University of Southern California Dissertations and Theses

Behavioral and demographic predictors of breast cancer stage at diagnosis

(USC Thesis Other)

Behavioral and demographic predictors of breast cancer stage at diagnosis

PDF

Download a page range

Download transcript

Copy asset link

Request this asset

Transcript (if available)

Content INFORMATION TO USERS
This manuscript has been reproduced from the microfilm master. UMI
fidms the text directly from the original or copy submitted. Thus, some
thesis and dissertation copies are in typewriter face, while others may be
from any type of computer printer.
The quality of this reproduction is dependent upon the quality of the
copy submitted. Broken or indistinct print, colored or poor quality
illustrations and photographs, print bleedthrough, substandard margins,
and improper alignment can adversely afreet reproduction.
In the unlikely event that the author did not send UMI a complete
manuscript and there are missing pages, these will be noted. Also, if
unauthorized copyright material had to be removed, a note will indicate
the deletion.
Oversize materials (e.g., maps, drawings, charts) are reproduced by
sectioning the original, beginning at the upper left-hand comer and
continuing from left to right in equal sections with small overlaps. Each
original is also photographed in one exposure and is included in reduced
form at the back of the book.
Photographs included in the original manuscript have been reproduced
xerographically in this copy. Kgher quality 6” x 9” black and white
photographic prints are available for any photographs or illustrations
appearing in this copy for an additional charge. Contact UMI directly to
order.
UMI
A Bell & Howell Information Company
300 North Zed) Road, Arm Arbor MI 48106-1346 USA
313/761-4700 800/521-0600
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
BEHAVIORAL AND DEMOGRAPHIC PREDICTORS
OF BREAST CANCER STAGE AT DIAGNOSIS
by
Linda Joyce Clark
A Dissertation Presented to the
FACULTY OF THE GRADUATE SCHOOL
UNIVERSITY OF SOUTHERN CALIFORNIA
In Partial Fulfillment o f the
Requirements for the Degree
DOCTOR OF PHILOSOPHY
(Education - Counseling Psychology)
December, 1998
Copyright 1998 Linda Joyce Clark
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
UMI Number: 9931870
UMI Microform 9931870
Copyright 1999, by UMI Company. All rights reserved.
This microform edition is protected against unauthorized
copying under Title 17, United States Code.
UMI
300 North Zeeb Road
Ann Arbor, MI 48103
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
UNIVERSITY OF SOUTHERN CALIFORNIA
THE GRADUATE SCHOOL
UNIVERSITY PARK
LOS ANGELES. CALIFORNIA 90007
This dissertation, zuritten by
under the direction of ..... Dissertation
Committee, and approved by all its members,
has been presented to and accepted by The
Graduate School in partial fulfillment of re
quirements for the degree of
DOCTOR OF PHTLOSOPHY
Dean of Gi Studies
1 5 , 1999
Date
DISSERTATION COMMITTEE
Chairperson
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
u
Dedication
To my Dad,
who convinced me I could accomplish the impossible, and who c —
judiciously omitted how long that can take;
and to my Sons,
who became very impressive men despite my repeated absences to
the library and inattention during finals weeks.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
lu
Acknowledgements
The author gratefully acknowledges
Dr. Annlia Paganini-Hill for generously allowing her data to be
used for this dissertation, and
Dr. Michael D. Newcomb for his assistance in producing and
improving the structural equation models.
This work was partially supported by U. S. Public Health Service Grant
CA 17054 from the National Cancer Institute, National Institutes o f
Health, Department o f Health and Human Services, and by the
California Public Health Foundation, subcontract 050-F-8709, which is
siqiported by the California Department o f Health Services as part o f its
statewide cancer-reporting program mandated by the California Health
and Safety Code Sections 210 and 211.3.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
IV
Table o f Contents
D edication........................................................................................................ ii
Acknowledgements......................................................................................... iii
List of T ab les.................................................................................................. v
List of F igures.................................................................................................. vi
A b stract........................................................................................................... vii
Introduction..................................................................................................... 1
Background and In ço rtan ce.................................................................. 1
Theoretical Bases for Predictive M odels................................................ 10
Hypotheses............................................................................................ 23
M ethods.............................................................................................................. 29
Particq>ants............................................................................................ 30
Variables and Latent F actors.................................................................. 40
A nalyses............................................................................................... 45
R esults.............................................................................................................. 48
Model 1.................................................................................................. 48
Model 2.................................................................................................. 64
D iscussion........................................................................................................ 77
Inq)lications on Hypotheses.................................................................. 77
Other Findings...................................................................................... 85
Theoretical Im plications........................................................................ 87
Intervention Inq)Iications........................................................................ 90
Limitations, and Directions for Future Research.................................... 91
Conclusion............................................................................................ 93
B ibliogr^hy ....................................................................................................... 94
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
List of Tables
Table 1. Subject Participation, by Breast Cancer Stage at Diagnosis,
for Group 1 Subjects....................................................................... 33
Table 2. Breast Cancer Stage at Diagnosis for the Sub-San^le and the
Full S a m p le .................................................................................................. 34
Table 3. Demographic Descrgtion o f the Sub-Sample and the Full Sanqile . . . 36
Table 4. Educational Description o f the Sub-Sançle and the Full Sample .... 37
Table 5. Socioeconomic Desciption o f the Sub-Sample and the
FuU Sam ple.................................................................................... 38
Table 6. Life-Style Description of the Sub-Sançle and the FuU Sam ple..... 39
Table 7. Descr^tion o f Access to Health Care by the Sub-Sample and the
FuU Sam ple.................................................................................... 41
Table 8. Fit Indices for the FuU Sample, Model 1 .......................................... 49
Table 9. Confirmatory Factor Analysis Intercorrelations for the FuU Sample . . 53
Table 10. Fit Indices for the Sub-Sanq)le, Model 2 ............................................. 64
Table 11. Confirmatory Factor Analysis Intercorrelations
for the Sub-Sample, n = 7 2 1 ......................................................... 68
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
V I
List o f Figures
Figure 1. Theoretical Model for the Full Sample, Combining Health Beliefe
Model and Problem Behavior Theory Components.............................. 24
Figure 2. Theoretical Model for the Sub-Sample, Combining Health Beliefe
Model and Problem Behavior Theory Components.............................. 27
Figure 3. Comfirmatory Factor Analysis Model 1, Full Sam ple........................... 51
Figure 4. Structural Equation Model I, Full Sample, Section 1:
Prediction of Health Care Practices..................................................... 59
Figure 5. Structural Equation Model 1, Full Sample, Section 2:
Direct Predictors of Mammogram and Stage......................................... 61
Figure 6. Confurmatory Factor Analysis Model 2, Sub-Sample........................... 66
Figure 7. Structural Equation Model 2, Sub-Sample, Section 1:
Prediction of Health Care Practices..................................................... 72
Figure 8, Structural Equation Model 2, Sub-Sample, Section 2:
Direct Predictors of Screening and Stage............................................ 74
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
vu
Abstract
In addition to the acknowledged medical effects, behavioral and demogr^hic
influences are shown to determine the stage o f breast cancer at diagnosis. Among a
population-based sample o f2,307 women, a set of latent fectors and variables is
identified that predicts stage. The participants, 55 to 69 year old residents of Los
Angeles County, California, U.S.A, had been diagnosed with primary breast cancer,
with histological confirmation o f st%e. Thirty variables hypothesized by the Health
Belief Model (HBM) and the Problem Behavior Theory (PBT) are analyzed using
Structural Equation Modeling. Screening variables, including the method of first
detection of breast cancer symptoms and utilization of mammography, show the
strongest effects on stage. Detection by mammography predicts early stage, and
detection by self-examination predicts later stage of breast cancer. However, femily
history of breast cancer, age, socioeconomic status, obesity, and hormone use also
directly predict stage, independent of the screening variables. Health care variables,
such as having a regular physician and that physician being an obstetrician or
gynecologist, predict more mammography. The Perceived Susceptibility component
of the HBM, including family history of breast cancer and personal history of benign
breast disease, predicts more mammography and diagnosis at earlier stages o f the
disease, as hypothesized. The Barriers conçonent of the HBM includes age, racial,
religious, marital status, education, and proxy income variables. Women in lower
socioeconomic levels receive less mammography and diagnosis at more advanced
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
vm
stage. However, age and race are not barriers to early diagnosis, once the effects
of other variables and latent fectors, such as socioeconomic status and access to
health care, have been statistically partialled ou t The Conventionality-Deviance
conqwnent of the PBT is applied for the first time to breast health behavior. Among
its variables, obesity directly predicts stage and caffeine consunq)tion predicts the
method o f detection o f breast cancer symptoms. Less caffeine and more exercise,
each predict more mammography. Alcohol consumption does not predict screening
or stage. A specific efect of smoking is noted: women who smoked more had fewer
mammograms, as hypothesized.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Introduction
Full recovery from breast cancer is highly dependent upon the severity of the
disease when treatment begins. Clarification o f elements that predict the stage of
breast cancer at diagnosis would help identify women at greater risk of more severe
disease and allow targeting these women for intervention programs.
Background and Importance
According to 1997 predictions, breast cancer will strike I out of 8 women in
the United States (American Cancer Society, 1997). Great advances in diagnosis and
treatment have kept the rate o f breast cancer deaths relatively constant since 1930,
despite increases in incidence rates (American Cancer Society, 1997; Kelsey, &
Gammon, 1990). Nevertheless, breast cancer is fatal in about 40% of cases (Mettler
et aL, 1996).
Many risk fectors for breast cancer have been identified and their impact is
often e^ressed as a relative risk. Relative risk is the ratio of the incidence rate in an
exposed group relative to that in an unexposed group. A relative risk of 2.0 denotes
the ejqposure doubles the incidence of breast cancer. For exan^le, breast cancer in
men is rare and incidence in women rises steeply with age, so sex and age are strong
risk fectors (relative risk above 4). A woman with a femily history o f both a mother
and sister with breast cancer also has a relative risk above 4. Relative risks between
2.1 and 4.0 include having a personal history o f breast cancer or having either a
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
mother or a sister with breast cancer. Relative risks between 1.1 and 2.0 include
several sociodemographic variables. Women in high socioeconomic groups, or who
were never married, or who are obese have a somewhat higher risk of developing
breast cancer than women who are in low socioeconomic groups, or who are
married, or are thin. In the U. S., race is also associated with breast cancer
incidence. African American women have the highest incidence of breast cancer
detected before 45 years old, and non-Hispanic Caucasian women have the highest
incidence o f breast cancer detected in older women (Kelsey, 1993).
Stage
As an aid in determining prognosis, systems have been developed which rank
breast cancer into stages, which correspond to increasing levels of severity.
Physicians may estimate stage prior to surgery, but histological confirmation is
required for reliable classification (American Joint Committee on Cancer, 1992).
Histological confirmation of the stage requires surgical biopsy of the entire tumor
and associated lymph nodes. Microscopic examination determines the tumor's exact
size, cell type, and degree o f invasion into other tissues. Similar examination o f the
lymph nodes assesses whether cancer cells have spread beyond the original site.
Pathologists use 15 classifications grouped under numerical labels from zero
to Roman numeral IV to describe breast tumors. However, in the risk fector
literature, it is more common to see these levels combined into four stages: in situ,
local, regional and distant (Berstein, Ross, Deapen, Boone, & Miu, 1991). The in
situ stage describes early localized tumors that have not invaded adjacent tissues.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Local stage breast cancer tumors are invasive, but their spread is restricted to
adjacent breast tissue, nipple, or areola, and no cancer cells are detected in the lymph
nodes. At the regional stage, direct extension o f the disease has invaded
subcutaneous tissue, skin, or the chest wall and/or cancer cells are detected in those
lymph nodes receiving the primary drainage from the breast. At the distant stage,
tumors have metastasized to remote sites. The most frequent métastasés sites for
breast cancer are bone, lung, brain, or liver, but it has been found in almost every
remote site (American Joint Committee on Cancer, 1992).
Vi% h proper treatment, 99% of U. S. women will be alive 5 years after a
diagnosis of breast cancer in situ, and 97% after a local diagnosis. If the breast
cancer has spread, the 5-year survival rates drop to 76% for regional and 20% for
distant stage diagnoses (American Cancer Society, 1997).
Stage risk factors. Breast cancer stage is a function o f disease
characteristics, the effectiveness o f detection methods, and tte time elapsed
between initial development of the cancer and its diagnosis (Gould-Martin,
Paganini-Hill, Casagrande, Mack, & Ross, 1982). The first o f these risk
factors is beyond the control of the patient. However, given adequate
information, a woman can act to increase the probability o f early diagnosis.
Most of the published reports that specify breast cancer stage, use it either as
a predictor of survival (Busetti, Miller, To, & Rohan, 1996; Ewertz, Gülanders,
Meyer, & Zedeler, 1991), or to examine the histopathological and cytochemical
markers which distinguish stages in order to guide treatment decisions. Research to
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
identify risk fectors that predict stage has been limited. Although there is some
overlap, risk fectors of developing breast cancer differ from risk frctors that predict
at what stage it will be diagnosed (Gould-h/fertin et aL, 1982). Survival and
mortality risk frictors also differ from breast cancer incidence risk fectors, but they
are not adequate predictors of stage because they include the effects of treatment
variables. The risk fectors that have been suggested as predictors of breast cancer
stage include age, socioeconomic status, including education and access to health
care, race, marital status, obesity, breast self-examination, delay in presenting
symptoms fr> r treatment, and fest growing tumors.
It is commonly accepted that breast cancer stage at diagnosis increases with
age (Holmes, & Heame, 1981; Satariano, Belle, & Swanson, 1986). The effect may
be underestimated, because frequently, elderly women with breast disease symptoms
are not referred for surgery, and so their breast cancer stage may be undetermined
(Goodwin et aL, 1986).
Socioeconomic status (SES) is variously operationalized by income,
education, occupational classification, census tract of residence, or insurance status
(private vs. subsidized). Lower SES is widely associated with later stage at
diagnosis (Farley, & Flannery, 1989). Living in a socioeconomically disadvantaged
neighborhood is a strong predictor of invasive versus in situ breast cancer diagnosis
and the effect is stronger than jfor low income alone.
Extensive studies have reported that minority women, especially African
American (Zaloznik, 1995) and Hispanic (Saint-Germain, & Longman, 1993)
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
women, are more often diagnosed at a later stage, even though the overall breast
cancer incidence rates are lower fi> r them than for Caucasian women (American
Cancer Society, 1997; Danigelis et aL, 1995; Satariano et aL, 1986). The reason for
these differences continues to be debated. SES, for exan^le, may confound the
effect of race. According to the 1990 U. S. Census, 11% o f Caucasians, 32% of
African Americans, and 28% of Hispanics were living at poverty levels (American
Cancer Society, 1997). Some studies report that when adjusted for SES (Breen, &
Figueroa, 1996; Dayal, Power, & Chiu, 1982), and especially for access to medical
care (Hunter et aL, 1993; Zaloznik, 1995), racial differences in stage are
insignificant.
Body mass index, BMT, the ratio o f weight divided by the square o f height, is
also a predictor of stage o f breast cancer at diagnosis. Obesity is associated with
higher stage (Ingram et aL, 1989; Zumofl^ & Dasgupta, 1983) and decreased
survival rates (Ewertz et aL, 1991; Mohle-Boetani et aL, 1988).
