level of functioning (Anderson-Hanley et al., 2003). This finding
underscores the importance of selecting an appropriate comparison group matched
on potential third variables, or from whom these variables can be measured and
statistically controlled, to minimize the chance of finding spurious effects.
In light of the possibility that the relationship between cancer and cognitive
problems might be spurious, one of the first tasks of an investigation of cognitive
functioning in long-term cancer survivors is to ascertain that cognitive deficits found
in long-term survivors cannot be explained by potentially confounding variables.
Controlling for potential confounds would help ensure that any cognitive deficits
found in cancer survivors are not due to third variables associated with both
cognitive dysfunction and cancer, and can help ascertain whether cancer treatment
has any additional utility in predicting cognitive problems.
The present research
Study 1 was designed to isolate the long-term effects of cancer or its
treatment on cognitive functioning from potential confounds by controlling for at
least 50% of genetics, many early environmental factors, and later environmental,
behavioral, and psychological factors such as exercise, alcohol use, and depression.
In a previous study, we found that long-term cancer survivors were 2.7 times more
likely to exhibit cognitive dysfunction than co-twins without a cancer history (Heflin
et al., 2005). In response to our study, one researcher observed that there may be
other reasons that long-term cancer survivors had such increased risk for cognitive

50
dysfunction (Grant, 2005). Such potentially confounding factors include alcohol
consumption, education level, and exercise. The analyses in Study 1 were intended to
clarify whether the cognitive deficits observed in our earlier study of long-term
cancer survivors remain after controlling for such potential confounds.
Researchers had also criticized our use of a gross measure of cognitive
dysfunction in our earlier study (Wefel and Meyers, 2005). Although the measure we
used to classify individuals as having cognitive dysfunction has been well validated
(Gatz et al., 2002; Gatz et al, 2005), the use of a gross measure of cognitive
dysfunction did not allow an exploration of the specific cognitive domains in which
cancer survivors are likely to experience deficits. To address this issue, this study
also examined the domains of cognitive functioning in which long-term cancer
survivors experience deficits compared to individuals who have never been
diagnosed with cancer.
The second study was conducted to expand upon the findings of the first
study by including longitudinal measures of cognitive functioning. I was interested
first in examining whether cognitive functioning in older adult long-term cancer
survivors declined at a more rapid rate than similar individuals without a cancer
history. This approach would begin to address the question of whether cancer and its
treatment might cause a one-time cognitive insult that lowers individuals’ cognitive
functioning, or whether cancer and its treatment might interact with pre-existing
biological factors to produce a more rapid rate of cognitive decline in older age than
individuals without a cancer history experience.

51
The second study also explored whether cognitive functioning changes
significantly after cancer diagnosis and treatment, compared to cognitive functioning
before individuals were diagnosed with cancer. Although it was not focused upon
long-term cancer survivors, this portion of the present research addressed the
question of whether cognitive functioning changes substantially following cancer
diagnosis and treatment more closely than previous research has been able to do.
Because it is difficult to predict who is going to be diagnosed with cancer, and thus
to obtain measures of cognitive functioning prior to cancer diagnosis, previous
research has used a baseline measure of cognitive functioning after cancer diagnosis
(Iconomou et al., 2004; Wefel et al., 2004a; Fliessbach et al., 2003; Ahles et al.,
1998; Kaasa, Olsnes, & Mastekaasa, 1988; Oxman & Silberfarb, 1980). Those
baselines used in previous research may thus have been influenced by the
psychological distress of a cancer diagnosis (Wefel et al., 2004a) or by disease-
related factors. The baseline measure of cognitive functioning in this study should
not have been influenced by psychological distress related to learning about a cancer
diagnosis, and possibly was less influenced by disease-related factors.
In order to address the research questions described above, I conducted
secondary database analyses using data from two distinct samples of cancer
survivors and their matched comparison groups. The following paper is organized
into two studies to reflect these two samples, with cancer survivors from the Swedish
Twin Registry (STR) comprising the first study, and cancer survivors from the

52
University of Southern California’s (USC) Alzheimer Disease Research Center
(ADRC) comprising the second study.
Chapter 2: Study 1
The purpose of the first study was to follow up on findings of increased
cognitive dysfunction in long-term older adult cancer survivors from our previous
research (Heflin et al., 2005) in order to ensure that this finding was not due to
confounding variables. A second goal of Study 1 was to examine more closely the
types of cognitive problems that older adult long-term cancer survivors might have
compared to individuals without a cancer history.
Using a sample of twin pairs discordant for cognitive dysfunction, the first
portion of this study examined whether the cancer survivors remain at higher risk for
cognitive dysfunction after controlling for several potential confounds. The second
portion of this study used individually matched pairs of long-term cancer survivors
and cancer history-free individuals who had completed a cognitive battery to
investigate specific types of cognitive deficits for which long-term cancer survivors
might be most at risk in older adulthood.
I had two primary hypotheses for this study. Firstly, I hypothesized that long-
term cancer survivors (those diagnosed with cancer 5 or more years before cognitive
testing) would have a higher rate of cognitive dysfunction than co-twin controls
without a cancer history, after controlling for education, alcohol use, physical
activity, and depression. Secondly, I hypothesized that, compared to individuals
without a cancer history, long-term cancer survivors would perform more poorly on

53
tests most dependent on frontal-subcortical systems, after controlling for covariates.
The tests most dependent on frontal-subcortical systems within this study were those
measuring information processing speed, verbal memory, and nonverbal memory. I
expected that long-term cancer survivors would not perform significantly worse than
controls on tests that were less dependent on frontal-subcortical systems—those
measuring verbal fluency, verbal functioning, or visuoconstructional skills.
Method
Procedures
Developed in the late 1950’s as a research resource, the STR is the largest
twin registry in the world, composed of monozygotic and dizygotic twins born in
Sweden. In 1963 and 1967, the STR sent to all same-sex twins born before 1926
questionnaires that assessed family constellation, education level, alcohol use, and a
number of health conditions. Twins who did not respond to the questionnaire in
1967, or who had a significant amount of missing data, were sent a similar
questionnaire in 1970. All same-sex twins born between 1926 and 1958 were sent a
similar questionnaire in 1973, which assessed exercise, alcohol consumption, and
education, among other health conditions, environmental exposures, and behaviors.
Beginning in 1998, all same- and opposite-sexed twins who were born before
1935 were contacted by telephone for the Screening Across the Lifespan Twin
(SALT) study and administered the standardized SALT telephone interview. The
SALT screening used a computer-based data collection system and trained
interviewers. The screening included questions about education level and exercise,

54
and assessed depression through the use of a standardized measure. Twins from
SALT who were 65 years or older were administered a cognitive screening as part of
the telephone interview. If a twin was unable to be interviewed, or performed poorly
on the cognitive screening, an informant was requested and interviewed.
As described below, the SALT telephone cognitive screening differentiated
between individuals who had cognitive dysfunction sufficiently severe to identify
them as dementia suspects, and individuals with less or no cognitive dysfunction. All
individuals who were identified as dementia suspects on the basis of the SALT
cognitive screening were referred for a complete dementia evaluation, which
included in-depth cognitive testing with a cognitive battery. If the twin was
diagnosed with dementia, his or her co-twin was also invited to participate in a
dementia work-up with the same cognitive battery. In addition, a smaller number of
twins from cognitively intact twin pairs were given the same work-up and
administered the cognitive battery. The work-up was administered at the twins’
homes by a physician and a nurse, with the cognitive battery administered by the
nurse who was trained to administer the cognitive battery.
Cancer data were obtained through an existing link between the STR and the
Swedish Cancer Registry, where information on cancer diagnoses, malignancy, and
date of diagnoses is updated regularly by medical staff at Swedish hospitals.
Individuals were considered to have cancer only if they had been diagnosed with a
malignant cancer. For individuals with multiple cancers, the time span between
cancer and dementia screening or cognitive testing was calculated using the

55
diagnosis date of the most recent cancer. This was done to ensure a focus on long-
term cancer survivors, rather than individuals who had had cancer more recently.
Measures
Education
Education data were compiled from a number of Swedish Twin sources,
including the 1963 and 1967 questionnaires. The 1963 questionnaire assessed
education level by asking, “Do you have any further education beyond compulsory
school?” and allowing participants to answer “no,” “O-level,” “A-level (high
school),” “folk high school,” “vocational training,” and “other training, namely ____
(participants were to write in their other training).” Similarly, the 1973 questionnaire,
sent to twins born in 1926 or later, asked, “Have you completed a course of
education at any schools beyond elementary school?” Responses included: ”no,” or
“yes.” If participants responded “yes,” they were asked to indicate “what levels of
schooling” they had completed.
Individuals participating in SALT were also asked to describe their
education, allowing collection of education level for those opposite-sexed twins who
participated in SALT. To determine years of education from the information
provided in the questionnaires and SALT interview, Swedish experts took into
account the changes in the education system that led to different numbers of years of
school required by different programs at different times and converted these to
comparable years. Education level used in this study was years of education as
determined by these experts.

56
Exercise
Exercise data were taken from two sources: the questionnaires, and the SALT
interview. In the questionnaires given to same-sexed twins born before 1926,
exercise was measured in questionnaires by asking participants, “How much physical
exercise have you had from age 25-50?” and allowing them the response options of
“hardly any physical exercise,” “light exercise, e.g., regular walks or light
gardening,” “regular exercise,” or “hard physical training.”
The 1973 questionnaire asked about exercise level with the question, “Here
are 7 alternatives to describe the exercise you get during your leisure time. Which
one applies best to you when considering the exercise you get during the year as a
whole?” Respondents then indicated, “virtually no exercise,” “exercise very little,”
“exercise rather little,” “do not exercise especially much,” “exercise rather much,”
“exercise much,” “exercise quite a lot.” A 4-point ordinal scale was created from all
the questionnaires based upon examination of the frequency distribution of responses
from the questionnaires, with higher numbers reflecting greater amounts of exercise.
1

Because the questionnaires were sent only to same-sexed twins, these earlier life
exercise data were not available for opposite-sexed twins.
The SALT interview assessed exercise with questions corresponding to the
questionnaires, but phrased in the present. Within the SALT interview, twins born

1
The 1967 and 1970 questionnaire responses were coded 1-4, in order, for the response options,
“hardly any physical exercise,” “light exercise, e.g., regular walks or light gardening,” “regular
exercise,” or “hard physical training.” The 1973 questionnaire responses were coded as follows:
“hardly any physical exercise,” and “light exercise, e.g., regular walks or light gardening,” were coded
as 1, “exercise rather little,” “do not exercise especially much,” were coded as 2, “exercise rather
much,” “exercise much,” were coded as 3, and exercise quite a lot,” was coded as 4.

57
before 1926 were asked, “How much do you exercise?” and allowing responses of
“hardly any physical exercise,” “light exercise, e.g., regular walks or light
gardening,” “regular medium exercise,” or “hard physical exercise.” Participants
born between 1926 and 1958 were asked, “Of these 7 alternatives, which fits your
annual exercise pattern?” and allowed response options of, “almost never,” “much
less exercise than average,” less exercise than average,” “average amounts of
exercise,” “more exercise than average,” “much more exercise than average,” or
“maximum amount of exercise.” A dichotomous variable was created from the
SALT exercise measure for use in matching for the second set of analyses.
2

Although the SALT interview questions were similar to the questionnaire questions,
the questionnaires assessed earlier- to mid-life exercise, whereas the SALT interview
assessed late life exercise.
Depression
Depression was assessed using the Composite International Diagnostic
Interview – Short Form (CIDI-SF; Kessler et al., 1998) that was administered as part
of the SALT interview. The CIDI-SF is a measure that assesses symptoms of major
depression based upon the DSM-IV (APA, 1994). A 3-level categorical variable was
created to differentiate between individuals who endorsed no symptoms of
depression, individuals who endorsed initial criteria but no further symptoms of
depression, and individuals who met diagnostic criteria for major depression. A score

2
For the dichotomous variable, responses of “hardly any physical exercise,” “light exercise, e.g.,
regular walks or light gardening,” “virtually no exercise,” “exercise very little,” “exercise rather
little,” or “do not exercise especially much,” resulted in a designation of “no-little exercise” and

58
of 0 indicated that the person had never experienced major depression, a score
between 1 and 3 was taken to indicate that the person had at some point in their life
experienced subclinical depression (met initial criteria only), and scores of 4 or
above indicated that the person had at some point experienced clinical depression.
Alcohol
Data on alcohol consumption were obtained from questions assessing
drinking habits in the 1967 and 1970 questionnaires. Alcohol use was calculated in
grams per month using a formula that incorporated specific information about the
quantity and frequency of drinking beer, wine, and liquor on weekdays and
weekends. Because risk for cancer and cognitive deficits is frequently decreased or
minimal with small to moderate amounts of alcohol consumption, but increased with
heavy amounts of alcohol consumption (Sesso et al., 2001; Zuccala et al., 2001), it
was important to test for non-linear relationships between alcohol consumption and
both cancer risk and cognitive deficits. To do this, the alcohol consumption variable
was squared to test for the nonlinear relationship that had been found between
alcohol consumption and cancer risk, and alcohol consumption and cognitive deficits
(Sesso et al., 2001; Zuccala et al., 2001). This squared alcohol consumption variable
was used in analyses. Because the questionnaires were sent only to same-sexed
twins, these alcohol consumption data, like the earlier life exercise data, were not
available for opposite-sexed twins.

responses of “regular exercise,” “hard physical training,” “exercise rather much,” “exercise much,” or
“exercise quite a lot” resulted in a designation of “moderate-heavy exercise.”

59
Data on whether an individual had ever in his or her lifetime met diagnostic
criteria for alcohol abuse or dependence were obtained from the SALT interview. In
this interview, participants were asked questions assessing DSM-IV (APA, 1994)
diagnostic criteria for alcohol abuse and alcohol dependence. If participants met
initial criteria to be administered the alcohol abuse interview, but did not meet
criteria for alcohol abuse or dependence, they were assigned an alcohol abuse
diagnosis of “none.” If participants met criteria for alcohol abuse, they were assigned
a diagnosis of “abuse;” if participants met criteria for alcohol dependence, they were
assigned a diagnosis of “dependence.”
Cognitive Functioning Scale
The SALT telephone cognitive screening included the 10 questions that
comprise the Mental Status Questionnaire (MSQ; Kahn et al., 1961), which measure
level of orientation (e.g. questions about the individual’s street address, country, age,
and birthday) and long-term memory (e.g. naming the prime minister and the former
prime minister). The screening also included questions measuring other aspects of
cognitive functioning, including attention (subtracting by 3’s from 20; from the Short
Portable Mental Status Questionnaire; Pfeiffer, 1975), short-term memory (a three-
item recall task; from the Mini-Mental State Examination; Folstein et al., 1975), and
abstraction (questions about similarity between words; Jonsson & Molander, 1964).
Informant interviews were requested if individuals performed poorly on the
cognitive screen or were unable to be interviewed (see Appendix A). The informant
interviews included the Blessed Dementia Rating scale, an established instrument for

60
indicating whether memory or cognitive problems are sufficiently severe to interfere
with activities of daily living (Blessed et al., 1968). Using the cognitive screening
scores and informant reports, an established algorithm was used to assign scores on a
4-point cognitive functioning scale (Gatz et al., 2005). A score of 0 identifies
cognitively intact individuals, 1 indicates minor errors, 2 represents poor
performance (i.e., problems in any of the cognitive domains tapped by screening, but
no confirmation that there was impairment in ability to perform activities of daily
living), and 3 indicates cognitive dysfunction (i.e., inability independently to perform
activities of daily living due to memory problems, or problems in multiple cognitive
domains). This cognitive functioning scale has been dichotomized such that scores of
0, 1, or 2 indicate intact cognitive functioning, and a score of 3 indicates dysfunction
(Heflin et al., 2005); this dichotomized scale was used for this study.
Age
Individuals’ age at the time of participating in the SALT interview was
calculated by subtracting individuals’ date of birth from the date of the SALT
interview, and is referred to in this paper as ‘age at cognitive screening.’ This age
variable was used for the first set of analyses that focus upon cognitive dysfunction
as measured by the SALT cognitive screening. All individuals who participated in
the SALT cognitive screening were aged 65 years or older.
For individuals tested with the cognitive battery, age at the date of testing
was also recorded. This age variable was used for the second set of analyses that
focus upon specific cognitive abilities.

61
Cognitive Battery
The cognitive battery was administered at the twins’ homes by individuals
trained in neuropsychological assessment. It consisted of a brief interview inquiring
into the individual’s memory difficulties, and of a series of tests developed to assess
various cognitive domains (see Appendix B).
The tests listed below are organized into broad cognitive domains on the
basis of a widely used neuropsychological text (Lezak, 2004).
Verbal Memory:
The Word List is a measure assessing short-term verbal memory by asking
individuals to learn and remember a list of visually and orally presented words.
Three scores from the test were used. The Learning score is the sum of the words
immediately recalled after each of the three learning trials, the Delayed Recall score
is the number of words recalled correctly after a delay, and the Recognition score is
the number of words correctly recognized after a delay. This test was taken from the
Consortium to Establish a Registry for Alzheimer’s Disease (CERAD; Morris et al.,
1989). It has 1-month test-retest reliabilities ranging from .43 (for moderately
demented individuals) to .64 (for normal controls), and was found to load heavily on
a factor measuring memory (Morris et al., 1989).
The Prospective Memory test assesses prospective memory by verbally
giving a future task to complete when a particular stimulus is given and ascertaining
whether a person completes the task at a later time when the stimulus is presented.
Poor performance on this task is shown to be associated with the presence of very

62
mild dementia (Huppert, Johnson, & Nickson, 2000). Prospective memory tasks
correlate positively with other tests of verbal memory, such as word list recall,
although these correlations appear to be weaker for individuals without any cognitive
impairment than for individuals who perform more poorly on cognitive tests
(Huppert & Beardsall, 1993).
Nonverbal Memory:
The Memory in Reality Test, also known as the Apartment Test, measures
nonverbal memory by asking the individual to place 10 household items in a model
of an apartment, and to later recall where the items were previously placed
(Johansson, 1988/1989). This test correlates positively (.42 to .60) with other
memory tests, suggesting it has good construct validity as a memory test (Goodman
& Zarit, 1995).
Visuoconstructional Skills:
Figure Copying is a test that measures visuospatial and visuoconstructional
skills by having individuals copy a series of increasingly difficult shapes. This
subtest was taken from the CERAD (Morris et al., 1989). It has 1-month test-retest
reliabilities ranging from .54 (for normal controls) to .81 (for moderately demented
individuals), and was found to load on a factor of praxis (Morris et al., 1989).
Block Design is a test assessing visuospatial/visuoconstructional abilities and
information processing speed that requires the individual to quickly reproduce
geometric models with blocks. This subtest was taken from the WAIS—Swedish
Version (Jonsson & Molander, 1964). The WAIS-R version of Block Design, which

63
is an identical task, has split-half reliability coefficients between .76 and .81 in older
adults (Wechsler, 1997), and has been found in a number of studies to load on a
factor of perceptual organization (Lezak, 2004).
Verbal Fluency:
Animals is a test measuring verbal output by asking the individual to name as
many animals as he/she can think of within a specified time limit. This subtest was
taken from the CERAD (Morris et al, 1989). It has 1-month test-retest reliabilities
ranging from .67 (for normal controls) to .76 (for moderately demented individuals),
and loaded on a language factor (Morris et al., 1989).
Verbal Functioning:
Judgment is a test assessing verbal comprehension, reasoning, and social
judgment. This subtest consists of 3 questions taken from the Comprehension
subtest of the WAIS-Swedish Version (Jonsson & Molander, 1964). The
comprehension subtest loads on a factor of verbal ability and is heavily influenced by
long-term knowledge, but poor scores on this test can indicate poor social judgment
(Lezak, 2004).
The Information Test measures long-term memory, fund of information, and
range of general factual knowledge by asking a series of factual questions. This
subtest was taken from the WAIS—Swedish Version, and loads on a factor of verbal
ability (Jonsson & Molander, 1964).

64
Information Processing Speed:
Symbol Digit is a test that assesses information processing speed, short-term
memory, and sustained attention by asking the individual orally to produce digits
corresponding to symbols. It was normed on a sample of 420 people aged 18-74,
and correctly identified 86% of patients with confirmed and chronic brain lesions
and 92% of normal controls using a cutoff of –1.5 standard deviations (Lezak, 2004).
Participants and Design
This study focused upon twins who had participated in the SALT telephone
dementia screening and subsequent cognitive testing, and used data from the earlier
questionnaires as covariates or matching variables. Participants for the first part of
the study were drawn from individuals who had completed the SALT telephone
dementia screening. To determine whether cancer is a significant risk factor for
cognitive dysfunction after controlling for covariates, a case-control design was
used, in which each control was the co-twin of a case. All twin pairs who met the
following criteria were included: 1) discordance for cognitive dysfunction, 2)
completion of the telephone dementia screen or informant reports of cognitive
functioning for both twins, 3) diagnosis with cancer must have preceded the
telephone dementia screen by 5 years or more. Exclusion criteria for the first portion
of Study 1 include: 1) diagnosis with brain cancer in the cancer-surviving twin, 2)
diagnosis with dementia prior to diagnosis with cancer, and 3) being an opposite-sex
twin pair. To ensure that only long-term cancer history was examined as a risk for
cognitive dysfunction rather than cancer history more generally, twin pairs were also

65
excluded if one or both members of the pair had been diagnosed with cancer less
than 5 years prior to cognitive screening. One hundred eighty-nine twin pairs were
included in the analyses to test Hypothesis 1.
Cancer survivors who had completed the cognitive battery 5 years or more
after the cancer diagnosis were identified for inclusion in the second sample, to test
the second hypothesis. Because the small sample size of twin pairs who completed
the cognitive battery would have prevented controlling for covariates, a co-twin
control design was not used for this second portion of the study. Instead, each of the
long-term cancer survivors who had been given the cognitive battery was matched to
an individual who had never had cancer on the basis of age, sex, education, current
and past exercise, past alcohol consumption, history of alcohol abuse diagnoses, and
current depression. This approach maintained a larger sample size and allowed for
the control of covariates through matching.
Inclusion criteria for cancer survivors in the second portion of Study 1
required: 1) diagnosis with a malignant cancer 2) completion of the cognitive battery
3) cancer diagnosis must have preceded completion of the cognitive battery by 5
years or more. Exclusion criteria for cancer survivors the second portion of Study 1
included: 1) diagnosis with brain cancer, and 2) diagnosis with dementia prior to
diagnosis with cancer. Inclusion criteria for the comparison group for the second
portion of Study 1 required: 1) completion of the cognitive battery, 2) absence of
diagnosis with malignant cancer, 3) match to a cancer survivor on the basis of
demographic and third variable characteristics. Forty-three long-term cancer

66
survivors and matched individuals without a cancer history were included in this
portion of the study.
The following minimum criteria were established a priori to match cancer-
history-free individuals to long-term cancer survivors: sex was matched exactly,
education level was matched within 2 years, age was matched within 5 years, alcohol
use was matched within 250 grams/mo, alcoholism diagnoses were matched by
presence of a diagnosis, depression level was matched exactly, past exercise was
matched within 1 point, and present exercise level (dichotomized) was matched
exactly. However, in addition to matching on all the data that were available for the
cancer survivor, some of the cancer-free individuals had additional data on variables
that the cancer survivors were missing. When possible, matches were found with
missing data that corresponded to the cancer survivor’s missing data. However, this
was not always possible. As would be expected, this resulted in a slightly greater
amount of missing data among the cancer survivors than among their matched
comparison group.
Data Analysis
Prior to hypothesis testing, the data were examined for missing data and
outliers. To determine whether data were missing randomly, McNemar’s chi square
tests were used to examine whether individuals with cognitive dysfunction were
more likely than individuals without cognitive dysfunction to be missing data on
potential confounding variables. McNemar’s chi square tests were similarly used to
examine whether individuals with cancer were more likely than individuals without

67
cancer to be missing data on potential confounding variables. In addition, missing
data caused subgroups of the original sample to be excluded from the logistic
regressions in which multiple variables were used to predict cognitive dysfunction.
These subgroups were compared to the groups of individuals who had complete data
using t-tests and chi square tests. In this way, potential biases that missing data might
have introduced could be examined.
Descriptive statistics for both samples were calculated using paired t-tests for
the continuous variables (age at cognitive screening, alcohol history, and education),
McNemar’s chi square test for the dichotomous variable (sex), and conditional
logistic regressions for the categorical variables (alcoholism diagnoses, exercise, and
depression). None of these statistical procedures requires independence of samples;
this characteristic is necessary given that the samples were either twin pairs or
matched pairs, and therefore non-independent. A Spearman’s correlation matrix was
also used to assess the relationship between each independent and dependent variable
and to check for multicollinearity.
Conditional logistic regression was used to test the hypothesis that cancer
history would remain a significant predictor of cognitive dysfunction after
controlling for potential third variables. Because logistic regression does not assume
normally distributed variables, homoscedasticity, or a linear relationship between the
predictor and outcome variables, these assumptions were not tested. However,
logistic regression does assume a linear relationship between the outcome variable
and the logit of the predictor variables (Tabachnick & Fidell, 2001). This

68
relationship was tested by creating a logit variable for each of the predictors and
entering them into a conditional logistic regression with the original variables
(Tabachnick & Fidell, 2001). If any of the logit variables are significant predictors of
the outcome variable, the linearity assumption is considered to be violated
(Tabachnick & Fidell, 2001).
Separate conditional logistic regressions without covariates were used to
determine whether each of the potential predictor variables—cancer, education, age
at cognitive testing, alcohol history, exercise, and depression—individually were
associated with cognitive dysfunction. To examine whether cancer remained a
significant predictor of cognitive dysfunction after controlling for potential third
variables, all of the significant predictors of cognitive dysfunction were then entered
into a single conditional logistic regression predicting cognitive dysfunction. P-
values less than .10 were considered statistically significant for deciding whether a
potentially confounding variable would be entered into the final conditional logistic
regression, and p-values less than .05 was used to determine statistical significance
for findings in the final conditional logistic regression.
A correlation matrix was used to assess whether tests within the cognitive
battery could be grouped into separate cognitive domains with good convergent and
discriminant validity. Because all the tests were highly correlated with one another,
providing poor discriminant validity between cognitive domains, each test was
examined individually rather than grouped together into cognitive domains. This
approach of using each test as a separate outcome variable, rather than creating a

69
single outcome score that incorporated performance on all the tests, was used in
order to examine the specific cognitive abilities on which long-term cancer survivors
might perform more poorly than cancer history-free individuals—a primary goal of
this study.
To test hypothesis 2, paired t-tests were used to compare performance on
each cognitive test between the long-term cancer survivors who were given the
cognitive battery and matched cancer history-free individuals. To ensure the
assumption of normality was met for the paired t-tests, difference scores for each
cognitive test were calculated, and these difference scores were examined for
normality. Paired t-tests were then used to compare the cognitive test performance of
cancer survivors to that of cancer history-free individuals. For tests on which
difference score distributions were non-normal, the Wilcoxon signed rank test, a
non-parametric equivalent of the paired t-test, was used instead of the paired t-test.
Because this hypothesis testing involved multiple comparisons of performance on
cognitive tests, a Bonferroni adjustment was used that established a p-value of .0035
as the level necessary for statistical significance.
Results
Sample Selection
The first sample focused upon twin pairs discordant for cognitive
dysfunction, as measured by the SALT dementia screening. Of the 4,350 twin pairs
who completed the SALT dementia screening, 609 twin pairs were discordant for
cognitive dysfunction. Two hundred and three twin pairs were excluded due to

70
being opposite-sex. Forty-seven twin pairs were excluded because one or both
members of the pair had been diagnosed with cancer fewer than five years prior to
the cognitive screening. Two pairs were also excluded because one member had been
diagnosed with brain cancer. The initial sample thus consisted of 357 pairs
discordant for cognitive dysfunction. Due to missing data for potential confounding
variables, the sample size was reduced to 189 pairs for the analysis that controlled
for all significant (p<.10) predictors of cognitive dysfunction, and reduced to 153
pairs for the analysis that tested the association between cancer and cognitive
dysfunction and adjusted for significant (p<.10) and non-significant potential
confounding variables.
The second sample focused upon long-term cancer survivors, and a matched
comparison group of cancer history-free individuals, who had been given the
cognitive battery. Of the 12,635 twins who completed the SALT dementia screening,
1,380 individuals were identified as having cognitive dysfunction on the basis of
telephone screening scores and informant reports. These individuals were referred
for the dementia work-up that included the cognitive battery; 70.2% of these
individuals participated in the work-up and cognitive battery (Gatz et al., 2005). The
participation rate in the work-up for individuals who were not able to be screened
due to dementia was 57.8%, that of co-twins of demented individuals was 77.2%,
and that of normal controls was 73.9% (Gatz et al., 2005).
In sum, cognitive battery data were available for 1,263 individuals. Forty-five
of these individuals who participated in cognitive battery testing were long-term

71
cancer survivors. Two of the long-term cancer survivors had been diagnosed with
brain cancer, and were excluded from the sample, resulting in a sample of 43 long-
term cancer survivors. As described earlier, each of these long-term cancer survivors
was matched to a cancer history-free individual on the basis of demographic and
other third variable characteristics, resulting in a sample size of 43 matched pairs.
Cancer survivors were not matched to their co-twins, even if the co-twin had
completed the cognitive battery and did not have a cancer history.
Initial Data Examination
There were no outliers on the cognitive tests or covariates. However, data for
one of the potential confounds appeared to be missing non-randomly. For the co-twin
sample used to test Hypothesis 1, McNemar’s chi square tests showed that the
cognitively impaired group was missing more data on depression (χ2(1)=87.37,
p<.0001) than was the cognitively intact group. There were no other significant
differences between cognitively impaired and cognitively intact twins on rate of
missing data for potential confounds (χ2(1)=0.62 – 2.46, ps >.10). McNemar’s chi
square tests showed no differences between cancer survivors and cancer history-free
twins on rate of missing data for any of the potential confounds, χ2(1)=0.05 – 2.79,
ps>.09.
Comparisons were performed between the twin pairs that had complete data
and those that had incomplete data in order to ascertain any biases that might exist
within the samples that had complete data for the primary analyses. Two sets of
comparisons were performed for this purpose. Firstly, twin pairs with complete data

72
for the significant (p<.10) potential confounds (n=189 pairs) were compared to twin
pairs with incomplete data for the significant potentially confounding variables
(n=168 pairs); the results of these comparisons are displayed in Tables 3 and 4.
Secondly, twin pairs with complete data for all the potentially confounding variables
(n=153 pairs) were compared to twin pairs who had complete data for the significant
predictors of cognitive dysfunction, but incomplete data for the potentially
confounding variables that were not significant predictors of cognitive dysfunction
(n=36 pairs). Tables 5 and 6 show the results of these comparisons. As shown in
Table 3, twins with complete data for the significant (p<.10) potential confounds
were significantly younger than twins with incomplete data, t(712)=9.46, p<.0001.

Table 4. Comparing individuals with complete data (for significant (p<.10) Hypothesis 1
predictors) to individuals with incomplete data: categorical variables

Individuals with
Complete Data
Individuals with
Incomplete Data
Variable
N
N χ2 df P-value
Cancer
378
336 0.06 1 0.81
Cognitive
Dysfunction

378 336 0.00 1 >0.99
Exercise
355
306 2.84 3 0.42
Depression
378
148 4.46 2 0.11

Note: Individuals with complete data tended to have lower rates of cancer, report less early-life
light exercise, and have never been depressed.
Table 3. Comparing individuals with complete data (for significant (p<.10) Hypothesis 1
predictors) to individuals with incomplete data: continuous variables

Individuals with
Complete Data
Individuals with
Incomplete Data
Variable N Mean SD N Mean SD t df P-value
Age at Cognitive
Screening 378 73.97 6.10 336 78.47 6.62 9.46 712 <.0001
Education 378 7.96 2.63 325 7.89 2.97 -0.31 653 0.76
Alcohol Use 344 107.27 219.64 279 97.76 353.97 -0.39 444 0.70

73
Table 6 shows that twins without complete data for all the potentially confounding
variables were more likely to have reported engaging in hard or very frequent past
exercise than individuals who had complete data for all the potentially confounding
variables, χ2(3)=13.30, p=.004.
Table 5. Comparing individuals with complete data (for all Hypothesis 1 predictors) to
individuals with incomplete data: continuous variables

Individuals with
Complete Data
Individuals with
Incomplete Data
Variable N Mean SD N Mean SD t df P-value
Age at Cognitive
Screening 306 73.82 6.08 72 74.60 6.17 0.98 376 0.33
Education 306 8.07 2.72 72 7.47 2.19 -1.74 376 0.08
Alcohol Use 306 109.97 226.63 38 85.58 152.68 -0.87 59.5 0.39

Note: Individuals with complete data reported significantly less early life exercise, and tended to have higher
rates of cancer and less depression.

For the matched sample (n=43 matched pairs) used to test Hypothesis 2,
missing cognitive test data were unrelated to cancer history (p>.09 for each cognitive
test). However, cancer survivors were missing significantly more depression data
than were matched individuals without a cancer history, χ2(1)=5.46, p=.02. This
finding of more missing depression data among cancer survivors was the result of the
Table 6. Comparing individuals with complete data (for significant and nearly significant
Hypothesis 1 predictors) to individuals with incomplete data: categorical variables

Individuals with
Complete Data
Individuals with
Incomplete Data
Variable
N
N χ2 df P-value
Cancer
306
72 0.01 1 0.92
Cognitive Dysfunction 306 72 0.00 1 >0.99
Exercise
306
49 13.30 3 0.004
Depression
306
72 1.66 2 0.44

74
matching process, which necessitated that the cancer-free individuals have at least
the same amount of data as the cancer survivors in order to obtain a match.
To assess for multicollinearity between the predictor variables used in
Hypothesis 1, a correlation matrix between each of the potential predictor variables
was calculated (see Table 7). Correlations ranged from -.19 to .27, indicating that
multicollinearity was not a problem for the logistic regression. The assumption of
linearity of the relationships between the logit predictor variables and the outcome
variable was then tested. None of the logit predictor variables were significant
predictors of the outcome variable (χ2(1)=0.4 – 1.8, ps>.10), indicating that none of
the predictor variables violated this assumption.

Table 7. Spearman’s correlations between Study 1 variables for Hypothesis 1 sample (N=357 twin
pairs).

Measure 1 2 3 4 5 6 7
1. Cognitive Dysfunction —
2. Cancer .10** —
3. Age at Cognitive Screening .003 .06 —
4. Education -.10** -.08* -.19† —
5. Alcohol Use .04 -.03 -.11** .27† —
6. Depression .03 .03 -.10* .11* .04 —
7. Exercise (STR) -.06 -.02 .03 .06 -.02 -.02 —
Note: Although the full sample of n=357 was used for the correlation, sample size for specific correlations varied
depending upon among of missing data for each variable. * p < 05. ** p < .01. † p < .001.

A correlation matrix of the cognitive tests for Hypothesis 2 was used to
examine whether the cognitive tests could be grouped statistically into subscales by
the functional domains described in the methods section (Table 8). Although there
was some evidence for grouping the tests by the primary function believed to be
measured by them, the evidence was not sufficient to combine them into subscales.

75

Table 8. Pearson’s correlations between cognitive battery tests (N=43 matched pairs§)

Measure 1 2 3 4 5 6 7 8 9 10 11 12 13

1. Word List Learning —
2. Word List Recall .89† —
3. Word List Recognition .78† .74† —
4. Prospective Memory .65† .57† .62† —
5. Apartment Naming .52† .39* .55† .36** —
6. Apartment Recall .80† .80† .66† .78† .37† —
7. Apartment Recognition .78† .78† .82† .67† .52† .77† —
8. Apartment Placement Recall .78† .67† .68† .61† .43† .74† .72† —
9. Figure Copying .61† .37* .61† .55† .60† .52† .53† .56† —
10. Block Design .72† .47** .59† .55† .36** .68† .58† .70† .63† —
11. Animals .76† .71† .69† .58† .48† .71† .65† .65† .62† .74† —
12. Judgment .57† .47** .56† .46† .47† .49† .54† .47† .53† .56† .63† —
13. Information .68† .70† .70† .49† .46† .54† .66† .59† .48† .59 .66† .69† —
14. Digit Symbol .78† .66† .67† .62† .36** .70† .72† .70† .58† .81† .73† .52† .68†

§ Although the full sample of n=43 matched pairs was used for the correlation, sample size for specific correlations varied depending upon among of
missing data for each variable.
* p < 05. ** p < .01. † p < .001.

Note: Correlation matrices were also calculated separately for cancer survivors and cancer history-free individuals, and showed a similar range and
patterns of correlations.

75

76
The main reason for this was that most of the cognitive tests were highly correlated
with one another; although the tests believed to primarily measure the same function
tended to be correlated more highly with similar tests, they were also highly
correlated with other tests, providing poor discriminant validity.
Difference scores (cancer survivor – matched cancer history-free individual)
were calculated for each of the 14 cognitive tests, and these difference scores were
examined for normality. All difference scores were normally distributed except
Judgment, Apartment Test Naming, and Apartment Test Recognition. Wilcoxon
signed rank tests were used to compare the performance on those three tests of long-
term cancer survivors to that of the matched comparison group.
Descriptive Statistics
Figure 1 shows the number of twins with and without cognitive dysfunction
who had a cancer history. Table 9 provides descriptive statistics for the first, co-twin
sample, used in this study.
Figure 1. Number of twin pairs with cognitive dysfunction who were cancer survivors

0 30

13 0

The 3 most common cancers among long-term cancer survivors for the first
sample used in this study were breast (33%), prostate (16%), and colorectal (12%),
Twin without cognitive dysfunction
Cancer
Cancer
No Cancer
No Cancer
Twin with cognitive dysfunction

77

Table 9. Demographic and covariate descriptive statistics for Hypothesis 1 - twin pairs discordant
for cognitive dysfunction

Sex
Group N Male Female
Cognitively impaired twins 357 143(40.1%) 214(59.9%)
Cognitively intact twins 357 143(40.1%) 214(59.9%)

Age at Cognitive Screening
Group N Mean SD
Cognitively impaired twins 357 76.10 6.72
Cognitively intact twins 357 76.06 6.75

Education
Group N Mean SD
Cognitively impaired twins 346 7.70 2.57
Cognitively intact twins 346 8.15 2.98

Alcohol Use (grams/mo)
Group N Mean SD
Cognitively impaired twins 283 117.24 355.56
Cognitively intact twins 283 86.71 218.00

Exercise (SALT)
Group N None/Minimal Moderate/High
Cognitively impaired twins 299 191(63.9%) 108(36.1%)
Cognitively intact twins 299 176(58.9%) 123(41.1%)

Exercise (STR)
Group N
Hardly Any/
None Light Regular Heavy
Cognitively impaired twins 313 52(16.6%) 198(63.3%) 36(11.5%) 27(8.6%)
Cognitively intact twins 313 51(16.3%) 178(56.9%) 56(17.9%) 28(8.9%)

Depression
Group N None Subclinical Clinical Depression
Cognitively impaired twins 190 166(87.4%) 4(2.1%) 20(10.5%)
Cognitively intact twins 190 177(93.2%) 2(1.1%) 11(5.8%)

78
with average time from diagnosis being 13.15 years (sd=8.21, min - max: 5.0-38.0).
Table 10 provides a frequency description of all the cancers with which long-term
cancer survivors had been diagnosed.
Table 10. Cancer diagnoses in long-term cancer survivors for Hypothesis 1 -- pairs
discordant for cognitive dysfunction

Diagnosis Frequency (% total cancers)
Breast 14 (32.6%)
Prostate 7 (16.3%)
Colorectal 5 (11.7%)
Bladder 2 (4.7%)
Uterine 2 (4.7%)
Melanoma 2 (4.7%)
Endocrine 2 (4.7%)
Salivary gland 2 (4.7%)
Ovarian 1 (2.3%)
Non-Hodgkin's 1 (2.3%)
Nose 1 (2.3%)
Soft tissue 1 (2.3%)
Eye 1 (2.3%)
Lip 1 (2.3%)
Female genital 1 (2.3%)

Statistics were calculated for the 43 long-term cancer survivors who had been
given the cognitive battery and their matched comparison group, in order to provide
a check on the precision of the match. Table 11 provides the results of these
analyses. The 5 most common cancers among these survivors in this sample were
breast (21%), uterine (12%), endocrine (12%), prostate (7%), and bladder (7%).
Average time since cancer diagnosis was 15.2 years (min - max: 5.0-38.0). Table 12
provides a frequency description of all the cancers with which these 43 long-term
cancer survivors had been diagnosed. Most of the individuals selected for cognitive

79
Table 11. Demographic and covariate statistics to check match precision for Hypothesis 2 - matched
pairs

Sex: N Male Female χ2 df p-value
Long-term survivors: 43 15 (34.9%) 28 (65.1%) 0.00 1 >0.99
Cancer-free individuals: 43 15 (34.9%) 28 (65.1%)

Age: N Mean SD Min-Max t df p-value
Long-term survivors: 43 76.79 5.82 67-88 -0.52 42 0.61
Cancer-free individuals: 43 76.86 5.73 66-88

Education: N Mean SD Min-Max t df p-value
Long-term survivors: 42 7.45 2.38 0-14 0.00 41 0.99
Cancer-free individuals: 43 7.42 2.37 0-14

Alcohol use (grams/mo): N Mean SD Min-Max t df p-value
Long-term survivors: 27 56.78 92.77 0-305 15.33 26 0.15
Cancer-free individuals: 30 33.03 56.00 0-238

Alcoholism diagnoses: N Abuse Dependence χ2 df p-value
Long-term survivors: 2 0 (0%) 2 (100%) 0.00 1 0.24
Cancer-free individuals: 2 1 (50.0%) 1 (50.0%)

Exercise (SALT): N No – Low Moderate – High χ2 df p-value
Long-term survivors: 30 19 (63.3%) 11 (36.6%) 0.00 1 >0.99
Cancer-free individuals: 35 23 (65.7%) 12 (34.3%)

Exercise (STR): N None Low Moderate High χ2 df p-value
Long-term survivors: 29 7(24.1%) 19(65.5%) 1(3.5%) 2(6.9%) 0.00 1 >0.99
Cancer-free individuals: 30 5(16.7%) 23(76.7) 0(0.0%) 2(6.7%)

Depression: N None Minimal Clinical χ2 df p-value
Long-term survivors: 29 26 (89.7%) 0 (0%) 3 (10.3%) 0.00 1 >0.99
Cancer-free individuals: 35 32 (91.4%) 0 (0%) 3 (8.6%)

Note: McNemar’s χ2 tests were used to examine sex and exercise; paired t-tests for age at cognitive screening,
education, and alcohol use; conditional logistic regressions for alcoholism diagnosis, STR exercise, & depression.

80

Table 12. Cancer diagnoses in long-term survivors for Hypothesis 2 – matched pairs (n=43 pairs)

Diagnosis Frequency (% total cancers)
Breast 9 (20.9%)
Uterine 5 (11.6%)
Endocrine 5 (11.6%)
Prostate 3 (7.0%)
Bladder 3 (7.0%)
Colon 2 (4.7%)
Kidney 2 (4.7%)
Lip 2 (4.7%)
Ovarian 2 (4.7%)
Skin 2 (4.7%)
Cervical 1 (2.3%)
Lung 1 (2.3%)
Melanoma 1 (2.3%)
Non-Hodgkin’s 1 (2.3%)
Rectal 1 (2.3%)
Salivary gland 1 (2.3%)
Soft tissue 1 (2.3%)
Thyroid 1 (2.3%)

testing had scored poorly on telephone cognitive screening (N=36 in cancer group
and 35 in comparison group).
Hypothesis 1: Co-twin sample controlling for potential third variables
Conditional logistic regressions were used to determine whether each of the
variables individually were associated with cognitive dysfunction. The results of
these regressions are displayed in Table 13. Cancer, lower education, depression, and
Table 13. Conditional logistic regressions predicting cognitive dysfunction with long-term cancer
history and each potential covariate

Model Predictors
N
(pairs) χ2 df
p-
value OR 95% CI
Model
p-value
1 Cancer 357 7.52 1 0.006 2.89 (1.35, 6.17) 0.003
2 Age at cognitive screening 357 3.21 1 0.07 1.67 (0.95, 2.91) 0.06
3 Education 346 7.28 1 0.007 0.90 (0.84, 0.97) 0.005
4 Alcohol 283 0.57 1 0.45 1.00 (1.00, 1.00) 0.26
5 Depression 190 3.59 1 0.06 1.49 (0.99, 2.24) 0.049
6 Exercise (STR) 313 1.89 1 0.17 0.85 (0.67, 1.07) 0.17

81
older age at cognitive screening were each significant (p<.10) predictors of cognitive
dysfunction in separate regression equations. Next, each of the significant
predictors—cancer, age at cognitive screening, education, and depression—was
entered into a logistic regression to determine whether cancer continued to be a risk
factor for cognitive dysfunction after controlling for potential confounds. As shown
in Table 14 (model 1), cancer remained a significant risk factor for cognitive
dysfunction after controlling for these covariates. When alcohol consumption and
past exercise were entered into the equation with the other covariates, however,
cancer was no longer a significant risk for cognitive dysfunction (Table 14, model 2).
Table 14. Conditional logistic regression predicting cognitive dysfunction with long-term
cancer history and covariates

Model: Predictors
N
(pairs) χ2 df p-value OR 95% CI
Model
p-value
1 Cancer 189 4.01 1 0.045 3.21 (1.03, 10.06) 0.004

Age at cognitive
screening 3.17 1 0.07 2.19 (0.92, 5.21)
Education 1.86 1 0.17 0.93 (0.83, 1.03)
Depression 3.76 1 0.05 1.52 (0.99, 2.33)

2 Cancer 153 3.18 1 0.07 2.91 (0.90, 9.41) 0.06

Age at cognitive
screening 0.79 1 0.37 1.51 (0.61, 3.72)
Education 0.59 1 0.44 0.96 (0.86, 1.07)
Alcohol 0.06 1 0.81 1.00 (1.00, 1.00)
Depression 2.44 1 0.12 1.49 (0.90, 2.45)
Exercise (STR) 1.88 1 0.17 0.75 (0.50, 1.13)

To better understand why the addition of alcohol into the regression equation
rendered cancer history a non-significant risk factor, I re-ran the individual logistic
regressions with the smaller (n=153) sample that was used in the last regression
equation. With the decreased power from this smaller sample, none of the variables

82
was a statistically significant predictor of cognitive dysfunction (χ2(1)=0.0001-3.62,
ps=0.06-0.32), even when each of the predictors was entered into separate regression
equations. This finding suggests that decreased statistical power is the reason that
cancer history is no longer a significant predictor of cognitive dysfunction when
alcohol and past exercise are entered into the logistic regression. In addition, the
correlations displayed in Table 7 show no association between cancer and alcohol
consumption, or cancer and past exercise.
Because the full sample size of n=357 pairs was not the actual sample size for
the logistic regression with significant predictors of cognitive dysfunction, individual
logistic regressions predicting cognitive dysfunction were calculated using the
sample that had complete data for the significant predictors of cognitive dysfunction
(n=189 pairs). Table 15 shows the results of these analyses. The odds ratios within
Table 15 provide a better comparison than those within Table 13 between the
predictive value of each significant predictor alone, and the predictive value of each
predictor when controlling for the other significant predictors of cognitive
dysfunction.

Table 15. Conditional logistic regressions predicting cognitive dysfunction with long-term
cancer history and each potential covariate (n=189 pairs)

Model: Predictors
N
(pairs) χ2 df
p-
value OR 95% CI
1 Cancer 189 4.88 1 0.03 3.50 (1.15, 10.63)
2 Age at cognitive screening 189 3.82 1 0.05 2.30 (1.00, 5.30)
3 Education 189 2.11 1 0.15 0.93 (0.83, 1.03)
4 Depression 189 3.59 1 0.06 1.49 (0.99, 2.24)

83

Hypothesis 2: Matched sample examining specific cognitive deficits
Paired t-tests were used to examine whether cancer survivors performed more
poorly than matched cancer history-free individuals on individual cognitive tests. For
the three tests on which difference scores were not normally distributed—Judgment,
Apartment Test Naming, and Apartment Test Recognition—the Wilcoxon signed
rank test was used instead of the paired t-test to compare the performance of cancer
survivors to their matched cancer history-free individuals. Table 16 provides the
results of these comparisons. After the Bonferroni adjustment, which set the p-value
at .0035, only the Word List Recognition and the Apartment Test Recognition tests
were statistically significant.
3

Exploratory analyses were conducted to ascertain whether there were higher
rates of dementia in the 43 cancer survivors than in the matched comparison group,
and whether groups differed in terms of types of dementia diagnosed. There were
approximately equal rates of dementia diagnosed between groups (12 of the cancer
survivors, and 13 of the controls), indicating that higher rates of dementia amongst
the cancer survivors could not be responsible for the differences observed in
cognitive tests. When types of dementia were compared between groups, there was a
non-significant trend for greater rates of vascular dementia within the cancer history-

3
Because of the high number of individuals in both groups who had performed poorly on the
telephone cognitive screen, exploratory analyses were conducted to compare the cognitive test
performance of those cancer survivors and cancer history-free individuals who did not perform poorly
on the telephone cognitive screen. With this sample of 7 cancer survivors and 8 cancer history-free
individuals, groups differed significantly only on performance on the Figure Copying test (p<05).
Cancer survivors performed significantly worse than cancer history-free individuals on this test.

84
Table 16. Comparing cognitive test performance of Swedish Twin long-term cancer survivors to matched cancer history-free controls

Long-term Cancer Survivors Cancer History-Free Controls
Cognitive Test N (pairs) Mean SD Mean SD Paired t df p-value
Word List Learning 42 11.69 7.84 14.21 7.19 2.16 41 0.04
Word List Recall 43 2.93 2.86 3.88 3.02 1.90 42 0.06
Word List Recognition 43 14.53 5.86 17.16 3.98 3.15 42 0.003*
Prospective Memory 39 1.72 1.41 2.05 1.36 1.36 38 0.18
Apartment Naming 42 9.48 1.33 9.69 1.26 15.00 § 0.21
Apartment Recall 39 4.31 3.50 4.95 3.35 0.99 38 0.33
Apartment Recognition 39 6.72 4.06 8.46 2.69 77.00 § 0.0026*
Apartment Placement Recall 38 5.89 3.42 6.16 3.69 0.38 37 0.70
Figure Copying 41 6.93 3.32 8.12 2.88 2.25 40 0.03
Block Design 35 11.57 9.26 11.86 8.95 0.21 34 0.83
Animals 41 12.05 7.02 14.00 7.40 1.51 40 0.14
Judgment 41 2.24 1.18 2.37 1.02 22.00 § 0.60
Information 40 11.81 6.55 12.93 6.36 1.03 39 0.31
Digit Symbol 31 22.68 14.43 24.87 12.84 1.00 30 0.32

*significant after Bonferroni adjustment, which set p-value at .0035
§ Wilcoxon signed rank test used, due to non-normal distribution

84

85
free individuals compared to the cancer survivors (χ2(1)=2.00, p=.15, OR=2.0), and a
similar non-significant trend for higher rates of Alzheimer’s disease (AD) among
cancer survivors compared to cancer history-free individuals (χ2(1)=0.33, p=.56,
OR=2.0).
Discussion
The results of the first portion of the study confirmed my hypothesis that
long-term cancer survivors remained at increased risk for cognitive dysfunction after
controlling for potential confounding factors. When the variables that were
significant risk or protective factors for cognitive dysfunction alone (namely,
education, age at testing, and depression) were included with cancer history in the
equation, cancer history remained a significant risk factor for cognitive dysfunction.
When all the potential predictor variables were included with cancer history in the
equation, however, cancer history was not a significant predictor of cognitive
dysfunction (p=.07). Of note was the reduced sample size when controlling for
alcohol use and past exercise, which included 153 twin pairs rather than the 189 twin
pairs included in analyses when controlling for education, age at testing, and
depression. Yet, when an exploratory analysis examined cancer as the sole predictor
of cognitive dysfunction using this reduced sample size of 153 pairs, it was not a
significant predictor of cognitive dysfunction, either (p=.12). Therefore, the smaller
sample size, rather than a relationship between cancer and cognitive dysfunction
caused by alcohol consumption, is responsible for reducing the effect of cancer to
non-significance. The stability of the odds ratio generated for cancer history when it

86
was a sole predictor of cognitive dysfunction (Table 15, OR=3.50), and when
controlling for the other significant predictors (Table 14, OR=3.21), provides further
evidence for this interpretation.
Although there remain other potential confounds, such as dietary factors, that
might increase risk of both cancer and cognitive dysfunction, the consistency of
long-term cancer survivorship as a risk factor for cognitive dysfunction while
controlling for exercise, depression, age at testing, and education is remarkable. This
finding suggests that factors more directly related to the cancer experience are likely
to contribute to cognitive problems later in life. These results are particularly
important in light of the previously described similarities between the types of
cognitive dysfunction seen with alcohol abuse (Hildebrandt et al., 2004; Sullivan et
al., 2002; Bruce & Jones, 2004), the cognitive abilities that may be preserved with
exercise (Colcombe & Kramer, 2003), and the cognitive deficits that previous
research suggests are common in cancer survivors (e.g., Ahles et al., 2002; van
Oosterhout et al., 1996; Cull et al., 1994; Castellon et al., 2004).
The results of the second portion of the first study, which found that long-
term cancer survivors performed more poorly on tests measuring verbal and
nonverbal recognition memory, partially confirmed my hypothesis. However, this
study did not find significant differences between long-term cancer survivors and
matched comparisons on a test of information processing speed. Moreover, the type
of verbal and nonverbal memory that was significantly poorer in long-term cancer
survivors was recognition memory, rather than recall memory. While previous

87
studies have found recall memory problems in cancer survivors, recognition memory
has consistently been found to be spared (Ahles et al., 2002; Schagen et al., 1999;
van Oosterhout et al., 1996). The present finding of poor recognition in cancer
survivors is thus surprising, and warrants consideration of why recognition memory
might be the distinguishing factor between these two groups of individuals who were
both, for the most part, experiencing significant cognitive dysfunction.
The sample in the present study differs from previous samples in ways that
are likely to have influenced the results. Probably most influential in producing the
current pattern of results was the large number of individuals who had been found to
experience severe gross cognitive problems prior to more sensitive cognitive testing.
In contrast with the present study, most previous studies focused upon cancer
survivors who showed mild cognitive problems rather than serious cognitive
impairment (e.g., Bender et al., 2006; Castellon et al., 2004; Ahles et al., 2002;
Schagen et al., 1999; van Oosterhout et al., 1996). Cancer survivors in these previous
studies have probably not been found to perform poorly on recognition memory
tasks because recognition memory is a relatively easy cognitive task (Johnson,
1990). When general cognitive functioning is intact, even individuals with memory
deficits often perform well on recognition memory tasks (Johnson, 1990).
Impairments in recognition memory therefore suggest such profound impairments in
memory systems that even relatively simple memory tasks cannot be performed. It is
therefore probable that, when comparing groups of individuals who are largely
cognitively impaired, the patterns of cognitive problems differ from those which

88
would be found in cognitively intact groups. Individuals with only mild levels of
cognitive problems, such as those in previous studies, would be expected to show
only mild memory problems, if any, rather than the profound memory problems
suggested by recognition memory deficits.
Due to this sampling bias towards individuals with gross cognitive
dysfunction, the present sample was likely biased in ways that other studies were not.
However, because there were equal rates of dementia between the long-term cancer
survivors and the comparison group, it is not true that the cancer survivors
demonstrated this pattern of deficits simply because they were selected to have more
cognitive dysfunction than the comparison group. Despite the limitation inherent in
focusing largely upon individuals with serious cognitive dysfunction, the comparison
done here is informative because, given that both groups are equally likely to have
cognitive problems, it is a comparison between the types of cognitive problems that
long-term cancer survivors may be more susceptible to, compared to individuals who
have never had cancer.
While recall memory involves both memory acquisition and retrieval
(Shallice et al., 1994), which can be disrupted during storage or retrieval of
information, recognition tests require storage but not retrieval of information.
Because recognition memory is primarily a hippocampal task (Manns et al., 2003),
the recognition differences observed in this study between long-term cancer
survivors and matched cancer history-free individuals suggest cognitive problems in
long-term cancer survivors may be due to temporal lobe dysfunction. Consistent with

89
the trends found in this study, poor performance on recognition tasks indicates an
amnestic memory problem, characteristic of AD (Cummings, 2003).
It may be of note that cancer survivors performed more poorly than matched
controls on Word List Learning and Figure Copying by standard significance-test
levels (p<.05), and there was a non-significant trend for the cancer survivors to
perform more poorly than matched controls on all cognitive tests. Although the
difference between long-term cancer survivors and cancer-free individuals on Figure
Copying was no longer statistically significant after the Bonferroni adjustment, and
may thus be simply a result of the large number of statistical tests performed, even
these non-significant findings are consistent with an Alzheimer’s-type neurological
deterioration. AD patients have been found to perform more poorly than controls on
Figure Copying (Morris et al., 1989), probably because copying tasks require good
parietal functioning (Rascovsky et al., 2002). AD typically affects the temporal
lobes, manifesting in learning and amnestic memory problems, and progresses to the
parietal lobes, manifesting in visuospatial problems (Cummings, 2003; Rascovsky et
al., 2002). The pattern found here in long-term cancer survivors, with no significant
differences in cognitive tasks except recognition memory and, prior to the
Bonferroni adjustment, learning and visuoconstructional functioning, thus appears
consistent with an Alzheimer-type or primarily temporal lobe dysfunction among
long-term cancer survivors.
This pattern of results was unexpected, given the more frontal-subcortical
pattern of attention, executive function, and recall memory dysfunction that had been

90
found in prior studies of cancer survivors (Ahles et al., 2002; van Oosterhout et al.,
1996; Cull et al., 1994). However, finding memory problems in cancer survivors is
certainly not unprecedented (Ahles et al., 2002; van Oosterhout et al., 1996),
particularly for cancer survivors who are fewer than 5 years post-diagnosis (Wefel et
al., 2004a; Brezden et al., 2000; Schagen et al., 1999; Wieneke & Dienst, 1995;
Meyers, 1992). There is corresponding evidence from brain imaging studies
suggestive of neuroanatomical damage within the temporal lobes. As mentioned
previously, Inagaki and colleagues (2006) found that one year following surgery,
breast cancer patients who received chemotherapy had significantly smaller
prefrontal, parahippocampal, and cingulate gyri, and the precuneus volumes than
breast cancer patients who had been treated only surgically. These areas are largely
within the temporal lobes and typically associated with learning, recall and
recognition memory, and working memory. At one year follow-up, these volumetric
differences were no longer evident, suggesting that the damage was temporary. The
deficits in recall memory and the observed temporal lobe damage that other studies
have found might occur as the result of initial damage to the temporal region of the
brain, which can recover over time but which the present study suggests could be at
risk for worsening in older age to produce the more serious deficits observed in the
present study.
Due to the high number of individuals with serious cognitive dysfunction in
both groups, the smaller between-group differences in verbal learning and recall are
not believed to be due to cancer survivors performing well on those tasks, but rather

91
on the cancer history-free individuals also performing poorly on the tasks. The non-
significant trend towards a greater number of diagnoses of vascular dementia, which
affects learning and recall but not recognition (Tierney et al., 2001), among the
comparison group compared to the cancer survivors supports this idea. The non-
significant trend towards a greater number of diagnoses of AD among cancer
survivors than the comparison group similarly supports this idea. That is because
both AD and vascular dementia can result in deficits in word list learning and recall
tasks (Tierney et al., 2001, Cummings, 2003). However, deficits in recognition are
more common in AD, due to the amnestic quality of the memory problems
(Cummings, 2003). Because the present analyses comparing types of dementia
between cancer survivors and cancer history-free individuals were not significant,
they are merely hypothesis-generating rather than conclusive. However, it may be
important for future research to include an examination of factors that might place
long-term cancer survivors at increased risk for damage to the temporal lobes, or
Alzheimer-type cognitive problems.
Because the present sample focused upon people with cognitive problems, it
is not likely to be representative of cancer survivors. As mentioned previously, long-
term cancer survivors without serious cognitive dysfunction—even those who are
older adults—may have a much different pattern of cognitive problems than the
present study would suggest. Due to the bias created by selecting mostly individuals
with cognitive dysfunction, the cancer survivors in the present study may differ in
important ways from many other cancer survivors. For instance, they may be the

92
cancer survivors at greatest genetic risk for severe cognitive problems in older
adulthood. Any genetic risk for later-life cognitive dysfunction may have interacted
with their cancer treatment to produce a different pattern of results than that of
cancer survivors without genetic risk for cognitive dysfunction. Put simply, long-
term older adult cancer survivors without other risk factors for serious cognitive
dysfunction may never show the serious memory impairments indicated by the
poorer recognition memory found in this study. The results of this study therefore
should not be generalized to all long-term older adult cancer survivors. Rather, it
should be noted that cancer survivors who begin showing severe cognitive problems
may have problems that are most pronounced in memory areas.
Although the difficulty in generalizing these findings is a limitation of the
study, this research may have clinical implications. Older adult long-term cancer
survivors who are concerned about their cognitive functioning should receive
cognitive testing that includes a comprehensive memory assessment, since this study
suggests that memory can be seriously affected in cancer survivors. Additionally,
medical professionals who provide care for older adults with cognitive impairments
should be aware that older adults with cognitive impairments and a history of cancer
may have more serious memory problems than other individuals. Care plans for such
individuals should be modified to compensate for potential memory problems such
that important information is provided in writing and to the individual’s spouse or
caregiver.