Breast self-examination (BSE) has been widely proposed as an inexpensive,
non-invasive, early detection strategy. Early studies reported lower stage, or longer
survivaL among women who reported regularly examining their own breasts for
lumps or other abnormalities, but later studies challenged the utility of BSE
(Auvinen, Elovainio, & Hakama, 1996). The majority o f women who report
performing BSE describe inadequate technique and/or frequency (Gould-Martin et
al., 1982). In addition, the BSE technique women say they use is not significantly
correlated with the technique that they demonstrate (Newcomb, Olsen, Roberts,
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Storer, & Love, 1995). The difficulty in standardizing BSE practice confounds
studies of the efficacy o f the detection method, which may explain why BSE has not
denronstrated mortality reduction in clinical trials ÇRakowski, Pearlman, Rimer, &
Ehrich, 1995). Various studies report only 8 to 69% of women who regularly
practice BSE actually detected their breast cancers during BSE (Greenwald et aL,
1978). BSE practicers tend to be more health conscious and participate in preventive
medical care more often. This selection bias may account for the linkage o f BSE and
improved outcomes reported by early studies (Auvinen et aL, 1996).
A delay between initial symptom identification and diagnosis allows time for
the disease to spread and so is associated with advanced stage (Feldman, Carter,
Nicastri, & Hosat, 1981; Vernon, Tilley, Neale, & Steinfeldt, 1985). Older women
are more likely to delay after breast syng)toms are noted (Green, & Roberts, 1977),
perhaps because early breast cancer synq)toms appear insignificant in conçarison to,
or must conçete for attention with, other painful and disabling chronic conditions
(Satariano et aL, 1986). The presence o f dementia or depression may also hamper
preventive care (Goodwin et aL, 1986). Unfortunately, benign lumps are often
presented ft> r diagnosis more quickly than those turning out to be malignant (Greer,
1974). This may be because benign problems tend to produce rapidly enlarging and
painful lumps, or may bleed, while malignant tumors are asymptomatic in the early
stages (Owens, & Ashcroft, 1986).
Breast cancer tumors grow at different rates. Estimates o f breast tumor
doubling times vary extensively, between averages o f a few days to many months
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
(Gould-Martin et aL, 1982). Younger women are more likely to have higher growth
rate tumors (Peer, Verbeek, Straatman, Hendriks, & Holland, 1996). Faster growing
tumors have been suggested as another reason some breast cancers are diagnosed at
later stage.
Screening
Because breast cancer cannot be prevented (American Medical Association
Council on Scientific Affairs, 1989) and women may not be able to significantly
alter their personal risk factors, early detection is the best available opportunity for
reducing breast cancer mortality (American Cancer Society, 1997; Vernon, LaviUe,
& Jackson, 1990). Early breast cancer has no symptoms, a feet unknown by nearly
half of women over 65 (King et aL, 1993). In addition, nearly 75% of those
diagnosed with breast cancer do not have anv o f the identified risk fectors, except
sex and age (Seidman, Gelb, Silverberg, LaVerda, & Lubera, 1987). Consequently,
no criteria, other than being female and over 40, adequately identify who should be
screened.
The best method of detection is periodic screening o f all asymptomatic older
women, so that cancer is caught at earlier stages (American Cancer Society, 1997;
Farley, & Flannery, 1989). Early detection results in treatment involving less
drastic surgery, less chemotherapy (Choong et aL, 1996), less hospitalization (Kaija,
hfetti, & Tapani, 1996), and improved survival (American Joint Committee on
Cancer, 1992; Lerman, Rimer, Trock, Balshem, & Engstrom, 1990; Owens, &
Ashcroft, 1986). With annual mammography, 62% o f breast cancer is diagnosed
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
8
before lymph node involvement; without annual mammography, the percentage is
only 40% (Tink, Shapiro, & Roester, 1972).
The American Cancer Society and the National Cancer Institute recommend
annual physical examinations by a physician, called clinical breast exams (CBE), and
annual mammography for women over 50. For women 40-49 years old, the
recommendation is for mammograms every 1-2 years and CBE every 3 years
(American Cancer Society, 1997). The National Cancer Institute’s Year 2000
Objective is to increase annual mammogram and CBE rates to 80% for all U. S.
women age 50 and older, as part o f its goal to reduce cancer mortality by 25-50% by
the turn o f the next century (Vernon et aL, 1990).
Mammography is a safe method of early detection (Vernon et al., 1990).
Current data suggest that for women receiving annual mammograms between the
ages of 50 and 75, the benefits o f reduced mortality firom early detection are almost
100 times greater than the increased risks from radiation-induced carcinogenesis
(Mettler et aL, 1996). Mammography is the most effective method available for
detecting early breast malignancies (Lerman et aL, 1990; Wertheimer et aL, 1986). It
can identify a malignant lesion while it is still too small to be detected by physical
examination, either by a physician or by the woman herself (McCarthy, Yood,
Mac William, & Lee, 1996; Peer et aL, 1996). It is significantly more effective than
CBE in decreasing the risk of cancer spread to lymph nodes (American Medical
Association Council on Scientific Affairs, 1989).
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
However, mammograms alone are not sufBcient. Mammograms have a felse
negative rate of about 15%. In one national study of annual screening, for example,
17% of breast cancers were detected by CBE in the interval between scheduled
mammograms ^aker, 1982). It is estimated that roughly a third of interval cancers
were missed on the previous mammogram either through error or because the tumor
was not imaged (Brekelmans, Peeters, Faber, Deurenberg, & Collette, 1994; Frisell,
Eklund, Hellstrom, & Somell, 1987). Inadequate mammogr^hic imaging o f a
tumor can occur either because its location is near the edge o f the breast or because
the higher density of young breast tissue obscures the view (Andersson, 1984). For
somewhat less than a third o f interval cancers, re-examination of the most recent
mammogram shows a previously unrecognized sign, some variation which was
insufficient to meet diagnostic criteria, but which can be identified as atypical in
retrospect (Erisell et aL, 1987). The remaining third or more (35-70% by various
estimates) are “true” interval cancers, which were too small mm) to be
recognized on the previous mammogram, but have such a rapid growth rate that they
were readily detected ( ^ 0 mm) by palpitation before the next scheduled
mammogram (Frisell et a l, 1987; Peer et al, 1996).
Interval breast cancers are biologically different, by histopathological and
clinical-course criteria, than screen-detected breast cancers. They do not show the
typical calcifications, are more often seen in younger patients, have much earlier
metastasis (Baker, 1982; Frisell et aL, 1987), and lower survival rates (Heuser et aL,
1984). They underscore the importance of annual CBE along with annual
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
10
mammograms. Despite the existence of these more virulent breast cancers, for
which mammogr^hic screening has limited utility, it is estimated that 20,000 lives
could be saved in the U. S. annually by conq)Iiance with screening recommendations
(Vogel, Bondy, Halabi, Lord, & LaviUe, 1993).
Breast screening, especiaUy mammography, is underutilized. Although the
rates have been increasing steadUy, in 1992 only 53% of U. S. women over 50
reported having a mammogram in the past 2 years and 33% of women that age had
never had one (Pearlman, Rakowski, & Ehrich, 1996). Even among women at
increased risk of developing breast cancer due to having a first degree relative with
the disease, only 50-84% had ever had a mammogram and only 15-65% had one in
past year (Lerman, & Schwartz, 1993). Older, lower SES and m inority women are
particularly under-served. Nearly half of breast lesions are diagnosed at later, more
life-threatening, stages of the disease (Bernstein, Ross, De^>en, Boone, & NCu,
1991).
Theoretical Bases for Predictive Models
Several theories have been proposed to explain health behaviors. In this
study, three conçonents fi’ om two theories wiU be compared: the Susceptibility and
Barriers conçonents from the Health BeUefe Model OHochbaum, 1958; Rosenstock,
1974) and the Conventionality-Deviance conq)onent from the Problem Behaviors
Model (lessor, 1987).
Both theories recognize that some variables are more proximal in causal or
te n d ra i linkages and, having a more direct influence on the outcome, yield stronger
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
11
correlations. The most proximal set o f variables for predicting breast cancer stage at
diagnosis is screening behavior. M ammography is the best detector of earliest stage
breast cancer (Baker, 1982) and is highly correlated with stage (Hedegaard,
Davidson, & Wright, 1996). However, this strong association may not entirely
reflect the e%ctiveness of mammography as a screening strategy. Many
mammograms are performed not for screening, but to assist in diagnosing a breast
problem, such as a lump, which was detected by other means (National Cancer
Institute Breast Cancer Screening Consortium, 1990).
The method of detection o f the original breast synçtom, which eventually led
to a diagnosis o f breast cancer, is an important component of the screening proximal
variables. Women who themselves detect their breast symptoms may be diagnosed
earlier than women whose symptoms are detected by a health care professional
during a routine medical examination (Auvinen, et aL, 1996). Combining
mammography and method o f detection may improve prediction o f stage.
Distal variables (such as Susceptibility, Barriers, and Conventionality) are
expected to show weaker correlation with the target outcome (Stage), and their
influence may be indirectly applied through mediating variables. However, these
theoretical variables are often more useful in understanding the choices people make.
For exanç)le, while use of mammography is the best predictor o f stage, the more
practical questions concern how the distal demographic and behavioral elements
influence the use of mammography. Health care practices, which are predicted to
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
12
mediate between the distal, theoretical variables and screening, will be discussed
after the two theories have been introduced.
Health Beliefe Model (HBMl
The Health Beliefe Model (Hochbaum, 1958; Rosenstock, 1974) is the most
highly influential and widely researched theory of health behaviors (Taylor, 1991).
It proposes that two perceptions are the major elements in predicting whether a
person practices a particular health behavior: 1) the perception of the degree of
personal health threat and 2) the perception that a particular practice will be effective
in reducing that threat. The first element, personal threat, has at least three
conqjonents: general health issues, which include interest and concern about health;
specific beliefe about personal susceptibility to a particular disorder; and estimates of
the seriousness of the disease. The second element, effectiveness, has two
conqx)nents: the belief in the ability of the behavior to reduce the threat, and the
belief that the benefits of undertaking the behavior overcome the barriers. In this
study, the Susceptibility conq)onent of element one and the Barriers conqx>nent of
element two are examined.
Susceptibilitv. The perception o f personal susceptibility to breast cancer is
predicted to be increased in women who have had a personal or femily history of
breast disease. Women who have had a previous breast problem serious enough to
be diagnosed by a physician may have feced days or weeks o f uncertainty and the
trauma o f a biopsy before learning their breast problem was not breast cancer. Those
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
13
who have a mother and/or sisters who have had breast cancer may have increased
awareness of the dangers of the disease.
A woman's personal or femily experience with breast disease may have
given her an opportunity to learn about her own increased risk of developing breast
cancer. Such women have a two-to-four-fold increased risk o f breast cancer (Kelsey,
1993). However, most studies (Kaplan, Weinberg, Small, & Herndon, 1991;
Vernon, et aL, 1990), although not aU (King, et aL, 1993), report that women at such
higher risk are no more likely to participate in mammographie screening than women
without these risk fectors.
Barriers. Barriers have been defined either specifically, such as the cost of
mammography or the difficulty o f arranging transportation to screening sites, or
broadly, such as the socio demographic effects of age, race, SES, or marital status,
which influence specific behaviors (Janz, & Becker, 1984; Mayer Oakes, et aL,
1996). In extensive research across several different health behaviors, health beliefs
have been shown to be a modest determinant of intentions regarding a health
behavior (Taylor, 1991). However, the best negative predictor of the actual practice
of health behaviors is the perception of the barriers (Janz, & Becker, 1984).
The barriers to mammography that will be examined in this study are age,
SES including education, race, religion, and marital status. Elements that decrease
mammography are predicted to advance breast cancer stage.
It has been well documented that older women are less likely to get
mammograms (Anda, Sienko, Remington, Gentry, & Marks, 1990; Vernon et aL,
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
14
1990). The National Health Interview Survey (NHIS) is a continuing, nationwide
survey that uses multistage probability methods to select a random sample of
households in the U.S.A. for interview. Li both the 1987 and 1992 NHIS, women
age 65 and over were the least likely to have ever had a mammogram or to have had
one in past year (Martin, Calle, Wingo, & Heath, 1996). It is unclear whether this is
a cohort effect. However, in one study of well educated Caucasian women with
above average SES, if physicians recommended a mammogram and reminders were
sent to encourage participation, age did not significantly effect mammography
participation (Taphn et aL, 1994).
Low SES correlates with lower probability o f ever having had a mammogram
and o f having a recent mammogram (Mayer-Oakes et aL, 1996; Rakowski et aL,
1995). Higher incomes, higher occupational status (Pink et aL, 1972) and having
private medical insurance (Mayer-Oakes et aL, 1996) all associate with increased
mammography use. However, in one study where preventive health services were
equitably available, the SES effect on mammography was insignificant (Lane,
Polednak, & Burg, 1992).
Lower education decreases participation in mammogr^hy (Rimer et aL,
1996; Vernon et aL, 1990). Ifaving less than a high school education is associated
with 40% fewer women having a recent (past year) mammogram (Martin et aL,
1996) and decreases the probability o f having a second mammogram (Rimer, Trock,
Engstrom, Lerman, & King, 1991). There were improvements in mammogram
utilization by many groups of women between the 1987 and 1992 NHIS reports.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
15
However, those women with low-income levels near the poverty level, less than high
school education, and no usual source o f medical care had the smallest gains and
continued to seriously under-utilize breast screening (Martin et aL, 1996).
Minority status is widely associated with less mammography. Many studies
have reported that Caucasians have more mammogram s than non-Caucasians, but
most of those studies did not control for access to health care (McCarthy et aL,
1996). Some reports suggest that the effects of race are diminishing and that now the
salient issues are SES and education (Mutchler, & Burr, 1991; Rimer et aL, 1996).
However, others have shown minority women receive fewer mammograms
regardless of SES, even if the m am m ograms are free (Pearhnan, Radowski, Ehrich,
& Clark, 1996). Between the 1987 and 1992 NHIS reports, the greatest gains in the
category ‘ Ever had a mammogram?” were in Hispanic, and to lesser degree, in
African American groups. However, the disparity between Caucasian and non-
Caucasian women continues to leave minority women under-protected (Martin et aL,
1996).
Marital status is also associated with mammography. Unmarried women,
whether never married or currently divorced, widowed, or separated, are less likely
to have had mammograms (McCarthy et aL, 1996; Vernon et aL, 1990). Among
women over 65 years old, mammography is especially low for single or widowed
women (Marchant, & Sutton, 1990). Age and SES may confound some o f the effect
o f marital status on mammogr^hy (Vernon et aL, 1990; Rimer et aL, 1991). For
example, mammogram use among women employees in a low-cost mammography
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
16
program showed no marital status differences. These women were between 40 and
65, and were better educated than the general population (Rutledge, Hartmann,
Kinman, & Winfield, 1988).
Problem Behavior Theory (PBT)
The Problem Behavior Theory was originally developed to explain
adolescent alcohol drinking behavior. PBT ençhasizes that behavior is a result of a
person-environment interaction, extending the genetic and biological considerations
of health behaviors by using a field theory perspective (lessor, 1987). Some
behaviors are culturally defined as problems and their occurrence usually elicits
some kind o f social control response. PBT suggests that even these problem
behaviors are shaped by the norms and expectations of the culture.