93
Several limitations of these studies should be noted in addition to the
limitation in generalizability discussed earlier. The amount of data that were missing
for the variables of interest for the first set of analyses, and the apparent non-
randomness of this missing data, are one limitation of this study. For the first set of
analyses, which used twin pairs discordant for cognitive dysfunction, cognitively
impaired twins were missing more education and depression data than cognitively
intact twins. This may have been due to a decreased ability for individuals with
cognitive impairment to respond to education and depression questions during the
SALT interview. Regardless of the cause, these missing data may have biased the
results found by preventing these potentially confounding variables from being
adequately controlled for statistically. In addition, individuals with complete data for
the logistic regression that controlled for significant covariates were significantly
younger than individuals with incomplete data. This finding suggests that the range
of individuals examined in this study was constricted by missing data; moreover, as
the focus of this study was on older adults, this unintentional exclusion of some of
the older individuals within the sample excluded individuals of greatest interest.
Secondly, this study was retrospective in nature, and is thus less well
controlled than prospective studies that can focus upon specific types of cancer
survivors and can collect specific data of interest, such as cancer treatment data. The
samples are also fairly homogenous, consisting of Caucasian Swedish twins, and so
do not adequately represent the ethnic range of individuals diagnosed with and
treated for cancer in the United States and other parts of the world. In addition, the

94
cancer survivors had been diagnosed at different ages and different time points prior
to the collection of neuropsychological functioning data; these patients were also
diagnosed with a variety of cancers, at various stages of disease severity, and
received a variety of different chemotherapies and other treatments. This
heterogeneity may have obscured the results, and makes it difficult to make
inferences about potential mechanisms of cognitive dysfunction in cancer survivors.
Furthermore, as alluded to earlier, treatment data were not available, and so an
examination of the association between cancer treatments and cognitive dysfunction
was not possible.
One additional limitation of the study deserves mention, particularly as it
relates to the field of neuropsychology more broadly. As observed by the high
correlations between many of the cognitive tests (Table 8), there appears to be
significant overlap between the functions necessary to complete the tests. That is, it
is difficult to discriminate statistically between many of the cognitive functions
measured by different tests. Although the idea that these tests measure discrete and
distinct cognitive functions is overly simplistic—a straw man easily knocked
down—the high correlation between the tests also casts doubt upon the more
fundamental assumption that the tests do a good job measuring the distinct cognitive
functions that they are purported to be primarily measuring. This concern provides
another reason that the present findings should be viewed as suggestive, rather than
conclusive.

95
Nevertheless, the tests do provide a sample of a person’s abilities. As a result,
they provide a sense of how the individual may be functioning in daily life. For
instance, if a person has difficulty immediately recalling a word list, it is evidence
that a person has difficulty quickly learning unstructured, non-contextual material.
Although it is impossible to know from just this one test whether the individual’s
difficulty arises from a learning deficit, slow processing speed, poor attention, or
difficulty with retrieval, it is nevertheless possible to get a glimpse of the types of
tasks that an individual might struggle with in real life. In addition, when a cognitive
battery includes multiple tests that appear to measure similar functions, it is possible
to get a more reliable picture of an individual’s functioning.
Despite the study’s limitations, it has a number of strengths. By using
cognitive data that were collected independently of cancer history, it avoided some
biases that may be operating in other studies of cognitive functioning in cancer
survivors (e.g., the comparison of those cancer survivors who are concerned about
their cognitive functioning to healthy controls who do not share these concerns). This
study also addressed the issue of potential confounding factors that could cause a
spurious association between cancer and cognitive dysfunction, a possibility that has
not been previously addressed in the empirical literature. By controlling for lifestyle
covariates that have not been previously examined in this line of research, this study
suggests neuropsychological deficits in cancer survivors are not likely due to
confounding factors that put individuals at risk for both cancer and
neuropsychological deficits. It also focused upon older adults who, although they

96
comprise the group most likely to experience neuropsychological deficits, have been
underrepresented in the study of neuropsychological deficits in cancer survivors.
Chapter 3: Study 2
The second study examined cognitive functioning at multiple time points to
discern whether the rate of cognitive decline differed between long-term cancer
survivors and matched individuals without a cancer history. This study also
examined whether the findings in Study 1 replicate using a different sample.
As with the first study, I hypothesized that long-term cancer survivors would
demonstrate poorer performance on neuropsychological tests most dependent upon
the frontal-subcortical neurological system: tests measuring attention, working
memory, and verbal memory, than individuals without a cancer history. I did not
expect long-term cancer survivors to demonstrate poorer performance on tests
measuring verbal functioning or visuoconstructional skills. Because little is known
about the rate of cognitive decline in older adult long-term cancer survivors, I was
also curious to determine whether long-term cancer survivors would show a steeper
rate of decline on neuropsychological tests measuring attention, working memory,
and verbal memory than individuals without a cancer history.
Lastly, I intended to explore whether cognitive functioning in cancer patients
declined from a baseline level of cognitive functioning prior to cancer diagnosis to
approximately 1 year post-treatment, compared to individuals who had never been
diagnosed with cancer. This pre-cancer baseline level of cognitive functioning
should be less influenced by disease-related factors and diagnosis-related

97
psychological distress than cognitive functioning baselines that are obtained after
individuals have been diagnosed with cancer, which other studies have used
(Iconomou et al., 2004; Wefel et al., 2004a; Fliessbach et al., 2003; Ahles et al.,
1998; Kaasa, Olsnes, & Mastekaasa, 1988; Oxman & Silberfarb, 1980).
Method
Procedure
Study 2 used data from the University of Southern California’s Alzheimer
Disease Research Center (ADRC). The ADRC was established at USC in 1984, and
has recruited community participants who have AD and those who are cognitively
intact for its longitudinal study. Participants completed an intake interview that
inquires about self-reported cognitive and neurological symptoms, medical history
(including cancer), age, and several risk factors for neuro-degenerative diseases.
Following the intake interview, participants were administered a neuropsychological
battery.
Participants were re-tested with the neuropsychological battery at pre-
determined intervals that vary depending on age and risk for developing AD.
Individuals complete neuropsychological assessment at more frequent intervals with
increasing age and AD risk: participants younger than 70 years old are assessed
every 5 years, participants aged 70 – 80 years old are assessed every 2 years, and
participants aged 80 years and older, or those with memory problems (such as a
diagnosis of dementia or mild cognitive impairment), are assessed annually. Many
tests within the neuropsychological assessment have been consistently administered

98
in the ADRC battery for 10 years or more. Tests that were consistently administered
for any 10 year span or more of the approximately 23 years that the longitudinal
study has been collecting data, and on which approximately 50% or more of the data
were available, were used as dependent variables measuring neuropsychological
functioning.
Participants within the ADRC were also given follow-up interviews that
inquired about new medical, neurological, or psychological problems that had
occurred since the initial interview. A small number of participants reported in those
interviews that they had been diagnosed with cancer since the intake interview.
Individuals who reported that they had been diagnosed with cancer were not re-
tested with the neuropsychological battery until a year following completion of their
cancer-related treatment. At that time, if cancer patients showed serious cognitive
deficits, they were dropped from the ADRC study.
To ensure accurate identification of all ADRC participants who had been
diagnosed with cancer in Los Angeles County, and to obtain diagnosis and treatment
information for these individuals, the ADRC data were linked to data from the
Cancer Surveillance Program (CSP). Established in 1972, the CSP is a population-
based cancer registry for Los Angeles County to which hospitals and medical
providers report diagnostic and treatment-related information for all Los Angeles
County residents diagnosed with cancer. Only the first course of a cancer patient’s
treatment generally was reported to the CSP.

99
A dataset from the ADRC that contained individuals’ full names, birthdates,
social security numbers, cancer diagnoses, and ADRC subject identification number
was released to the CSP to identify and confirm cancer survivors and to ensure that
individuals who had not reported a history of cancer had not been diagnosed with
cancer. The CSP merged this dataset with their own dataset, ensuring accurate match
of individuals through checking the consistency of all available identifying
information. The CSP then deleted all unique identifiers from this new dataset except
date of birth, and released the cancer data with the ADRC’s subject identification
numbers. The ADRC also released a dataset to me that contained each participant’s
subject identification number, and intake, follow-up, and neuropsychological
assessment variables, but lacked unique identifiers except date of birth. I merged the
CSP and ADRC datasets together by subject identification number to have cancer
and neuropsychological data in one dataset.
Participants and Design
This study used participants from 2 groups of cancer survivors: 1 group of
long-term cancer survivors who entered the study with a history of cancer, and 1
group who were diagnosed with and treated for cancer during the course of their
participation in the ADRC longitudinal study.
ADRC study participants who were long-term cancer survivors were
identified through two means. Participants were first identified as cancer survivors
through the match with the CSP data. If the CSP had records of an ADRC participant
being diagnosed with malignant cancer, that person was identified as a cancer

100
survivor. If the cancer survivor had been diagnosed with cancer 5 years or more prior
to his or her first ADRC neuropsychological assessment, he or she was considered a
long-term cancer survivor. Additional participants were identified as cancer
survivors if they had reported a history of malignant cancer during the ADRC intake
interview. Again, if the cancer had been diagnosed 5 years or more prior to the first
neuropsychological assessment, the individual was considered a long-term survivor.
Inclusion criteria for long-term cancer survivors were: 1) history of malignant
cancer prior to entry into ADRC study, 2) most recent cancer diagnosis at least 5
years prior to first neuropsychological assessment, and 3) participation in at least 2
ADRC neuropsychological assessment batteries. Exclusion criteria for long-term
cancer survivors were: 1) diagnosis with brain cancer, 2) diagnosis with only benign
cancers, 3) diagnosis with cancer in childhood, or 4) diagnosis with dementia prior to
diagnosis of cancer. Individuals who did not have a history of cancer were eligible
for the comparison group if they: 1) had never been diagnosed with malignant
cancer, and 2) had participated in at least 2 ADRC neuropsychological assessment
batteries.
For the second set of analyses, participants who were diagnosed with cancer
during the course of their participation in the ADRC study were identified through
the CSP. Neuropsychological assessments of these participants were postponed until
1 year after their cancer treatments ended, at which time they resumed their normal
neuropsychological assessment schedule. Inclusion criteria for these cancer survivors
include: 1) diagnosis with first malignant cancer during participation in ADRC

101
study, and 2) completion of at least one neuropsychological battery prior to cancer
diagnosis, and one after cancer diagnosis. Exclusion criteria for these cancer
survivors were: 1) diagnosis with brain cancer, 2) diagnosis with only benign
cancers, or 3) diagnosis with dementia prior to diagnosis of cancer. Individuals who
did not have a history of cancer were eligible for the comparison group if they: 1)
had never been diagnosed with malignant cancer, and 2) had participated in at least 2
ADRC neuropsychological assessment batteries.
Each long-term cancer survivor was individually matched to a participant
without a cancer history on the basis of age (within 4 years), education (within 2
years), ethnicity (matched exactly, unless no precise ethnic match could be found
within the other matching parameters), sex (matched exactly, unless no same sex
match could be found within the other matching parameters), and date of first
neuropsychological assessment (within 10 years). Cognitive status (demented versus
not demented) was not used as a matching characteristic, due to the possibility that
matching on it would, in effect, control for the very outcome—cognitive
functioning—that this study was investigating. Each of the cancer patients who had
been diagnosed with cancer in Los Angeles County during his or her participation in
the ADRC study was matched to an individual without a cancer history on the basis
of age (within 4 years) at the assessment immediately before cancer diagnosis (for
the comparison group, age at the time of first cognitive assessment was used),
education (within 2 years), ethnicity (matched exactly, unless no precise ethnic
match could be found within the other matching parameters), sex (matched exactly,

102
unless no same sex match could be found within the other matching parameters), and
date (within 10 years) of first neuropsychological assessment following cancer
diagnosis (for the comparison group, this date was simply their second
neuropsychological assessment).
Because depression was a significant covariate in the first, Swedish Twin,
study, the inclusion of depression as a matching variable was considered. However,
because the relationship between cancer history, depression, and cognitive
functioning was not clear from the first study—depression could be prodromal, or
could be a symptom of cognitive decline—the relationship between cancer and
depression over time was first explored. To determine whether to match on
depression or include it as an outcome variable, a brief exploratory analysis was
conducted. Repeated measures Analyses of Variance (ANOVA) using two, three,
and four time points of data were used to explore the relationship between cancer
and Geriatric Depression Scale scores over time, using the entire sample of long-
term cancer survivors and individuals without a cancer history. These different
lengths of follow-up were used to achieve a balance between examining depression
over a length of time, and recognizing the loss of statistical power that resulted with
the smaller sample of individuals who had 3 and 4 time points of data. There were no
significant group differences between cancer survivors as a group and individuals
without a cancer history using two, three, or four time points (e.g., using 3 time
points, F(2, 286)=0.86, p=.42). However, there were significant group-by-time
interactions using three and four time points, with cancer survivors showing

103
increased depression over time while individuals without a cancer history showed
decreased depression over time (e.g., using 3 time points, F(2, 286)=4.72, p=.0096).
Due to this interaction and the desire to avoid unnaturally constricting a potentially
interesting outcome variable by matching on it, I decided to examine depression as
an outcome variable rather than as a matching variable.
Measures
Intake interview
The intake interview for the ADRC is a standardized, structured interview
that obtains demographic information such as age, education, ethnicity, and sex, and
self-report information about participants’ cognitive functioning, any dementia
diagnoses, and medical history (see Appendix C). The intake interview was
conducted either with the participant, or with a spouse, family member, or
professional caregiver. The intake interview provided the variables necessary for
matching the ADRC data to the CSP: subject ID, last name, first name, date of birth,
address, and social security number. Age at first neuropsychological testing,
education, and ethnicity were also used to match cancer survivors to cancer-history-
free individuals. Education was recorded as years of formal education, with a
master’s degree equal to 18 years of education, law degree equal to 19 years of
education, and M.D. or Ph.D. equal to 20 years of education. Ethnicity was recorded
as “American Indian/Alaskan Native,” “Asian/Pacific Islander,” “Black,”
“Hispanic,” or “White.”
Mood

104
The Geriatric Depression Scale (GDS) is a 30 item measure of depression
developed for and normed on older adults. This measure has internal consistency and
split-half reliability coefficients of approximately .9 (Brink et al., 1982), and has
high correlations with other measures of depression (Lezak et al., 2004).
Neuropsychological assessment
The tests listed below are organized into broad cognitive domains on the
basis of a widely-used neuropsychological text (Lezak et al., 2004).
Gross Cognitive Functioning
The Mini Mental State Examination (MMSE) is a widely-used 11 item
measure of gross cognitive functioning that was designed to measure orientation,
registration, attention and calculation, recall, and language (Folstein, Folstein, &
McHugh, 1975). Scores can range from 0 – 30, with scores below 24 considered to
indicate cognitive impairment.
Verbal Functioning:
As described in Study 1, Animals was used to assess semantic verbal fluency.
FAS, also called the Controlled Oral Word Association test, is a test that
measures phonemic verbal fluency by asking the individual to name as many words
within a specified time limit that begin with the letters F, A, and S. One year test-
retest reliability coefficients tend to be above .7 (Snow et al., 1988).
Boston Naming Test (BNT) is a task of confrontation naming taken in which
individuals are asked to name 60 increasingly obscure objects presented in drawings.

105
It correlates highly with tests of verbal ability, such as the WAIS-R Vocabulary test
(Killgore and Adams, 1999).
Verbal Memory:
As described in Study 1, the CERAD Word List was used to measure short-
term verbal memory. The Learning score is the sum of an individual’s scores on the
three learning trials of the word list. The Delayed Recall score represents the number
of words correctly recalled from the list after a delay. The Recognition score is the
number of words correctly recognized from a list that includes intrusions and the
words initially presented.
Visuoconstructional Skills:
As described in Study 1, Figure Copying is a measure taken from the
CERAD that assesses visuoconstructional ability. This test has been found to be
sensitive to changes in cognitive functioning in individuals with AD when the test is
re-administered after a year (Morris et al., 1989).
Attention and Working Memory:
Digit Span Forward is a measure of simple attention taken from the WAIS-R.
Individuals are asked to repeat verbally presented strings of numbers of increasing
length. Test-retest reliability coefficients are between .66 and .89 (Matarazzo and
Herman, 1984).
Digit Span Backward is a measure of mental tracking taken from the WAIS-
R that asks individuals to repeat in reverse order increasingly long strings of
numbers. Individuals in their 70’s and older typically can repeat 1 fewer digits

106
backwards than can younger individuals (Lezak et al., 2004). Norms are available up
to age 89 (Wechsler, 1997).
Data Analysis
Prior to hypothesis testing, the data were examined for missing data and
outliers. Because the neuropsychological test battery changed over time, some data
were unavailable if a neuropsychological test was not in the battery at a time a
participant was tested. To examine data that were truly missing (where a person
should have completed a neuropsychological test, but did not), variables were
created to indicate for each variable whether data were missing, available, or not
available due to the neuropsychological test not being administered during the date
of the assessment. McNemar’s chi square tests were used to compare the amount of
missing data by cognitive status or cancer history. The number of
neuropsychological assessments was compared between cancer survivors and
matched cancer history-free individuals to examine whether cancer survivors
dropped out of the study at a significantly higher rate than cancer history-free
individuals.
Data were examined for non-normality and heteroscedasticity. Linearity was
not examined, because the predictor, cancer, is a dichotomous variable, and “a linear
model is de facto acceptable for dichotomous predictors” (Singer & Willet, 2003,
p.128). For distributions that were non-normal or had heteroscedasticity, log
transformations were computed and examined for improvements in normality and
homoscedasticity. If the log transformations produced improvements, the

107
transformed data were used in the analyses. The fixed effects from those analyses
were then exponentiated for ease of interpretation. If the log transformation did not
produce improvements, the original data were used in the analyses.
The quality of the match of cancer survivors to cancer-free individuals was
investigated for both samples, by using the demographic variables on which
individuals were matched as outcome variables. For continuous demographic
matching variables (age and education), paired t-tests were used to compare cancer
survivors and matched cancer history-free individuals. For the dichotomous
demographic matching variable (sex), McNemar’s chi square test was used to
compare cancer survivors and matched cancer history-free individuals. For the
categorical demographic matching variable (ethnicity) conditional logistic regression
was used to compare cancer survivors and matched cancer history-free individuals.
Each of these statistical procedures accommodates non-independence of
observations; this characteristic is necessary given that the samples were matched,
and therefore non-independent.
Prior to data analyses, a number of plots were created to examine the data.
For each outcome variable, a separate plot was created for each individual of his or
her performance over time, by age, with a linear Ordinary Least Squares (OLS)
regression line superimposed. These plots allowed for an examination of the linearity
of each individual’s trajectory over time. These plots were also examined to see
whether types of trajectories (e.g., steep decline versus flat or mild decline) varied
between different groups of people (e.g., by cancer, age tested, and time since

108
diagnosis). Next, the empirical performance of all cancer survivors were plotted
together for each outcome variable (one plot per outcome variable for all long-term
cancer survivors, and one plot per outcome variable for all individuals diagnosed
with cancer during the ADRC), as were those of the respective comparison groups,
to examine cognitive performance trajectories by group. Similar group plots were
also created using the OLS regressions that had been calculated for each individual.
These plots were examined for variability in performance, and to see whether clear
between-group trends were evident. Summary statistics were then calculated to
describe the exploratory analyses. Using the OLS regression equations that had been
calculated for each individual, group means for both samples’ initial status and rate
of change were calculated for each outcome variable.
Linear mixed effect models with random effects (Laird & Ware, 1982) were
used to compare trajectories of neuropsychological functioning of cancer survivors
to individuals without a cancer history. The use of random effects models
accommodated the unbalanced data (Laird, 1988) that resulted from individuals
being given different neuropsychological batteries at different points in calendar time
(e.g., due to batteries changing over time, one individual might have received the
FAS at his or her first assessment, and another individual might not have received
the FAS until his or her fourth assessment). To account for data dependencies in the
repeated measures data and the matched pairs, time and pair number were entered as
random effects. Separate models were created for each neuropsychological test. A
time variable, in years, was calculated by subtracting the date of an individual’s first

109
follow-up assessment from each subsequent assessment and dividing by 365. For
individuals who were diagnosed with cancer during the ADRC study, the assessment
immediately preceding the cancer diagnosis was considered the first assessment, and
earlier pre-cancer assessments were excluded from the analyses.
Model-building for each outcome variable was conducted in several steps,
following the guidelines set forth by Singer and Willett (2003). In the first step, an
unconditional means model, a model without any predictors, was fit to examine the
variation in outcomes across people regardless of time or cancer history. In the
second step, an unconditional growth model, a model with only time as a predictor,
was fit to examine the variation in outcomes across people and time, regardless of
cancer history. In the third step, a model with time and cancer as a predictor of
neuropsychological functioning was fit, testing variation in levels (intercepts) of the
outcome variable related to cancer. In the fourth step, a model was fit with time,
cancer, and a time by cancer interaction term as predictors, testing variation in rates
of change (slopes) of the outcome variable related to cancer.
The fit of the models were compared primarily using the deviance statistic,
with reductions in deviance statistics indicating improvements in fit, although the
Akaike Information Criterion (AIC) and Bayesian Information Criterion (BIC) were
also examined as measures of fit. If the model with cancer as a predictor was a
significant improvement over the unconditional growth model, then that model was
considered the best fit. If the model with cancer and the cancer by time interaction
was a significant improvement over the unconditional growth model (significance

110
was determined using the change in deviance statistic, which follows a χ2
distribution), and had a significant (p<.05) group x time interaction, then that model
was considered the best fit. If the models that included cancer as a predictor were not
significant improvements over the unconditional growth model, cancer was not a
significant correlate of the outcome, and the unconditional growth model was
considered the best fit.
With the exception of the unconditional means models (for which a
covariance structure does not need to be specified), the models for each outcome
variable were fit using five different covariance structures, and the resulting deviance
statistics were compared in order to find the covariance structure with the best fit.
The following covariance structures were included in this process: unstructured,
compound symmetric, autoregressive of order 1, toeplitz, and spatial power. The
unstructured covariance structure estimates each correlation using the data provided,
making no assumptions about the data; this covariance structure typically provides
the best fit, but cannot always be fit due to the number of parameters that are
estimated (Moser, 2004). The compound symmetric covariance structure has
homogenous variance and correlations that are constant regardless of how far apart
measurements are (Moser, 2004). Both the autoregressive and toeplitz covariance
structures have homogenous variance and larger correlations for data closer together
in time (Moser, 2004). The spatial power covariance structure is related to the
autoregressive covariance structure, and models larger correlations for data that are
closer in time by using spatial modeling estimation for one dimension (Littell et al.,

111
1996). The suitability of the covariance structure with the best fit for the data was
examined using a correlation matrix of the neuropsychological test across timepoints.
If the correlation matrix confirmed the suitability of the best covariance structure, it
was used throughout the model-building process for the outcome variable.
In three situations, mixed modeling was unable to be used. If the correlation
matrix for an outcome variable did not meet the assumptions of any of the
“structured” covariance structures, if models for an outcome variable were unable to
converge using any of the covariance structures, or if all covariance structures
produced unreliable models (models without positive covariance matrices), the
following strategy was used to examine the data. OLS regressions were calculated
for each individual, and the resulting rate of change (slope) was used as the outcome
variable for a 1-way ANOVA. In these ANOVAs, matched pair was entered as a
within-subjects variable in order to control for data dependencies that resulted from
the matching process.
For the long-term cancer survivors, the relationship between length of cancer
survival and neuropsychological functioning was examined using mixed models that
excluded the matched comparison group. Time between cancer diagnosis and first
neuropsychological assessment was calculated, and then used as a predictor in
separate models for each of the neuropsychological test outcome variables. The same
4-step model-building process described above was used for these analyses. An
unstructured covariance structure was used for these analyses.

112
To examine whether the results obtained through the mixed modeling
replicated more traditional statistical analyses, the primary hypotheses were also
examined using repeated measure ANOVAs. Each neuropsychological test was used
as a separate outcome variable in a repeated measure ANOVA, with time and
matched pair entered as within-subjects variables. To help accommodate the
unbalanced data for these analyses, pairs of cancer survivors and their matches were
only included if both had data for a cognitive test at the first and second assessment
point. The results of the mixed models were also compared to the mean slope and
intercepts of the individual OLS regressions.
Due to the large number of analyses run, a Bonferroni correction was
calculated to adjust the p-value and reduce the Type I error rate. With a Bonferroni
correction for the 11 outcome variables, the p-value was .0045. This adjusted p-value
was used for analyses with the long-term cancer survivors. Because analyses using
the cancer survivors who had been diagnosed with cancer during their participation
in the ADRC study were exploratory, a traditional p-value of .05 was considered
statistically significant. Results from these analyses should be viewed as hypothesis-
generating.
Results
Sample Selection
Identification of individuals who met inclusion and exclusion criteria for the
long-term survivor portion of the study is shown in Figure 2, and was as follows. Of
the 1213 participants in the ADRC’s longitudinal study, 223 were identified by the

113
cancer registry match as having a history of cancer. Thirty-four participants who had
a history of benign, in situ, or borderline invasive cancer, but no history of malignant
cancer, were excluded from the study. Forty-two of the 189 survivors of malignant
cancer were long-term survivors. Twenty-five of those 42 had received more than
one neuropsychological assessment. Two long-term cancer survivors had been
diagnosed with dementia prior to their diagnosis of cancer, and were excluded from
the study. None had been diagnosed with brain cancer or cancer in childhood. In
sum, 23 long-term cancer survivors were identified by the CSP and met all inclusion
and exclusion criteria.
In addition, 186 individuals not found within the CSP had reported a history
of malignant cancer during the ADRC intake interview. Because those individuals
had likely been diagnosed with cancer outside of LA County, or before the CSP
began collecting diagnosis data in 1972, but were still cancer survivors, those who
met eligibility criteria were also included in the study. Eighty-five of these 186
individuals were excluded from the study due to having been diagnosed only with
non-melanoma skin cancer. Eighty-seven of the remaining cancer survivors were
long-term cancer survivors. Fifty-four of the 87 long-term cancer survivors had been
given more than one neuropsychological assessment. Five of those 87 long-term
cancer survivors had been diagnosed with dementia prior to their diagnosis of cancer,
and were excluded from the study. One cancer survivor had been diagnosed with
cancer prior to age 18, and was excluded from the study. In all, 48 of the individuals

114
Figure 2. Sample selection for ADRC long-term cancer survivors

1213 participants within
ADRC

42 were long-term
cancer survivors according
to CSP

87 not in CSP, but self-reported a
cancer history 5 or > years before
neuropsychological testing
25 had 2 or more time points of
neuropsychological assessments

17 were excluded due to
having only 1 time point

2 were excluded due to having
been diagnosed with dementia
prior to cancer diagnosis

23 met all inclusion and
exclusion criteria

54 had 2 or more time points of
neuropsychological
assessments

33 were excluded due to
having only 1 time point

5 were excluded due to
having been diagnosed with
dementia prior to cancer
1 was excluded due to having
been diagnosed with cancer
as an adolescent

48 met all inclusion and exclusion
criteria
114

115
who had reported a cancer history during the ADRC intake, but who were not
identified by the CSP, were eligible for the study.
The first sample thus consisted of 71 pairs of cancer survivors and matched
cancer-free individuals. These individuals participated in the ADRC study in a
manner similar to participants without a cancer history, undergoing
neuropsychological assessment at the pre-determined time intervals described above.
Of these 71 long-term cancer survivors who met eligibility criteria for this study,
none were known to have been treated with chemotherapy. Because of this, analyses
examining long-term cancer survivors treated with chemotherapy could not be
conducted.
Identification of individuals who met inclusion criteria for the second portion
of the study, which compared the cognitive functioning of cancer patients before and
after cancer diagnosis to that of individuals who had never had cancer, was as
follows. Thirty-one individuals were identified by the CSP as having been diagnosed
with cancer during the time period that they participated in the ADRC longitudinal
study, and had neuropsychological data from before and after their cancer diagnosis.
Of these, 11 cancer survivors identified through the CSP had also been diagnosed
with cancer previously, and were excluded from the sample in order to avoid the
potentially additive or multiplicative effects that having been diagnosed with and
treated for multiple cancers could have on cognitive functioning. None of the
remaining 20 participants had a diagnosis of dementia prior to their diagnosis of
cancer. Each of these 20 individuals was matched to a cancer-free individual. The

116
second sample of cancer survivors thus consisted of 20 participants who were
diagnosed with their first cancer during the course of their participation in the
longitudinal study, and 20 matched individuals who had never had cancer.
Initial Data Examination
Prior to hypothesis testing, the data were examined for outliers. No data were
identified as outliers. Data were then examined for completeness. Data on
neuropsychological tests were incomplete for two reasons. Firstly, the battery of
neuropsychological tests changed over time, so each test was used during a span of
only a certain number of years. Individuals did not have data on those tests,
therefore, which were not included in the battery on the dates that they were tested.
Data which were not available for this reason would not be considered missing by
traditional standards, because the individuals never had the opportunity to provide
those data. Secondly, individuals were sometimes missing data for a specific test
even when that test was cited by the ADRC as within the battery for the date of an
individual’s testing. These data were considered missing, because individuals should
have had the opportunity to provide data, but did not. Table 17 provides a description
of when the tests were given, the percentage of data that were not available (due to
the test not being in the battery), and the percentage of data that were missing by
traditional standards (where the test was in the battery, but there were no data for an
individual) for the first and second time points for the sample of long-term cancer
survivors and their matched comparison group.

117
Table 17. Date range tests were included in the ADRC battery, and % missing and not administered data for long-term cancer survivors

Test Dates administered
% Missing
T1
% Missing
T2
% Not Administered
T1
% Not Administered
T2
Mini-Mental State Examination 11/8/1982 - 3/27/2004 15.49% 18.31% 0.00% 0%
Animals 10/5/1987 - 8/11/1998 7.04% 10.00% 7.04% 12.68%
FAS 10/15/1984 - 8/22/2001 15.49% 21.13% 1.41% 1.41%
Boston Naming Test 10/15/1984 - 5/28/2001 19.72% 25.35% 4.23% 2.82%
Figure Copying 10/5/1987 - 5/28/2001 8.45% 2.82% 33.80% 33.80%
Word List Learning 10/5/1987 - 5/28/2001 9.86% 4.23% 35.21% 35.21%
Word List Recall 10/5/1987 - 5/28/2001 9.86% 2.82% 35.21% 35.21%
Word List Recognition 10/5/1987 - 5/28/2001 11.27% 8.45% 35.21% 35.21%
Digit Span Forward score 10/17/1984 - 3/24/2004 11.27% 8.45% 1.41% 0.00%
Digit Span Backward score 10/17/1984 - 10/18/2003 12.68% 7.04% 1.41% 0.00%
Geriatric Depression Scale 10/15/1984 - 3/24/2004 15.49% 23.94% 0.00% 0.00%

Note: T1 = 1st neuropsychological assessment; T2 = 2nd neuropsychological assessment. Data were considered “Missing” if they were
missing at a time during which the neuropsychological test was in the battery. Data were considered “Not Administered” if they were
missing at a time during which the neuropsychological test was not in the battery, and so the test had not been administered.