The theory suggests a dimension o f Conventionality to summarize the degree
of commitment to and involvement in the prevailing values, standards and
institutions of society (Donovan, lessor, & Costa, 1991). In the original application
o f the theory to adolescents, problem behaviors, such as getting drunk, having
precocious sexual activity, and smoking cigarettes, are predicted to be interrelated
behaviors on the deviant end of the dimension. These are also predicted to be
negatively correlated with conventional behaviors, such as church attendance and
involvement in school activities. Research with various adolescent problem
behaviors has supported the theory (lessor, 1987), and shown that it accounts for
between a third and a half of the variance in the relationshq> between problem
behaviors and conventional behaviors in adolescents (Donovan et aL, 1991).
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
17
McGee and Newcomb (1992) confirmed the integrity o f the norm-violating
syndrome as related to drug use and academic, sexual, and criminal behaviors. They
identified latent constructs including social conformity and a second order construct
o f general deviance which predicted various problem behaviors at dififerent
developmental stages from adolescence to adulthood.
Recently the PBT has been extended to show conventional adolescents are
more likely than deviant ones to practice health maintenance behaviors (Donovan et
al., 1991; Donovan, lessor, & Costa, 1993). A fiuther extension of the theoiy may
assist in understanding adult health behaviors. PBT’s prediction of patterning among
behaviors and the concept of conventionality as an underlying dimension may
explain a syndrome o f health behaviors. A conventional person is predicted to be
more likely to adopt the patterns o f behavior promoted by conventional institutions.
In the case o f women's health behaviors, these would include mammographie
screening and exercise. Such a person is predicted to be less likely to adopt
behaviors discouraged by conventional institutions, such as smoking, heavy alcohol
and caffeine consmnption, having abortions, and being overweight.
Health care research provides some support fr> r this theory. People do report
social pressures to comply with health care system values. Patients report they often
expect health care providers to reprimand them after they admit to an unhealthy
(deviant) behavior and feel blamed for disease because of unhealthy prior habits
(Strecher, Becker, Clark, & Prasada-Rao, 1989).
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
18
A patterning of health care behaviors has also been documented. People
appear to exhibit propensities for particular health behavior orientations (Ungemack,
1994) and a disease-prevention orientation is associated with increased
mammography (Nfeyer-Oakes et aL, 1996). Interventions focusing on an overall
preventive orientation in older women are based on the assumption that a synergistic
approach may inq)rove patient outcomes (Mayer-Oakes et aL, 1996).
Conventionalitv. Conventionality can be represented by behaviors at either
end of the dimension. Smoking, heavy alcohol and caJ^eine consumption, and
overeating are behaviors that are treated by the medical care system as problem or
deviant behaviors. At the other end o f the conventionality spectrum, exercise is
widely recommended for its health benefits.
Costa et aL (1987) reported that having abortions clustered with deviant
behaviors in a study of women in their mid-20s in 1981. However, Stein, Newcomb
and Bentler (1994) did not find that general social conformity predicted abortions in
a group of mid-20s women in 1988. Specific attitudes about abortions and early and
frequent sexual activity were more important predictors o f abortions than the general
social-conformity latent construct
For women in the present study, abortion is expected to be a more strongly
deviant behavior than for cohorts that are more recent. The women in this study
were bom between 1923 and 1937. When abortions were legalized in the United
States in 1973, these women were between 36 and 50 years old. Therefore, the only
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
19
abortions to which they may have had access, during most o f their reproductive
years, were illegal abortions.
The relationship between conventionality and mammography is not as well
documented as the relationships between sociodemographic barriers and
mammography. Smoking has been shown to correlate with a less healthy lifestyle
relative to cardiovascular health (Castro, Newcomb, McCreary, & Baezconde-
Garbanati, 1989). Moderate to heavy smokers (7+ cigarettes per day) were less
health conscious and engaged in other coronary heart disease risk behaviors more
often Some studies (Pearhnan, Rakowski, & Ehrich, 1996; Rimer et aL, 1991;
Taplinet aL, 1994), but not others (Bondy, Vogel, Halabi, & Lustbader, 1992;
Sutton, Bidder, Sancho-Aldridge, & Saidi, 1994), have associated smoking with
less screening mammography In one study, daily alcohol drinkers were less likely
to attend screening than more moderate drinkers were (Sutton et aL, 1994). Obesity
has been shown to increase breast cancer stage (Ingram et aL, 1989; Zumof^ &
Dasgupta, 1983). Obesity and exercise are ejqjected to be negatively correlated.
Their relationship with mammography will be examined in this study.
Mediating Variables
Elements reflecting the awareness o f Inreast health practices, specifically, or
the utilization o f health care, generally, are predicted to mediate between the
proximal variables of screening and the distal variables o f susceptibility, barriers and
conventionality. A woman who has been pregnant, especially if she breast-fed her
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
20
child(ren) may have increased awareness of breast health, practices and be more
likely to self-monitor, perhaps leading to earlier detection of breast symptoms.
Health care utilization has been reported to be a predictor of screening. In
fact, the group of variables that most consistently associates with participation in
breast screening is the practice of other health behaviors. As ejqpected, the
association is stronger than for the more distal sociodemographic elements
(McCarthy et aL, 1996; Vernon et aL, 1990). Health care utilization has been
operationalized in various breast cancer screening studies as the recommendation
from a physician for a mammogram, the specialty of the physician seen, having a
regular source of medical care, or the number of visits to a physician.
The strongest predictor of having had a mammogram in the past year (King
et aL, 1993) and of having regular, annual mammography (Friedman et aL, 1995), is
having a recommendation for a mammogram from a physician. In one study,
patients who reported having a physician’s recommendation and access to free
mammograms were 12 times as likely to have received a mammogram as those who
had neither (Rimer et aL, 1991). Many women report that lack of a physician
recommendation is the reason they have never had a mammogram (National Cancer
Institute Breast Cancer Screening Consortium, 1990). These data have been used to
support the essential role o f physicians in influencing asymptomatic women to
undergo breast screening (Friedman, et aL, 1995). The high correlation has been
interpreted to mean that more recommendations would result in more screening.
Because only 37% of physicians report following the American Cancer Society’s
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
21
guidelines in their recommendations of screening mammography (American Cancer
Society, 1990), it has been suggested that programs to increase physician
recommendations for mammograms would have a great positive impact on public
health (National Cancer Institute Breast Cancer Screening Consortium, 1990).
However, an equally plausible explanation challenges the causal
directionality of these interpretations. The correlation o f physician recommendation
and mammography may not be solely due to the physician's influence on women’s
intentions to use preventive health care. The high correlation may be partially the
result o f restriction, by allowing only those who have a prescription from a physician
to have a mammogram. Physicians have been the gatekeepers o f breast screening for
most women until recent years. The high correlation between mammography and
physician recommendation may reflect that most women could not get a
mammogram any other way.
The specialty o f the woman's physician is another health care utilization
element. Seeing a gynecologist, rather than some other specialist, predicts that a
woman will talk to her physician about mammography. Gynecologists prescribe up
to four times more mammograms than other physicians (King et aL, 1993; McCarthy
et aL, 1996) and more often offer them routinely (Bergner, AUison, Diehr, Ford, &
Feigl, 1990). Older women are less likely to be currently seeing, or to have ever
seen a gynecologist (Celentano, Shapiro, & Weisman, 1982; Celentano, Klassen,
Weisman, & Rosenshein, 1988). Women after ‘reproductive age’ (18-44 years old)
or who have never married, or who are divorced, separated or widowed, are also less
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
22
likely to see gynecologists (Martin et aL, 1996). Thus, part o f the effect o f the distal
sociodemographic variables, age and marital status, on the frequency o f
mammography may be mediated by the specialty of the physician.
Having a regular source of medical care increases the probability o f
mammography (Martin et aL, 1996; National Cancer Institute Breast Cancer
Screening Consortium, 1990). In addition, those women who do not have a regular
physician require many more reminders to encourage their attendance fijr
mammography, even when the mammograms are free (Vemon et aL, 1990).
Many studies have reported that mammography increases directly with the
number o f physician visits (Hedegaard et aL, 1996; King et aL, 1993), especially
with the number of healthy visits (Rimer et aL, 1991). For example, in one study of
10,000 low income, minority. Community Health Center patients, women who had
four or more doctor visits in the past year were three times as likely to have received
a mammogram as those women with fewer doctor visits (Bedegaard et aL, 1996).
The number o f visits may mediate SES effects on mammography^ because poor
women report fewer doctor visits (Vernon et aL, 1990).
In this cohort of peri-menopausal women, the use o f hormones may predict
frequent physician visits. Estrogen and progesterone are prescribed for menopausal
symptoms and often require regular check-iq)s to receive the prescr^tions. The role
of hormone use in the prediction of breast cancer stage is explored in this study.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
23
Hypotheses
The current study analyzes behavioral and demographic data from a large
sample of U. S. women to identify predictors of stage. Additional information,
concerning the method of discovery of the breast cancer syrcçtoms, is only available
from a portion o f these women, so a sub-sample o f the data is analyzed separately to
examine the role of that variable.
Hypotheses from two theories are compared to determine their relative
importance in predicting stage of breast cancer at diagnosis. Each theory predicts
that the most proximal variable, screening, will be the strongest predictor of stage
and that the mediating variables conprising the utilization of health care and
awareness o f breast health practices wül be the best predictors o f screening. In this
context, the Health Beliefs Model predicts that the strongest predictors of health care
and mammography will be the perception o f susceptibility, as reflected by personal
and family history of breast disease, and the barriers of age, SES, including
education, race, religion, and marital status. The Problem Behavior Theory predicts
the behaviors that reflect the dimension of conventionahty-deviance will have the
strongest effect on health care, mammography and subsequently, on stage at
diagnosis. The dimension will include exercise on the conventional side, and
smoking, alcohol and caffeine misuse, abortion and obesity on the deviant side.
Figure 1 presents the first model, which examines the full sample. It shows
the combination o f the two theories, where heavier arrows represent the predicted
stronger path coefficients. The heaviest arrow runs from the most proximal variable.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
24
Figure 1. Theoretical Model for the Full Sample, Combining
Health Belief Model and Problem Behavior Theory Components:
Arrows represent hypothesized paths. Heaviest arrows represent paths hypothesized
to be stroi^est; open arrows = paths to be examined in this study; dashed arrows =
paths hypothesized to be weak associations; ObGyn = obstetrician or gynecologist.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
CD
■ D
O
Q .
C
g
Q .
■ D
CD
C/)
C/)
8
" O
CD
3 .
3 "
CD
CD
■ D
O
Q .
C
a
O
3
■ D
O
CD
Q .
■ D
CD
C/)
C/)
HBM
Susceptibility:
Family histwy
Breast Disease
HBM Barriers:
Age
Race
Socioeconomic;
Residence &
Occupation
Education
Marital status
Religion
Health Care
Regular Dr,
ObGyn
Hormones
Pregnancy
Nursing:
Childrai nursed
Nursing duration
Stage at
diagnosis
Screening:
Mammograms
PBT Conventionality:
Smoking
Alcohol: at 25, at 40 &
Recent
Caffeine
Obesity
Abortiwis
Exercise
26
Screening, to the outcome variable. Stage at Diagnosis, because this path is predicted
to be the strongest. The relationship between Health Care and Screening is predicted
to be almost as strong. The paths of the distal variables. Susceptibility, Barriers and
Conventionality, to Health Care and to Screening directly are the primary focus of
this study. They are denoted by open arrows. The dashed arrows, from the distal
variables directly to Stage, independent o f the proximal and mediating variables,
predict weak associations.
Figure 2 is very similar to Figure I and the meanings of the arrows are the
same. The same combined theories are shown, but Model 2 includes the method of
discovery data. Since this data is only available from a portion of the women in the
study, the sub-sample is analyzed separately.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
27
Figure 2. Theoretical Model for the Sub-Sample, Combining
Health Belief Model and Problem Behavior Theory Components:
Arrows represent hypothesized paths. Heaviest arrows represent paths hypothesized
to be strongest; open arrows = paths to be examined in. this study; dashed arrows =
paths hypothesized to be weak associations; ObGyn = obstetrician or gynecologist. By
MD = by physician. By Mammog. = by mammography.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
■ o
o
Q.
C
g
Q.
■ O
CD
C/)
C/)
CD
8
■ O
v <
( § ■
p .
3 "
CD
CD
■ O
O
Q.
C
a
O
3
" O
O
CD
Q.
■ O
CD
C/)
C/)
HBM
Susceptibility:
Family histwy
Breast Disease
HBM Barriers:
Age
Race
Socioeconomic
Residence &
Occupation
Education
Marital status
Religion
Health Care:
Regular Dr.
ObGyn
Hormones
Pregnancy
Nursing:
Children nursed
Nursing duration
Screening:
Mammograms
Discovery:
By Self
By MD
By Mammog,
Stage at
diagnosis
PBT Conventionality:
Smoking
Alcohol: at 25, at 40 &
Recent
Caffeine
Obesity
Abortiais
Exercise
to
0 0
29
Methods
Archival data were analyzed in this study to compare the relative contributions
o f the Susceptibility and Barriers conç)onents of the Health Belief Model to the
Conventionality-Deviance conc^nent of the Problem Behavior Theory m the
prediction o f breast cancer stage at diagnosis. The chosen data set provided au
excellent opportunity to examine these relationships for several reasons. A wide range
o f demographic and behavioral elements were assessed for each participant, allowing
clarification o f their multiple influences. The data were collected between 1987 and
1995. The large sample is population-based, eliminating selection bias that might
arise if participants were self-selected volunteers or if only patients from particular
hospitals were recruited.
Annlia Paganini-Hill, Ph.D., of the Department of Preventive Medicine at the
University of Southern California School o f Medicine, directed the collection o f these
data as part of her project to elucidate the association between hormone replacement
therapy and breast cancer (Carpenter, Ross, Paganini-Hül, & Bernstein, 1998;
Longnecker, Bernstein, Paganini-Hill, Enger & Ross, 1996; Longnecker, Paganini-
Hill, & Ross, 1995). Data were collected after Institutional Review Board approval of
the study’s human subjects' procedures. The project was supported by U. S. Public
Health Service Grant CA 17054 from the National Cancer Institute, National Institutes
o f Health, Department o f Health and Human Services. Dr. Paganini-Hill has given
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
30
permission for a portion o f her archival data to be examined in this dissertation and for
publication of the results.
Participants
The Cancer Surveillance Program o f Los Angeles County gathers data on
cancer diagnoses in all the county’s hospitals and cancer-diagnostic laboratories
(Bernstein et aL, 1991). Its work is funded by the California Public Health
Foundation, subcontract 050-F-8709, which is supported by the California Department
o f Health Services as part o f its statewide cancer-reporting program mandated by
California Health and Safety Code Sections 210 and 211.3. From its records, target
groups can be identified for population-based studies. The population chosen for this
study was all English-speaking women living in Los Angeles County, California,
U.S.A., V iho were diagnosed for the first time with primary breast cancer of epithelial
origin between 1987 and 1992. This excluded women who had had breast cancer
before the selection diagnosis and were experiencing it again.
Two groiq)S of participants were targeted from this population. Groiq) 1
consisted of all Caucasian (mcluding Hispanic) women bom in the U.S.A., Canada or
Western Europe who were age 55-64 when diagnosed between March 1, 1987 and
December 31, 1989. Group 2 consisted o f women diagnosed between January 1, 1992
and December 31, 1992 who were age 55-69, Caucasian (including Hispanic) or
African American, and bom in the U.S j V .