117

118
Missing cognitive data (data that should have been available because the test
was in the battery at the time of the assessment, but were not) were examined to
determine whether they were randomly missing. The comparison group for long-
term cancer survivors was missing significantly more data than the long-term cancer
survivors on the Word List Learning (χ2(1)=7.35, p<0.05), Word List Delayed Recall
(χ2(1)=11.00, p<0.05), and Word List Recognition tests (χ2(1)=4.45, p<0.05).
Multiple imputation was considered to impute data that were missing. If data were
missing completely at random (MCAR) or missing at random (MAR), it would not
be necessary to impute missing data. Because it is plausible that the probability of
data being missing was related to cognitive status—that is, cognitively impaired
individuals may have received missing scores if they could not attempt a test, or
testing could have been concluded early with individuals who were cognitively
impaired—data could not be assumed to be MCAR. However, data could be
assumed to be MAR, since the probability of data being missing is likely not related
to cognitive status after controlling for prior time points of cognitive functioning
(which would likely show any cognitive impairment that could increase chances of
having missing data later). In addition, missing data are not as problematic for mixed
models as they are for repeated measures ANOVA; individuals with missing data are
deleted in ANOVAs, whereas mixed modeling analyzes all available data, although
if data are not MAR or MCAR the models will produce biased estimates
(Gueorguieva & Krystal, 2004). For these reasons, multiple imputation was not used.

119
Data were then examined visually and through formal tests for non-normality
and heteroscedasticity. For the sample of long-term cancer survivors and their
comparison group, all of the neuropsychological variables were significantly non-
normal by the Shapiro-Wilk statistical standard. These variables also showed
heteroscedasticity. Despite the statistical significance of the non-normality, however,
several of the distributions closely resembled normal distributions (e.g., for Digit
Span Forward, skew=.03, kurtosis=.24; for Digit Span Backward, skew=.03,
kurtosis=-.07; for Animals, skew=-.08, kurtosis=-.62; for FAS, skew=.03, kurtosis=-
.45). Lognormal transformations were computed for the neuropsychological
variables, and the skew and kurtosis of the transformed variables were compared to
those of the original variables. Only the distribution of the GDS was improved by a
lognormal transformation (original GDS skew=1.41, kurtosis=2.17; log-transformed
skew=-.30, kurtosis=-.59). As a result, the log-transformed GDS variable and the
other original, non-transformed neuropsychological variables were used in the mixed
modeling analyses.
Similar results were obtained when examining the distributions of the cancer
survivors who had been diagnosed with cancer during their participation in the
ADRC study, and their comparison group. For this sample, only Animals and FAS
were normally distributed and had homoscedasticity. Lognormal transformations
were computed for the neuropsychological variables that were significantly non-
normal, and the skew and kurtosis of the transformed variables were compared to
those of the original variables. Again, only the distribution of the GDS was improved

120
by lognormal transformation (original GDS skew=1.43, kurtosis=2.09; log-
transformed skew=-.28, kurtosis=-.02), and so the log-transformed GDS was used in
the mixed modeling analyses.
Descriptives and matching check
The 5 most common cancers among long-term cancer survivors in this study
were breast (32%), uterine (19%), prostate (14%), colorectal (6%), and melanoma
(4%), with average time from diagnosis being 19.4 years (min - max: 5.0-45.0).
Seventeen percent of the long-term cancer survivors were diagnosed with localized
cancer, 4% had regional cancer with extension, 3% had regional cancer with nodal
involvement, and for 76% of the long-term cancer survivors, the summary staging
was unknown. None of the long-term cancer survivors was known to have been
treated with chemotherapy, although chemotherapy was recommended to 13% of the
long-term survivors. None were known to be treated with hormones, although
hormone therapy was recommended to 13% of the long-term cancer survivors, nor
were they known to be treated with immunotherapy or central nervous system
radiation. Table 18 compared the demographic characteristics of long-term cancer
survivors and their comparison groups in order to check the precision of the match.
There were no significant differences between the groups on any of the demographic
variables, suggesting that the match was successful.
The empirical trajectory of each person within the sample was plotted for
each neuropsychological test. Visual inspection of these plots revealed a range of
trajectories, with all tests showing good variation in trajectories and scores. Separate

121

Table 18. Check of match precision for ADRC long-term cancer survivors and cancer history-free comparison group

Age at First Cognitive Testing
Group N Mean (years) SD t df p-value
Long-term cancer survivors 71 75.86 8.83 0.82 70 0.42
Cancer-free individuals 71 75.69 8.57

Gender
Group N Male Female χ2 df p-value
Long-term cancer survivors 71 23.94% 76.06% 2.00 1 0.16
Cancer-free individuals 71 26.76% 73.24%

Education
Group N Mean (years) SD t df p-value
Long-term cancer survivors 65 13.74 3.01 1.17 64 0.25
Cancer-free individuals 65 13.57 3.01

Ethnicity
Group N non-Hispanic Black Hispanic non-Hispanic White χ2 df p-value
Long-term cancer survivors 70 4.29% 8.57% 87.14% 0.00 1 >0.99
Cancer-free individuals 71 4.23% 9.86% 85.92%

Note: p-values were calculated using paired t-tests for age at first cognitive testing and education, McNemar's chi square for
gender, and conditional logistic regression for ethnicity

121

122
within-person ordinary least square regressions were fitted for each long-term cancer
survivor and cancer history-free comparison individual, to examine the linear
trajectory of cognitive decline for each person on each neuropsychological test.
Table 19 shows the mean initial score and slope for each neuropsychological test,
using these individual regressions.
The 3 most common cancers among long-term cancer survivors for the
sample of cancer survivors first diagnosed with cancer during participation in the
ADRC study were colorectal (25%), prostate (25%), and cancer not otherwise
specified (NOS, 20%); there were one each of the following cancers: bladder, breast,
cervical, liver, lung, and uterine. Sixty-five percent of the cancer survivors in this
sample had localized cancer, 20% had distal involvement, 5% had regional cancer
with extension, 5% had regional cancer with extension and node involvement, and
for 5% the summary staging was unknown. None of the cancer survivors was known
to have been treated with chemotherapy, although it was recommended as a
treatment for 60% of the sample. None of the survivors was known to be treated with
central nervous system radiation or immunotherapy. One person (5% of the sample)
was treated with hormone therapy, and an additional 60% were recommended
hormone therapy, but did not receive it as an initial treatment. Table 20 compared the
demographic characteristics of cancer survivors diagnosed with cancer during
participation in the ADRC study and their comparison group in order to check the
precision of the match. There were no significant differences between the groups on
any of the demographic variables, suggesting that the match was successful. A paired

123
Table 19. Descriptive statistics for individual growth parameters for separate within-person
regressions as a function of years since 1st test - long-term cancer survivors and comparison group

Long-term Cancer Survivors Cancer History-Free Individuals
Min - Initial Rate of Min - Initial Rate of
N max status change N max status change
MMSE 66 0 - 30 70 0 - 30
Mean 23.82 -1.68 22.65 -1.61
Standard deviation 7.66 3.14 8.11 3.02

Animals 61 0 - 36 57 0 - 29
Mean 15.38 -1.01 14.05 -0.93
Standard deviation 8.05 2.23 6.95 2.06

FAS 60 1 - 80 57 0 - 78
Mean 35.10 -1.84 33.13 -1.69
Standard deviation 17.26 4.19 19.89 5.04

Boston Naming Test 64 0 - 60 57 2 - 60
Mean 41.09 -1.59 41.48 -2.22
Standard deviation 18.75 5.90 17.52 6.87

WL Learning 56 0 - 30 45 0 - 30
Mean 17.14 -0.80 15.91 -0.98
Standard deviation 8.19 3.26 8.15 1.68

WL Delayed Recall 56 0 - 10 45 0 - 10
Mean 5.18 -0.31 4.52 -0.29
Standard deviation 3.51 0.68 3.74 1.22

WL Recognition 55 8 - 20 45 0 - 20
Mean 18.13 -0.41 17.05 -0.73
Standard deviation 3.28 1.41 3.65 1.70

Figure Copying 58 0 - 11 52 0 - 11
Mean 8.98 -0.44 9.29 -0.40
Standard deviation 2.73 2.56 1.81 0.62

Digit Span F 63 0 - 14 59 0 - 14
Mean 6.84 -0.44 6.86 -0.26
Standard deviation 2.48 1.42 2.32 0.88

Digit Span B 63 0 - 13 59 0 - 13
Mean 5.80 -0.59 5.37 -0.30
Standard deviation 2.74 1.32 2.93 1.40

GDS 63 0 - 26 66 0 - 23
Mean 5.28 0.09 6.25 -0.25
Standard deviation 4.53 1.93 5.45 3.96

124

Table 20. Check of match precision for cancer survivors diagnosed with cancer during their
participation in the ADRC and cancer history-free comparison group

Age at First Cognitive Testing
Group N Mean SD t df p-value
Long-term cancer survivors 20 77 10.53 -0.45 19 0.66
Cancer-free individuals 20 77.2 10.01

Gender
Group N Male Female χ2 df p-value
Long-term cancer survivors 20 70.00% 30.00% 0.00 1 >0.99
Cancer-free individuals 20 70.00% 30.00%

Education
Group N Mean SD t df p-value
Long-term cancer survivors 18 13.61 3.5 -0.82 17 0.42
Cancer-free individuals 18 13.78 3.66

Ethnicity
Group N Hispanic non-Hispanic White χ2 df p-value
Long-term cancer survivors 20 20.00% 80.00% 0.00 1 >.99
Cancer-free individuals 20 20.00% 80.00%

Note: p-values were calculated using paired t-tests for age at first cognitive testing and education,
McNemar's chi square for gender, and conditional logistic regression for ethnicity

125
t-test showed that cancer survivors and their comparison group did not differ in the
length of time between their first and second assessments, t(19)=0.65, p=.52.
The empirical trajectory of each person who was diagnosed with cancer
during the ADRC, and their matches, was plotted for each neuropsychological test.
Again, visual inspection of these plots revealed a range of trajectories, with all tests
showing good variation in trajectories and scores. Separate within-person ordinary
least square regressions were fitted for each person diagnosed with cancer during
participation in the ADRC study, and each cancer history-free individual in the
comparison group, to examine the linear trajectory of cognitive decline for each
person on each neuropsychological test. Table 21 shows the mean initial score and
slope for each neuropsychological test, using these individual regressions.
Hypothesis 3 and Question 1: Trajectory of Cognitive Function in Long-term
Cancer Survivors
For each of the neuropsychological tests, the full sample of 142 individuals
(71 matched pairs) was included in analyses. In general, an unstructured covariance
structure provided the best model fit, because it modeled the covariance structure to
the data without making assumptions. The models for 7 of the 11 outcome
variables—MMSE, FAS, Boston Naming Test, Word List Delayed Recall, Figure
Copying, Digit Span Backwards, and Geriatric Depression Scale—were successfully
fit using the unstructured covariance matrix. However, the unstructured covariance
matrix was not always tenable; some outcome variables lacked sufficient data to

126
Table 21. Descriptive statistics for individual growth parameters from separate within-person
regressions as a function of years since 1st test - people diagnosed with cancer during study and
comparison group)
Cancer Survivors Cancer History-Free Individuals

N
Min -
max
Initial
status
Rate of
change N
Min -
max
Initial
status
Rate of
change
MMSE 19 13 - 30 19 14 - 30
Mean 28.71 -0.63 26.83 0.00
Standard deviation 1.40 1.05 3.68 1.48

Animals 16 3 - 29 17 4 - 28
Mean 17.98 -0.69 18.94 -1.22
Standard deviation 5.41 0.87 3.35 1.30

FAS 16 10 - 69 17 14 - 68
Mean 43.59 -0.90 39.05 0.28
Standard deviation 13.04 3.18 9.95 2.28

Boston Naming Test 16 31 - 60 17 43 - 60
Mean 54.38 -0.77 53.83 -0.45
Standard deviation 7.31 1.41 4.30 1.62

WL Learning 14 5 - 29 14 9 - 30
Mean 22.29 -0.98 20.34 -0.52
Standard deviation 4.31 1.03 4.32 2.25

WL Delayed Recall 14 0 - 10 14 0 - 10
Mean 7.37 -0.51 6.92 -0.31
Standard deviation 1.96 0.85 2.93 0.84

WL Recognition 14 10 - 20 14 13- 20
Mean 19.96 -0.32 19.21 -0.21
Standard deviation 0.62 0.63 1.34 0.82

Figure Copying 18 6 - 11 17 4 - 11
Mean 9.82 -0.22 8.89 0.16
Standard deviation 1.49 0.45 2.22 0.65

Digit Span F 16 3 - 14 16 4 - 14
Mean 8.40 -0.13 8.74 -0.13
Standard deviation 1.99 0.29 2.69 0.71

Digit Span B 16 0 - 14 16 0 - 13
Mean 8.18 -0.09 7.28 -0.16
Standard deviation 2.58 0.47 2.72 0.35

GDS 19 1 - 25 20 0 - 24
Mean 7.52 0.56 5.53 1.32
Standard deviation 6.44 0.90 1.32 4.50

127
Recognition—which were fit using the covariance structure with the next best fit, the
autoregressive order 1. Two additional outcome variables—Animals and Digit Span
Forward—could not be fit using the unstructured covariance matrix when the group-
by-time interaction term was included as a predictor. However, examination of the
correlation matrices for these variables over time showed violations of the
assumptions of the other covariance structures attempted. These violations included
heterogeneity of variance, and correlations that were not larger among closer time
points than among more distant time points.
Because other covariance structures could not adequately model the data for
Animals and Digit Span Forward, an unstructured covariance structure was used in
the first 3 steps of the model-building process to examine the unconditional means
model, the unconditional growth model, and the model with cancer as a predictor.
Then, to test a group-by-time interaction for these two outcome variables, I
examined whether cancer history predicted slope from individual OLS regressions,
using a 1-way ANOVA. This analysis found no significant group-by-time interaction
for Animals (F(1, 52)=0.00, p=.97) or Digit Span Forward (F(1, 54)=.55, p=.46).
For the mixed models, the mean (SD) number of assessment measures per subjects
included for each test was as follows: MMSE: 5.64(4.39), Animals: 5.58(5.22), FAS:
4.95(4.75), Boston Naming Test: 4.41(3.86), Word List Learning: 4.15(4.21), Word
List Delayed Recall: 4.16(4.24), Word List Recognition: 4.02(4.16), Figure Copying:
4.35(4.21), Digit Span Forward: 5.20(4.14), Digit Span Backward: 5.52(4.47), and
Geriatric Depression Scale: 5.13(4.67). Tables 22 - 32 show the results of the mixed

128
models for long-term cancer survivors compared to matched individuals who had
never had cancer. Cancer history was not a significant predictor of any of the
neuropsychological tests. Time was a significant predictor of all of the measures
except for the Geriatric Depression Scale, with individuals’ scores on the all of the
measures except for the Geriatric Depression Scale decreasing over time. There were
no significant group-by-time interactions, indicating that long-term cancer survivors
and individuals who had never had cancer did not differ significantly in trajectory of
cognitive functioning over time. Although Model 4 in Table 24 appears to suggest a
significant (p=.05) group-by-time interaction for FAS, examination of the deviance
statistics shows that the model that included this interaction was not a significantly
better fit for the data than the unconditional means model.
To provide a comparison to more classical statistical approaches, the analyses
were also run using repeated measures ANOVA, using the first two assessment time
points. Table 33 displays the results of those analyses. Using the Bonferroni adjusted
p-value, none of the group or time effects was statistically significant using
ANOVAs. There was a non-significant trend for the long-term cancer survivors to
perform better on the Word List Recognition test, however (F(1,23)=5.33, p=.03),
and non-significant trends for the Mini Mental State Examination and Figure
Copying scores to decrease over time (F(1,45)=5.68, p=.02; & F(1,27)=5.30, p=.03,
respectively). No group-by-time interaction was statistically significant.

129

Table 22. Estimates of fixed effects, variance components, and fit indices for individual growth
models predicting Mini Mental State Examination (MMSE) scores for long-term cancer survivors
and comparison group
Model 1 Model 2 Model 3 Model 4

Fixed Effects

Main effects
Intercept 21.36(0.70)*** 23.49(0.61)*** 23.30(0.78) *** 23.16(0.85)***
Time -1.71(0.20)*** -1.71(0.20)*** -1.78(0.28)***
Cancer 0.38(1.01) 0.65(1.22)

Interactions
Cancer x Time 0.16(0.40)

Variance
Components

Level 1
Within-person 12.02(0.66)*** 4.26(0.28)*** 4.26(0.28)*** 4.27(0.28)***
Level 2
In initial status 63.09(8.20)*** 92.77(17.40) *** 91.48(17.46) *** 91.13(17.40) ***
In rate of change 4.18(0.85)*** 4.19(0.86)*** 4.17(0.85)***
In matched pair 0.00(.) 0.00(.) 0.00(.)

Covariance (initial
status, time) 13.12(3.66)*** 13.12(3.63) *** 13.05(3.64)***

Covariance (initial
status, pair) -0.50(0.16)** -0.49(0.16)** -0.48(0.16)**

Covariance (time,
pair) -0.07(0.06) -0.07(0.06) 0.00(.)

Fit Indices

Goodness of Fit
Deviance 4458 4269.1 4269.1 4260.7
AIC 4466 4285.1 4287.1 4280.7
BIC 4477.8 4308.7 4313.7 4310.3
Model Comparison
Δ deviance Model 1 188.9*** 188.9*** 439.8***
Δ deviance Model 2 0 0.3

Note: Model 1 is an unconditional means model with no predictors of MMSE scores, Model 2 is an
unconditional growth model with time (in years since 1st testing) predicting MMSE scores, Model 3
includes both time and cancer as predictors of MMSE scores, and Model 4 includes time, cancer,
and a time x cancer interaction as predictors of MMSE scores. Smaller Deviance, AIC, and BIC
scores indicate better model fit. * p < .05; ** p<.01; *** p<.001.

130

Table 23. Estimates of fixed effects, variance components, and fit indices for individual growth
models predicting Animals scores for long-term cancer survivors and comparison group

Model 1 Model 2 Model 3 Model 4 §

Fixed Effects

Main effects
Intercept 13.06(0.65)*** 14.17(0.66)*** 14.05(0.94)***
Time -0.77(0.09)***
Cancer 0.25(1.29)

Interactions
Cancer x Time

Variance Components

Level 1
Within-person 14.09(0.79)*** 9.13(0.55)*** 9.13(0.55)***
Level 2
In initial status 45.37(10.78)*** 42.93(14.04)** 43.56(14.70)**
In rate of change 0.44(0.13)*** 0.44(0.13)***
In matched pair 0.00(.) 0.00(.)

Covariance (initial
status, time) 4.46(1.22)*** 4.41(1.26)***

Covariance (initial
status, pair) 0.12(0.18) 0.11(0.19)

Covariance (time,
pair) -0.08(0.03)* -0.08(0.03)*

Fit Indices

Goodness of Fit
Deviance 4458 4269.1 4269.1
AIC 4466 4285.1 4287.1
BIC 4477.8 4308.7 4313.7
Model Comparison
Δ deviance Model 1 188.9*** 188.9***
Δ deviance Model 2 0

Note: Model 1 is an unconditional means model with no predictors of Animals scores, Model 2 is
an unconditional growth model with time (in years since 1st testing) predicting Animals scores,
Model 3 includes both time and cancer as predictors of Animals scores, and Model 4 includes
time, cancer, and a time x cancer interaction as predictors of Animals scores. Smaller Deviance,
AIC, and BIC scores indicate better model fit. * p < .05; ** p<.01; *** p<.001. § Model unable to
be fit

131

Table 24. Estimates of fixed effects, variance components, and fit indices for individual growth
models predicting FAS scores for long-term cancer survivors and comparison group

Model 1 Model 2 Model 3 Model 4

Fixed Effects

Main effects
Intercept 31.72(1.59)*** 32.85(1.57)*** 31.60(2.26)*** 31.29(2.25) ***
Time -1.08(0.21)*** -1.06(0.21)*** -1.51(0.31)***
Cancer 2.39(3.13) 2.94(3.13)

Interactions
Cancer x Time 0.81(0.41)*

Variance Components

Level 1
Within-person 51.57(3.02)*** 35.87(2.32)*** 35.90(2.32)*** 35.90(2.32) ***
Level 2
In initial status 279.48(38.75)***
400.11(114.82)
***
404.60(114.92)
***
401.05(113.73)
***
In rate of change 2.55(0.83)** 2.51(0.82)** 2.37(0.78)**
In matched pair 0.00(.) 0.00(.) 0.00(.)

Covariance (initial
status, time) 27.12(9.47)** 26.35(9.71)** 26.36(9.05)**

Covariance (initial
status, pair) -1.32(1.28) -1.40(1.27) -1.36(1.26)

Covariance (time,
pair) -0.54(0.20)** -0.53(0.20)** -0.54(0.19)**

Fit Indices

Goodness of Fit
Deviance 5145.7 5040 5039.5 5035.6
AIC 5151.7 5056 5057.5 5055.6
BIC 5160.6 5079.7 5084.1 5085.1
Model Comparison
Δ deviance Model 1 105.7*** 106.2*** 110.1***
Δ deviance Model 2 0.5 4.4

Note: Model 1 is an unconditional means model with no predictors of FAS scores, Model 2 is an
unconditional growth model with time (in years since 1st testing) predicting FAS scores, Model 3
includes both time and cancer as predictors of FAS scores, and Model 4 includes time, cancer, and a
time x cancer interaction as predictors of FAS scores. Smaller Deviance, AIC, and BIC scores
indicate better model fit. * p < .05; ** p<.01; *** p<.001.

132
Table 25. Estimates of fixed effects, variance components, and fit indices for individual growth
models predicting Boston Naming Test (BNT) scores for long-term cancer survivors and
comparison group
Model 1 Model 2 Model 3 Model 4

Fixed Effects

Main effects
Intercept 39.26(1.66)***
41.87(1.60)
***
42.69(2.22)
*** 41.54(2.33)***
Time
-2.01(0.33)
***
-2.02(0.33)
*** -2.56(0.45)***
Cancer -1.72(2.90) 0.51(3.20)

Interactions
Cancer x Time 1.13(0.61)

Variance Components

Level 1
Within-person 33.04(2.08)***
15.06(1.31)
***
15.18(1.36)
*** 15.67(1.45)***
Level 2
In initial status
324.23(43.30)
***
392.83
(98.36)***
381.83
(100.11)***
391.76(101.48)
***
In rate of change 8.90(2.87)** 8.63(2.86)** 7.08(2.54)**
In matched pair 0.00(.) 0.00(.) 0.00(.)

Covariance (initial
status, time) 22.57(15.32) 23.85(15.55) 29.52(16.29)

Covariance (initial
status, pair) -1.25(1.19) -1.09(1.24) -1.28(1.23)

Covariance (time,
pair) 0.21(0.39) 0.20(0.40) -0.00(0.44)

Fit Indices

Goodness of Fit
Deviance 4414.9 4206.5 4206.2 4203.2
AIC 4420.9 4222.5 4224.2 4223.2
BIC 4429.8 4246.1 4250.8 4252.8
Model Comparison
Δ deviance Model 1 208.4*** 208.7*** 211.7***
Δ deviance Model 2 0.3 3.3

Note: Model 1 is an unconditional means model with no predictors of BNT scores, Model 2 is an
unconditional growth model with time (in years since 1st testing) predicting BNT scores, Model
3 includes both time and cancer as predictors of BNT scores, and Model 4 includes time, cancer,
and a time x cancer interaction as predictors of BNT scores. Smaller Deviance, AIC, and BIC
scores indicate better model fit. * p < .05; ** p<.01; *** p<.001

133
Table 26. Estimates of fixed effects, variance components, and fit indices for individual growth
models predicting Word List (WL) Learning scores for long-term cancer survivors and comparison
group

Model 1 Model 2 § Model 3 § Model 4 §

Fixed Effects

Main effects
Intercept 14.88(0.79)*** 20.84(0.91)*** 19.59(1.37)*** 19.58(1.37)***
Time -0.70(0.07)*** -0.69(0.07)*** -0.73(0.12)***
Cancer 2.23(1.82) 2.25(1.82)

Interactions
Cancer x Time 0.06(0.15)

Variance Components

Level 1
Within-person 10.94(0.70)*** 6.19(0.44)*** 6.18(0.44)*** 6.18(0.44)***
Level 2
In initial status 60.54(8.94)***
In rate of change
In matched pair

Covariance (initial
status, time)

Covariance (initial
status, pair)

Covariance (time,
pair)

Fit Indices

Goodness of Fit
Deviance 3418.3 3294.1 3292.6 3292.4
AIC 3424.3 3304.1 3304.6 3306.6
BIC 3433.1 3318.9 3322.3 3327.1
Model Comparison
Δ deviance Model 1 124.2*** 125.7*** 125.9***
Δ deviance Model 2 1.5 1.7

Note: Model 1 is an unconditional means model with no predictors of WL Learning scores, Model 2
is an unconditional growth model with time (in years since 1st testing) predicting WL Learning
scores, Model 3 includes both time and cancer as predictors of WL Learning scores, and Model 4
includes time, cancer, and a time x cancer interaction as predictors of WL Learning scores. Smaller
Deviance, AIC, and BIC scores indicate better model fit. * p < .05; ** p<.01; *** p<.001; § Models
fit using autoregressive order 1. Using this covariance structure, only 1 variance component is
estimated.

134

Table 27. Estimates of fixed effects, variance components, and fit indices for individual growth
models predicting Word List Recall scores for long-term cancer survivors and comparison group

Model 1 Model 2 Model 3 Model 4

Fixed Effects

Main effects 4.31(0.33)*** 4.69(0.36)*** 4.34(0.52)*** 4.29(0.53)***
Intercept -0.22(0.04)*** -0.22(0.04)*** -0.19(0.06)**
Time 0.58(0.70) 0.67(0.71)
Cancer

Interactions
Cancer x Time -0.05(0.08)

Variance Components

Level 1
Within-person 1.94(0.12)*** 1.25(0.08)*** 1.25(0.08)*** 1.25(0.08)***
Level 2
In initial status 10.67(1.57)*** 15.99(5.91)** 16.59(6.11)** 16.67(6.14)**
In rate of change 0.05(0.02)*** 0.05(0.02)*** 0.05(0.02)***
In matched pair 0.00(.) 0.00(.) 0.00(.)

Covariance (initial
status, time) 0.24(0.49) 0.21(0.50) 0.21(0.51)

Covariance (initial
status, pair) -0.04(0.07) -0.05(0.07) -0.05(0.07)

Covariance (time,
pair) -0.01(0.01) -0.01(0.01) -0.01(0.01)

Fit Indices

Goodness of Fit
Deviance 2399.7 2260.2 2259.5 2259.2
AIC 2405.7 2276.2 2277.5 2279.2
BIC 2414.6 2299.8 2304.1 2308.7
Model Comparison
Δ deviance Model 1 139.5*** 140.2*** 140.5***
Δ deviance Model 2 0.7 1.0

Note: Model 1 is an unconditional means model with no predictors of WL Recall scores, Model 2 is
an unconditional growth model with time (in years since 1st testing) predicting WL Recall scores,
Model 3 includes both time and cancer as predictors of WL Recall scores, and Model 4 includes
time, cancer, and a time x cancer interaction as predictors of WL Recall scores. Smaller Deviance,
AIC, and BIC scores indicate better model fit. * p < .05; ** p<.01; *** p<.001

135
Table 28. Estimates of fixed effects, variance components, and fit indices for individual growth
models predicting Word List Recognition scores for long-term cancer survivors and comparison
group

Model 1 Model 2 § Model 3 § Model 4 §

Fixed Effects

Main effects
Intercept 16.99(0.35)*** 18.87(0.34)*** 18.41(0.51)*** 18.47(0.52)***
Time -0.17(0.03)*** -0.16(0.03)*** -0.20(0.06)**
Cancer 0.78(0.67) 0.70(0.69)

Interactions
Cancer x Time 0.05(0.07)

Variance Components

Level 1
Within-person 3.26(0.12)*** 2.71(0.19)*** 2.72(0.19)*** 2.71(0.19)***
Level 2
In initial status 11.36(1.74)***
In rate of change
In matched pair

Covariance (initial
status, time)

Covariance (initial
status, pair)

Covariance (time,
pair)

Fit Indices

Goodness of Fit
Deviance 2577.7 2554 2552.7 2552.2
AIC 2583.7 2564 2564.7 2566.2
BIC 2592.5 2578.8 2582.4 2586.9
Model Comparison
Δ deviance Model 1 23.7*** 25.0*** 25.5***
Δ deviance Model 2 1.3 1.8

Note: Model 1 is an unconditional means model with no predictors of WL Recognition scores,
Model 2 is an unconditional growth model with time (in years since 1st testing) predicting WL
Recognition scores, Model 3 includes both time and cancer as predictors of WL Recognition scores,
and Model 4 includes time, cancer, and a time x cancer interaction as predictors of WL Recognition
scores. Smaller Deviance, AIC, and BIC scores indicate better model fit. * p < .05; ** p<.01;
*** p<.001; § Models fit using autoregressive order 1. Using this covariance structure, only 1
variance component is estimated.

136

Table 29. Estimates of fixed effects, variance components, and fit indices for individual growth
models predicting Figure Copying scores for long-term cancer survivors and comparison group

Model 1 Model 2 Model 3 Model 4

Fixed Effects

Main effects
Intercept 8.35(0.21)*** 8.90(0.21)*** 9.04(0.30)*** 9.05(0.30)***
Time -0.34(0.04)*** -0.23(0.04)*** -0.33(0.07)***
Cancer -0.28(0.42) -0.29(0.42)

Interactions
Cancer x Time -0.02(0.09)

Variance Components

Level 1
Within-person 2.26(0.14)*** 1.34(0.09)*** 1.34(0.09)*** 1.34(0.09)***
Level 2
In initial status 4.41(0.70)*** 12.86(8.42) 11.35(8.41) 11.36(8.35)
In rate of change 0.12(0.04)** 0.12(0.04)** 0.12(0.04)**
In matched pair 0.00(0.00) 0.00(0.00) 0.00(0.00)

Covariance (initial
status, time) 1.13(0.47)* 1.03(0.48)* 1.05(0.49)*

Covariance (initial
status, pair) -0.14(0.18) -0.11(0.18) -0.11(0.18)

Covariance (time,
pair) -0.02(0.01) -0.02(0.01) -0.02(0.01)

Fit Indices

Goodness of Fit
Deviance 2509.2 2338.3 2337.9 2337.9
AIC 2515.2 2356.3 2357.9 2359.9
BIC 2524 2382.9 2387.5 2392.4
Model Comparison
Δ deviance Model 1 170.9*** 171.3*** 171.3***
Δ deviance Model 2 0.4 0.4

Note: Model 1 is an unconditional means model with no predictors of Figure Copying scores,
Model 2 is an unconditional growth model with time (in years since 1st testing) predicting Figure
Copying scores, Model 3 includes both time and cancer as predictors of Figure Copying scores,
and Model 4 includes time, cancer, and a time x cancer interaction as predictors of Figure
Copying scores. Smaller Deviance, AIC, and BIC scores indicate better model fit. * p < .05; **
p<.01; *** p<.001.

137

Table 30. Estimates of fixed effects, variance components, and fit indices for individual growth
models predicting Digit Span Forward scores for long-term cancer survivors and comparison
group
Model 1 Model 2 Model 3 Model 4 §

Fixed Effects

Main effects
Intercept 6.60(0.22)*** 6.92(0.21)*** 7.04(0.30)**
Time -0.08(0.02)*** -0.08(0.02)***
Cancer -0.23(0.42)

Interactions
Cancer x Time

Variance Components

Level 1
Within-person 1.90(0.11)*** 1.84(0.11)*** 1.84(0.11)***
Level 2
In initial status 5.19(0.73)*** 8.52(2.09)*** 8.64(2.12)***
In rate of change 0.00(0.00) 0.00(0.00)
In matched pair 0.00(.) 0.00(.)

Covariance (initial
status, time) 0.17(0.09)* 0.18(0.09)*

Covariance (initial
status, pair) -0.04(0.02)* -0.05(0.02)*

Covariance (time,
pair) -0.00(0.00) -0.00(0.00)

Fit Indices

Goodness of Fit
Deviance 2970.6 2950.8 2950.5
AIC 2976.6 2966.8 2968.5
BIC 2985.5 2990.4 2995.1
Model Comparison
Δ deviance Model 1 19.8*** 20.1***
Δ deviance Model 2 0.3

Note: Model 1 is an unconditional means model with no predictors of Digit Span Forward scores,
Model 2 is an unconditional growth model with time (in years since 1st testing) predicting Digit
Span Forward scores, Model 3 includes both time and cancer as predictors of Digit Span Forward
scores, and Model 4 includes time, cancer, and a time x cancer interaction as predictors of Digit
Span Forward scores. Smaller Deviance, AIC, and BIC scores indicate better model fit.
* p < .05; ** p<.01; *** p<.001; § Model unable to be fit

138
Table 31. Estimates of fixed effects, variance components, and fit indices for individual growth
models predicting Digit Span Backward scores for long-term cancer survivors and comparison
group

Model 1 Model 2 Model 3 Model 4

Fixed Effects

Main effects
Intercept 5.02(0.24)*** 5.39(0.23)*** 5.33(0.34)*** 5.32(0.34)***
Time -0.26(0.04)*** -0.26(0.04)*** -0.27(0.06)***
Cancer 0.10(0.47) 0.12(0.47)

Interactions
Cancer x Time 0.03(0.08)

Variance Components

Level 1
Within-person 2.32(0.13)*** 1.77(0.11)*** 1.77(0.11)*** 1.77(0.11)***
Level 2
In initial status 6.38(0.09)*** 9.47(2.52)*** 9.37(2.54)*** 9.30(2.53)***
In rate of change 0.08(0.03)** 0.08(0.03)** 0.08(0.03)**
In matched pair 0.00(.) 0.00(.) 0.00(.)

Covariance (initial
status, time) 0.80(0.24)** 0.79(0.24)** 0.78(0.24)**

Covariance (initial
status, pair) -0.04(0.03) -0.04(0.03) -0.03(0.03)

Covariance (time,
pair) -0.02(0.01)** -0.02(0.01)** -0.02(0.01)**

Fit Indices

Goodness of Fit
Deviance 3061.1 2977.6 2977.6 2977.4
AIC 3067.1 2993.6 2995.6 2997.4
BIC 3076 3017.3 3022.2 3026.9
Model Comparison
Δ deviance Model 1 83.5** 83.5** 83.7**
Δ deviance Model 2 0.0 0.2

Note: Model 1 is an unconditional means model with no predictors of Animals scores, Model 2 is
an unconditional growth model with time (in years since 1st testing) predicting Animals scores,
Model 3 includes both time and cancer as predictors of Animals scores, and Model 4 includes
time, cancer, and a time x cancer interaction as predictors of Animals scores. Smaller Deviance,
AIC, and BIC scores indicate better model fit. * p < .05; ** p<.01; *** p<.001.

139
Table 32. Estimates of fixed effects, variance components, and fit indices for individual growth
models predicting Geriatric Depression Scale scores for long-term cancer survivors comparison
group

Model 1 † Model 2 † Model 3 † Model 4 †

Fixed Effects

Main effects
Intercept 4.48(1.06)** 4.39(1.07)*** 4.39(1.09)*** 4.66(1.11)***
Time -1.00(1.01) -1.00(1.01) 1.00(1.02)
Cancer 0.86(1.14) 0.89(1.15)

Interactions
Cancer x Time 1.01(1.03)

Variance Components

Level 1
Within-person 0.34(0.02)*** 0.26(0.02)*** 0.26(0.02)*** 0.26(0.02)***
Level 2
In initial status 0.37(0.06)*** 0.68(0.27)** 0.65(0.27)** 0.65(0.27)**
In rate of change 0.01(0.00)** 0.01(0.00)** 0.01(0.00)**
In matched pair 0.00(0.00) 0.00(0.00) 0.00(0.00)

Covariance (initial
status, time) -0.03(0.02) -0.03(0.02) -0.03(0.02)

Covariance (initial
status, pair) -0.01(0.01) -0.01(0.01) -0.01(0.01)

Covariance (time,
pair) 0.00(0.00) 0.00(0.00) 0.00(0.00)

Fit Indices

Goodness of Fit
Deviance 1294.7 1224.8 1223.4 1223.1
AIC 1300.7 1242.8 1243.4 1245.1
BIC 1309.6 1269.4 1272.9 1277.6
Model Comparison
Δ deviance Model 1 69.9*** 71.3*** 71.6***
Δ deviance Model 2 1.4 1.7

Note: Model 1 is an unconditional means model with no predictors of Geriatric Depression Scale
scores, Model 2 is an unconditional growth model with time (in years since 1st testing) predicting
Geriatric Depression Scale scores, Model 3 includes both time and cancer as predictors of
Geriatric Depression Scale scores, and Model 4 includes time, cancer, and a time x cancer
interaction as predictors of Geriatric Depression Scale scores. Smaller Deviance, AIC, and BIC
scores indicate better model fit. * p < .05; ** p<.01; *** p<.001. † The natural log of Geriatric
Depression Scores was used as the outcome variable for these models. The Fixed Effects
estimates were exponentiated following the log-transformed analyses for ease of interpretation.

140
Table 33: Repeated measures ANOVA comparing neuropsychological functioning of long-term cancer survivors to cancer-free individuals

Cancer survivors Cancer-free individuals p-values §
N Time 1 Time 2 Time 1 Time 2 Group Time Interaction Group Time Interaction
pairs Mean(SD) Mean(SD) Mean(SD) Mean(SD) F(df) F(df) F(df) effects effects effects
Gross Cognitive Functioning
MMSE 45 25.38(5.89) 24.80(6.18) 23.98(7.04) 23.31(7.25) 0.93(1) 5.68(1) 0.02(1) 0.34 0.02 0.88

Verbal Functioning
Animals 40 16.05(7.98) 16.08(7.67) 13.63(6.71) 13.33(6.74) 2.46(1) 0.08(1) 0.11(1) 0.12 0.78 0.75
FAS 35 33.51(16.14) 33.66(17.56) 34.80(18.56) 32.80(17.84) 0.00(1) 1.09(1) 1.66(1) 0.96 0.30 0.21
BNT 31 42.87(18.92) 44.74(17.96) 42.81(15.16) 44.00(16.58) 0.01(1) 1.59(1) 0.12(1) 0.93 0.22 0.73

Learning and Memory
WL Learning 25 16.28(7.33) 15.28(8.37) 14.68(7.95) 15.68(8.30) 0.06(1) 0.00(1) 3.28(1) 0.81 >0.99 0.08
WL Recall 25 4.52(3.12) 4.52(3.40) 3.96(3.69) 3.96(4.10) 0.31(1) 0.00(1) 0.00(1) 0.58 >0.99 >0.99
WL
Recognition 23 17.74(3.45) 18.26(2.12) 15.96(3.95) 15.91(4.13) 5.33(1) 0.30(1) 0.94(1) 0.03 0.59 0.34

Visuospatial
Figure Copying 27 9.00(2.17) 8.41(2.69) 8.93(2.00) 8.63(2.36) 0.01(1) 5.30(1) 0.64(1) 0.91 0.03 0.43

Attention & Executive Functioning
DS Forward 47 6.02(1.70) 5.74(1.81) 5.98(1.33) 6.09(1.41) 0.28(1) 0.57(1) 1.75(1) 0.60 0.46 0.19
DS Backward 48 4.56(1.97) 4.58(1.82) 4.08(1.51) 4.46(1.69) 0.90(1) 3.63(1) 1.83(1) 0.35 0.06 0.18

Emotion
GDS 41 5.56(5.20) 4.98(5.57) 6.05(5.63) 5.59(5.42) 0.28(1) 1.96(1) 0.02(1) 0.60 0.17 0.89

Note: MMSE=Mini Mental State Examination; BNT=Boston Naming Test; WL=Word List; DS=Digit Span; GDS=Geriatric Depression Scale. § Bonferroni-
adjusted p-value=.0045

140

141
The relationship between length of cancer survival and neuropsychological
performance was investigated using mixed models that focused only on the cancer
survivors, with length of cancer survival entered as a predictor of each of the
neuropsychological tests. For all of the outcome variables, models were able to be fit
using an unstructured covariance structure. Tables 34 – 44 show the results of this
modeling. Model 2, the unconditional growth model, was the best fit for all of the
outcome variables except the Geriatric Depression Scale. For the Geriatric
Depression Scale, shown in Table 44, Model 4 was the best fit. Although not
statistically significant after a correction for the number of significance tests run,
individuals who had survived cancer for a longer period of time showed slightly less
depression than individuals who had been diagnosed with cancer more recently, but
showed slightly more rapid increases in depression over time. For all other tests,
length of cancer survival was not a significant predictor of neuropsychological
performance.
Question 2: Cognitive functioning before and after cancer diagnosis
With the exception of data from the neuropsychological assessment
immediately before cancer diagnosis, all observations for cancer survivors that were
before the cancer diagnosis were discarded for mixed modeling analyses. This
allowed a focus on the trajectory of cognitive functioning from just before cancer
diagnosis throughout the remainder of these survivors’ participation in the ADRC
study.

142

Table 34. Mixed models predicting MMSE with time since long-term cancer diagnosis

Model 1 Model 2 Model 3 Model 4
Fixed Effects

Main effects
Intercept 21.96(0.99)*** 23.80(0.90)*** 23.47(1.46)*** 23.06(1.88)***
Time -1.75(0.28)*** -1.75(0.28)*** -1.93(0.58)**
Time since diagnosis 0.02(0.06) 0.04(0.09)

Time*Time since
diagnosis interaction

0.01(0.03)

Variance Components

Level 1
Within-person 9.83(0.74)*** 3.75(0.33)*** 3.75(0.33)*** 3.75(0.33)***
Level 2
In initial status 62.46(11.50)*** 51.34(9.45)*** 51.19(9.43)*** 51.17(9.42)***
In rate of change 4.25(1.19)*** 4.26(1.19)*** 4.25(1.19)***

Covariance (initial
status, time) 12.10(2.81)*** 12.08(2.81)*** 12.08(2.80)***

Fit Indices

Goodness of Fit
Deviance 2380 2163.4 2163.3 2163.2
AIC 2386 2175.4 2177.3 2179.2
BIC 2392.8 2189 2193.1 2197.3
Model Comparison
Δ deviance Model 1 216.6*** 216.7*** 216.8***
Δ deviance Model 2 0.1 0.2

Note: Model 1 is an unconditional means model with no predictors of MMSE scores, Model 2 is an
unconditional growth model with time (in years since 1st testing) predicting MMSE scores, Model 3
includes both time and cancer as predictors of MMSE scores, and Model 4 includes time, cancer,
and a time x cancer interaction as predictors of MMSE scores. Smaller Deviance, AIC, and BIC
scores indicate better model fit. Δ deviance Model 1 = Model 1 deviance - current model deviance;
Δ deviance Model 2 = Model 2 deviance - current model deviance. * p < .05; ** p<.01; *** p<.001

143

Table 35. Mixed models predicting Animals with time since long-term cancer diagnosis

Model 1 Model 2 Model 3 Model 4
Fixed Effects

Main effects
Intercept 13.74(0.93)*** 15.05(1.02)*** 13.98(2.13)*** 14.04(2.14)***
Time -0.61(0.12)*** -0.60(0.12)*** -0.76(0.26)**
Time since diagnosis 0.05(0.10) 0.05(0.10)

Time*Time since
diagnosis interaction 0.01(0.01)
Variance Components

Level 1
Within-person 14.39(1.09)*** 9.76(0.79)*** 9.77(0.79)*** 9.77(0.80)***
Level 2
In initial status 49.21(9.64)***
58.49(11.47)
***
58.20(11.43)
*** 58.39(11.47) ***
In rate of change 0.38(0.15)*** 0.38(0.15)*** 0.36(0.15)**

Covariance (initial
status, time) 0.60(1.30) 0.50(1.32) 0.41(1.31)

Fit Indices

Goodness of Fit
Deviance 2447.4 2371.9 2371.6 2371.1
AIC 2453.4 2383.9 2385.6 2387.1
BIC 2460.2 2397.5 2401.4 2405.2
Model Comparison
Δ deviance Model 1 75.5*** 75.8*** 76.3***
Δ deviance Model 2 0.3 0.8

Note: Model 1 is an unconditional means model with no predictors of Animals scores, Model 2 is an
unconditional growth model with time (in years since 1st testing) predicting Animals scores, Model 3
includes both time and cancer as predictors of Animals scores, and Model 4 includes time, cancer,
and a time x cancer interaction as predictors of Animals scores. Smaller Deviance, AIC, and BIC
scores indicate better model fit. Δ deviance Model 1 = Model 1 deviance - current model deviance; Δ
deviance Model 2 = Model 2 deviance - current model deviance. * p < .05; ** p<.01; *** p<.001

144

Table 36. Mixed models predicting FAS with time since long-term cancer diagnosis

Model 1 Model 2 Model 3 Model 4
Fixed Effects

Main effects
Intercept 33.33(2.17)*** 34.43(2.25)*** 28.18(4.59)*** 28.72(4.63)***
Time -0.51(0.25)* -0.46(0.25) -0.88(0.52)

Time since
diagnosis 0.32(0.20) 0.29(0.21)

Time*Time since
diagnosis interaction 0.02(0.02)
Variance Components

Level 1 47.07(3.68)*** 34.22(2.89)*** 34.28(53.72)*** 34.19(2.88)***
Within-person
Level 2
In initial status
270.09(51.82)
***
287.26(55.37)
***
277.90(53.72)
***
278.11(53.73)
***
In rate of change 1.53(0.60)** 1.51(0.59)** 1.51(0.58)**

Covariance (initial
status, time) 0.18(5.40) -1.27(5.48) -1.37(5.42)

Fit Indices

Goodness of Fit
Deviance 2798.5 2749.2 2747 2746.1
AIC 2804.5 2761.2 2761 2762.1
BIC 2811.3 2774.8 2776.8 2780.2
Model Comparison
Δ deviance Model 1 49.3*** 51.5*** 52.4***
Δ deviance Model 2 2.2 3.1

Note: Model 1 is an unconditional means model with no predictors of FAS scores, Model 2 is an
unconditional growth model with time (in years since 1st testing) predicting FAS scores, Model 3
includes both time and cancer as predictors of FAS scores, and Model 4 includes time, cancer, and a
time x cancer interaction as predictors of FAS scores. Smaller Deviance, AIC, and BIC scores
indicate better model fit. Δ deviance Model 1 = Model 1 deviance - current model deviance; Δ
deviance Model 2 = Model 2 deviance - current model deviance. * p < .05; ** p<.01; *** p<.001

145

Table 37. Mixed models predicting Boston Naming Test (BNT) with time since long-term cancer
diagnosis

Model 1 Model 2 Model 3 Model 4
Fixed Effects

Main effects
Intercept 39.83(2.40)*** 40.90(2.26)*** 41.69(3.05)*** 42.69(4.83)***
Time -1.31(0.22)*** -1.31(0.22)*** -1.20(0.48)*
Time since diagnosis -0.04(0.10) -0.09(0.21)

Time*Time since
diagnosis interaction -0.01(0.02)
Variance Components

Level 1
Within-person 16.64(1.42)*** 13.13(1.16)*** 13.13(1.16)*** 13.13(1.16)***
Level 2
In initial status
361.84(65.11)
***
321.74(58.30)
***
321.77(58.30)
***
321.27(58.23)
***
In rate of change 2.69(0.98)** 2.62(0.98)** 2.65(0.98)**

Covariance (initial
status, time) 29.42(6.85)*** 29.06(6.86)*** 29.16(6.85)***

Fit Indices

Goodness of Fit
Deviance 2202.8 2155.3 2155.2 2155.1
AIC 2298.8 2167.3 2169.2 2171.1
BIC 2215.6 2180.9 2185 2189.2
Model Comparison
Δ deviance Model 1 47.5*** 47.6*** 47.7***
Δ deviance Model 2 0.1 0.2

Note: Model 1 is an unconditional means model with no predictors of BNT scores, Model 2 is an
unconditional growth model with time (in years since 1st testing) predicting BNT scores, Model 3
includes both time and cancer as predictors of BNT scores, and Model 4 includes time, cancer, and
a time x cancer interaction as predictors of BNT scores. Smaller Deviance, AIC, and BIC scores
indicate better model fit. Δ deviance Model 1 = Model 1 deviance - current model deviance; Δ
deviance Model 2 = Model 2 deviance - current model deviance. * p < .05; ** p<.01; *** p<.001

146

Table 38. Mixed models predicting Word List (WL) Learning with time since long-term cancer
diagnosis

Model 1 Model 2 Model 3 Model 4
Fixed Effects

Main effects
Intercept 15.30(1.05)*** 17.20(1.09)*** 17.10(2.21)*** 15.81(2.31)***
Time -0.97(0.15)*** -0.96(0.15)*** -1.51(0.31)***
Time since diagnosis 0.01(0.10) 0.07(0.10)

Time*Time since
diagnosis interaction 0.03(0.01)*
Variance Components

Level 1
Within-person 10.67(0.90)*** 5.07(0.47)*** 5.07(0.47)*** 5.04(0.46)***
Level 2
In initial status
58.83(11.75)
***
63.55(12.73)
***
63.47(12.81)
***
64.10(12.88)
***
In rate of change 0.71(0.26)** 0.71(0.26)** 0.60(0.23)**

Covariance (initial
status, time) 3.51(1.63)* 3.49(1.69)* 2.94(1.57)

Fit Indices

Goodness of Fit
Deviance 1937.9 1795.4 1795.4 1791.1
AIC 1943.9 1807.4 1809.4 1807.1
BIC 1950.7 1821 1825.2 1825.2
Model Comparison
Δ deviance Model 1 142.5*** 142.5*** 146.8***
Δ deviance Model 2 0 4.3

Note: Model 1 is an unconditional means model with no predictors of WL Learning scores, Model
2 is an unconditional growth model with time (in years since 1st testing) predicting WL Learning
scores, Model 3 includes both time and cancer as predictors of WL Learning scores, and Model 4
includes time, cancer, and a time x cancer interaction as predictors of WL Learning scores.
Smaller Deviance, AIC, and BIC scores indicate better model fit. Δ deviance Model 1 = Model 1
deviance - current model deviance; Δ deviance Model 2 = Model 2 deviance - current model
deviance. * p < .05; ** p<.01; *** p<.001

147

Table 39. Mixed models predicting Word List (WL) Delayed Recall with time since long-term
cancer diagnosis

Model 1 Model 2 Model 3 Model 4
Fixed Effects

Main effects
Intercept 4.52(0.44)*** 5.17(0.48)*** 3.98(0.99)*** 4.26(1.01)***
Time -0.25(0.05)*** -0.23(0.05)*** -0.40(0.10)***
Time since diagnosis 0.06(0.04) 0.05(0.04)

Time*Time since
diagnosis interaction 0.01(0.00)*
Variance Components

Level 1
Within-person 1.97(0.17)*** 1.26(0.11)*** 1.26(0.11)*** 1.26(0.11)***
Level 2
In initial status 10.05(2.00)*** 12.09(2.43)*** 11.92(2.41)*** 12.08(2.43)***
In rate of change 0.04(0.02)** 0.04(0.02)** 0.03(0.01)*

Covariance (initial
status, time) 0.05(0.21) -0.05(0.23) -0.06(0.20)

Fit Indices

Goodness of Fit
Deviance 1376.1 1292.4 1290.9 1286.7
AIC 1382.1 1304.4 1304.9 1302.7
BIC 1388.9 1318.0 1320.7 1320.8
Model Comparison
Δ deviance Model 1 83.7*** 85.2*** 89.4***
Δ deviance Model 2 1.5 5.7

Note: Model 1 is an unconditional means model with no predictors of WL Recall scores, Model 2
is an unconditional growth model with time (in years since 1st testing) predicting WL Recall
scores, Model 3 includes both time and cancer as predictors of WL Recall scores, and Model 4
includes time, cancer, and a time x cancer interaction as predictors of WL Recall scores. Smaller
Deviance, AIC, and BIC scores indicate better model fit. Δ deviance Model 1 = Model 1 deviance
- current model deviance; Δ deviance Model 2 = Model 2 deviance - current model deviance.
* p < .05; ** p<.01; *** p<.001

148

Table 40. Mixed models predicting Word List (WL) Recognition with time since long-term cancer
diagnosis

Model 1 Model 2 Model 3 Model 4
Fixed Effects

Main effects
Intercept 17.52(0.42)*** 18.00(0.41)*** 17.94(0.82)*** 17.61(0.87)***
Time -0.29(0.07)*** -0.29(0.07)*** -0.45(0.14)**
Time since diagnosis 0.00(0.04) 0.02(0.03)

Time*Time since
diagnosis interaction 0.01(0.01)
Variance Components

Level 1
Within-person 2.85(0.24)*** 1.94(0.18)*** 1.94(0.18)*** 1.96(0.18)***
Level 2
In initial status 8.87(1.89)*** 8.06(1.81)*** 8.05(1.81)*** 8.11(1.83)***
In rate of change 0.13(0.07)* 0.13(0.07)* 0.11(0.06)*

Covariance (initial
status, time) 0.69(0.33)* 0.69(0.34)* 0.64(0.33)

Fit Indices

Goodness of Fit
Deviance 1435.8 1371.0 1371.0 1369.1
AIC 1441.8 1383.0 1385.0 1385.1
BIC 1448.6 1396.5 1400.8 1403.3
Model Comparison
Δ deviance Model 1 64.8*** 64.8*** 66.7***
Δ deviance Model 2 0.0 1.9

Note: Model 1 is an unconditional means model with no predictors of WL Recognition scores,
Model 2 is an unconditional growth model with time (in years since 1st testing) predicting WL
Recognition scores, Model 3 includes both time and cancer as predictors of WL Recognition
scores, and Model 4 includes time, cancer, and a time x cancer interaction as predictors of WL
Recognition scores. Smaller Deviance, AIC, and BIC scores indicate better model fit. Δ deviance
Model 1 = Model 1 deviance - current model deviance; Δ deviance Model 2 = Model 2 deviance -
current model deviance. * p < .05; ** p<.01; *** p<.001

149

Table 41. Mixed models predicting Figure Copying with time since long-term cancer diagnosis

Model 1 Model 2 Model 3 Model 4
Fixed Effects

Main effects
Intercept 8.13(0.33)*** 8.80(0.36)*** 8.89(0.72)*** 8.67(0.74)***
Time -0.33(0.06)***
-
0.34(0.06)*** -0.50(0.13)***
Time since diagnosis -0.00(0.03) 0.01(0.03)

Time*Time since
diagnosis interaction 0.01(0.01)
Variance Components

Level 1
Within-person 2.62(0.22)*** 1.64(0.15)*** 1.64(0.15)*** 1.64(0.15)***
Level 2
In initial status 5.62(1.23)*** 6.38(1.39)*** 6.39(1.39)*** 6.40(1.40)***
In rate of change 0.10(0.07) 0.11(0.07) 0.10(0.06)

Covariance (initial
status, time) 0.50(0.31) 0.51(0.33) 0.50(0.31)

Fit Indices

Goodness of Fit
Deviance 1434.9 1354.9 1354.9 1353.0
AIC 1440.9 1366.9 1368.9 1369.0
BIC 1447,7 1380.4 1384.7 1387.1
Model Comparison
Δ deviance Model 1 80.0*** 80.0*** 81.9***
Δ deviance Model 2 0.0 1.9

Note: Model 1 is an unconditional means model with no predictors of Figure Copying scores,
Model 2 is an unconditional growth model with time (in years since 1st testing) predicting Figure
Copying scores, Model 3 includes both time and cancer as predictors of Figure Copying scores,
and Model 4 includes time, cancer, and a time x cancer interaction as predictors of Figure
Copying scores. Smaller Deviance, AIC, and BIC scores indicate better model fit. Δ deviance
Model 1 = Model 1 deviance - current model deviance; Δ deviance Model 2 = Model 2 deviance -
current model deviance. * p < .05; ** p<.01; *** p<.001

150

Table 42. Mixed models predicting Digit Span Forward with time since long-term cancer
diagnosis

Model 1 Model 2 Model 3 Model 4
Fixed Effects

Main effects
Intercept 6.58(0.31)*** 6.74(0.31)*** 7.03(0.65)*** 6.98(0.65)***
Time -0.06(0.02)* -0.06(0.02)* -0.02(0.06)
Time since diagnosis -0.02(0.03) -0.01(0.03)

Time*Time since
diagnosis interaction -0.00(0.00)
Variance Components

Level 1
Within-person 1.83(0.14)*** 1.74(0.14)*** 1.74(0.14)*** 1.73(0.14)***
Level 2
In initial status 5.45(1.05)*** 5.49(1.10)*** 5.47(1.10)*** 5.48(1.10)***
In rate of change 0.00(0.01) 0.00(0.01) 0.00(0.01)

Covariance (initial
status, time) 0.06(0.08) 0.05(0.08) 0.04(0.08)

Fit Indices

Goodness of Fit
Deviance 1604.6 1596.0 1595.8 1595.0
AIC 1610.6 1608.0 1609.8 1611.0
BIC 1617.4 1621.6 1625.6 1629.1
Model Comparison
Δ deviance Model 1 8.6* 8.8* 9.6*
Δ deviance Model 2 0.2 1.0

Note: Model 1 is an unconditional means model with no predictors of Digit Span Forward scores,
Model 2 is an unconditional growth model with time (in years since 1st testing) predicting Digit
Span Forward scores, Model 3 includes both time and cancer as predictors of Digit Span Forward
scores, and Model 4 includes time, cancer, and a time x cancer interaction as predictors of Digit
Span Forward scores. Smaller Deviance, AIC, and BIC scores indicate better model fit. Δ
deviance Model 1 = Model 1 deviance - current model deviance; Δ deviance Model 2 = Model 2
deviance - current model deviance. * p < .05; ** p<.01; *** p<.001

151

Table 43. Mixed models predicting Digit Span Backward with time since long-term cancer
diagnosis

Model 1 Model 2 Model 3 Model 4
Fixed Effects

Main effects
Intercept 5.12(0.33)*** 5.49(0.34)*** 4.99(0.72)*** 5.00(0.72)***
Time -0.19(0.04)*** -0.18(0.04)*** -0.26(0.08)**
Time since diagnosis 0.03(0.03) 0.03(0.03)

Time*Time since
diagnosis interaction 0.00(0.00)
Variance Components

Level 1
Within-person 2.24(0.17)*** 1.79(0.15)*** 1.79(0.15)*** 1.80(0.15)***
Level 2
In initial status 6.26(1.21)*** 6.66(1.33)*** 6.61(1.32)*** 6.58(1.31)***
In rate of change 0.03(0.01)* 0.03(0.01)* 0.02(0.01)

Covariance (initial
status, time) 0.14(0.14) 0.12(0.14) 0.14(0.13)

Fit Indices

Goodness of Fit
Deviance 1653.9 1609.8 1609.2 1608.2
AIC 1659.9 1621.8 1623.2 1624.2
BIC 1666.6 1635.4 1639 1642.3
Model Comparison
Δ deviance Model 1 44.1*** 44.7*** 45.7***
Δ deviance Model 2 0.6 1.6

Note: Model 1 is an unconditional means model with no predictors of Digit Span Backward
scores, Model 2 is an unconditional growth model with time (in years since 1st testing)
predicting Digit Span Backward scores, Model 3 includes both time and cancer as predictors of
Digit Span Backward scores, and Model 4 includes time, cancer, and a time x cancer interaction
as predictors of Digit Span Backward scores. Smaller Deviance, AIC, and BIC scores indicate
better model fit. Δ deviance Model 1 = Model 1 deviance - current model deviance; Δ deviance
Model 2 = Model 2 deviance - current model deviance. * p < .05; ** p<.01; *** p<.001

152

Table 44. Mixed models predicting Geriatric Depression Scale (GDS) with time since long-term
cancer diagnosis

Model 1 Model 2 Model 3 Model 4
Fixed Effects

Main effects
Intercept 5.09(0.50)*** 5.28(0.58)*** 6.73(1.10)*** 7.76(1.16)***
Time -0.09(0.08) -0.09(0.09) -0.51(0.16)**
Time since diagnosis -0.07(0.05) -0.12(0.05)*

Time*Time since
diagnosis interaction 0.02(0.01)*
Variance Components

Level 1
Within-person 8.31(0.64)*** 4.77(0.39)*** 4.75(0.39)*** 4.82(0.40)***
Level 2
In initial status 13.03(2.64)*** 18.64(3.64)*** 17.37(3.44)*** 16.96(3.34)***
In rate of change 0.19(0.06)*** 0.20(0.06)*** -0.61(0.27)*