The utility o f mammography for women under 50 and over 70 years old
remains a controversial issue. The iq)per age limit used in this study reflects the
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
31
guidelines of the American College of Preventive Medicine which reserves its
recommendation for annual mammography to women 50-69 years old (Ferrini,
Mannino, Ramsdell, & Hill, 1996).
Procedures
All women in both target groups were entered into a two-step recruitment
process. First, permission to invite the patient to participate was requested from her
attending physician. If the physician did not object, women were invited by letter to
give informed consent to be interviewed. No condensation was offered to the women
for their participation.
Semi-structured, self-report interviews were administered by trained personnel,
an average of 15 months after diagnosis, and usually in the participant’s home, to
gather demographic and lifestyle data, and hormonal, reproductive and disease
histories.
It was assumed that just before diagnosis, some women might have suspected
they had breast disease and altered their usual behaviors. To avoid bias by including
such changed behaviors, a cutoff called the reference date, which was one year prior to
her diagnosis, was determined for each participant. No dataware collected about
behaviors occurring after the reference date, with one exception: Group 2 women were
asked how they first discovered the symptom that led to their diagnosis of breast
cancer.
The data for all the predictor variables in this study were self-reported. Self-
reports have established predictive validity with respect to long term health benefits
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
32
(Velicer, Prochaska, Rossi, & Snow, 1992). However, all studies that rely on
retrospective self-report data must consider the possibility o f memory problems,
selective recall, and intentional misreporting. The pressures on participants to
misreport behaviors are called demand characteristics (Welte, & Russell, 1993). In
this study, several techniques were utilized to decrease demand and increase the
accuracy of self-reports. Assurances of confidentiality, and non-threatening,
unappraising and sympathetic questions were used to decrease the pressure to respond
in socially desirable ways (Strecher et aL, 1989; Velicer et aL, 1992). Interviewer-
administered questionnaires have been shown to increase sensitivity (true positive
responses) and specificity (true negative responses) over self-administered
questionnaires (Patrick et aL, 1994). A timeline o f personal major life events was used
as a memory aid to give a tenqx>ral anchor and improve recall (Sobell, Toneatto, &
SobeU, 1990). Questions were objective and clearly defined, to minimize ambiguity
and increase accuracy (Toneatto, SobeU, & Sobell, 1992).
Non-participants
Sixty-two percent of the 3,709 women identified by the Cancer Surveillance
Program as potential subjects did participate in this study. Twenty percent were never
invited to partic^ate because their physicians refused us permission to invite them^
they could not be located, they had moved too fer away for a personal interview, they
were too ill, or they had died. O f those who were invited to particÿate, 78% were
interviewed.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
33
Table 1 shows the reasons for non-participation of Group 1 women, by stage at
diagnosis. Stage information for non-participants was only available for Group 1
women. As expected, women diagnosed at the distant stage were less likely to be
interviewed. More than half o f these potential subjects had died before the invitation
to participate in the study arrived.
Table 1
Subject Participation, bv Breast Cancer Stage at Diagnosis, for Group 1 Subjects
Stage Inter
viewed
Reason Not Interviewed
Unable
To
Locate
Moved Deceased Too
1 1 1
Physician
Refused
Subject
Refused
In situ 74.9% 0.4% 2.6% 1.3% 0.0% 4.6% 15.9%
Local 70.6% 0.9% 4.2% 2.4% 0.9% 4.9% 16.2%
Regional 63.5% 0.8% 4.8% 7.5% 1.1% 5.4% 16.9%
Distant 23.5% 0.8% 7.6% 52.3% 2.3% 3.8% 9.9%
(stage
unknown)
13.9% 0.8% 7.4% 37.7% 1.6% 9.0% 29.5%
Table 1 also includes women who were diagnosed with breast cancer, but for
vdiom stage was unknown. These women may have included patients who were too
ill to benefit from further diagnostic procedures, as well as those who refused further
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
34
treatment. This unknown-stage group had nearly twice the percentage of refusal-to-
participate o f any other group, and three times the refusal rate of those who were
diagnosed at the distant stage.
Sample Description
Interviews of 1,586 participants in Group 1 were conpleted during 1987-1992;
721 women in Group 2 were interviewed during 1992-1995. Data from Groups 1 and
2 are combined in the fuU sample. Some analyses were performed on a sub-sample of
Group 2 participants only. Table 2 contains stage data for the sub-sample and the full
sample.
Table 2
Breast Cancer Stage at Diagnosis
for the Sub-Sample and the Full Sample
Stage Sub-Sample Full Sample
n % n %
in situ 69 9.6 239 10.4
Local 453 62.8 1,348 58.4
Regional 181 25.1 654 28.3
Distant 10 1.4 41 1.8
(stage
unknown)
8 1.1 25 1.1
Totals 721 100.0 2,307 100.0
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
35
Tables 3 through 7 present descriptions of the participants in the sub-sample
and the full sample. As expected, they are very similar, except that the sub-sample is
more ethnically diverse. Table 3 contains demographic data. Both. Hispanic and
African American women were under-represented in the full sanq>le. The racial-
ethnic distribution o f the entire 1980 Los Angeles County population was 56%
non-Hispanic Caucasian, 24% Hispanic, and 13% African American (Bernstein et aL,
1991). Speaking English was an inclusion criterion for this study and women were
ineligible if bom in Mexico, Central, or South America These criteria excluded many
Hispanic women. African American women conçrised 11.4% of Group 2, but were
not included in Group 1, which limited their percentage representation in the total
sample. Other racial and ethnic groups were excluded by selection criteria The mean
age, at diagnosis, o f the participants was 60.6 years for the full sample and 62.3 years
for the sub-sample.
The religious preference most often reported by women in this sample was
Protestant Catholic and Jewish preferences were also well represented. These three
conçrised 87.7% o f the full sample.
The majority o f the participants in the full sample were currently married,
63.6%, or had been married, 30.5%. In this cohort o f women, bom between 1923 and
1937, reports of a cohabiting relationship were uncommon.
As shown in Table 4, this was a well-educated sangle. Only 12.1% of the
women in the full sample lacked a high school diploma or its equivalent, and 55.3%
had some college educatioit
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
36
Tables
Demographic Description o f the Sub-Sample and Full Sample
Characteristic Response Sub-Sairçle
n = 721
%
Full Sample
n = 2,307
%
Racial and ethnic Non-Hispanic Caucasian. 79.6 88.8
self-designation
Hispanic 9.0 7.6
African American 11.4 3.6
Religious Protestant 50.5 46.3
Preference
Catholic 25.9 27.8
Jewish 12.2 13.5
Mormon 0.8 1.4
Jehovah's Witness 1.4 0.7
Christian Scientist 0.1 0.2
Religions Not Listed 2.5 3.0
None 6.5 7.0
Marital Status Married 58.5 63.6
Divorced 16.4 16.2
Widowed 16.6 13.0
Single 6.0 5.4
Separated 1.9 1.3
Cohabiting 0.6 0.5
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
37
Table 4
Educational Description of the Sub-Sample and the Full Sample
Educational Level Sub-Sattçle
n = 721
%
Full Sample
n = 2,307
%
Less than 7th grade 1.4 1.0
7th through 9th 3.1 3.5
10th through 11th 7.6 7.6
BQgh School Graduate 31.8 32.7
Partial College 33.4 32.9
College Graduate 14.3 13.2
Graduate/Professional Training 8.5 9.2
Table 5 presents a socioeconomic description of the sample. First, the
occupational levels (HoUingshead, 1957) o f the participants are shown. More women
reported being homemakers, retired or otherwise unemployed than being in any of the
other occupational classifications. Both the ages of these women, 55-69 years old, and
their cohort may have contributed to fewer women being employed prior to their
diagnosis of breast cancer.
Another pro^qr variable for income is the census tract o f residence. Based on
the 1980 census, residential tracts were ranked on annual femily income and average
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
38
Table 5
Socioeconomic Description o f the Sub-Sample and the Full Sample
Characteristic Sub-SatEple
n=721
%
FuU Sample
n = 2,307
%
Occupational Level “
Executives, Major Professionals 1.7 1.8
Managers, Lessor Professionals 12.1 12.9
Administrators, Semi-Professionals 12.8 15.0
Clerical, Sales, Technicians 16.8 19.3
SkUled Manual 1.0 1.4
Machine Operators, Semi-SldUed 6.4 5.2
Unskilled 2.5 2.1
Not Ençtloyed, Homemaker, Retired 46.9 42.2
Estimated Income bv Census Tract o f Residence
High 15.8 18.0
Moderately High 32.3 33.0
Medium 23.0 23.8
Moderately Low 24.3 21.8
Low 4.6 3.3
Note, from the HoUingshead Occi^ational Scale (HoUingshead, 1957).
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
39
education. The majority of the participants were living in medium to upper scale
neighborhoods: 74.8% o f the full sample, 71.1% o f the sub-sample.
Table 6 presents the responses to life-style questions that defined some of the
variables o f the study. Just over half o f these women had exercised three or more
hours per week for a year or more, sometime in their adult lives. Most had been
pregnant. The majority had never had breast cysts, lumps, or other symptoms of
breast disease, prior to the symptoms that lead to their diagnosis of breast cancer.
Table 6
Life-Stvle Description o f the Sub-Sample and the Full Sample
Variable Question Re
sponse
Sub-Sanple
n = 712
%
Full Sample
n = 2,307
%
Exercise Have you ever exercised 3 Yes 50.6 55.0
times/week for a year or more
during your adult life? No 49.4 45.0
Pregnancy Have you ever been pregnant? Yes 89.6 88.4
No 10.4 11.6
Personal Prior to the symptoms which Yes 39.5 38.0
History led to this diagnosis, had you
ofBreast ever had breast disease, cysts No 60.5 62.0
Problems or lumps in the breast which
were diagnosed by a doctor?
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
40
Access to health care is presented in Table 7. Most of the participants reported
having a regular physician, who was most often a general practitioner or an internist.
Almost two-thirds of the participants had taken estrogen or progesterone, or both, but
only one-third or fewer of them had ever used birth control pills.
Variables and Latent Factors
Distal Theoretical Variables
Susceptibilitv. The Health Belief Model conçonent, perceived susceptibility,
included two elements: personal history o f breast disease and family history of breast
cancer. The personal history item asked whether the woman had ever had breast
disease, cysts, or lumps diagnosed by a physician before the reference date. Family
history was operationalized as the sum of the participant's female first-degree
relatives, le., mother and sisters, who had been diagnosed with breast cancer.
Barriers. The Barriers conçonent o f the Health Belief Model included these
social stmctures: age, race, SES, education, marital status, and religion. Age was the
participant's age in years at diagnosis. Each participant's racial or ethnic group was
self-classified as African American, non-Hispanic Caucasian or Hispanic, then dummy
coded as Caucasian versus other, and Hispanic versus other. Afiican American
women were included only in Group 2. Selection criteria excluded other racial
groxQ)s.
The latent factor SES was c^tured in two proxy variables that were used in
place o f annual household income. The first was the census tract o f residence at the
time o f diagnosis. An index of five levels has been assigned to each census tract based
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
41
Table 7
Description o f Access to Health Care bv the SnH-Sampie and the Full Sample
Question Response Sub-Sanç)le
n = 712
%
Full Sample
n = 2,307
%
Do you have a regular Yes 85.4 85.5
physician?
No 14.6 14.5
What is the specialty o f General Practitioner 33.8 35.1
your regular physician?
Internist 33.8 32.7
Obstetrician or
Gynecologist
18.0 18.5
Family Practitioner 7.8 7.0
Cardiologist 1.8 1.6
Surgeon 1.5 1.4
Oncologist 0.3 0.5
Other Specialty 3.1 3.2
Have you ever taken the Yes 65.7 64.8
hormones estrogen or
progesterone (not including No 34.3 35.2
birth control pills)?
Have you ever taken birth Yes 33.3 31.9
control pills?
No 66.7 68.1
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
42
on 1980 U.S. Census data of median household income and average educational level.
The second was the combined occupational level o f the participant and her husband, if
any, during the reference year (the year ending one year before diagnosis), based on
the HoUingshead Occupational Scale (HoUingshead, 1957).
Education was the average level of formal education conq)leted by the
participant and her husband, if any, up to the reference date. Marital status was coded
according to whether the woman was in a relationship, including married or
cohabiting, during the reference year.
Religious preferences were dummy coded as Catholic versus other, and Jewish
versus other. When more than one answer was given, the participant was asked to
choose the one which best described her during recent years.
Conventionalitv. The Conventionality conqwnent o f the Problem Behavior
Model included five elements from the deviant end o f the dimension, and from the one
conventional end. Smoking, alcohol intake, caffeine consun^tion, obesity, and
number of induced abortions were deviant behaviors. Exercise was included as a
conventional behavior.
The latent fector of Licit Drug Use was captured in smoking and alcohol intake
measures. Smoking experience was coded to reflect total lifetime consumption. It
was the product o f the number o f years the woman had smoked and the average
number of cigarettes she smoked per day. Alcohol intake was calculated by
combining the average weekly consun^tion of beer, wine, and distilled liquor at each
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
43
of three ages: age 25, age 40, and during the reference year. Each 12 ounces of beer or
4 ounces o f wine was assumed to contain the alcohol equivalent of 1 ounce of liquor.
Caffeine consumption was the sum o f the reported average daily intake of
caffeine-containing coffee, tea, and colas during the reference year. It was calculated
by assigning a cup or glass of coffee or tea as one unit, a mug as 1.5, and 12 ounces of
cola as two units.
Information on abortions was obtained during a detailed reproductive history
review to increase disclosure. The number of induced abortions performed for birth
control, rather than medical necessity, was included in this study.
Obesity was operationalized as the body mass index. It was calculated as the
participant's reported weight, in pounds, during the reference year, divided by the
square o f her height, in inches. The exercise item asked vdiether the w oman had,
during her adult life, participated in a sport, dance, or exercise program in which she
exercised at least an hour, three times a week for a year or more.
Mediating Variables
Health care practices were predicted to mediate between the proximal
screening behaviors and the distal theoretical variables. Participants were asked if
they had a regular physician, and if so, the specially o f that physician.
The use of the hormones estrogen or progesterone for non-contraceptive
purposes was examined as potentially reflecting these women's recent health care
utilization. Participants were asked whether they had ever taken these hormones,
other than as con^onents o f birth control pills. Estrogen and progesterone are
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
44
prescribed for peri-menopausal symptoms. For the women included in this study, who
ranged from pre-menopausal to late-menopausal, hormone use may reflect recent
regular visits to a physician.
Three reproductive history variables were included to explore the relationships
among pregnancy, breast-feeding eîqperiences, and stage at diagnosis. The pregnancy
item asked whether the w oman had ever been pregnant. The nursing latent fector was
captured in two measures, the number o f children the woman had nursed and the
combined total number o f months she nursed.
Proximal Variables
Breast cancer screening is the behavior most obviously linked to early
discovery of the disease. In this model, screening was placed as the most proximal
predictor of stage at diagnosis. It included the number o f mammogram s the woman
had ijp to the reference date.
For Group 2 participants only, a second model was created, including how the
symptom that led to the breast cancer diagnosis was first detected. Women may have
discovered the problem themselves, or it may have been detected during a physical
examination by a physician or other health professional, by routine mammography, or
by other means.
Criterion variable
Breast cancer stage at diagnosis was the predicted variable. The stage was
histologically confirmed for 99% of the participants after surgical removal o f the
tumor and required lymph nodes. The size o f the tumor and the degree o f its invasion
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
45
into other tissues were measured microscopically, and reported on the hospital
abstracts collected by the Cancer Surveillance Program. For this study, the
histological categories were collapsed into four levels: in situ, local, regional, and
distant stage (see Table 2).
Breast cancer stage is an ordinal scale, reflecting increasing severity of the
disease. Although stage was coded 1 through 4, and treated as a continuous variable,
the numerical coding does not imply that the stage levels are equally spaced.
Stage could not be definitively classified for 1% of the participants. A
conservative method was followed to retain these data in the analyses. They were
assigned the modal value for stage, which corresponded to the local level. As seen in
Table 2, 63% o f the participants were diagnosed at the local stage.
Analyses
The descriptive and univariate analyses o f these data were accomplished using
the SAS Statistical Package, version 6.12 (SAS Institute, Inc., 1996). The multivariate
analyses were performed using the EQS Structural Equations Program (Bentler,
1995a). EQS was chosen for its ability to examine the relationshq)s of the predictor
variables to the outcome, despite the expected interrelationsh^s among the predictor
variables.
Skew and kurtosis estimates for the measured variables in the full sample
suggest that several were not normally distributed. Skewness and kurtosis values o f
zero indicate normal distributions. Eighty-five percent of the variables had skewness
estimates between -2 and + -5 ; the alcohol and abortion variables were highly.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
46
positively, skewed with values between +6 and +10. Over half of the variables were
fairly mesokurtic with kurtosis estimates between -2 and +2. Another third were
leptokurtic with values between +2 and +9. Five variables: months of nursing,
number of abortions, and all three alcohol consumption measures had extreme scores
and a peaked distribution, with kurtosis values over 20. The maximum likelihood
structural-model estimator used here does assume both univariate and multivariate
normally distributed data, but it has been shown to be quite robust over normality
violations (Newcomb, 1990).
Confirmatory fector analysis (CFA) assessed the adequacy of the hypothesized
measurement modeL It is quite common for CFA with many variables and many
subjects to produce initial models which do not adequately reflect the data (Newcomb,
1994). In these situations, correlated residuals are added to improve the fit. These
residuals are chosen using the Legrange Multiplier Modification Index to insure that
the modifications do not disturb the critical features of the modeL They are only
included if they make sense theoretically.
Once an adequate CFA is produced, a structural equation model (SEM) is
developed to examine the relationships among observed variables and latent fectors.
The model is over-fit with the addition of the correlated residuals identified in the
CFA, to capture all possible association. Then the SEM is trimmed by systematically
deleting aU non-significant paths using the multivariate Wald test ÇMacCallum, 1986)
to make it parsimonious and easier to interpret
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
47
The chi-square statistic is commonly used to assess how well a model reflects
its data. Maximum likelihood chi-square estimates that have non-significant p values
are often accepted as criteria for not rejecting an SEM. However, the chi-square
statistic is sensitive to the number of variables and to sample size (Newcomb, 1994).
In the present study with a large sançle size and 30 variables, other goodness of fit
estimators were necessary: the comparative fit index (CFI, Bentler, 1990) that adjusts
for sanq)le size, the normed fit index (NFI, Bentler & Bonett, 1980), and the non-
normed fit index (NNFI, Bentler, 1989). These indices range in value between 0 and 1
and indicate the proportion o f sample covariations accounted for by the structural
equation modeL Values greater than 0.90 signify an adequate fit of the model
(Newcomb, 1990). The NNFI corrects for degrees of freedom and is identical to the
Tucker-Lewis index (Bentler, 1995b).
Two sets o f models were produced. The first set contains data from all
subjects, but does not include information about method o f discovery o f the breast
problem, because only women in Group 2 were asked that question. The second set o f
models analyzes a sub-sample o f data from the 721 Group 2 women only, but includes
the data on method o f discovery.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
48
Results
Multivariate analyses were performed in four steps on each o f two models.
Data from the full sample o f2,307 participants was examined in Model 1, and from
the sub-sample o f 721 participants in Model 2.
Model 1
Measurement Model
Step 1. In the first step, an initial confirmatory fector analysis (CFA) model
assessed whether the latent fectors were captured by the chosen measured variables
in a statistically reliable way and determined the latent fector intercorrelations. All
fector loadings were freed, latent factor variances were constrained at 1.00, and all
latent fectors were allowed to correlate freely with each other and with the measured
variables not c^tured in latent fectors.
There were three latent fectors. The socioeconomic status (SES) latent fector
was reflected in three measured variables: the occupational levels of the subject and
her spouse, if any; the census tract o f residence using 1980 census data for median
income; and the average level of education of the subject and spouse. The Licit
Drug Use latent fector was captured by measures of alcohol consumption at ages 25
and 40 years old and during the reference year. Le., the year ending one year before
diagnosis, and by lifetime cigarette smoking experience. The Nursing latent fector
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
49
was reflected in the total months of breast-feeding and the number of children
nursed.
Table 8 presents the fit indices for Model 1: the ratio of chi-square to degrees
of freedom, (ÿ\ the Conçarative Fit Index (CFI, Bentler, 1990); the Normed Fit
Index (N ET, Bentler & Bonett, 1980); and the Non-Normed Fit Index (NNFI,
Bentler, 1995b). The initial CFA for the full model did not meet the standards for a
well-fitting modeL Acceptable values for models with high number of subjects and
many variables are <^less than 2, and CFI, N ET, and NNFI greater than 0.90
(Newcomb, 1990).
Table 8.
Fit Indices for the Full Sample. Model 1
Analysis
(n = 2,307)
CFI NFI NNFI
Initial CFA 9.29 0.90 0.89 0.73
Final CFA 4.68 0.96 0.95 0.88
Final SEM 2.80 0.96 0.94 0.94
Note. CFA = Confirmatory Factor Analysis; SEM =
Structural Equation Model; CFI = Conçarative Fit Index
(Bentler, 1990); N E T = Normed Fit Index (Bentler &
Bonett, 1980); NNFI = Non-Normed Fit Index gentler,
1995b).
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
50
Step 2. The fit o f the initial CFA model was improved in step 2. Two
covariances, suggested by the Lagrange multiplier test (Chou & Bentler, 1990) were
added in the final CFA model between the error residuals and a measured variable.
Each of these covariances linked the marital status o f the subject and the residual o f
another variable which involved the subject's spouse, if she had one. The two
residuals that were allowed to correlate with marital status in the final CFAs were the
couple's combined occupational level and the couple's average educational level
The fit indices in the final CFA models were greatly improved (see Table 8), and
with the exception of % ^ / df, indicated the models fit the data reasonably well.
The final CFA for Model 1 is graphically presented in Figure 3. The latent
factors were captured well by their manifest measures. The standardized latent
factor loadings were aU significant (p < .001) and substantial The residual variances
of the observed variables are shown in small circles. The two correlated residuals
are displayed as double headed arrows. None of the latent fectors significantly
correlated with each other, le.. Socioeconomic Status, Licit Drug Use, and Nursing
experience were independent latent fectors.
The intercorrelations among measured variables and latent fectors for the full
sample final CFA are shown in Table 9. Both significant and non-significant
correlations are tabulated. Most o f the significant correlations are in the
hypothesized directions.
As expected, race, religion, and marital status correlated with the latent
fector, SES, which reflected formal education, and residence and occiq>ational status
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
51
Fietire 3. Confirmatory Factor Analysis Model 1, Full Sanq)ie
Factor structure from the confirmatory fector analysis (CFA) model for ail
subjects: rectangles are measured variables, large circles are latent constructs,
and small circles are residuals; single beaded arrows are fector loadings;
double beaded arrows are correlations; parameters are standardized.
Significance: *** p<.001. Family H = femily bistory; B Problem= breast
problems; SES = socioeconomic status; Alcohol, 25 = alcohol consumption
at 25 years old; Alcohol 40 = alcohol consunq)tion at 40 years old; Recent
Ale = alcohol consumption during the year ending one year before the
diagnosis; Reg. Dr. = regular physician; Ob/Gyn = obstetrician or
gynecologist; # Nursed = number of children nursed.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
52
Family H
Reg. Dr.
B Problem
Ob/Gyn
Residence
Occupation
Education
Marital
Caucasian
EQspanic
Catholic
Jewish
Mammogram Stage
Smoking
Alcohol, 25
Alcohol, 40
1. 00*** Obesity
# Nursed
Abortions
77**tl Nursing
Duration
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
CD
■ D
O
Q .
C
g
Q .
■ D
CD
C/)
W
o'
3
O
3
CD
8
"D
( O '
3 "
Ï
3
CD
"n
c
3 .
3 "
CD
CD
■ D
O
Q .
C
a
O
3
■D
O
CD
Q .
■ D
CD
( / )
cn
Table 9.
Confirmatory Factor Analysis Intercorrelations for the Full Sample (part 1 of 3)
Latent Factor or 1 . 3 . 4. 7 . 8. 10.
measured yariable
1, family history Parts o f
Table
2, breast problems
3. age .02 -.03
4. SES Factor
5. Caucasian
6. Hispanic
7. Catholic
8. Jewish
9. marital status
10. Licit Drug Use
Factor
11. caffeine
12. obesity
.02
.04*
-.03
-.02
.03*
.02
-.01
.01
-.00
.11*** -.18*** —
.06 *** -.02 .47***
.04* -.01
.04* -.04*
.08*** -.04
.04* .01
.02 .00 -.03
,44*** -.81*** —
,20*** -.25*** .35*** —
25*** .14*** -.11*** -.24*** —
32*** .09*** -.06 ** -.03 .05**
. 10*** - .12*** -.01
.04* -.08*** -.12*** .05* -.02 .03
.15*** .03 -.25 *** -.17*** .11 *** .05*
- .10* * * - .1 1 * * * —
-.11*** -.06*** .12***
-.06 ** -.05** -.07**
■ o
o
Q.
C
g
Q.
■ O
CD
( / )
o "
3
O
8
■ D
3.
( O
3 "
i
3
CD
3 .
3 "
CD
CD
T D
O
Q.
C
a
O
3
" O
O
CD
Q.
" O
CD
C/)
C/)
Table 9.
Confirmatory Factor Analysis Intercorrelations for the Full Sample (Continued, part 2 of 3 1
Latent Factor or
measured yariable
1. 2. 3. 4. 5. 6. 7. 8. 9. 10.
13. abortions .01 .02 -.01 -.01 -.04 .02 -.04* .05*
.06***
.14***
14. exercise .02 .03 -.03 .24 *** .10*** -.09 *** -.05** .07 ***
07 ***
-.04
15. regular Dr. .00 .06*** -.00 .11*** .03* -.06 ** -.02 .10*** .04** .01
16. ob/gyn .00 .05* -.10***
14***
.04 -.04 .02 .03 .02 .02
17. hormones -.03 .12*** .03
10***
.08*** -.05* -.05* .02 .06 *** -.02
18. mammograms .08*** .34*** .08*** .25*** .08*** -.11***
.09***
.19*** .06*** -.03
19. pregnancy .01 -.02 .01 -.05* -.03 .03 -.04* .04*
08 ***
-.02
20. Nursing Factor .00 -.07 *** .05* -.00 -.05* .06 ** .03 -.15*** .08 *** -.03
21. stage at diagnosis -.05** -05** -.06** -.08*** -.04* .04* .02 -.04* -.02 -.02
CD
" O
O
Q.
C
g
Q.
■ D
CD
C/)
o "
3
O
8
■ D
(û ‘
3 "
i
3
CD
3 .
3 "
CD
CD
■ D
O
Q.
C
a
O
3
■ D
O
CD
Q.
■ D
CD
(/)
C/)
Table 9.
Confirmatory Factor Analysis Intercorrelations for the Full Sample (Continued, part 3 of 3 1
Latent Factor or
measured variable
11. 12. 13. 14. 15. 16. 17. 18. 19. 20.
11. caffeine
12. obesity .01
13. abortions -.06 ** -.00
. . .
14. exercise -.09*** -.07*** -.00
15. regular Dr. -.04 -.02 -.00 .04*
. . .
16. ob/gyn -.06** -.10*** .04* .03
18***
— -
17. hormones -.03 -.15*** .01 .02 .13*** .15*** ---
18. mammograms -.10*** -.11*** .01 .10*** .19*** .12*** .16*** ---
19. pregnancy .03 .07*** .08*** -.04 .01 .02 .04 .00 —
20. Nursing Factor ,03 .06 ** -.06** .01 -.02 -.02 -.02 -.04 .24*** —
21. stage at diagnosis .02 .07*** -.02 -.02 -.03 -.05** -.09*** -.11*** .04 .03
Note. *p < .05, ** p <.01, *** p <.001; ob/gyn = obstetrician or gynecologist; empty rows not shown.
L A
L n
56
as proxy measures of income. Caucasian and Jewish variables correlated positively
with each other and positively with SES. Hispanic and Catholic correlated positively
with each other and negatively with SES. Being married, or cohabiting, during the
reference year was also linked to higher SES. These variables and the SES latent
factor follow the expected pattern in correlating with other variables in the table.
Higher SES and its correlates were linked with more breast problems, less obesity,
more exercise, and with more utilization of medical care: having a regular physician,
having obstetrician or gynecologist, using hormones, and having more
mammograms.
Significant correlations with lower stage included having a fomily history o f
breast cancer, having a personal history o f benign breast disease, higher SES, and
being Caucasian, or Jewish. Having an obstetrician or gynecologist as her regular
physician, having taken hormones, and more mammograms were also significantly
associated with lower stage. Being Hispanic or obese was associated with more
advanced stage. All these relationships were as hypothesized. An unexpected
correlation involved the age of the subject. Age was positively correlated with
number of mammograms and negatively correlated with stage, contrary to published
reports linking older women with fewer mammograms and stage that is more
advanced.
Table 9 contains three other significant correlations, which were not
ejq)ected. Being Catholic is negatively correlated with whether a woman had ever
been pregnant. It was expected that more, not fewer. Catholic women would have
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
57
had at least one pregnancy. Being obese was negatively correlated with smoking and
the use of alcohol in the Licit Drug Use latent fector. High consumption o f can in e
was negatively correlated with a history o f elective abortions. Both of these
correlations were contrary to the Conventionality-Deviance prediction that deviant
behaviors would intercorrelate positively.
Since the effect of the other variables was not statistically partiaUed out of
these correlations, the table cannot show whether an association is in feet the result
o f both variables being confounded with a third. This level o f analysis is
acconçlished by the Structural, or Path, Models.
Structural Models
Step 3. Once the adequacy of the CFA models was determined, a saturated
structural equation model (SEM) was constructed to test influences among latent
fectors and measured variables. The paths in the structural model showed the unique
effects, with the influences o f other variables and fectors statistically partialled out.
Similar to the backward or step-down process in stepwise multiple regression, all
paths were initially included in step 3, whether hypothesized or not, and aH
correlations were allowed (Chou, & Bentler, 1990). Each level predicted the next as
a path. That is, the distal theoretical variables and latent factors classified under the
theoretical headings of susceptibility, barriers, and conventionality predicted the
health care measured variables and latent fectors. Health care in turn predicted the
screening variables. Finally, screening predicted stage at diagnosis. Paths firom
earlier levels were also allowed to predict screening and stage directly.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
58
Step 4. In the fourth step, the over-fit model was trimmed of non-significant
paths (MacCallum, 1986). When a significant path coefficient and its corresponding
significant correlation had opposite signs, it demonstrated a siq)pressor effect.