Covariance (initial
status, time) -0.87(0.33)** -0.77(0.33)* 0.13(0.05)**

Fit Indices

Goodness of Fit
Deviance 2071.4 1946.3 1944.2 1937.2
AIC 2077.4 1958.3 1958.2 1953.2
BIC 2084.2 1971.8 1974.0 1971.3
Model Comparison
Δ deviance Model 1 125.1*** 127.2*** 134.2***
Δ deviance Model 2 2.1 9.1*

Note: Model 1 is an unconditional means model with no predictors of GDS scores, Model 2 is an
unconditional growth model with time (in years since 1st testing) predicting GDS scores, Model 3
includes both time and cancer as predictors of GDS scores, and Model 4 includes time, cancer, and
a time x cancer interaction as predictors of GDS scores. Smaller Deviance, AIC, and BIC scores
indicate better model fit. Δ deviance Model 1 = Model 1 deviance - current model deviance; Δ
deviance Model 2 = Model 2 deviance - current model deviance. * p < .05; ** p<.01; *** p<.001

153
For each of the neuropsychological tests, the full sample of 40 individuals (20
matched pairs) was included in analyses. None of the outcome variables was able to
be modeled using the unstructured covariance structure, probably due to the small
sample size. Five of the outcome variables—MMSE, Animals, FAS, Word List
Recall, and Geriatric Depression Scale—were modeled using the compound
symmetric covariance structure. One outcome variable, Word List Learning, was
modeled best using the autoregressive of order 1 covariance structure. The remaining
five outcome variables were unable to be modeled using any of the covariance
structures. For these outcome variables, a group-by-time interaction was tested by
examining whether cancer predicted slope from individual OLS regressions, using a
1-way ANOVA. Cancer was not a significant predictor of slope in any of these
analyses, indicating that there were no significant group-by-time interactions on the
Boston Naming Test (F(1, 14)=0.15, p=.70), Word List Recognition (F(1, 11)=0.00,
p=.95), Figure Drawing (F(1, 15)=4.25, p=.06), Digit Span Forward (F(1, 14)=0.01,
p=.94), or Digit Span Backward (F(1, 14)=0.38, p=.55).
For the six outcome variables that were able to be modeled, the mean (SD)
number of assessment times per subjects included in the mixed modeling test was as
follows: MMSE: 6.58(4.41), Animals: 6.85(5.33), FAS: 6.13(4.83), Word List
Learning: 4.88(4.24), Word List Delayed Recall: 4.90(4.22), and Geriatric
Depression Scale: 6.55(4.43). Tables 45 – 50 show the results of the mixed models
predicting MMSE, Animals, FAS, Word List Learning, Word List Recall, and
Geriatric Depression Scale for cancer survivors who were diagnosed with their first

154
Table 45. Estimates of fixed effects, variance components, and fit indices for individual growth
models predicting Mini Mental State Examination (MMSE) scores for individuals diagnosed with
cancer during the ADRC study, and matched comparison group

Model 1 Model 2 § Model 3 § Model 4 §
Fixed Effects

Main effects
Intercept 27.45(0.39)*** 28.92(0.38)*** 28.40(0.52)*** 28.41(0.53)***
Time -0.36(0.06)*** -0.35(0.06)*** -0.33(0.08)***
Cancer 0.90(0.73) 0.92(0.74)

Interactions
Cancer x Time -0.05(0.13)

Variance Components

Level 1
Within-person 4.50(0.48)*** 2.91(0.42)*** 2.99(0.43)*** 2.95(0.44)***
Level 2
In initial status 1.47(1.20)
In rate of change
In matched pair 0.02(0.01)*

Covariance (initial
status, time)

Covariance (initial
status, pair)

Covariance (time,
pair)

Fit Indices

Goodness of Fit
Deviance 996.0 947.5 946.1 945.9
AIC 1004.0 957.5 958.1 959.9
BIC 1010.7 965.9 968.2 971.7
Model Comparison
Δ deviance Model 1 48.5*** 49.9*** 50.1***
Δ deviance Model 2 1.4 1.6

Note: Model 1 is an unconditional means model with no predictors of MMSE scores, Model 2 is
an unconditional growth model with time (in years since 1st testing) predicting MMSE scores,
Model 3 includes both time and cancer as predictors of MMSE scores, and Model 4 includes time,
cancer, and a time x cancer interaction as predictors of MMSE scores. Smaller Deviance, AIC,
and BIC scores indicate better model fit. * p < .05; ** p<.01; *** p<.001. § Model fit using
compound symmetric covariance structure. Using this covariance structure, only 1 variance
component is estimated.

155
Table 46. Estimates of fixed effects, variance components, and fit indices for individual growth
models predicting Animals scores for individuals diagnosed with cancer during the ADRC study,
and matched comparison group

Model 1 Model 2 § Model 3 § Model 4 §
Fixed Effects

Main effects
Intercept 16.38(0.78)*** 18.70(0.71)*** 20.22(1.00)*** 20.36(0.99)***
Time -0.65(0.13)*** -0.65(0.12)*** -0.79(0.16)***
Cancer -2.92(1.37)* -3.21(1.38)*

Interactions
Cancer x Time 0.33(0.24)

Variance Components

Level 1
Within-person 13.43(1.46)*** 7.98(0.97)*** 7.92(0.95)*** 7.92(0.95)***
Level 2
In initial status 16.84(8.54)*
In rate of change
In matched pair 0.00(0.02)

Covariance (initial
status, time)

Covariance (initial
status, pair)

Covariance (time,
pair)

Fit Indices

Goodness of Fit
Deviance 1171.5 1120.5 1116.1 1114.4
AIC 1179.5 1130.5 1128.1 1128.4
BIC 1186.3 1139.0 1138.3 1140.2
Model Comparison
Δ deviance Model 1 51.8*** 55.4*** 57.1***
Δ deviance Model 2 4.4 6.1*

Note: Model 1 is an unconditional means model with no predictors of Animals scores, Model 2 is an
unconditional growth model with time (in years since 1st testing) predicting Animals scores, Model
3 includes both time and cancer as predictors of Animals scores, and Model 4 includes time, cancer,
and a time x cancer interaction as predictors of Animals scores. Smaller Deviance, AIC, and BIC
scores indicate better model fit. * p < .05; ** p<.01; *** p<.001. § Model fit using compound
symmetric covariance structure. Using this covariance structure, only 1 variance component is
estimated.

156

Table 47. Estimates of fixed effects, variance components, and fit indices for individual growth
models predicting FAS scores for individuals diagnosed with cancer during the ADRC study, and
matched comparison group

Model 1 Model 2 § Model 3 § Model 4 §
Fixed Effects

Main effects
Intercept 39.35(1.92)*** 43.69(1.70)*** 42.17(2.46)*** 42.42(2.44)***
Time -1.01(0.28)** -1.03(0.28)*** -0.72(0.37)
Cancer 3.01(3.36) 2.50(3.35)

Interactions
Cancer x Time -0.71(0.56)

Variance Components

Level 1
Within-person 38.70(4.33)*** 30.32(4.01)*** 30.47(4.03)*** 30.48(4.00)***
Level 2
In initial status
111.51(29.95)
***
In rate of change
In matched pair 0.00(.)

Covariance (initial
status, time)

Covariance (initial
status, pair)

Covariance (time,
pair)

Fit Indices

Goodness of Fit
Deviance 1386.6 1348.8 1348.1 1346.5
AIC 1392.6 1358.8 1360.1 1360.5
BIC 1397.7 1367.3 1370.2 1372.3
Model Comparison
Δ deviance Model 1 37.8*** 38.5*** 40.1***
Δ deviance Model 2 0.7 2.3

Note: Model 1 is an unconditional means model with no predictors of FAS scores, Model 2 is an
unconditional growth model with time (in years since 1st testing) predicting FAS scores, Model 3
includes both time and cancer as predictors of FAS scores, and Model 4 includes time, cancer, and
a time x cancer interaction as predictors of FAS scores. Smaller Deviance, AIC, and BIC scores
indicate better model fit. * p < .05; ** p<.01; *** p<.001. § Model fit using compound symmetric
covariance structure. Using this covariance structure, only 1 variance component is estimated.

157
Table 48. Estimates of fixed effects, variance components, and fit indices for individual growth
models predicting Word List Learning scores for individuals diagnosed with cancer during the
ADRC study, and matched comparison group

Model 1 Model 2 § Model 3 § Model 4 §
Fixed Effects

Main effects
Intercept 19.94(0.76)*** 23.13(0.77)*** 21.43(0.95)*** 21.07(0.94)***
Time -0.67(0.13)*** -0.66(0.12)*** -0.36(0.17)*
Cancer 3.58(1.41)* 4.20(1.40)**

Interactions
Cancer x Time -0.49(0.23)*

Variance Components

Level 1
Within-person 11.41(1.25)*** 7.98(0.97)*** 8.23(1.00)*** 8.14(0.96)***
Level 2
In initial status 13.92(7.05)*
In rate of change
In matched pair 0.00(0.02)

Covariance (initial
status, time)

Covariance (initial
status, pair)

Covariance (time,
pair)

Fit Indices

Goodness of Fit
Deviance 1089.3 1055.7 1050.5 1045.9
AIC 1097.3 1065.7 1062.5 1059.9
BIC 1104.1 1074.1 1072.6 1071.7
Model Comparison
Δ deviance Model 1 33.6*** 38.8*** 43.4***
Δ deviance Model 2 5.2 9.8**

Note: Model 1 is an unconditional means model with no predictors of WL Learning scores, Model
2 is an unconditional growth model with time (in years since 1st testing) predicting WL Learning
scores, Model 3 includes both time and cancer as predictors of WL Learning scores, and Model 4
includes time, cancer, and a time x cancer interaction as predictors of WL Learning scores.
Smaller Deviance, AIC, and BIC scores indicate better model fit. * p < .05; ** p<.01; *** p<.001;
§ Models fit using autoregressive order 1. Using this covariance structure, only 1 variance
component is estimated.

158

Table 49. Estimates of fixed effects, variance components, and fit indices for individual growth
models predicting Word List Delayed Recall scores for long-term cancer survivors and
comparison group

Model 1 Model 2 § Model 3 § Model 4 §
Fixed Effects

Main effects
Intercept 6.65(0.45)*** 8.16(0.37)*** 8.11(0.50)*** 8.11(0.50)***
Time -0.28(0.06)*** -0.28(0.06)*** -0.28(0.09)**
Cancer 0.11(0.73) 0.11(0.73)

Interactions
Cancer x Time 0.00(0.11)

Variance Components

Level 1
Within-person 2.58(0.28)*** 1.95(0.23)*** 1.95(0.23)*** 1.95(0.23)***
Level 2
In initial status 3.31(2.65)
In rate of change
In matched pair 0.01(0.01)

Covariance (initial
status, time)

Covariance (initial
status, pair)

Covariance (time,
pair)

Fit Indices

Goodness of Fit
Deviance 817.8 778.4 778.3 778.3
AIC 825.8 788.4 790.3 792.3
BIC 832.6 796.8 800.5 804.2
Model Comparison
Δ deviance Model 1 39.4*** 39.5*** 39.5***
Δ deviance Model 2 0.1 0.1

Note: Model 1 is an unconditional means model with no predictors of WL Recall scores, Model 2
is an unconditional growth model with time (in years since 1st testing) predicting WL Recall
scores, Model 3 includes both time and cancer as predictors of WL Recall scores, and Model 4
includes time, cancer, and a time x cancer interaction as predictors of WL Recall scores. Smaller
Deviance, AIC, and BIC scores indicate better model fit. * p < .05; ** p<.01; *** p<.001;
§ Model fit using compound symmetric covariance structure. Using this covariance structure, only
1 variance component is estimated.

159
Table 50. Estimates of fixed effects, variance components, and fit indices for individual growth
models predicting Geriatric Depression Scale (GDS) scores for individuals diagnosed with cancer
during the ADRC study, and matched comparison group

Model 1 † Model 2 †§ Model 3 †§ Model 4 †§
Fixed Effects

Main effects 5.47(1.12)*** 3.97(1.09)*** 2.94(1.13)*** 4.38(1.14)***
Intercept 1.06(1.02)** 1.06(1.02)*** 1.05(1.02)*
Time 1.72(1.16)*** 1.60(1.20)**
Cancer

Interactions
Cancer x Time 1.02(1.03)

Variance Components

Level 1
Within-person 0.23(0.03)*** 0.19(0.02)*** 0.18(0.02)*** 0.18(0.02)***
Level 2
In initial status 0.20(0.12)*
In rate of change
In matched pair 0.00(0.00)

Covariance (initial
status, time)

Covariance (initial
status, pair)

Covariance (time,
pair)

Fit Indices

Goodness of Fit
Deviance 329.0 320.4 310.1 309.6
AIC 337.0 330.4 322.1 323.6
BIC 343.7 338.9 332.2 335.5
Model Comparison
Δ deviance Model 1 8.6 18.9 19.4
Δ deviance Model 2 10.3 10.8

Note: Model 1 is an unconditional means model with no predictors of GDS scores, Model 2 is an
unconditional growth model with time predicting GDS scores, Model 3 includes both time and
cancer as predictors of GDS scores, and Model 4 includes time, cancer, and a time x cancer
interaction as predictors of GDS scores. Smaller Deviance, AIC, and BIC scores indicate better
model fit. * p < .05; ** p<.01; *** p<.001; § Model fit using autoregressive of order 1 covariance
structure; † The natural log of GDS was used as the outcome variable for these models. The Fixed
Effects estimates were exponentiated following the log-transformed analyses.

160
cancer during their participation in the ADRC study compared to matched
individuals who had never had cancer. In Table 46, Model 3, the best-fit model
predicting Animals, shows that cancer was a significant predictor of number of
Animals generated on this verbal fluency task (p<.05). Individuals diagnosed with
cancer during their participation in the ADRC study performed more poorly than
matched cancer history-free individuals. There was also a significant effect for time
on Animals performance (p<.001), with both groups showing declines on this test
over time. There was no significant group-by-time interaction, indicating that cancer
survivors did not experience a different rate of decline than cancer history-free
individuals on the Animals test after being diagnosed with and treated for cancer.
Figure 3 shows a plot of the prototypical performance trajectories on Animals for
individuals diagnosed with cancer during the ADRC study and the cancer history-
free individuals.
Figure 3. Prototypical performance trajectories on Animals for individuals diagnosed with cancer
during ADRC study and matched cancer history-free comparison group.
0
5
10
15
20
25
0 1 2 3 4
Years since cancer diagnosis
Animals score

Note: Higher scores indicate better performance.

No cancer
Cancer

161
In Table 48, Model 4, the best-fit model predicting Word List Learning,
shows that cancer was a significant predictor of Word List Learning (p<.01), with the
individuals who were diagnosed with cancer during the study performing better on
this measure of verbal learning than individuals without a cancer history. This model
also shows a significant group-by-time interaction (p<.05); the individuals who were
diagnosed with cancer during the study had a steeper rate of decline on Word List
Learning than matched individuals without a cancer history. A plot of the
prototypical performance trajectories for these cancer survivors and cancer history-
free individuals is shown in Figure 4.
Figure 4. Prototypical performance trajectories on Word List Learning for individuals diagnosed with
cancer during ADRC study and matched cancer history-free comparison group.

0
5
10
15
20
25
30
0 1 2 3 4
Years since cancer diagnosis
Word List Learning score

Note: Higher scores indicate better performance.

In Table 50, Model 3, the best-fit model predicting Geriatric Depression
Scale scores, shows that cancer was a significant predictor of depression scores
(p<.001). Individuals who were diagnosed with cancer during the ADRC study
reported higher levels of depression than did individuals without a cancer history.
Cancer
No cancer

162
Both groups reported increased levels of depression over time, as indicated by the
main effect for time (p<.001). There was no significant group-by-time interaction,
suggesting that cancer survivors did not report greater increases in depressive
symptoms following cancer diagnosis and treatment than did cancer history-free
individuals as they aged. Figure 5 shows a plot of the prototypical depression
trajectories for these cancer survivors and cancer history-free individuals.
Figure 5. Prototypical depression trajectories on Geriatric Depression Scale for individuals diagnosed
with cancer during ADRC study and matched cancer history-free comparison group.

0
1
2
3
4
5
6
7
8
0 1 2 3 4
Years since cancer diagnosis
Geriatric Depression Scale score

Note: Higher scores indicate more depressive symptoms.

Because there were significant main effects for cancer on both the Animals
test and the Geriatric Depression Scale, and depression can diminish verbal fluency
scores (Henry & Crawford, 2005), the possibility that depression confounded the
association between cancer and poorer semantic fluency was explored. Geriatric
Depression Scale scores were entered into a model as a level-1, time-varying,
predictor of Animals scores, with time included as an additional predictor. Next,
Geriatric Depression Scale scores, cancer, and time were entered into a model as
Cancer
No cancer

163
predictors of the Animals test. Table 51 shows the results of these two models
(Models 2 and 3), with the original unconditional means model included in the table
(Model 1) for comparative purposes. Model 2 shows that depression was a
significant predictor of Animals scores. Model 3 shows that when both depression
and cancer were included as predictors of Animals, neither was a significant
predictor (cancer p=.08; GDS p=.07).
Repeated measures Analysis of Variance were also used to examine whether
neuropsychological functioning of cancer survivors declines following cancer
treatment compared to individuals never diagnosed with and treated for cancer.
Table 52 displays the results of these analyses. None of the results was statistically
significant.
Power analyses
Power analyses were conducted using GPower 3 (Erdfelder, Faul, &
Buchner, 1996) to examine whether the study had sufficient power to find significant
effects. For analyses comparing long-term cancer survivors to their comparison
group with repeated measures ANOVA for two time points, the power to detect a
small main effect (effect size of .10) ranged from .80 to .88, depending upon the
sample size and the strength of the correlation between variables over time. For
analyses comparing cancer patients who were diagnosed with cancer during the
ADRC study to their comparison group with repeated measures ANOVAs, the power
to detect a medium effect (effect size of .25) with a sample this size ranged from .13
to .43, depending upon the sample size and strength of the correlation between

164

Table 51. Estimates of fixed effects, variance components, and fit indices for individual growth
models predicting Animals scores and controlling for depression (for individuals diagnosed
with cancer during the ADRC study, and matched comparison group)

Model 1 Model 2 § Model 3 §
Fixed Effects

Main effects
Intercept 16.38(0.78)*** 20.10(0.97)*** 21.16(1.15)***
Time -0.60(0.14)*** -0.61(0.13)***
GDS -1.07(0.50)* -0.93(0.50)
Cancer -2.36(1.38)

Variance Components

Level 1
Within-person 13.43(1.46)*** 7.54(1.02)*** 7.47(1.00)***
Level 2
In initial status 16.84(8.54)*
In rate of change
In matched pair 0.00(0.02)
Covariance (initial status, time)
Covariance (initial status, pair)
Covariance (time, pair)

Fit Indices

Goodness of Fit
Deviance 1171.5 966.7 963.8
AIC 1179.5 978.7 977.8
BIC 1186.3 988.4 989.1
Model Comparison
Δ deviance Model 1 204.8*** 207.7***
Δ deviance Model 2 2.9

Note: Model 1 is the unconditional means model with no predictors of Animals scores, Model 2
predicts Animals scores with time and Geriatric depression scores, Model 3 includes time,
cancer, and depression scores as predictors of Animals scores. Smaller Deviance, AID, and BIC
scores indicate better model fit. * p < .05; ** p<.01; *** p<.001; § Model fit using compound
symmetric covariance structure. Using this covariance structure, only 1 variance component is
estimated.

165
Table 52: Repeated measures ANOVA comparing neuropsychological functioning of cancer survivors diagnosed with cancer during the ADRC study
and matched cancer-free individuals

Cancer survivors Cancer-free individuals p-values
N Time 1 Time 2 Time 1 Time 2 Group Time Interaction Group Time Interaction
pairs Mean(SD) Mean(SD) Mean(SD) Mean(SD) F(df) F(df) F(df) effects effects effects
Gross Cognitive Functioning
MMSE 15 28.40(1.50) 27.60(4.17) 27.00(3.61) 27.47(3.38) 0.50(1,14) 0.09(1,14) 2.53(1,14) 0.49 0.76 0.13

Verbal Functioning
Animals 14 19.07(5.09) 19.21(6.55) 18.43(3.34) 16.86(4.69) 0.51(1,13) 1.19(1,13) 0.67(1,13) 0.49 0.30 0.43
FAS 14 43.71(13.75) 45.43(12.62) 36.86(10.08) 39.79(11.15) 2.42(1,13) 1.94(1,13) 0.26(1,13) 0.14 0.19 0.62
BNT 10 49.30(13.35) 51.90(6.31) 53.70(4.11) 54.10(4.48) 1.79(1,9) 2.15(1,9) 0.63(1,9) 0.21 0.18 0.45

Learning and Memory
WL Learning 8 24.38(4.75) 21.50(3.63) 19.25(3.01) 20.75(4.17) 2.07(1,7) 0.71(1,7) 2.95(1,7) 0.19 0.43 0.13
WL Recall 9 7.55(2.79) 7.00(2.12) 6.89(2.20) 6.33(2.74) 0.27(1,8) 2.17(1,8) 0.00(1,8) 0.62 0.18 >.99
WL Recognition 9 19.44(1.13) 19.67(0.50) 19.22(1.30) 18.22(2.77) 1.09(1,8) 1.23(1,8) 3.90(1,8) 0.33 0.30 0.08

Visuospatial
Figure Copying 12 9.50(1.57) 9.83(1.27) 8.75(1.96) 9.33(1.87) 0.70(1,11) 2.58(1,11) 0.34(1,11) 0.42 0.14 0.57

Attention & Executive Functioning
DS Forward 15 6.93(1.22) 7.07(1.79) 6.53(1.64) 6.47(1.77) 0.85(1,14) 0.01(1,14) 0.22(1,14) 0.37 0.91 0.65
DS Backward 16 5.63(1.31) 5.56(1.41) 4.94(1.77) 4.56(2.45) 3.23(1,15) 0.70(1,15) 0.66(1,15) 0.09 0.42 0.43

Emotion
GDS 16 4.56(2.48) 5.81(3.85) 5.13(5.92) 5.19(5.67) 0.00(1,15) 0.87(1,15) 0.79(1,15) 0.98 0.37 0.39

Note: MMSE=Mini Mental State Examination; BNT=Boston Naming Test; WL=Word List; DS=Digit Span; GDS=Geriatric Depression Scale.

165

166
variables over time. Because the power for the second sample is so low, the null findings
should be interpreted cautiously.
Discussion
The results of the first portion of this second study failed to confirm my
hypothesis: long-term cancer survivors did not perform more poorly as a group than
individuals without a cancer history on tests of attention, working memory, and verbal
memory. Indeed, there were no significant group differences between long-term cancer
survivors and individuals without a cancer history on any of the neuropsychological tests,
indicating that long-term cancer survivors were no more likely to show any type of
cognitive deficits than similar individuals without a cancer history. Although this finding
was unexpected, this study was not the first to fail to find significant cognitive
differences between long-term cancer survivors and individuals without a cancer history.
Buckwalter and colleagues (2005) also failed to find cognitive problems among older
adult long-term cancer survivors, compared to individuals without a cancer history.
However, Buckwalter et al.’s study used only a gross measure of cognitive abilities,
similar to the MMSE; such gross measures of cognitive dysfunction have been suggested
to be insensitive to cognitive deficits, particularly for executive functioning deficits
(Meyers & Wefel, 2003). Additional studies that failed to find significant deficits in long-
term cancer survivors using more sensitive cognitive measures may well exist, but remain
unpublished due to the difficulty in publishing null findings.
The lack of group-by-time interactions in the analyses that focused on long-term
cancer survivors suggest that cancer survivors do not experience cognitive declines at a
more rapid rate than individuals who have never had a history of cancer. It remains

167
possible that there are subgroups of cancer survivors who experience a more rapid rate of
cognitive decline than individuals who have never had cancer. Indeed, previous
longitudinal research suggests that only subgroups of cancer survivors, rather than cancer
survivors as a whole, seem to experience cognitive declines (Schagen et al., 2006; Mar
Fan et al., 2005; Wefel et al., 2004b) or neurological changes after cancer treatment
(Fliessbach et al., 2003; Johnson et al., 1990). In this study, visual inspection of the plots
of cancer survivors’ trajectories on each neuropsychological test failed to find any
common factors, such as age, cancer type, or length of cancer survival, among survivors
who had steeper decline trajectories. Consistent with the visual inspection, analyses
found that cancer survival length, or the length of time between cancer diagnosis and 1
st

cognitive testing, was not a predictor of performance on any of the neuropsychological
tests, although it had a very modest relationship with depression scores. Future research
that has more comprehensive treatment data and that measures some of the genetic
characteristics, such as APOE genotype, that Ahles & Saykin (2007) suggest may interact
with cancer treatment, could more fruitfully explore predictors of subgroups at risk for
cognitive decline.
As noted previously, longitudinal studies of cognitive functioning in cancer
survivors have failed to show significant declines in most cancer survivors following
treatment (Anderson-Hanley et al. 2003). There is also evidence that, prior to treatment,
cancer survivors demonstrate cognitive dysfunction (Hermelink et al., 2007; Iconomou et
al., 2004; Wefel et al., 2004a; Komaki et al., 1995; van Oosterhout et al., 1995), raising
the possibility that cognitive deficits in cancer survivors are pre-existing rather than due
to treatment. This possibility would explain the lack of significant cognitive decline that

168
cancer survivors show in longitudinal studies, as well as the main effect for group found
in the second portion of this study.
Individuals who were diagnosed with cancer during the ADRC study performed
more poorly as a whole than individuals without a cancer history on the measure of
semantic verbal fluency (Animals). They did not, however, demonstrate more rapid
declines on this test than individuals without a cancer history, indicating that their poorer
performance on this test was not a change from pre- to post-cancer. Examination of the
intercepts shown in descriptive analyses on the semantic fluency task is consistent with
this idea, showing that cancer survivors had poorer semantic fluency prior to cancer
diagnosis than did individuals without a cancer history at their first assessment. Indeed,
although not statistically significant, cancer survivors tended to decline less over time on
semantic fluency than individuals without a cancer history. These results suggest that
cancer survivors do not experience significant declines in semantic fluency following
cancer diagnosis and treatment, but may be susceptible to semantic fluency deficits even
prior to cancer diagnosis.
Because cancer survivors also showed greater levels of depression as a group than
individuals without a cancer history, the possibility that depression confounded the
association between cancer and poorer semantic fluency was explored. Previous research
has certainly demonstrated a link between depressive symptoms and deficits in semantic
fluency (Henry & Crawford, 2005). However, because semantic fluency is not uniquely
affected in depressed individuals, some researchers have speculated that diminished
semantic fluency may be a marker of general cognitive dysfunction in depression (Henry
& Crawford, 2005). This idea is consistent with the research suggesting that depression is

169
prodromal for dementia (Jorm, 2001), an idea which, if true here, would indicate that the
cancer survivors were demonstrating serious risk for dementia even before their cancer
diagnosis. If this idea were true for this sample, however, multiple domains of cognitive
functioning would be expected to be diminished; cancer survivors as a group would have
performed more poorly than individuals without a cancer history on most of the
neuropsychological tests. In the present study, individuals who were diagnosed with
cancer during the ADRC study did not show group effects indicating poorer performance
on any of the other neuropsychological tests, suggesting that the difficulty with semantic
fluency was not a marker for general cognitive dysfunction among these cancer
survivors.
Within this sample, examining the effects of depressive symptoms and recent
cancer diagnosis on semantic fluency yielded unremarkable results. Neither cancer nor
depressive symptoms were significant predictors of semantic fluency when both were
entered into a single model. The fact that both depression and cancer were barely
rendered non-significant predictors (p<.10) of semantic fluency when entered into a
model together, and that their β weights changed little, suggests that poor statistical
power was the primary factor in making the predictors non-significant, a suggestion that
the power analysis supports. Depression therefore does not appear to be a compelling
cause of the relationship between cancer and semantic fluency that existed even before
treatment.
However, it is unclear what can account for this relationship between cancer and
semantic fluency, given that cancer survivors performed more poorly than the
comparison group even prior to their cancer diagnosis. Cimprich and colleagues (2004)

170
investigated several pre-treatment factors that could conceivably influence cognitive
functioning in breast cancer survivors, including age, education, menopausal status, and
other chronic health problems. When examining all these predictors together, the authors
found only age and education to be significant predictors of cognitive problems
(Cimprich et al., 2004), factors that were controlled for in the present study through
matching. Because the analyses in the second part of the present study did not adjust for
the number of significance tests run, it remains possible that the finding of diminished
semantic fluency scores among cancer survivors is merely the result of chance; therefore,
and, in light of the lack of compelling explanations, this finding should not be considered
conclusive. Rather, future research may benefit from exploring reasons that could put
cancer survivors at increased risk for such cognitive problems even prior to cancer
treatment. In a thoughtful review, Ahles and Saykin (2007) recently proposed several
genetic factors common to cancer and cognitive problems, including “deficits in DNA-
repair mechanisms that lead to greater DNA damage, and deregulation of the immune
response,” (p. 199) that could be mechanisms of such cognitive problems. The present
study’s findings of semantic fluency deficits that exist prior to cancer diagnosis suggest
that such mechanisms are deserving of future research attention.
One interesting finding that emerged from this second study was the significant
group by time interaction on the verbal learning test, Word List Learning. Individuals
who were diagnosed with cancer during their participation in the ADRC study declined
more rapidly on the verbal learning task than individuals without a cancer history. While
verbal memory deficits have been found frequently in previous studies comparing cancer
survivors to norms or control groups, they have not consistently been found in

171
longitudinal studies (see, e.g., Anderson-Hanley, 2003, for a meta-analytic review). In
addition, the present study found that cancer survivors declined more rapidly on the
learning section of the Word List task, rather than the memory portion. Although learning
is clearly related to memory, they are distinct constructs—with learning understood as the
process of encoding and storing information, and memory understood as the storage and
retrieval of information (Squire & Zola-Morgan, 1991). These different constructs may
have different clinical and research implications.
Adequate verbal learning is important for verbal memory, but the two constructs
can be affected by separate processes (Fletcher et al., 1998; Squire & Zola-Morgan, 1991;
Zola-Morgan & Squire, 1993). For instance, verbal learning can be diminished by poor
learning strategies such as a failure to organize information (Fletcher et al., 1998), or by
amnestic deficits that prevent learning of new verbal information from taking place
(Milner et al., 1998). These functional distinctions belie some anatomical distinctions,
although, as noted in the introduction, significant overlap exists between neurological
systems. Generally, cancer survivor’s more rapid rate of decline on the word list learning
task compared to individuals without a cancer history would appear to implicate
dysfunction in the medial temporal lobe systems, which are instrumental in learning new
information (Fernandez et al., 1998; Squire, 1992). Damage to the temporal lobe system
can result in amnestic memory deficits, characterized by a pronounced difficulty learning
new information and rapid forgetting (Milner et al., 1998). However, due to the role the
frontal lobe systems appear to play in word list learning strategies (Fletcher et al., 1998;
Stuss & Benson, 1986), dysfunction in these systems could also result in the observed
decline in cancer survivors’ verbal learning skills (Fletcher et al., 1998). Without

172
significant differences between cancer survivors and cancer history-free individuals in
rate of decline on other measures of neuropsychological functioning that might
demonstrate a pattern of deficits, the present findings are suggestive of temporal system
dysfunction, but neither specific nor conclusive.
Considering the close relationship between learning and memory, the observed
accelerated decline in verbal learning among cancer survivors is consistent with cancer
survivors’ self-reports of memory difficulties (Bender et al., 2006; Schagen et al., 1999;
van Dam et al., 1998; Wieneke & Dienst, 1995; Johnson et al., 1985). Cancer survivors,
and indeed many people who perceive cognitive problems, may lack the clinical
sophistication necessary to distinguish between learning and memory difficulties, and
simply lump them together as “memory problems.” The present findings should therefore
not be construed as irrelevant to these voiced memory complaints, even though the cancer
survivors did not differ significantly from the comparison group in rate of decline on the
measures of recall. Instead, cancer survivors who are concerned about their memory
should have their concerns validated, and receive thorough neuropsychological
evaluations that include measures of learning and memory.
Several plausible treatment mechanisms exist that previous research suggests
could have produced the learning difficulties observed in the present study.
Chemotherapy in particular has been the subject of much investigation in this area, with
researchers often finding memory problems in cancer patients who were treated with
chemotherapy (Downie et al., 2006; Castellon et al., 2004; Ahles et al., 2002; Brezden et
al., 2000; Schagen et al., 1999; van Oosterhout et al., 1996; Wieneke & Dienst, 1995;
Meyers et al., 1992; Kaasa et al., 1988). Emerging animal research provides additional

173
evidence that chemotherapy can cause memory problems (Winocur et al., 2006). One
experimental study randomly assigned healthy mice to receive combination
chemotherapy (methotrexate and 5-fluorouracil) or to an untreated control group, and
found that the mice treated with chemotherapy took longer to learn a spatial memory
task, and made more errors, than the control mice, as well as performing more poorly on
a conditional rule learning task (Winocur et al., 2006). Another treatment modality that is
beginning to garner research attention for its potential relationship with cognitive deficits
is hormone treatment, which reduces sex hormone levels that could otherwise contribute
to cancer growth. A brief longitudinal study by Green and colleagues (2002) compared
older adult prostate cancer patients randomly assigned to treatment with one of two types
of hormone therapy or close monitoring of the cancer with no treatment. They found that
men treated with hormone therapy showed declines over time on a word list learning and
recall task, while the performance of men who weren’t treated remained stable (Green et
al., 2002). Other studies, however, have found no significant memory or other cognitive
deficits in patients treated with hormone therapy (Salminen et al., 2003; Schagen et al.,
2002; Schagen et al., 1999; van Dam et al., 1998), rendering the relationship between
hormone treatment and cognitive functioning unclear, and in need of continued research.
None of the individuals who were diagnosed with cancer during the ADRC study
were known from the CSP records to have received systemic treatment such as
chemotherapy or hormone therapy. Because both chemotherapy and hormone therapy
were recommended as treatments for twelve of the patients, but refused, it is possible that
these patients later received chemotherapy or hormone therapy. However, it is impossible
to know with any certainty whether these patients were exposed to systemic treatments.