Suppressing paths were eliminated from the modeL The final SEM for the full
sample fit the data quite well (see Table 8).
The standardized solutions for the final SEM for Model 1 are presented
diagrammaticaUy in a pair of figures, because they are too conqjlex to present clearly
on one page. Figures 4 and 5 display the SEM 1 for the full sample. It is important
to note that the two diagrams represent the halves o f one structural modeL and they
must be considered in conjunction with each other. The first half o f Model 1 (Figure
4) presents the measured variables and latent fectors that predict the mediating
variables o f health care practices. The second half of Model 1 (Figure 5) shows the
paths that directly predict mammograms or stage.
Figure 4 shows variables and fectors in both the Susceptibility and Barriers
conçonents of the HEM predicted whether the partic^ant had a regular physician.
As ejq)ected, having a personal history o f breast problems, being Jewish, or having a
higher SES predicted having a regular physician.
Variables or fectors fiom both theories predicted the regular physician being
an obstetrician or gynecologist. Predictors from the Barriers component were SES,
or being Catholic, or younger. Le., age was a negative path. From the
Conventionality component, drinking less caffeine, being thinner or having had
induced abortions, predicted seeing an obstetrician or gynecologist.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
59
Figure 4. Structural Equation Modal 1, Full Sample, Section 1 :
Prediction of Health Care Practices
First half of the final path model for all subjects; arrows are paths, rectangles are
measured variables, large circles are latent fectors, and small circles are residuals;
parameter estimates are standardized. Significance: * p< .05; ** p< .01; *** p< .001.
Family H = femily history; B Problem = breast problems; SES = socioeconomic
status; Reg. Dr. = regular physician; Ob/Gyn = obstetrician or gynecologist.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
60
- . 04*
Family H Reg, Dr.
-.05** B Problem
04* Age
SES
Ob/Gyn
07*
m il
Marital
Caucasian
Hispanic Hormones
07**
.04*
Catholic
07***
Jewish
\ .05*
\ - .1 5 * * *
Licit
Drug Use
Pregnant
-.0 5 * *
Caffeine
Obesity
jlfl***
Abortions
Nursing
Exercise
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
61
Figure S. Structural Equation Model 1, Full Sample, Section 2:
Direct Predictors o f Mammogram and Stage
Second half of the final path model for aU subjects: arrows are paths, rectangles are
measured variables, large circles are latent fectors, and small circles are residuals; the
dashed line is a specific, nonstandard, path; parameter estimates are standardized.
Significance: * p< .05; ** p< .01; *** p< .001. Family H = femily history;
B Problem = breast problems; SES = socioeconomic status; smok = smoking; Reg. Dr.
= regular physician; Ob/Gyn = obstetrician or gynecologist.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
62
liJSl
Family E L
. 12*** Reg. Dr.
.29***
B.ProbIem
Ob/Gyn
.15***
Manta!
Caucasian
-.07***
Mammogram
Buspanic
.08***
Catholic
Hormone
.07***
Jewish
Smok I
Caneme
Obesity
( Nursing )
Abortions
Exercise
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
63
One additional modification was made of SEM 1. A special effect was
identified and added to the model At Step 1, three latent fectors were identified by
CFA. For exançle, the Licit Drug Use latent fector represented the shared variance
among the four measured variables o f alcohol use at three ages, and smoking
experience. Throughout subsequent analyses, the latent fector was used, and the
specific effects of its individual variables were relegated to error residuals. In the
final modification of the models, these specific effects were examined to determine
whether any o f them might be a significant predictor individually, distinct and
independent from the latent fector (Newcomb, 1994). Guided by the Legrange
Multiplier Modification index, one specific, or nonstandard, effect was foxmd and
added to the full sample model It is shown as a dashed square and line in Figure 5.
The smoking component o f the Licit Drug Use latent fector was a specific negative-
predictor o f mammograms and this influence was significant even when the shared
influence among alcohol use and smoking had been statistically partialled out.
Women who smoked more had received fewer mammograms.
Figure 5 shows the direct predictors o f mammogram and stage. Ten standard
paths (shown as solid lines) and one specific effect (shown as a dashed line) predict
the number o f mammograms, aU in the hypothesized directions, except that being
older predicts more mammograms. In addition to mammograms, five other paths
directly predict stage: family history, age, SES, hormone use, and obesity. These
paths are also in the hypothesized directions, except that being older predicts lower
stage.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
64
Model 2
The data analyzed in Model 2 comes from the sub-sample of 721 participants
(Group 2) and has three variables more than Model 1. These variables present
information about the method of discovery o f breast cancer symptoms, vdiich was
only obtained from women in Group 2. The analyses of Model 2 followed the same
process already described for Model 1.
Measurement Model
Step 1. The initial CFA for Model 2 was produced in the same manner as for
Model 1. Table 10 shows the fit indices for Model 2. The smaller sample size
allowed the ratio of chi-square to degrees o f freedom to be Iowa: than for the full
sample (see Table 8), and the Conçarative Fit Index (Bentler, 1990) and Normed Fit
Table 10.
Fit Indices for the Sub-Sample. Model 2
Analysis
(n=721)
CFI NFI NNFI
Initial CFA 3.65 0.93 0.91 0.80
Final CFA 2.48 0.96 0.94 0.89
Final SEM 1.61 0.96 0.91 0.95
Note. CFA = Confirmatory Factor Analysis; SEM =
Structural Equation Model; CFI = Conqjarative Fit Index
(Bentler, 1990); N E T = Normed Fit Index (Bentler & Bonett,
1980); NNFI = Non-Normed Fit Index (Bentler, 1995b).
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
65
Index (Bentler & Bonett, 1980) were over 0.90, but the Non-Normed Fit Index
(Bentler, 1995b) was not.
Step 2. As in Model 1, the fit of the CFA for Model 2 was improved by
adding covariances between the marital status variable and the residuals o f two
variables that included data about the participant's spouse, if she had one. This
modification improved the fit o f the model to acceptable values (see Table 10).
Figure 6 gives a graphic representation o f the CFA for Model 2. The
standardized foctor loadings were all significant (p < .001) and substantial, and the
latent fectors were independent, as in Model 1.
Table 11 presents the intercorrelations, significant or not, of the measured
variables and latent foctors in the final CFA o f the Model 2 sub-sample. The smaller
number o f subjects gave lower power to these analyses, and so fewer correlations
were significant. However, with only one exception, the significant correlations in
Table 11 agree in sign with their counterparts on Table 9. The exception was the
negative correlation of age and number of mammograms in Table 11.
Table 11 also contains the method o f discovery data not available for the
entire sample. For example, it shows the breast symptoms o f high SES women and
of married women were more often discovered by mammogram. It shows that
Caucasian women and women in higher SES levels are less likely to discover their
breast symptoms themselves. Women with regular physicians were more likely to
have their symptoms discovered by a mammogram. Obstetricians and gynecologists
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
6 6
Figure 6. Confirmatory Factor Analysis Model 2, Sub-Sanq)le
Factor structure from the confirmatory Actor analysis (CFA) model fi>r Group 2:
rectangles are measured variables, large circles are latent constructs, and small
circles are residuals; single headed arrows are Actor loadings; double headed arrows
are correAtions; parameters are standardized. Significance: *** p<.001. Family H =
Amily history; B Problem = breast problems; SES = socioeconomic status; Alcohol,
25 = alcohol consumption at 25 years old; Alcohol 40 = alcohol consumption at 40
years old; Recent Ale = alcohol consumption during the year ending one year before
the diagnosis; Reg. Dr. = regular physician; Ob/Gyn = obstetrician or gynecologist;
# Nursed = number of children nursed; By M.D. = by physician; By Mmgm = by
mammogram.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
67
Family H
Reg. Dr.
B Problem
Ob/Gyn
Residence
Occnpation
Education
Marital
Caucasian
EQspanic
Catholic
Jewish
Mammogram
By Self
Stage
ByM .D.
By Mmgm
Smoking
Alcohol, 25
Alcohol, 40
# Nursed
Abortions
Nursmg
.77***
Duration
Exercise
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
CD
■ D
O
Q .
C
g
Q .
■ D
CD
C/)
W
o'
3
0
3
CD
8
■D
( O '
3 "
1
3
CD
"n
c
3 .
3 "
CD
CD
■ D
O
Q .
C
a
O
3
■D
O
CD
Q .
■ D
CD
( /)
( /)
Tablell.
Confirmatory Factor Analysis Intercorrelations for the Sub-Sample, n = 721 (Part 1 of 3)
Latent Factor or
measured yariabie
1 . 2. 3. 5. 6. 7. 8. 9. 10. 1 1 .
1. family history
2. breast problems .02
3. age .02
4. SES -.01
5. Caucasian .04
6. Hispanic -.02
7. Catholic -.03
8. Jewish -.00
9. marital status .00
10. Licit Drug Use .02
11. caffeine .07
12. obesity .05
13. abortions -.02
06 —
17*** -.18***
13*** .07*
07* -.01
05 -.02
09** -.07
07*
05
04
-.01
-.03
- .10**
15*** .02
01 -.02
.52***
-.39***
-.15***
.31***
.35***
-.05
-.13**
-.28***
-.02
-.62***
-.17***
1 9 * * *
.15***
.08*
.04
-.23***
-.09*
Parts of
Table
.39***
- .12**
-.04
- .12 * *
.03
.08*
.04
-.22*** - - -
.00 .12***
-.00 -.07 -.13***
------
.10** -.11** -.04 .07 —
.03 -.13*** -.08** -.01 .02
-.05 .02 -.09** .04 -.13***
&
■ D
O
Q.
C
g
Q.
■ O
CD
C/)
W
o'
3
3
CD
8
( O '
3 "
i
3
CD
"n
c
3 .
3 "
CD
CD
" O
O
Q.
C
g
O
3
" O
O
CD
Q.
■ D
CD
C/)
C/)
Table 11.
Confirmatory Factor Analysis Intercorrelations for the Sub-Sample, n = 721 ('Continued. Part 2 of 3 ^
Latent Factor or
measured variable
1. 2. 3. 4. 5. 6. 7. 8. 9. 10. 11.
14. exercise .01 .08* -.02 .31*** .12*** -.12*** -.04 .12*** .13*** -.07
-.14***
15. regular Dr. .01 .04 -.07* .16*** .02 -.08* -.04 .11** .11*** .05 -.03
16. ob/gyn .01 .05 -.07* .21*** .04 -.05 -.02 .03 .06* .04 -.11**
17. hormones -.06* .16*** .00 .23*** JU*** .00 -.05 .10** .09** -.00 -.04
18. mammograms .05 .30*** -.07* .38*** .22*** -.18*** -.09** .19*** .13*** -.03 -.13***
19. pregnancy .04 -.00 .02 -.05 -.03 .03 -.06 .06 .06 -.05 .05
20. Nursing .02 -J2*** .10** .01 -.03 .07 .04 -.15*** .04 .02 .05
21. by self .04 .03 .02 -.09* -.09* .04 -.02 -.02 -.04 -.04 -.08*
22. by MD -.04 .05 .05 -.01 .03 -.01 -.05 .03 -.04 -.00 -.00
23. by mmgm -.02 -.06 -.05 .08* .05 -.03 .05 .01 .06* .03 .06
24. stage -.04 -.03 -.04 -.08* -.06* .05 .01 -.05 -.05 .01 .06
■ o
I
I
■ o
CD
C/)
œ
o'
3
8
c 5 '
3
i
3
CD
C
p .
3 "
CD
CD
■ o
O
Q .
O
3
■ o
O
CD
Q .
■ D
CD
C/)
(/)
Tablell.
Confirmatory Factor Analysis Intercorrelations for the Sub-Sample, n = 721 (Continued. Part 3 of 3 ^
Latent Factor or
measured variable
12. 13. 14. 15. 16. 17. 18. 19. 20. 21. 22. 23.
13. abortions .01
14. exercise -.06 -.04 —
15. regular Dr. .02 -.01 .09**
16, ob/gyn -A \* * .11** .05 .18*** ---
17. hormones -.20*** .00 .05 .17*** .20***
18. mammograms -.14*** -.04 .19*** .30*** .17*** .26*** —
19. pregnancy .05 .08* -.06 .03 .01 .02 .05 ---
20. Nursing .06 -.05 -.01 -.07 -.00 -.06 -.09* .23*** ---
21. by self -.04 .00 -.01 -.09** -.08* -.06 -.08* .00 -.02 ---
22. by MD -.05 .03 -.01 .00 .09** .03 -.01 .02 -.03 -.20*** - -
23. by mmgm .06 -.02 .02 .09* .04 .05 .09* -.00 -.01 -.84*** -.32*** —
24. stage .10** -.07 -.02 -.06 -.08* -.12*** -.13*** -.00 .03
20***
.05 -.25***
Note. *p < .05, ** p <.01, *** p <.001; ob/gyn = obstetrician or gynecologist; by MD = by physician; by mmgm = by
mammogram; empty rows not shown.
71
discovered breast cancer symptoms by physical examination more often than
physicians in other specialties did.
Structural Model
Steps 3 and 4. A saturated SEM was generated for Model 2 and trimmed o f
non-significant and suppressing paths by the same processes described for Model 1.
Figures 7 and 8 present the two halves of the final SEM for Model 2.
There are 45 significant standard paths in the full sanqjle SEM 1, and 36 in
the sub-sample SEM 2. This loss of nme paths in the sub-sample SEM 2 occurred
despite the addition of seven paths associated with the method o f discovery
variables. There are three explanations for having fewer paths in the sub-sample
model First, fewer subjects gave lower power to the sub-sample analyses,
preventing some paths from reaching statistical significance. Second, the discovery
paths may have claimed some o f the variance attributed to other paths in the full
sample SEM. Third, the sub-sample's demographic distribution was different and
the second model may have accurately presented those differences. The sub-san^le
was more ethnically diverse, containing African American women not included in
Groiqj 1, however Tables 2 through 7 show the full sample and sub-sample are
otherwise very similar. There are 27 significant paths that appear in both the full
sample and the sub-sample models. In every case, these corresponding paths agree
in sign, leading to the same interpretation of relationships among the variables and
factors.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
72
Figure 7. Structural Equation Model 2, Sub-Sample, Section 1:
Prediction o f Health Care Practices
First half of the final path model for Group 2 subjects: arrows are paths, rectangles are
measured variables, large circles are latent fectors, and small circles are residuals;
parameter estimates are standardized. Significance: * p< .05; ** p< .01; *** p< .001.
Family H = family history; B Problem = breast problems; SES = socioeconomic
status; Reg. Dr. = regular physician; Ob/Gyn = obstetrician or gynecologist.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
73
Reg. Dr.
Family H
B Problem
. 10**
Age
17***
.9 5
SES
Ob/Gyn
07*
Marital
Caucasian
- . 10*
Hormones
Hispanic
Catholic
N .07*
-.1 6 * * *
Jewish
Licit
Drug Use
Pregnant
- .0 7 *
Caffeine
07*
I D * *
Abortions
^ Nursing
Exercise
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
74
Figure 8. Structural Equation Model 2, Sub-Sample, Section 2:
Direct Predictors of Screening and Stage
Second half o f the final path model fijr Group 2 subjects: arrows are paths, rectangles
are measured variables, large circles are latent fectors, and small circles are residuals;
parameter estimates are standardized. Significance: * p< .05; ** p< .01; *** p< .001.
Family H = family history; B Problem = breast problems; SES = socioeconomic
status; Reg. Dr. = regular physician; Ob/Gyn = obstetrician or gynecologist; By M.D.
= by physician; By Mmgm = by mammogram.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
75
. 06*
Family E L
Reg. Dr.
B.Problem
Age
-.05*
Ob/Gyn
. 10**
SES
.7 3
- . 08*
Mammogram
Marital
.9 8
Caucasian
stage Hispanic By Self
Hormone
-.0 7 * Catholic
.9 1
Jewish ByMJ>.
.99
Pregnant Licit
Drag Use
By Mmgm
Caffeine
.07*
Obesity
Abortions
Nursing
Exercise
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
76
The discovery of breast cancer symptoms by mammography is the strongest
(negative) predictor of stage in either model Health care variables predict discovery
variables, as expected. Having a regular physician predicts symptoms will be
discovered by mammography rather than by self-examination.
Other than the method-of-discovery paths, only two paths appear in the sub
sample Model 2 that were not present in the full sample Model 1. In the full sample
SEM 1, a suppressor path (Caucasian predicting not having a regular physician) was
eliminated because its sign was negative on the path coefficient and positive on the
corresponding significant intercorrelation. When it was removed, the path from
Hispanic to Regular Dr. lost significance and no longer appeared in Model 1. In the
sub-sample SEM 2, the path from Caucasian to Regular Dr. was also negative, but its
corresponding intercorrelation was not significant. It was assumed that the path
played a suppressor role in both models, but it was retained in Figure 7 because it did
not meet the full criteria for removal. Consequently, the path from Hispanic to
Regular Dr. was also retained in the sub-sample Model 2.
The addition o f the discovery variables did inçrove the prediction o f stage,
as hypothesized. The residual variance for stage in Figure 5 was 0.97 and in Figure
8 is 0.91. Nine percent of the variance in stage was accounted for by the sub-sarrçle
model, conçared to three percent by the full sanç)le modeL
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
77
Discussion
Among a population-based sançie o f women who have been diagnosed with
primary breast cancer, a set of demographic and behavioral fectors and variables is
identified that predicts the stage at which the disease is diagnosed. Because long-term
survival firom breast cancer is highly dependent upon early diagnosis, these findings
can provide important, perhaps life-saving, information in guiding intervention
programs to identify women at risk of late stage diagnosis.
Implications on Hvpotheses
The theoretical models in Figures 1 and 2 suggest that the Health Beliefs
Model conçonents o f Perceived Susceptibility and Barriers, and the Problem
Behaviors Theory dimension o f Conventionality would guide the prediction of breast
cancer stage. These distal variables are hypothesized to be mediated by health care
practices, with screening behavior being the most proximal predictor of stage.
The paths o f the structural equation models (SE A /Q present the unique
contribution each variable or foctor makes to the predicted variable. The effects o f
other correlated variables and fectors have been statistically partialled out All the
paths in these final models show significant but modest to moderate effect size,
despite the large sample. This may reflect the complexity o f the influences that
determine health care, screening, and stage, many of which are beyond the scope o f
this study.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
78
The full sample and the sub-sample models generally agree in the patterns of
influence they present. The sub-sample is more ethnically diverse, includes method-
of-discovery variables, and fewer subjects, and its model has fewer sigmficant paths.
However, all the paths in the sub-sample SEM agree in sign with the corresponding
paths in the full sangle model Unless otherwise stated, the predictions discussed
below are seen in the full sample model
Screening
As expected, the SEM for the sub-sanq)le shows the strongest predictor of
stage at diagnosis is the method o f first discovery o f the symptoms that lead to the
diagnosis of breast cancer. Women who discover the disease by means of
mammography, rather than by self-examination or during a physician's examination^
are diagnosed at an earlier stage.
Also as hypothesized, both models show that having more mammograms
predicts lower stage, even when the influence o f detection by mammogram has been
statistically partialled out. However, the strength of the mammography effect is
smaller than anticipated. It had been ejq)ected that behavioral and demographic
elements would affect stage indirectly by influencing utilization of mammography
But, in addition, femily history, age, socioeconomic status (SES), obesity, and
hormone use predict stage directly in the full sample SEM In feet, these variables and
fectors are as important as the number of mammograms in predicting stage.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
79
Health Care Practices
The proposed theoretical models hypothesize that mediating variables
involving health care practices would predict screening. As ejq)ected, having a regular
physician, seeing an obstetrician or gynecologist, and taking hormones each predicts
more manunography. Hormone use appears to be a useful proxy indicator o f access to
medical care in this population o f peri-menopausal women.
The detection o f breast cancer symptoms by mammography is predicted, in
the sub-sample SEM, by having a regular physician, as e je c te d . But, contrary to
expectations, whether or not that physician is an obstetrician or gynecologist is not
significant. Without a regular physician or an obstetrician or gynecologist, symptoms
are more often detected by self examination.
In the multivariate analysis, only hormone use, o f all the health care practices
variables, predicts stage. Women who use estrogen and/or progesterone are diagnosed
at an earlier stage. Perhaps the known association o f estrogen use and breast cancer
(American Cancer Society, 1997) prompts physicians prescribing it, or women taking
it, to be especially diligent in screening for breast cancer.
Women who have a family history o f breast cancer are less likely to use
hormones. The recognition o f hormone use as a breast cancer risk fector may have
influenced these women, whose mothers or sister(s) had had breast cancer, to avoid
further increasing their own individual risk.
A history of breast problems predicts more hormone use, but the interpretation
of this finding is unclear. The causal direction may be reversed. Taking estrogen
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
80
increases breast tenderness and other cyclic breast changes in some women, which
may have lead to more diagnoses o f breast problems.
Pregnancy and nursing experiences are ejqplored in this study to determine
whether they might have increased a woman's body-awareness, and her self
monitoring, leading to lower stage at breast cancer diagnosis. No support is found for
either effect. Neither having been pregnant nor the latent fector reflecting nursing
experience significantly predicts mammography, method of detection, or stage in this
sample.
Susceptibüitv
The perceived susceptibility hypothesis o f the Health Belief Model is
supported in this study. It suggests women who feel more vulnerable to breast cancer
wül increase their health care practices, especially screening behaviors, and and that
will indirectly facilitate diagnosis at a lower stage. Perceived susceptibility is
operationalized by the variables o f family history of breast cancer, and personal
history o f breast problems or benign breast disease. In this sample, family history o f
breast cancer predicts less hormone use, more mammography, and earlier stage at
diagnosis. It is encouraging to see that contrary to earlier reports (Turman &
Schwartz, 1993), women vdio are at higher risk of developing breast cancer are taking
steps to improve their survival.
The full sample SEM also documents a modest association between previous
breast disease and the number of mammograms, as hypothesized. However, this
finding may be biased. Previous breast problems, as defined in this study, include the
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
81
stipulation of diagnosis by a physician. Since mammography is a typical early step in
such diagnoses, the association can be interpreted in two ways. It may mean that
women who had previous breast problems sought more mammography, as
hypothesized. However, it could partially reflect that undiagnosed breast problems are
excluded by this definition of breast disease.
Barriers
The Barriers component of the Health Beliefs Model is supported in several
ways. The theory hypothesizes various social constructs would act as barriers to
health care practices, indirectly reducing screening, and increasing the stage of breast
cancer at diagnosis. In this study, the Barriers conçonent contains age, racial and
ethnic, religious, and marital variables, and the latent Actor, SES, which reflects proxy
family income measures, and education.
SES. Caucasian, Jewish, and married women are wealthier in this sanq)le.
Hispanic and Catholic women, and women vdio were not married during the reference
year. Le., the year ending one year before diagnosis, are poorer. SES and most of its
associated variables predict health care practices and mammography in the expected
directions. Wealthier women have more health care utilization, including more
mammograms. The exception to the ejq)ected pattern is that being Catholic predicts
the participant is seeing an obstetrician or gynecologist as her regular physician.
Women who detected their breast cancer symptoms by mammography are also
married and have a higher SES level. Self-detection is associated with lower SES and
non-Caucasian women, and having fewer mammograms. Despite the feet that high
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
82
SES increases the risk of developing breast cancer, high SES also directly predicts
earlier detection o f the disease, even after the effects of health care access and
mammography are partialled out.
Age. It has been widely reported that older women are diagnosed at a higher
stage of breast cancer (Goodwin, 1986). Common explanations for the finding include
lower access to health care in general, and lower utilization of breast cancer screening
in particular (Satariano et aL, 1986). The older women in this sample are poorer, and
fewer of them see an obstetrician or gynecologist, consistent with published reports.
However, age of these women is not associated with having a regular physician.
Contrary to published reports, in this sample o f women, increasing age predicts lower
stage, possibly because the older women in this study have received more, not less,
mammography.
Education and income do not ^p ear to be confounding the age effect. This is
a rather highly educated sample of women: 55% have received some college
education. In addition, the distribution of census tract of residence data is skewed
toward including more high than low income women. However, in multivariate
analysis, age predicts lower stage even after the effects of income and education have
been statistically partialled out to remove their influence from the age effect. These
findings do not challenge the Barriers hypothesis. Rather, they suggest age may not
have functioned as a barrier to breast health care in this sample. The optimistic
interpretation o f these findings is that programs such as Medi-Care have fecilitated
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
83
access to medical care for older women and that efforts to encourage them to utilize
preventive screening for breast cancer have been successful.
Race. These results do not agree with several studies that have reported racial
differences in stage (Hunter, et aL, 1995; Zaloznic, 1995). African American women
conç)rise 11.4%, and Hispanic women, 9.0% o f this sub-sample and 3.6% and 7.6%,
respectively, of the full sample. Being Caucasian is correlated with lower stage in
univariate analyses. However, there are no significant racial or ethnic paths, in either
SEM, directly predicting stage in the multivariate analyses, contrary to the hypothesis
that race would act as a barrier to early stage diagnosis. Racial variables are highly
correlated with the SES latent fector. It may be that SES explains the variance
between race and stage in the structural models, as has been suggested by other studies
(Hreen & Figueroa, 1996; Dayal et aL, 1982). With the influence of other variables
and factors, including SES, partialled out, race is not a barrier to early diagnosis in this
sample.
Conventional itv-Deviance
The Problem Behavior Theory hypothesizes that a deviant lifestyle is a
negative predictor o f health care and screening, leading to diagnosis at a higher stage.
It receives mixed support in this study, with some variables and fectors predicting as
expected and others non-significant.
The dimension of conventionality-deviance is operationalized by alcohol,
tobacco, and caffeine consumption, obesity, and abortion at the deviant end, and
exercise at the conventional end o f the spectrum. It is hypothesized that the deviant
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
84
behaviors would correlate positively with each other and negatively with conventional
behaviors.
Licit Drug Use. As e^gected, smoking and alcohol use significantly load onto
a latent fector called Licit Drug Use. This factor is positively correlated with caffeine
use and abortions, but contrary to the Conventionality-Deviance hypothesis. Licit
Drug Use did not predict health care practices, screening or stage. However, a specific
effect from smoking is noted. Women who smoke receive significantly fewer
mammograms.
Exercise. Exercise is negatively correlated with caffeine use and obesity, as
hypothesized, but is not significantly associated with Licit Drug Use or abortions.
Exercise predicts more mammography, providing some support for the
conventionality hypothesis that socially prescribed behaviors would associate
positively with each other.
Obesitv. Contrary to the conventionality-deviance hypothesis, obesity is
negatively correlated with the Drug Use fector. It may be that obesity is not a deviant
characteristic in this population. AJtematively, it may be that smoking or heavy
alcohol consumption lowered body weight in a subgroup of participants enough to
mask the group effect of obesity associating with other deviant behaviors.
Obese women are less likely to have regular care by an obstetrician or
gynecologist, as hypothesized by the Problem Behavior Theory. They have also used
fewer hormones. Since some estrogen is produced by fet tissue, obese women may
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
85
have had less need to supplement their estrogen levels when menopause decreased
their ovarian output of the hormone.
Other Findings
Obesitv
As often reported (Zumoff & Dasgupta, 1983), breast cancers o f obese women
in this sample are detected at a more advanced stage. However, the usual explanation
does not seem to fit the findings o f this study. It is often suggested (Mohle-Boetani et
al., 1988) that it is more difficult to palpitate small lumps in large breasts, so breast
cancer in obese women may have delayed detection.
In this sample o f women, however, lower stage is associated with
mammography, not with detection by physical examination. The higher percentage of
fatty tissue in obese breasts would fecüitate detection of breast tumors by
mammogr^hy. Fatty tissue is less dense, and therefore more radio logically
transparent than the glandular breast tissue o f thinner women. In addition, obesity
predicts higher stage in the second model, even A^en the influence of the method of
detection has been partialled out. These women did see fewer
obstetricians/gynecologists, but obesity did not have a significant effect on the number
o f mammograms they received. Difficulties in detection do not eîqplain why these
obese women are at increased risk of higher stage at diagnosis.
Caffeine
La this study, results involving caffeine conflict with each other, and with its
theorized role as an indicator of deviance. Less caffeine predicts symptom discovery
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
86
by the woman herself; more caffeine predicts discovery by mammography. However,
more caffeine also predicts less lifetime mammography. Intuitively, it would appear
that having more mammograms should be linked with more, not less, discovery by
mammography. It is also noted that more caffeine is associated with less previous
breast disease.
The apparent contradictions found in the relationships among caffeine use and
breast health may be the result of combining medical and behavioral influences. A
possible scenario, outside the scope of the original theorized model, is offered to
attenpt to provide a rationale o f how low mammography and high detection by
mammography can both be linked with high caffeine use. It is widely suspected that
caffeine increases production o f benign breast cysts in some womeiL These cysts may
be painful and enlarge rapidly, pronçting the woman to seek diagnosis and treatment.
Often the diagnostic procedures involve mammography, which is painful for some
women. Because this study design excluded any woman who had a previous history
of breast cancer, all such diagnoses would have been negative fi)r these subjects.
Alternately, such cysts may have been first discovered by screening mammography,
then biopsied to confirm their benign nature.
Either way, it might be speculated that women with previously diagnosed
benign breast disease may have had weeks o f anxiety about the possible outcome,
coupled with expensive and perhaps painful diagnostic procedures, which produced
negative, albeit relieving, results. The negative results may have been interpreted by
some women as showing the whole traumatic experience was unnecessary. Such an
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
87
ejq)erience may have adversely conditioned her to avoid regular mammographie
screening when she was asymptomatic.
This scenario would explain how more past mammography could be linked
with fewer discoveries by mammogr^hy, why is recent caffeine consumption the
link? Women who had never had a previously diagnosed breast problem report
consuming more caffeine. The caffeine variable utilized in this study measured recent
consumption. Assuming that coffee drinking was pervasive for this cohort m their
earlier years, women with a history o f breast cysts may have reduced their coffee
intake, or switched to decaffeinated beverages, in order to avoid another cyst. Thus,
women with previous breast problems and more mammography would have less
recent caffeine use and less detection o f breast cancer symptoms by mammography.
Women who did not get breast cysts may have continued their caffeine consumption,
and remained more willing to use mammography for screening, since for them it was
not paired to frightening memories o f past diagnostic procedures.