174
Without confirmation that these patients received systemic treatment, ideas about
treatment-related mechanisms for the present study’s findings are highly speculative.
The limited amount of treatment data available was not entirely unexpected, since
only treatments given within the first 4 months following diagnosis are reported to the
CSP. However, an additional, unexpected, factor contributing to the unavailability of
treatment data for the long-term cancer survivors was that a fairly large number of the
long-term survivors were identified through the ADRC’s intake interview, rather than
through the CSP. These individuals were not within the CSP’s dataset, possibly because
they had been diagnosed with cancer outside of Los Angeles County or before 1972, and
because the ADRC intake interview does not inquire about cancer treatment, those data
were not available. Regardless of the reasons why the present study had no cancer
survivors who were known to be treated with chemotherapy, this limitation meant that the
relationship between cognitive problems and cancer treatment could not be explored. My
inability to meaningfully investigate potential treatment effects on neuropsychological
functioning is a disappointing limitation of this study.
An additional limitation of this study is that, although one of the larger research
questions that this study was attempting to address was whether cancer treatments appear
to result in discrete, lasting neurological damage or a process of progressive neurological
decline, pre-cancer baseline assessments of neuropsychological functioning were
unavailable for the long-term cancer survivors. This limitation would have required
extrapolating findings backwards in time to infer whether it appeared, from a main effect
for group or a group x time interaction, that cancer survivors experienced discrete
neurological damage, or whether it appeared, from a group x time interaction, that cancer

175
survivors experienced accelerated neurological decline. This inferential approach, and the
problems associated with it, turned out to be somewhat irrelevant, since the long-term
cancer survivors did not differ either in group effect or rate of change from their
comparison group, suggesting that, regardless of whatever cognitive changes may have
occurred soon after cancer treatment, no significant cognitive deficits persisted.
The statistical modeling techniques used in this study conferred both advantages
and disadvantages. The advantage, which has been noted previously, is that the modeling
techniques allowed for the incorporation of a good deal more data than did the more
traditional repeated measure ANOVAs. They accurately reflect the descriptive individual
change analyses, and are therefore believed to be better representations of the data than
the repeated measure ANOVAs. However, due to the matching, and the additional data
dependencies that were created as a result, accurately specifying the models to account
for these data dependencies became a difficult statistical task that even experts in
biostatistics appear to have difficulty resolving with certainty. A brief consultation with
an expert in longitudinal data analysis indicated that the approach used in this study was a
“good way to deal with the problem” (J. McArdle, personal communication, August 26,
2007), suggesting that this difficulty was resolved as well as possible, and that the models
generated were trustworthy.
This study has a number of strengths. As with the first study, this second study is
one of the few that has focused upon long-term cancer survivors in older adulthood
(others include Hurria et al., 2006; Buckwalter, Crooks, & Petitti, 2005). It is the first
study in my knowledge to examine the trajectory of cognitive functioning in older adult
cancer survivors, and may thus represent a first step towards understanding the course of

176
cognitive functioning in older adult long-term cancer survivors. This study also provided
a unique opportunity, albeit with a small sample, to examine changes in cognitive
functioning following cancer treatment compared to a baseline that was obtained prior to
cancer diagnosis. Previous studies that included multiple measures of neuropsychological
functioning had not only similarly small samples (Wefel et al., 2004; Oxman &
Silverfarb, 1980; van Oosterhout et al, 1996), but also relied on baseline assessments
acquired after cancer was diagnosed. These previous studies’ post-diagnosis baselines
may be affected by the stress of being recently diagnosed with cancer or by immediate
disease-related factors. Due to these strengths, this study is believed to contribute a small
piece towards a better understanding of cognitive functioning in long-term cancer
survivors, despite its limitations.
Chapter 4: General Discussion
Taken together, the findings from Studies 1 and 2 suggest several answers to my
initial hypotheses and questions. Firstly, both samples of long-term cancer survivors
within Study 1 found significant cognitive problems in cancer survivors after controlling
for potentially confounding factors, which suggest that cognitive problems found in long-
term cancer survivors are not the result of confounding variables. Secondly, recognition
memory and verbal functioning were found to be poorer in one sample of older adult
long-term cancer survivors, suggesting that these areas—and the neurological systems
most directly responsible for producing these abilities—might be particularly sensitive to
cancer-related cognitive problems. The second sample of long-term cancer survivors,
however, did not perform more poorly than individuals without a cancer history on any of
the neuropsychological tests examined, which raises the question of whether

177
methodological factors could have differentially influenced the results of the two studies.
Finally, word list learning ability appeared to decline at a more rapid rate after
individuals were diagnosed with cancer than it did for individuals without a cancer
history. This finding suggests that older adults diagnosed with and treated for cancer may
experience a more rapid rate of decline in verbal learning abilities.
The null findings from the long-term cancer survivors in Study 2 differed from
those in Study 1, which invited further investigation to adequately understand the
meaning of the results. These differences in results between the two studies occurred
despite some similarities in research design: both studies focused primarily upon older
adults, focused upon long-term cancer survivors, controlled for demographic factors,
excluded brain cancer and non-melanoma skin cancer survivors, and included an
otherwise heterogeneous group of cancer survivors. To investigate differences between
the Study 1 and 2 samples, comparisons between the two groups of long-term cancer
survivors were conducted on several of the variables that were common to the two
studies. For these comparisons, data from the first assessment were used for the ADRC
long-term cancer survivors.
The two groups of long-term cancer survivors did not differ significantly in age
(t(111)=0.68, p=0.50), with the mean age of Swedish Twin cancer survivors being non-
significantly older than that of ADRC cancer survivors (76.79 vs. 75.86 years). The two
groups did differ significantly in level of education (t(105)=-11.43, p<0.0001); long-term
cancer survivors in the ADRC had completed more years of education than the Swedish
Twin cancer survivors (13.74 vs. 7.45 years). This finding is largely the result of different

178
educational standards between the United States and Sweden. Specifically, Sweden
required 7 years of basic education at the time the study’s sample was in school.
Table 53 displays the results of the comparisons between the two groups on
neuropsychological tests. The two samples differed significantly on four out of the five
common neuropsychological tests. Long-term cancer survivors from the ADRC
demonstrated better performance than the Swedish Twin long-term cancer survivors on
all three subtests of the Word List, and on Figure Copying. Long-term cancer survivors
from the ADRC also demonstrated better mean performance on Animals, but this finding
did not reach statistical significance.
Table 53. Comparing cognitive test performance of Swedish Twin long-term cancer survivors to
ADRC long-term cancer survivors at Time 1

Swedish Twin
Cancer Survivors
ADRC Cancer
Survivors
Cognitive Test N Mean SD N Mean SD t df
P-
value
Word List Learning 42 11.69 7.84 39 15.31 7.31 -2.14 79 0.04
Word List Delayed Recall 43 2.93 2.86 39 4.26 3.03 -2.04 80 0.04
Word List Recognition 43 14.54 5.86 38 17.47 3.50 -2.77 69.8 0.007
Figure Copying 42 6.81 3.37 41 8.54 2.52 -2.65 76 0.01
Animals 41 12.05 7.02 59 14.68 8.12 -1.68 98 0.10

The poor cognitive performance of the Swedish Twin long-term cancer survivors
is not surprising, given that they were largely selected to undergo cognitive testing
because they demonstrated cognitive dysfunction. In addition, the Swedish Twin cancer
survivors had slightly less education, factors that are associated with poorer cognitive
performance. However, the ADRC sample is also comprised mostly of a group of
individuals at high risk for cognitive dysfunction, and specifically, AD. Indeed, there
were similar rates of dementia between the cancer survivors from the ADRC (at their first
cognitive testing) and the Swedish Twin cancer survivors (25.0% vs. 27.9%).

179
It is unknown why, given these similarities, the results from the long-term cancer
survivors in Study 1 were not replicated with the sample of long-term cancer survivors in
Study 2. Perhaps the most substantial difference between the two studies was that the
Swedish Twin long-term cancer survivors were a population-based sample, whereas the
long-term cancer survivors within the ADRC were from a sample of community
volunteers. A population-based sample, which undertook cognitive testing of everyone
within the population of older adult Swedish Twins who consented to it, would be
expected to have fewer selection biases than a study using community volunteers, which
relied upon individuals concerned about their cognitive functioning to present themselves
to the ADRC.
Of note, the results of both studies suggest possible temporal lobe dysfunction
(Lezak et al., 2004; Nolte, 2002). Although it is unclear why measures of verbal memory
and recognition were not diminished in the cancer survivors in Study 2, the overlap
between the brain systems implicated in the two studies is interesting. Future research
will be important in better understanding if long-term cancer survivors are, indeed, at risk
for dysfunction in this system in older age, and, if so, what the mechanisms for such
dysfunction may be.
These studies may have clinical implications for long-term cancer survivors. The
present studies’ findings of poorer cognitive functioning among long-term cancer
survivors validate cancer survivors who report experiencing cognitive problems. In
addition, the cognitive problems found here occurred after controlling for depression. The
presence of cognitive dysfunction after controlling for depression indicates that cancer
survivors are not experiencing cognitive problems solely as a result of depression.

180
Indeed, it is possible that cancer survivors who are distressed and who report cognitive
problems, but who do not demonstrate cognitive deficits on neuropsychological tests
(e.g., Castellon et al., 2004), may be more sensitive to their own deficits than are the
neuropsychological tests. It would be important for future research to address this
possibility through longitudinal research that measures self-report of cognitive
functioning along with neuropsychological testing.
The results of the first study appear to exonerate cancer survivors from the
suggestion that, through behavioral or lifestyle choices they made, they are responsible
for any cognitive deficits they may experience after cancer. These findings, along with
increased rate of decline on the verbal learning task for individuals diagnosed with cancer
during the ADRC study, also suggest that researchers should continue to focus
investigations into potential mechanisms of cognitive dysfunction on factors more
directly related to cancer, such as treatment or systemic physiological changes that occur
when individuals have malignant cancer. In light of the finding that cancer survivors had
poorer semantic fluency prior to cancer diagnosis, however, research should also
continue to investigate pre-treatment factors that could increase cancer patients’ risk for
cognitive deficits.
Although imperfect, these studies addressed several important gaps in the
literature on neuropsychological deficits in cancer survivors, and highlighted areas in
which future research would be helpful. In particular, the phenomenon of cognitive
functioning in cancer survivors would benefit from future research in identifying and
predicting subgroups at greatest risk for cognitive problems. Identifying the subgroups at
greatest risk for cognitive problems could conceivably alert the cancer survivors who are

181
at risk for cognitive dysfunction to be vigilant for cognitive problems and to learn
cognitive compensation strategies or engage in cognitive remediation to cope with
cognitive problems. It might also allow individuals who are not at high risk to worry less
about their cognitive functioning, and could allow medical professionals to focus
resources on individuals who need them most. Finally, future research should explore the
mechanisms through which cancer and its treatment might result in cognitive problems
broadly, and in learning and memory problems, and temporal lobe system damage, more
specifically.

182
References
Adams, J.H. (1984). Head injury. In J.H. Adams, J.A.N. Corsellis, & L.W. Duchen
(Eds.), Greenfield’s Neuropathology. New York: Wiley.

Afifi, A.K., Bergman, R.A. (1998). Functional neuroanatomy: Text and atlas. New
York, NY: McGraw-Hill Companies, Inc.

American Psychiatric Association. (1994). Diagnostic and statistical manual of
mental disorders (4
th
Ed.). Washington, D.C.: American Psychiatric Association.

Ahles, T.A., & Saykin, A.J. (2007). Candidate mechanisms for chemotherapy-
induced cognitive changes. Nature Reviews Cancer, 7, 192-201.

Ahles, T.A., Saykin, A.J., Furstenberg, C.T., Cole, B., Mott, L.A., Skalla, K.,
Whedon, M.B., Bivens, S., Mitchell, T., Greenberg, E.R., & Silberfarb, P.M.
(2002). Neuropsychologic impact of standard-dose systemic chemotherapy in
long-term survivors of breast cancer and lymphoma. Journal of Clinical
Oncology, 20, 485-93.

Ahles, T.A., & Saykin, A.J. (2001). Cognitive effects of standard-dose chemotherapy
in patients with cancer. Cancer Investigation, 19, 812-820.

Anderson-Hanley, C., Sherman, M.L., Riggs, R., Agocha, V.B., & Compas, B.E.
(2003). Neuropsychological effects of treatments for adults with cancer: a
meta-analysis and review of the literature. Journal of the International
Neuropsychological Society, 9, 967-982.

Anstey, K., & Christensen, H. (2000). Education, activity, health, blood pressure and
Apolipoprotein E as predictors of cognitive change in old age: A review.
Gerontology, 46, 163-177.

Baile, W.F. (1996). Neuropsychiatric disorders in cancer patients. Current Opinions
in Oncology, 8, 182-7.

Bagnardi, V., Blangiardo, M., La Vecchia, C., & Corrao, G. (2001). A meta-analysis
of alcohol drinking and cancer risk. British Journal of Cancer, 85, 1700-5.

Baldo, J.V., Shimamura, A.P., Delis, D.C., Kramer, J., & Kaplan, E. (2001). Verbal
and design fluency in patients with frontal lobe lesions. Journal of the
International Neuropsychological Society, 7, 586-96.

Basso, M.R., Bornstein, R.A., & Lang, J.M (1999). Practice effects on commonly
used measures of executive functioning across twelve months. The Clinical
Neuropsychologist, 13, 283-292.

183

Baune, B.T., Suslow, T., Arolt, V., & Berger, K. (2007). The relationship between
psychological dimensions of depressive symptoms and cognitive functioning
in the elderly—The MEMO Study. Journal of Psychiatric Research, 41, 247-254.

Bender, C.M., Sereika, S.M., Berga, S.L., Vogel, V.G., Brufsky, A.M., Paraska,
K.K. (2006). Cognitive impairment associated with adjuvant therapy in breast
cancer. Psycho-Oncology, 15, 422-30.

Black, J.E., Isaacs, K.R., Anderson, B.J., Alcantara, A.A., & Greenough, W.T.
(1990). Learning causes synaptogenesis, whereas motor activity causes
angiogenesis in cerebellar cortex of adult rats. Proceedings of the National
Academy of Sciences, USA, 87, 5568-72.

Blessed, G., Tomlinson, B. E., & Roth, M. (1968). The association between
quantitative measures of dementia and of senile change in the cerebral grey
matter of elderly subjects. British Journal of Psychiatry, 114, 797-811.

Braaten, T., Weiderpass, E., Kumle, M., & Lund, E. (2005). Explaining the
socioeconomic variation in cancer risk in the Norwegian women and cancer
study. Cancer Epidemiology, Biomarkers, and Prevention, 14, 2591-7.

Bradley, C.J., Neumark, D., Luo, Z., Bednarek, H., & Schenk, M. (2005).
Employment outcomes of men treated for prostate cancer. Journal of the National
Cancer Institute, 97, 958-965.

Brazis, P.W., Masdeu, J.C., & Biller, J. (2007). Localization in clinical neurology,
5
th
Edition. Philadelphia, PA: Lippincott Williams & Wilkins.

Brezden, C.B., Phillips, K.A., Adolell, M., Bunston, T., & Tannock, I.F. (2000).
Cognitive function in breast cancer patients receiving adjuvant chemotherapy.
Journal of Clinical Oncology, 18, 2695-701.

Brink, T.L., Yesavage, J.A., Lum, O., et al. (1982). Screening tests for geriatric
depression. Clinical Gerontologist, 37-44.

Brown, M.S., Stemmer, S.M., Simon, J.H., Stears, J.C., Jones, R.B., Cagnoni, P.J., et
al. (1998). White matter disease induced by high-dose chemotherapy:
Longitudinal study with MR imaging and proton spectroscopy. American
Journal of Neuroradiology, 19, 217-221.

Bruce, G., & Jones, B.T. (2004). A pictorial Stroop paradigm reveals an alcohol
attentional bias in heavier compared to lighter social drinkers. Journal of
Psychopharmacology, 18, 527-33.

184
Brunetti, M., Belardinelli, P., Caulo, M., Del Gratta, C., Della Penna, S., Ferretti, A.,
et al. (2005). Human brain activation during passive listening to sounds from
different locations: An fMRI and MEG study. Human Brain Mapping, 26,
251-61.

Buckwalter, J.G., Crooks, V.C., Petitti, D.B. (2005). Cognitive performance of older
women who have survived cancer. International Journal of Neuroscience,
115, 1307-14.

Busch, R.M., Booth, J.E., McBride, A., Vanderploeg, R.D., Curtiss, G., Jennifer, J.
et al. (2005). The role of executive functioning in verbal and visual memory.
Neuropsychology, 19, 171-80.

Butler, R.N., Ahronheim, J., Fillit, H., Rapoport, S.I., Tatemichi, J.K. (1993).
Vascular dementia: Stroke prevention takes on a new urgency. Geriatrics, 48,
32-4, 40-2.

Castellon, S.A., Ganz, P.A., Bower, J.E., Petersen, L., Abraham, L., Greendale, G.A.
(2004). Neurocognitive performance in breast cancer survivors exposed to
adjuvant chemotherapy and tamoxifen. Journal of Clinical and Experimental
Neuropsychology, 26, 955-969.

Cimprich, B., Heeyoung, S., Ronis, D.L., & Trask, C. (2005). Pre-treatment factors
related to cognitive functioning in women newly diagnosed with breast cancer.
Psycho-oncology, 14, 70-8.

Colcombe, S., & Kramer, A.F. (2003). Fitness effects on the cognitive function of
older adults: A meta-analytic study. Psychological Science, 14, 125-130.

Condon, B., Montaldi, D., Lindsay-Wilson, J.T., Hadley, D. (1997). The relation
between MRI neuroactivation changes and response rate on a word-fluency task.
Applied Neuropsychology, 4, 201-7.

Cummings, J.L. (2003). The neuropsychiatry of Alzheimer’s Disease and related
dementias. London, U.K: Taylor & Francis.

Cull, A., Gregor, A., Hopwood, P., Macbeth, F., Karnicka-Mlodkowska, H.,
Thatcher, N., Burt, P., Stepniewska, K., & Stewart, M. (1994). Neurological and
cognitive impairment in long-term survivors of small cell lung cancer. European
Journal of Cancer, 30a, 1067-74.

Damasio, H., & Damasio, A.R. (1989). Lesion analysis in neuropsychology. New
York, New York: Oxford University Press.

185
DeCarli, C., Murphy, D.G., Tranh, M., Grady, C.L., Haxby, J.V., Gillette, J.A. et al.
(1995). The effect of white matter hyperintensity volume on brain structure,
cognitive performance, and cerebral metabolism of glucose in 51 healthy adults.
Neurology, 45, 2077-84.

de le Monte, S.M. (1988). Disproportionate atrophy of cerebral white matter in
chronic alcoholics. Archives of Neurology, 45, 990-992.

DeLeon, J., Gottesman, R.F., Kleinman, J.T., Newhart, M., Davis, C., Heidler-Gary,
J. et al. (2007). Neural regions essential for distinct cognitive processes
underlying picture naming. Brain, 130, 1408-22.

Dishman, R.K., Berthoud, H-R., Booth, F.W., Cotman, C.W., Edgerton, R., Fleshner,
M.R., et al. (2006). Neurobiology of exercise. Obesity, 14, 345-356.

Downie, F.P., Mar Fan, H.G., Houede-Tchen, N., Yi, Qilong, & Tannock, I.F.
(2006). Cognitive function, fatigue, and menopausal symptoms in breast cancer
patients receiving adjuvant chemotherapy: Evaluation with patient interview after
formal assessment. Psycho-Oncology, 15, 921-30.

Doyon, J., & Milner, B. (1991). Right temporal-lobe contribution to global visual
processing. Neuropsychologia, 29, 343-60.

Du, A.T., Schuff, N., Chao, L.L., Kornak, J., Ezekiel, F., Jagust, W.J. et al. (2005).
White matter lesions are associated with cortical atrophy more than entorhinal and
hippocampal atrophy. Neurobiology of Aging, 26, 553-559.

Du, A.T., Schuff, N., Zhu, X.P., Jagust, W.J., Miller, B.L., Reed, B.R. et al. (2003).
Atrophy rates of entorhinal cortex in AD and normal aging. Neurology, 60,
481-6.

Eberhardt, B., Dilger, S., Musial, F., Wedding, U., Weiss, T., & Miltner, W.H.R.
(2006). Medium-term effects of chemotherapy in older cancer patients. Support
Care Cancer, 14, 216-22.

Elderkin-Thompson, V., Mintz, J., Haroon, E., Lavretsky, H., Kumar, A. (2007).
Executive dysfunction and memory in older patients with major and minor
depression. Archives of Clinical Neuropsychology, 22, 261-70.

Erdfelder, E., Faul, F., & Buchner, A. (1996). GPOWER: A general power analysis
program. Behavior Research Methods, Instruments, & Computers, 28, 1-11.

186
Fernandez, G., Weyerts, H., Schrader-Bolsche, M., Tendolkar, I., Smid, H.G.O.M.,
Tempelmann, C. et al. (1998). Successful verbal encoding into episodic memory
engages the posterior hippocampus: A parametrically analyzed functional
magnetic resonance imaging study. The Journal of Neuroscience, 18, 1841-7.

Fletcher, P.C., Frith, C.D., Grasby, P.M., Shallice, T., Frackowiak, R.S.J., & Dolan,
R.J. (1995). Brain systems for encoding and retrieval of auditory-verbal memory:
An in vivo study in humans. Brain, 118, 401-16.

Fliessbach, K., Urbach, H., Helmstaedter, C., Pels, H., Glasmacher, A., Kraus, J.A.,
Klockgether, T., Schmidt-Wolf, I., Schlegel, U. (2003). Cognitive performance
and magnetic resonance imaging findings after high-dose systemic and
intravenous chemotherapy for primary central nervous system lymphoma.
Archives of Neurology, 60, 563-8.

Folstein, M., Folstein, S., & McHugh, P. (1975). “Mini-mental state”. A practical
method for grading the cognitive state of patients for the clinician. Journal of
Psychiatric Research, 12, 189-198.

Gatz, M., Fratiglioni, Johansson, B., Berg, S., Mortimer, J.A., Reynolds. C.A., et al.
(2005). Complete ascertainment of dementia in the Swedish Twin Registry: the
HARMONY study. Neurobiology of Aging, 26, 439-47.

Gatz, M., Reynolds, C., John, R., Johansson, B., Mortimer, J.A., Pedersen, N.L.
(2002). Telephone screening to identify potential dementia cases in a population-
based sample of older adults. International Psychogeriatrics,14, 273-89.

Gatz, M., Reynolds, C., Nikolic, J., Lowe, B., Karel, M., Pedersen, N. (1995). An
empirical test of telephone screening to identify potential dementia cases.
International Psychogeriatrics, 7, 429-37.

Goodman, C.R. & Zarit, S.H. (1995). Ecological measures of cognitive functioning:
A validation study. International Psychogeriatrics, 7, 39-50.

Gordis, L. (2000). Epidemiology, 2
nd
Edition. Philadelphia, PA: W.B. Saunders
Company.

Grant, W.B. (2005). Re: Cancer as a risk factor for long-term cognitive deficits and
dementia. Journal of the National Cancer Institute, 97, 1549.

Green, H.J., Pakenham, K.I., Headley, B.C., Yaxley, J., Nicol, D.L., Mactaggart,
P.N. et al. (2002). Altered cognitive function in men treated for prostate cancer
with luteinizing hormone-releasing hormone analogues and cyproterone acetate: a
randomized controlled trial. British Journal International, 90, 427-32.

187
Green, S.B. (1991). How many subjects does it take to do a regression analysis?
Multivariate Behavioral Research, 26, 449-510.

Gregory, C., Lough, S., Stone, V., Erzinclioglu, S., Martin, L., Baron-Cohen, S. et al.
(2002). Theory of mind in patients with frontal variant frontotemporal dementia
and Alzheimer’s disease: Theoretical and practical implications. Brain, 125, 752-
64.

Gueorguieva, R. & Krystal, J.H. (2004). Move over ANOVA: Progress in analyzing
repeated-measures data and its reflection in papers published in the Archives of
General Psychiatry. Archives of General Psychiatry, 61, 310-317.

Hamberger, M.J., Goodman, R.R., Perrine, K., & Tamny, T. (2001). Anatomic
dissociation of auditory and visual naming in the lateral temporal cortex.
Neurology, 56, 56-61.

Heflin, L.H., Meyerowitz, B.E., Lichtenstein, P., Johansson, B., Pedersen, N.L.,
Gatz, M. (2005). Cancer as a risk factor for long-term cognitive deficits and
dementia. Journal of the National Cancer Institute, 97, 854-6.

Hemminki, K. & Li, Xinjun. (2003). Level of education and risk of cancer in
Sweden. Cancer Epidemiology, Biomarkers, & Prevention, 12, 796-802.

Henry, J., & Crawford, J.R. (2005). A meta-analytic review of verbal fluency deficits
in depression. Journal of Clinical and Experimental Neuropsychology, 27, 78-
101.

Hermelink, K., Untch, M., Lux, M.P., Krelenberg, R., Beck, T., Bauerfeind, I. et al.
(2007). Cognitive function during neoadjuvant chemotherapy for breast cancer:
Results of a prospective, multicenter, longitudinal study. Cancer, 109, 1905-13.

Herzig, R.H., Wolff, S.N., Lazarus, H.M., Philips, G.L., Karanes, C., & Herzig, G.P.
(1983). High-dose cytosine arabinoside therapy for refractory leukemia.Blood, 62,
361-9.

Hildebrandt, H., Brokate, B., Eling, P., & Lanz, M. (2004). Response shifting and
inhibition, but not working memory, are impaired after long-term heavy alcohol
consumption. Neuropsychology, 18, 203-11.

Holmes, A.J., & Pizzagalli, D.A. (2007). Task feedback effects on conflict
monitoring and executive control: Relationship to subclinical measures of
depression. Emotion, 7, 68-76.

188
Howe, H.L., Xiaocheng, W., Ries, L.A.G., Cokkinides, V., Ahmed, F., Jemal, A.
(2006). Annual report to the nation on the status of cancer, 1975-2003,
featuring cancer among U.D. Hispanic/Latino populations. Cancer, 107, 1711-42.

Hughes, C. P., Berg L., Danziger, W. L., Coben, L. A., Martin, R. L. (1982). A new
clinical scale for the staging of dementia. British Journal of Psychiatry, 140, 566-
72.

Huppert, F.A. & Beardsall, L. (1993). Prospective memory impairment as an early
indicator of dementia. Journal of Clinical and Experimental Neuropsychology,
15, 805-821.

Huppert, F. A., Johnson, T., & Nickson, J. (2000). High prevalence of prospective
memory impairment in the elderly and in early-stage dementia: Findings from a
population-based study. Applied Cognitive Psychology, 14, S63-S81.

Hurria, A., Rosen, C., Hudis, C., Zuckerman, E., Panageas, K.S., Lachs, M.S., et al.
(2006). Cognitive function of older patients receiving adjuvant
chemotherapy for breast cancer: A pilot prospective longitudinal study. Journal of
the American Geriatrics Society, 54, 925-31.

Iconomou, G., Mega, V., Koutras, A., Iconomou, A.V., Kalofonos, H.P. (2004).
Prospective assessment of emotional distress, cognitive function, and quality of
life in patients with cancer treated with chemotherapy. Cancer, 101, 404-411.

Inagaki, M., Yoshikawa, E., Matsuoka, Y., Sugawara, Y., Nakano, T., Akechi, T. et
al. (2007). Smaller regional volumes of brain gray and white matter demonstrated
in breast cancer survivors exposed to adjuvant chemotherapy. Cancer, 109, 146-
56.

Janowsky, J.S., Shimamura, A.P. Kritchevsky, M., & Squire, L.R. (1989). Cognitive
impairment following frontal lobe damage and its relevance to human amnesia.
Behavioral Neuroscience, 103, 548-60.

Jason, G.W. (1990). Disorders of motor function following cortical lesions: Review
and theoretical considerations. In G.R. Hammond (Ed.), Cerebral control of
speech and limb movements. Amsterdam: Elsevier.

Jemal, A., Clegg, L.X., Ward, E., Ries, L.A.G., Wu, X., Jamison, P.M., Wingo, P.A.,
Howe, H.L., Anderson, R.N., Edwards, B.K. (2004). Annual Report to the Nation
on the Status of Cancer, 1975-2001, with a Special Feature Regarding Survival.
Cancer, 1, 3-27.

Jetter, W., Poser, U., Freeman, R.B., Jr., & Markowitsch, H.J. (1986). A verbal long-
term memory deficit in frontal lobe damaged patients. Cortex, 22, 229-242.

189

Johansson, B. (1988/89). The MIR – Memory-in-Reality Test. Stockholm, Sweden:
Psykologif`rlaget.

Jonsson, C.O., & Molander, L. (1964). Manual till CVB-skalan [Manual of the CVB
Scales]. Stockholm, Sweden: Psykologif`rlaget.

Johnson, B.E., Becker, B., Goff, W.B., II, Petronas, N., Krehbiel, M.A., Makuch,
R.W., et al. (1985). Neurologic, neuropsychologic, and computer cranial
tomography scan abnormalities in 2- to 10-year survivors of small-cell lung
cancer. Journal of Clinical Oncology, 3 (12), 1659-1667.

Johnson, B.E., Patronas, N., Hayes, W., Grayson, J., Becker, B., Gnepp, D., et al.
(1990). Neurologic, computed cranial tomographic, and magnetic resonance
imaging abnormalities in patients with small-cell lung cancer: further follow-
up of 6- to 13-year survivors. Journal of Clinical Oncology, 8 (1), 48-56.

Jorm, A.F. (2001). History of depression as a risk factor for dementia: An updated
review. Australian and New Zealand Journal of Psychiatry, 35, 776-781.

Kaasa, S., Olsnes, B.T., & Mastekaasa, A. (1988). Neuropsychological evaluation of
patients with inoperable non-small cell lung cancer treated with combination
chemotherapy or radiotherapy. Acta Oncologica, 27, 241-246.

Kahn, R. L., Pollack, M., & Goldfarb, A. I. (1961). Factors related to individual
differences in mental status of institutionalized aged. In P. Hoch & J. Zubin
(Eds.), Psychopathology of Aging (pp. 104-113), New York: Grune & Stratton.

Kalska, H., Punamaki, R.L., Makinen-Pelli, T., & Saarinen, M. (1999). Memory and
metamemory among depressed patients. Applied Neuropsychology, 6, 96-107.

Kandel, E.R., Schwartz, J.H., & Jessell, T.M. (1995). Essentials of neural science
and behavior. Norwalk, Connecticut: Appleton & Lange.

Katzman, R. (1993). Education and the prevalence of dementia and Alzheimer’s
disease. Neurology, 43, 13-20.

Kaufman, A.S., & Lichtenberger, E.O. (1999). Essentials of WAIS-III assessment.
New York, NY: John Wiley & Sons, Inc.

Kessler, R.C., Andrews, G., Mroczek, D., Ustun, B. & Wittchen, U.-L. (1998). The
World Health Organization Composite International Diagnostic Interview Short
Form (CIDI-SF). International Journal of Methods in Psychiatric Research, 7,
171-185.

190
Kieseppa, T., Tuulio-Henriksson, A, Haukka, J., van Erp, T., Glahn, D., Cannon,
T.D., Partonen, T., Kaprio, J., & Lonnqvist, J. (2005). Memory and verbal
learning functions in twins with bipolar-I disorder, and the role of
information-processing speed. Psychological Medicine, 35, 205-215.

Killgore, W.D. & Adams, R.L. (1999). Prediction of Boston Naming Test
performance from vocabulary scores: Preliminary guidelines for
interpretation. Perceptual and Motor Skills, 89, 327-337.

Kliegel, M., McDaniel, M.A., & Einstein, G.O. (2000). Plan formation, retention,
and execution in prospective memory: A new approach and age-related effects.
Memory and Cognition, 28, 1041-1049.

Kohout, F. J., Berkman, L. F., Evans, D. A., & Cornoni-Huntley, J. (1993). Two
shorter forms of the CES-D depression symptoms index. Journal of Aging and
Health, 5, 179-193.

Komaki, R., Meyers, C.A., Shin, D.M., Garden, A.D., Byrne, K., Nickens, J.A., &
Cox., J.D. (1995). Evaluation of cognitive function in patients with limited
small cell lung cancer prior to and shortly following prophylactic cranial
irradiation. International Journal of Radiation Oncology, Biology and
Physics, 33, 179-182.

Korrte, K.B., Horner, M.D., & Windham, W.K. (2002). The trail making test, part B:
Cognitive flexibility or ability to maintain set? Applied Neuropsychology, 9, 106-
109.

Kramer, A.F., Bherer, L., Colcombe, S.J., Dong, W., Greenough, W.T. (2004).
Environmental influences on cognitive and brain plasticity during aging. Journal
of Gerontology, 59A, 940-57.

Kramer, A.F., Hahn, S., Cohen, N.J., Banich, M.T., McAuley, E., Harrison, C.R., et
al. (1999). Ageing, fitness and neurocognitive function. Nature, 400, 418-9.

Kramer, J.H., Rosen H.J., Du, A.-T., Schuff, N., Hollnagel, C., Weiner, M.W. et al.
(2005). Dissociations in hippocampal and frontal contributions to episodic
memory performance. Neuropsychology, 19, 799-805.

Kropp, S., Becher, H., Nieters, A., & Chang-Claude, J. (2001). Low-to-moderate
alcohol consumption and breast cancer risk by age 50 years among women in
Germany. American Journal of Epidemiology, 154, 624-34.

Laird, N.M. (1988). Missing data in longitudinal studies. Statistics in Medicine, 7,
305-15.

191
Laird, N.M., & Ware, J.H. (1982). Random-effects models for longitudinal data.
Biometrics, 38, 963-974.

Le Carret, N., Lafont, S., Letenneur, L., Dartigues, J-F., Mayo, W., & Fabrigoule, C.
(2003). The effect of education on cognitive performances and its implication for
the constitution of the cognitive reserve. Developmental Neuropsychology, 23,
317-337.

Lee, J.H. (2003). Genetic evidence for cognitive reserve: Variations in memory and
related cognitive functions. Journal of Clinical and Experimental
Neuropsychology, 25, 594-613.

Leris, C., & Mokbel, K. (2001). The prevention of breast cancer: an overview.
Current Medical Research and Opinion, 16, 252-7.

Lezak, M.D., Howieson, D.B., & Loring, D.W. (2004). Neuropsychological
assessment. New York, NY: Oxford University Press.

Liebenthal, E., Binder, J.R., Spitzer, S.M., Possing, E.T., Medler, D.A. (2005).
Neural substrates of phonemic perception. Cerebral Cortex, 15, 1621-1631.

Littell, R.C., Milliken, G.A., Stroup, W.W., & Wolfinger, R.D. (1996). SAS system
for mixed models. Cary, NC: SAS Institute.

Malloy, P.F., Cohen, R.A., & Jenkins, M.A. (1998). Frontal lobe function and
dysfunction. In: P.J. Snyder & P.D. Nussbaum, (Eds.). Clinical
neuropsychology: A pocket handbook for assessment. (1998). Washington,
D.C.: American Psychological Association.

Manns, J.R., Hopkins, R.O., Reed, J.M., Kitchener, E.G., & Squire, L.R. (2003).
Recognition memory and the human hippocampus. Neuron, 37, 171-180.

Mar Fan, H., Houese-Tchen, N., Yi, Q.L., Chemerynsky, I., Downie, F.P., Sabate,
K., et al. (2005). Fatigue, menopausal symptoms, and cognitive function in
women after adjuvant chemotherapy for breast cancer: 1- and 2- year follow-up of
a prospective controlled study. Journal of Clinical Oncology, 23, 8025-8032.

Mar Fan, H.G., Vardy, J., Xu, W., & Tannock, I.F. (2006). Reply to: Does cognitive
impairment after chemotherapy for breast cancer improve over time or does
practice make perfect? Journal of Clinical Oncology, 24, 5171-5172.

Martinez, M.E. (2005). Primary prevention of colorectal cancer: lifestyle, nutrition,
exercise. Recent Results in Cancer Research, 166, 177-211.

192
Massie, M.J. (2004). Prevalence of depression in patients with cancer. Journal of the
National Cancer Institute Monographs, 32, 57-71.

Matarazzo, J.D. & Herman, D.O. (1984). Base rate data for the WAIS-R: Test-retest
stability and VIQ-PIQ differences. Journal of Clinical Neuropsychology, 6, 351-
366.

Meyers, C.A. (2000). Neurocognitive dysfunction in cancer patients. Oncology, 14,
75-81.

Meyers, C.A., & Abbruzzese, J.L. (1992). Cognitive functioning in cancer patients:
Effect of previous treatment. Neurology, 42, 434-6.

Meyers, C.A., & Wefel, J.S. (2003). The use of the Mini-Mental State Examination
to assess cognitive functioning in cancer trials: No ifs, ands, or buts, or
sensitivity. Journal of Clinical Oncology, 20, 485-93.

Milner, B., Squire, L. R., & Kandel, E.R. (1998). Cognitive neuroscience and the
study of memory. Neuron, 20, 445-468.

Morley, G.K., Lundgren, S., & Haxby, J. (1979). Comparison and clinical
applicability of auditory comprehension scores on the Behavioral Neurology
Deficit Evaluation, Boston Diagnostic Aphasia Examination, Porch Index of
Communicative Ability, and Token Tests. Journal of Clinical
Neuropsychology, 1, 249-258.

Morris, J. C., Heyman, A., Mohs, R. C., Hughes, J. P., van Belle, G., Fillenbaum, G.,
Mellits, E. D., & Clark, C. (1989). The consortium to establish a registry for
Alzheimer’s disease (CERAD): I. Clinical and neuropsychological assessment of
Alzheimer’s disease. Neurology, 39, 1159-1165.

Moser, E.B. (May, 2004). Repeated measures modeling with PROC MIXED.
Proceedings of the 29
th
SAS Users Group International Conference. Montreal,
Canada.

Mukamal, J., Kuller, L.H., Fitzpatrick, A., Longstreth, W.T., Mittleman, M.A.,
Siscovick, D.S. (2003). Prospective study of alcohol consumption and risk of
dementia in older adults. Journal of the American Medical Association, 1405-
1413.

Nolte, J. (2002). The human brain: An introduction to its functional anatomy. St.
Louis, MO: Mosby, Inc.

Olson, I.R., Plotzker, A., & Ezzyat, Y. (2007). The enigmatic temporal pole: A
review of findings on social and emotional processing. Brain, 130, 1718-31.

193

Oxman, T.E. & Silberfarb, P.M. (1980). Serial cognitive testing in cancer patients
receiving chemotherapy. American Journal of Psychiatry, 137, 1263-1265.

Perry, R.J., & Hodges, J.R. (2000). Differentiating frontal and temporal variant
frontotemporal dementia from Alzheimer’s disease. American Academy of
Neurology, 54, 2277-84.

Pfeiffer, E. (1975). A Short Portable Mental Status Questionnaire for the assessment
of organic brain deficits in elderly patients. Journal of the American Geriatrics
Society, 23, 433-441.

Pihlajamaki, M., Tanila, H., Ka-Snninen, T., Kononen, M., Laakso, M., Partanen, K.
et al. (2000). Verbal fluency activates the left medial temporal lobe: A functional
magnetic resonance imaging study. Annals of Neurology, 47,
470-6.

Qualls, C.E., Bliwise, N.G., & Stringer, A.Y. (2000). Short forms of the Benton
Judgment of Line Orientation Test: Development and psychometric properties.
Archives of Clinical Neuropsychology, 15, 159-163.

Raison, C.L, & Miller, A.H. (2003). Depression in cancer: new developments
regarding diagnosis and treatment. Society of Biological Psychiatry, 54, 283-94.

Rascovsky, K., Salmon, D.P., Ho, G.J., et al. (2002). Cognitive profiles differ in
autopsy-confirmed frontotemporal dementia and AD. Neurology, 5, 1801-8.

Reid, L.M., & MacLullich, A.M.J. (2006). Subjective memory complaints and
cognitive impairment in older people. Dementia and Geriatric Cognitive
Disorders, 22, 471-85.

Reischies, F.M. (1998). Age-related cognitive decline and the dementia threshold. In
J. Lomranz (Ed.) Handbook of Aging and Mental Health: An Integrative
Approach, (pp. 435-448). New York: Plenum Press.

Richards, M.A., Stockton, D., Babb, P., & Coleman, M.P. (2000). How many deaths
have been avoided through improvements in cancer survival? British Medical
Journal, 320, 895-8.

Roman, G.C., Erkinjuntti, T., Wallin, A., Pantoni, L., & Chui, H.C. (2002).
Subcortical ischaemic vascular dementia. Neurology, 1, 426-436.

Ropacki, M.T., & Elias, J.W. (2003). Preliminary examination of cognitive reserve
theory in closed head injury. Archives of Clinical Neuropsychology, 18, 643-
54.

194

Rosenstein, L.D. (1998). Differential diagnosis of the major progressive dementias
and depression in middle and late adulthood: a summary of the literature of
the early 1990s. Neuropsychology Review, 8, 109-167.

Ryan, J.J., Prifitera, A., & Powers, L. (1983). Scoring reliability on the WAIS-R.
Journal of Consulting and Clinical Psychology, 51, 149-150.

Salminen, E., Portin, R., Korpela, J., Backman, H., Parvinen, L.-M., Helenius, H. et
al. (2003). Androgen deprivation and cognition in prostate cancer. British Journal
of Cancer, 89, 971-6.

Satz, P. (1993). Brain reserve capacity on symptom onset after brain injury: A
formulation and review of evidence for the threshold theory. Neuropsychology, 7,
273-295.

Sawrie, S.M., Martin, R.C., Gilliam, F.G., Faught, R.E., Maton, B., Hugg, J.W. et al.
(2000). Visual confrontation naming and hippocampal function: A neural network
study using 1H magnetic resonance spectroscopy. Brain, 123, 770-80.

Scarmeas, N., & Stern, Y. (2003). Cognitive reserve and lifestyle. Journal of Clinical
and Experimental Neuropsychology, 25, 625-33.

Schagen, S.B., Muller, M.J., Boogerd, W., Mellenbergh, G.J., & van Dam, F.S.A.M.
(2006). Change in cognitive function after chemotherapy: a prospective
longitudinal study in breast cancer patients. Journal of the National Cancer
Institute, 98, 1742-1745.

Schagen, S.B., Muller, M.J., Boogerd, W., Rosenbrand, R.M., van Rhijn, D.,
Rodenhuis, S., & van Dam, F.S.A.M. (2002). Late effects of adjuvant
chemotherapy on cognitive function: a follow-up study in breast cancer patients.
Annals of Oncology, 13, 1387-97.

Schagen, S.B., Hamburger, H.L., Muller, M.J., Boogerd, W., van Dam, F.S.A.M.
(2001). Neuropsysiological evaluation of late effects of adjuvant high-dose
chemotherapy on cognitive function. Journal of Neuro-Oncology, 51, 159-165.

Schagen, S.B., van Dam, F.S.A.M., Muller, M.J., Boogerd W., Lindeboom J.,
Bruning P.F. (1999). Cognitive deficits after postoperative adjuvant
chemotherapy for breast carcinoma. Cancer, 85, 640-50.

Schmolck, H., Kensinger, E.A., Corkin, S., & Squire, L.R. (2002). Semantic
knowledge in patient H.M. and other patients with bilateral medial and lateral
temporal lobe lesions. Hippocampus, 12, 520-533.

195
Sesso, H.D., Paffenbarger, R.S., & Lee, I-M. (2001). Alcohol consumption and the
risk of prostate cancer: The Harvard Alumni Health Study. International Journal
of Epidemiology, 30, 749-55.

Shallice, T., Fletcher, P., Frith, C.D., et al. (1994). Brain regions associated with
acquisition and retrieval of verbal episodic memory. Nature, 368, 633-635.

Silberfarb, P.M. (1983). Chemotherapy and cognitive deficits in cancer patients.
Annual Review of Medicine, 34, 35-46.

Silberfarb, P.M., & Oxman, T.E. (1988).The effects of cancer therapies on the
central nervous system. Advances in Psychosomatic Medicine, 18, 13-25.

Silberfarb, P.M., Philibert, D., & Levine, P.M. (1980). Psychosocial aspects of
neoplastic disease: II. Affective and cognitive effects of chemotherapy in
cancer patients. American Journal of Psychiatry, 137, 597-601.

Silverman, D.H.S., Dy, C.J., Castellon, S.A., Lai, J., Pio, B.S., Abraham, L., et al.
(2007). Altered frontocortical, cerebellar, and basal ganglia activity in adjuvant-
treated breast cancer survivors 5-10 years after chemotherapy. Breast Cancer
Research and Treatment, 103, 303-11.

Singer, J.D., & Willett, J.B. (2003). Applied longitudinal data analysis: Modeling
change and event occurrence. New York, New York: Oxford University
Press.

Snell, R.S. (2001). Clinical neuroanatomy for medical students, 5
th
Edition.
Baltimore, Maryland: Lippincott Wlliams & Wilkins.

Snow, W.G., Tierney, M.C., Zorzitto, M.L., et al. (1988). WAIS-R test-retest
reliability of selected neuropsychological tests in older adults. Journal of Clinical
and Experimental Neuropsychology, 10, 60.

Smith, A. (1982). Symbol digit modalities test (SDMT). Manual. (Revised). Los
Angeles: Western Psychological Services.

Squire, L.R. (1992). Memory and the hippocampus: A synthesis from findings with
rats, monkeys, and humans. Psychological Review, 99, 195-231.

Squire, L.R., Stark, C.E.L., & Clark, R.E. (2004). The medial temporal lobe. Annual
Review of Neuroscience, 27, 279-306.

Squire, L.R., & Zola-Morgan, S. (1991). The medial temporal lobe memory system.
Science, 253, 1380-6.

196
Stein, C.J., & Colditz, G.A. (2004). Modifiable risk factors for cancer. British
Journal of Cancer, 90, 299-303.

Stern, Y. (2003). The concept of cognitive reserve: A catalyst for research. Journal
of Clinical and Experimental Neuropsychology, 25, 589-93.

Stern, Y. (2002). What is cognitive reserve? Theory and research application of the
reserve concept. Journal of International Neuropsychological Society, 8, 448- 60.

Stuss, D.T., & Benson, D.F. (1986). The frontal lobes. New York: NY, Raven Press.

Sullivan, E.V., Fama, R., Rosenbloom, M.J., & Pfefferbaum, A. (2002). A profile of
neuropsychological deficits in alcoholic women. Neuropsychology, 16, 74-83.

Sullivan, E.V., Harding, A.J., Pentney, R., Dlugos, C., Martin, P.R., Parks, M.H., et
al. (2003). Disruption of frontocerebellar circuitry and function in alcoholism.
Alcohol Clinical Experiemental Research, 27, 301-309.

Sullivan, E.V., Marsh, L., Mathalon, D.H., Lim, K.O., & Pfefferbaum, A. (1995).
Anterior hippocampal volume deficits in nonamnestic, aging chronic alcoholics.
Alcohol Clinical Experiemental Research, 19, 110-122.

Tabachnic, B.G., & Fidell, L.S. (2001). Using multivariate statistics (4th Edition).
Boston, MA: Allyn & Bacon.

Tchen, N., Juffs, H.G., Downie, F.P., Ti, Q.L., Hu, H., Chemerynsky, I., Clemons,
M., Crump, M., Goss, P.E., Warr, D., Tweedale, M.E., Tannock, I.F. (2003).
Cognitive function, fatigue, and menopausal symptoms in women receiving
adjuvant chemotherapy for breast cancer. Journal of Clinical Oncology, 21, 4175-
83.

Tierney, M.C., Black, S.E., Szalai, J.P., Snow, G., Fisher, R.H., Nadon, G. et al.
(2001). Recognition memory and verbal fluency differentiate probably Alzhiemer
disease from subcortical ischemic vascular dementia. Archives of Neurology, 58,
1654-9.

Troster, A. (1998). Assessment of movement and demyelinating disorders. In: P.J.
Snyder & P.D. Nussbaum, (Eds.). Clinical neuropsychology: A pocket handbook
for assessment. (1998). Washington, D.C.: American Psychological Association.

Troy, L., McFarland, K., Littman-power, S., Kelly, B.J., Walpole, E.T., Wyld, D. &
Thomson, D. (2000). Cisplatin-based therapy: a neurological and
neuropsychological review. Psycho-oncology, 9, 29-39.

197
Tucker, J., Prior, P.F., Green, C.R., Ede, G.M., Stevenson, J.F., Gawler, J. Jamal,
G.A., Charlesworth, M., Thakkar, C.M., Patel, P., & Lister, T.A. (1989). Minimal
neuropsychological sequelae following prophylactic treatment of the central
nervous system in adult leukemia and lymphoma. British Journal of Cancer, 60,
775-780.

Tullberg, M., Fletcher, E., DeCarli, C., Mungas, D., Reed, B.R., Harvey, D.J.,
Weiner, M.W., Chui, H.C., & Jagust, W.J. (2004). White matter lesions impair
frontal lobe function regardless of their location. Neurology, 63, 246-253.

Tuxen, M.K., & Hansen, S.W. (1994). Neurotoxicity secondary to antineoplastic
drugs. Cancer Treatment Reviews, 20, 191-214.

van Dam, F.S.A.M., Schagen, S.B., Muller M.J., Boogerd W., v.d. Wall, E.,
Droogleever Fortuyn M.E., & Rodenhuis, S. (1998). Impairment of cognitive
function in women receiving adjuvant treatment for high-risk breast cancer: high-
dose versus standard-dose chemotherapy. Journal of the National Cancer Institute
90, 210-8.

van Oosterhout, A.G.M., Boon, P.J., Houx, P.J., ten Velde, G.P.M., & Twijnstra, A.
(1995). Follow-up of cognitive functioning in patients with small cell lung cancer.
International Journal of Radiation Oncology, Biology, and Physics, 31, 911-914.

van Oosterhout, A.G.M., Ganzevles, P.G.J., Wilmink, J.T., de Geus, B.W.J., van
Vonderen, R.G.M.W., & Twijnstra, A. (1996). Sequelae in long-term survivors of
small cell lung cancer. International Journal of Radiation Oncology, Biology, and
Physiology 34, 1037-1044.

van Praag, H., Kempermann, G., & Gage, F.H. (1999). Running increases cell
proliferation and neurogenesis in the adult mouse dentate gyrus. Nature
Neuroscience, 2, 266-70.

Waxman, S.G. (2000). Correlative neuroanatomy, 24
th
Edition. New York, New
York: McGraw-Hill Companies, Inc.

Wechsler, D. (1987). Wechsler Memory Scale-Revised. San Antonio: The
Psychological Corporation.

Wechsler, D. (1997). Wechsler Adult Intelligence Scale-III. San Antonio: The
Psychological Corporation.

Wefel, J.S., & Meyers, C.A. (2005). Cancer as a risk factor for dementia: a house
built on shifting sand. Journal of the National Cancer Institute, 97, 788-9.

198
Wefel, J.S., Lenzi, R., Theriault, R., Buzdar, A.U., Cruickshank, S., & Meyers, C.A.
(2004a). ‘Chemobrain’ in breast carcinoma?: A prologue. Cancer, 101, 466-
75.

Wefel, J.S., Lenzi, R., Theriault, R.L., Davis, R.N., & Meyers, C.A. (2004b). The
cognitive sequelae of standard-dose adjuvant chemotherapy in women with breast
carcinoma: results of a prospective, randomized, longitudinal trial. Cancer, 100,
2292-9.

Wefel, J.S., & Meyers, C.A. (2005). Cancer a risk factor for dementia: a house built
on shifting sand. Journal of the National Cancer Institute, 97, 788-9.

Wessinger, C.M., VanMeter, J., Tian, B., Van Lare, J., Pekar, J., & Rauschecker, J.P.
(2001). Hierarchical organization of the human auditory cortex revealed by
functional Magnetic Resonance Imaging. Journal of Cognitive Neuroscience, 13,
1-7.

Wieneke, M.H., & Dienst, E.R. (1995). Neuropsychological assessment of cognitive
function following chemotherapy for breast cancer. Psychooncology, 4, 61-6.

Winocur, G., Moscovitch, M., & Stuss, D.T. (1996). Explicit and implicit memory in
the elderly: Evidence for double dissociation involving medial temporal- and
frontal-lobe functions. Neuropsychology, 10, 57-65.

Yoshikawa, E., Matsuoka, Y., Inagaki, M., Nakano, T., Akechi, T., Kobayakawa,
M., et al. (2005). No adverse effects of adjuvant chemotherapy on hippocampal
volume in Japanese breast cancer survivors. Breast Cancer Research and
Treatment, 92, 81-4.

Zakzanis, K.K., Kaplan, E.F., & Leach, L. (1999). Neuropsychological differential
diagnosis. Lisse, Netherlands: Swets & Zeitlinger.

Zelazo, P.D., Carter, A., Reznik, J.S., Frye, D. (1997). Early development of
executive function: A problem-solving framework. Review of General
Psychology, 1, 198-26.

Zola-Morgan, S., & Squire, L.R. (1993). Neuroanatomy of Memory. Annual Reviews
of Neuroscience, 16, 547-63.

Zuccala, G., Onder, G., Pedone, C., Cesari, M., Landi, F., Bernabei, R., et al. (2001).
Dose-related impact of alcohol consumption on cognitive function in
advanced age: Results of a multicenter survey. Alcoholism: Clinical &
Experimental Research, 25, 1743-8.

Abstract (if available)

Abstract Cancer and its treatment have been associated with a range of neuropsychological deficits, but the presence and temporal pattern of such problems in long-term cancer survivors are poorly understood. Moreover, although cognitive problems are likely to become most pronounced in older age, few studies have focused upon older cancer survivors or controlled for potential confounding variables. This research investigated cognitive functioning in older adult long-term cancer survivors in two studies. Using a co-twin control design, the first study found long-term cancer survival remained a significant risk for cognitive dysfunction, controlling for exercise, education, and depression. The first study also found long-term cancer survivors performed significantly worse than matched controls on verbal and visual recognition, but not on recall, verbal functioning, visuoconstruction, or information processing speed. The second study failed to replicate this pattern of deficits, finding no differences on any neuropsychological tests between long-term cancer survivors and a matched comparison group of individuals without a cancer history. However, individuals for whom pre- and post-cancer neuropsychological assessments were available showed significant declines on word list learning, compared to matched individuals without a cancer history. Implications of these findings are discussed.

Linked assets

University of Southern California Dissertations and Theses

Conceptually similar

PDF

Cognitive functioning and dementia following cancer: A Swedish twin study

PDF

Cognitive functioning following ovarian removal before or after natural menopause

PDF

Life transitions, leisure activity engagement, and cognition among older adults

PDF

Personality and cognitive aging: are positive traits protective?

PDF

Subclinical carotid atherosclerosis, psychosocial measures, and cognitive function in middle- to older-aged adults

PDF

Multilevel influences of care engagement and long-term survival among childhood, adolescent, and young adult cancer survivors

PDF

Association of carotid artery stiffness with measures of cognition in older adults

PDF

Cardiovascular disease risk factors and cognitive function

PDF

The effects of time and space on health status of older adults in China

PDF

Long term effects of air pollution on lung function in the European community respiratory health survey

PDF

The effects of physical functioning and public stigma on psychological distress as mediated by cognitive and social factors among Korean survivors of childhood cancer

PDF

Longitudinal relationships between bereavement and physical health‐related quality of life in middle‐ to older‐aged women

PDF

Arm lymphedema in a multi-ethnic cohort of female breast cancer survivors

PDF

Factors associated with long-term-care facility residence of patients with Alzheimer's disease from Alzheimer's disease research centers of California

PDF

Cerebrovascular mechanisms of cognitive enhancement after periodized resistance training in older adults

PDF

Thinking generatively versus acting generatively: exploring the associations of generative self-concept and generative activity with cognitive function among older adults

PDF

Dementia, co-morbid depressive symptoms, and mortality in older adults: a twin study

PDF

Early childhood health experience & adult phenotype in twins

PDF

Association between informed decision-making and mental health-related quality of life in long term prostate cancer survivors

PDF

Survival trends and related outcomes of survivors of childhood and young adult cancer

Asset Metadata

Creator Heflin, Lara Heather (author)

Core Title Neuropsychological functioning in older adult long-term cancer survivors

School College of Letters, Arts and Sciences

Degree Doctor of Philosophy

Degree Program Psychology

Publication Date 12/01/2007

Defense Date 05/18/2007

Publisher University of Southern California (original), University of Southern California. Libraries (digital)

Tag cancer,cognitive,long-term,neuropsychological,OAI-PMH Harvest,older adult

Language English

Advisor Meyerowitz, Beth E. (committee chair), Chambers, Kathleen (committee member), Gatz, Margaret (committee member), Mack, Wendy J. (committee member), Schwartz, David (committee member)

Creator Email lheflin@memory.ucsf.edu

Permanent Link (DOI) https://doi.org/10.25549/usctheses-m959

Unique identifier UC1296965

Identifier etd-Heflin-20071201 (filename),usctheses-m40 (legacy collection record id),usctheses-c127-487783 (legacy record id),usctheses-m959 (legacy record id)

Legacy Identifier etd-Heflin-20071201.pdf

Dmrecord 487783

Document Type Dissertation

Rights Heflin, Lara Heather

Type texts

Source University of Southern California (contributing entity), University of Southern California Dissertations and Theses (collection)

Repository Name Libraries, University of Southern California

Repository Location Los Angeles, California

Repository Email cisadmin@lib.usc.edu