Theoretical Implications
Health Belief Model fHBMl
The HBM is among the most widely studied theories o f asymptomatic health
behaviors. Its conqx)nents o f Barriers and Susceptibility are well supported in this
study. Demographic barriers related to socioeconomic status decrease a woman's use
of mammogr^hy, and increase her stage at diagnosis. Perceived Susceptibility
increases screening, and lowers stage, as hypothesized. Two variables, age and race,
did not function as independent barriers in this study. Perhaps SES is the salient
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
88
barrier, rather than race, per se. The effects previously reported for age may have
been confounded by access to health care and screening. Controlling for these
variables may have prevented their influences from posing as an age effect in this
study.
Problem Behavior Theorv (PBT)
This study is the first time, to the author's knowledge, that the PBT has been
utilized to examine breast cancer related behaviors. Some of the variables chosen to
operationalize the Conventionality-Deviance dimension are not significantly
associated with each other or with the outcome variable, stage. Neither Licit Drug
Use nor history of abortions predicts stage or screening in this sample. Other
Conventionality-Deviance variables do predict as ejqpected, providing some support
for the hypothesis. Obesity predicts higher stage directly and exercise predicts more
mammography. Caffeine and obesity are negative predictors of having an obstetrician
or gynecologist.
The Conventionality-Deviance hypotheses, as applied to breast cancer, are not
yet precise enough to specify whether relationships are e^gected to be found at the
latent foctor level, or among the specific measured variables which are represented by
the latent fectors. In order to test the theory more comprehensively, the model is
examined to find significant specific or non-standard effects (Newcomb, 1994). A
specific effect is identified for smoking. Cigarette smoking predicts fewer
mammograms, in support o f the Conventionality-Deviance hypothesis.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
89
With a better-informed selection o f deviance variables, the theory may add
new aspects to our understanding o f screening behavior. None of the Health Belief
Model variables or fectors predict whether the breast cancer symptoms would be
detected by mammography, physician examination or self^xamination, but one o f the
conventionality-deviance behaviors does. Women who consumed more caffeine
discovered their breast cancer by self-examination. Those who consumed less,
discovered it through mammography. Since method of discovery is the strongest
predictor o f stage, the Problem Behavior Theory's contribution is especially importanL
Summary
In summary, this study provides support for the Health Beliefs Model and the
Problem Behavior Theory. The HBM conponents of Perceived Susceptibility and
Barriers predict mammography and stage as hypothesized. However, age and race are
not independent barriers to early diagnosis. In feet, being older predicts more
mammography and earlier diagnosis in this study. While mammography is shown to
be important, femily history, age, SES, lower body weight, and hormone use each
predicts lower stage, independent of screening behaviors.
The study provides early support for applying the PBT to adult health
behaviors, despite the absence of screening or stage prediction by alcohol use.
Obesity is an independent predictor of higher stage. Less caffeine and less cigarette
smoking and more exercise each predict more mammography, as expected. Caffeine
is the only variable in the study, other than type o f physician, that predicts method of
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
90
& st discovery o f breast cancer symptoms, which, in turn, is the most inqwrtant
predictor o f stage.
Intervention Implications
The importance of mammography in reducing the stage of breast cancer
diagnoses, and there&re increasing survival, is supported in this study. However, it is
noteworthy that other variables and Actors have a direct effect on stage after the
influence o f mammography, and of detection method in the second model, have been
partialled ouL The magnitude of the path coefiBcient between mammography and
stage is similar to the paths from the other direct predictors. This implies that judging
an intervention by increases in mammography alone may miss important issues.
Decreases in stage may be achieved through several paths.
Obesity does not significantly predict mammography, in either SEM, but
directly predicts higher stage in both. In the sub-sample model, hormone use is also a
significant direct predictor o f stage. In the full sample path model, family history, age,
SES, and hormone use aU have direct effects on stage, in addition to their indirect
effects through mammography. Having a history of mother or sisters with breast
cancer, using hormones, being older, or wealthier each directly decreases the stage at
diagnosis, regardless o f mammography. Note that these latter four are all known risk
factors for acquiring the disease. This leads to the hopeful speculation that women
who know they are at increased risk are somehow more diligent, and detect their
disease at an earlier stage, perhaps by presenting symptoms for diagnosis with less
delay. It raises hopes that an adequate educational program, which alerts other groups
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
91
of women to their increased risk, might also result in more diligent screening, and
in^roved prognosis for breast cancer patients.
Another possible intervention implication concerns physician-patient
communications. Critical reactions from physicians to non-conventionai or deviant
behaviors are suggested as one reason that some women receive less screening.
Women in the study who smoked often reported anecdotally that they avoided doctor's
office visits unless they were truly ill because they did not want to be criticized and
blamed for their smoking. Whether this explanation would be substantiated by
errçirical research is not known.
T.imitations, and Directions for Future Research
These data were collected primarily from non-Hispanic Caucasian women.
They were between the ages of 55 and 69 when they had their first diagnosis of
primary breast cancer. Generalization to non-Caucasian groiçs, to younger women or
to the recurrence o f breast cancer will be questionable. In this sample, poor, and
poorly educated women are underrepresented. Efforts to increase the participation o f
disadvantaged women in future sançles would improve the generalizability of the
findings, especially since socioeconomic influences played an important role in
predicting stage. Prospective longitudinal studies could help clarify some of the
causal directions left unclear by this study, such as the effects o f previous breast
disease on the number of mammograms and on hormone use.
Several avenues o f promising future research are identified. The direct effect
of femily history, hormone use, age, and socioeconomic status on stage, regardless o f
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
92
mammography or utilization of medical care, requires further study to identify how
these influences might be manipulated to improve prognosis. The results o f this study
suggest exclusive focus on increasing mammogram screening may miss some
important influences.
The implication that obesity's effect on stage cannot be explained by difficulty
in detection, opens questions about the effect of body mass on breast cancer diagnosis.
Such questions are especially important since increasing numbers of women in the
U.S.A. are overweight. The effect o f supportive, non-blaming physician interactions
with patients who have non-conventional or deviant behaviors, also awaits further
study.
A hopeful interpretation that older women are no longer at increased risk of
more life-threatening late diagnoses o f breast cancer needs to be validated in other
studies. Known confounds of income, education, and race are controlled in this study,
but some other, unexpected, correlate may account for the effect. Further studies with
women from other regions will be necessary to determine whether older women in
general are receiving better preventive care, or whether some sampling bias, or
geographically limited influence, gave this sample of older women in Los Angeles
County an advantage.
Perhaps the most intriguing question raised in this study is the speculation
about a possible adverse conditioning effect of past traumatic benign diagnostic
procedures on recent mammogram use. To investigate the possible effect of adverse
conditioning on mammography, future research into mammogram utilization should
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
93
include three types of data that are not available in this study. First, whether previous
mammograms were for screening or diagnostic purposes. Second, whether surgical
biopsies acconçanied the mammogram. And third, personological measures to
distinguish which women would be more anxious, or otherwise more prone to
avoidance after a negative experience. Such studies may open our understanding of
screening behaviors to new interpretations.
Conclusion
In addition to the acknowledged medical factors, a complex blend of
behavioral and demogr^hic influences determines the stage of breast cancer at
diagnosis. Further elucidation of these influences could improve survival for the
nearly 200,000 women in the U.S. who will be diagnosed with breast cancer in the
coming year (American Cancer Society, 1997).
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
94
Bibliography
American Cancer Society. (1990). 1989 survey o f physicians’ attitudes and practices
in early cancer detection. CA- A Cancer Journal fhr Clinicians. 40. 77-101.
American Cancer Society. (1997). Cancer Facts and Figures - 1997. Atlanta:
Author.
American Joint Committee on Cancer. (1992). Manual for Stafiny of Cancer (4th
ed., pp. 149-154). Philadelphia: J. B. Lippincott.
American Medical Association Council on Scientific Affairs. (1989).
Mammographie screening in asymptomatic women aged 40 years and older.
JAMA. 261.2535-42.
Anda, R. F., Sienko, D. G., Remington, P. L., Gentry, E. M., & N(farks, J. S. (1990).
Screening mammography for women 50 years of age and older: Practices and
trends, 1987. American Journal of Preventive Medicine. 6 (3), 123-129.
Andersson, I. (1984). What can we leam from interval carcinomas? Recent Results
in Cancer Research. 90. 161-163.
Auvinen, A., Elovainio, L., & Hakama, M (1996). Breast selfiexamination and
survival from breast cancer: A prospective follow-up study. Breast Cancer
Research and Treatment. 38. 161-168.
Baker, L. H. (1982). Breast cancer detection demonstration project: Five year
summary report CA- A Cancer Journal for Clinicians. 32 (4), 194-225.
Bentler, P. M. (1990). Comparative fit indexes in structural models. Psvchological
Bulletin. 107.238-246.
Bentler, P. M. (1995a). EQS, A Structural Equations Program (version 5.7b)
[Computer software]. Encino, CA: Multivariate Software.
Bentler, P. M. (1995b). EOS Structural Equations Program Manual. Encino, CA:
Multivariate Software.
Bentler, P. M., & Bonett, D. G. (1980). Significance tests and goodness o f fit in the
analysis of covariance structures. Psvchological Bulletin. 88. 588-606.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
95
Bergner, M., Allison, C. J., Diehr, P., Ford, L. G., & Feigl, P. (1990). Early detection
and control o f cancer in clinical practice. Archives of Internal Medicine. 150,
431-436.
Bernstein, L., Ross, R. K., Deapen, D., Boone, J., & Miu, A. (Eds.). (1991). Cancer
in Los Angeles Countv: A Portrait of Incidence and Mortality. 1972-1987.
Los Angeles: Kenneth Norris Jr. Comprehensive Cancer Center.
Bondy, M., Vogel, V. G., Halabi, S., & Lustbader, E. D. (1992). Identification o f
women at increased risk for breast cancer in a population-based screening
program. Cancer Epideminlngv. Binmarkers. & Prevention. 143-147.
Breen, N., & Figueroa, J. B. (1996). Stage of breast and cervical cancer diagnosis in
disadvantaged neighborhoods: A prevention policy perspective. American
Journal of Preventive Medicine. 12 (5), 319-326.
Brekelmans, C. T. M., Peeters, P. H. M., Faber, J. A. J., Deurenberg, J. J. M., &
Collette, H. J. A. (1994). The epidemiological profile of women with an
interval cancer in the DOM screening -programme. Breast Cancer Research
and Treatment. 30.223-232.
Busetti, M. C., Miller, A. B., To, T., & Rohan, T. E. (1996). Risk factors for breast
cancer mortality among the National Breast Screening Study o f Canada
participants. Cancer Detection and Prevention. 20 (2), 122-129.
Carpenter, C. L., Ross, R. K , Paganini-Hill, A., Bernstein, L (1998). Weight gain,
family history, and breast cancer risk among postmenopausal women
[Abstiact]. American Journal o f Epidemiology 147 (Supplement), S 8.
Castro, F. G., Newcomb, M D., McCreary, C., & Baezconde-Garbanati, L. (1989).
Cigarette smokers do more than just smoke cigarettes. Health Psychology.
8(1), 107-129.
Celentano, D. D., Klassen, A. C., Weisman, C. S., & Rosenshein, N. B. (1988).
Cervical cancer screening practices among older women: Results from the
Maryland cervical cancer case-control study. Journal o f Clinical
Epidemiology. 41. 531-541.
Celentano, D. D., Shapiro, S., & Weisman, C. (1982). Cancer preventive screening
behavior among elderly women. Preventive Medicine. 11.454-463.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
96
Choong, P. L., de Silva, C. J. S., Dawkins, H. J. S., Sterrett, G. F., Robbins, P., &
Harvey, J. M. (1996). Predicting axillary lymph node métastasés in breast
carcinoma patients. Breast Cancer Research and Treatment 3 7 .135-149.
Chou, C.-P., & Bentler, P. M. (1990). Model modification in. covariance structure
modeling: A conçarison among likelihood ratio, Lagrange Multiplier, and
Wald tests. Multivariate Behavioral Research. 25. 115-136.
Costa, F., lessor, R., & Donovan, J. E. (1987). Psychosocial correlates and
antecedents o f abortion: An exploratory study. Population and Environment.
9,3-22.
Danigelis, N. L., Roberson, N. L., Worden, J. K., Flynn, B. S., Dorwaldt, A L.,
Ashley, J. A., SkeUy, J. M., & Mickey, R. M. (1995). Breast screening by
African-American women: Insights from a household survey and focus
groups. American Journal o f Preventive Medicine. 11 (5), 311-317.
Dayal, H. H., Power, R. N., & Chiu, C. (1982). Race and socioeconomic status in
survival from breast cancer. Journal of Chronic Diseases. 35. 675-683.
Donovan, J. E., lessor, R., & Costa, F. M (1991). Adolescent health behavior and
conventionality-unconventionality: An extension of Problem Behavior
Theory. Health Psvcholoev. 10 (1), 52-61.
Donovan, 1. E., lessor, R., & Costa, F. M. (1993). Structure of health-enhancing
behavior in adolescence: A latent-variable approach. Journal of Health and
Social Behavior. 34. 346-362.
Ewertz, M , Gillanders, S., Meyer, L., & Zedeler, K. (1991). Survival of breast
cancer patients in relation to fectors which affect the risk o f developing breast
cancer. International Journal of Cancer. 49. 526-530.
Farley, T. A , & Flannery, J. T. (1989). Late-stage diagnosis of breast cancer in
women of lower socioeconomic status: Public health implications. American
Journal o f Public Health. 79 I I 1). 1508-1512.
Feldman, J. G., Carter, A C., Nicastri, A D., & Hosat, S. T. (1981). Breast self-
examination, relationship to stage of breast cancer at diagnosis. Cancer. 47.
2740-2745.
Ferrini, R., Mannino, E., Ramsdell, E., & Hill, L. (1996). Screening mammography
ft> r breast cancer: American College of Preventive Medicine practice policy
statement. American Journal of Preventive Medicine. 12 (5), 340-341.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
97
Fink, R., Shapiro, S., & Rooster, R. (1972). Impact o f efforts to increase participation
in repetitive screenings for early breast cancer detection. American Journal of
Pubhc Health. 62, 328-336.
Friedman, L. C., Woodruff A., Lane, M., Weinberg, A. D., Cooper, H. P., & Webb,
J. A. (1995). Breast cancer screening behaviors and intentions among
asymptomatic women 50 years o f age and older. American Journal of
Preventive Medicine, ü (4), 218-223.
Frisell, J., Eklund, G., Hellstrom, L., & Somell, A. (1987). Analysis o f interval breast
carcinomas in a randomized screening trial in Stockholm. Breast Cancer
Research and Treatment. 9, 219-225.
Goodwin, J. S., Samet, J. M., Key, C. R., Humble, C., Kutvirt, D., & Hunt, C.
(1986). Stage at diagnosis o f cancer varies with the age of the patient. Journal
of the American Geriatrics. 34. 20-26.
Gould-Martin, K., Paganini-Hill, A., Casagrande, C., Mack, T., & Ross, R. K.
(1982). Behavioral and biological determinants of surgical stage of breast
cancer. Preventive Medicine. 11. 429-440.
Green, L. W., & Roberts, B. J. (1977). The research literature on why women delay
in seeking medical care for breast symptoms. Health Education Monographs.
2, 129-177.
Greenwald, P., Nasca, P. C., Lawrence, C. E., Horton, J., McGarrah, R. P., Gabriele,
R., Carlton, K. (1978). Estimated effect o f breast self-exarnination and
routine physician examinations on breast cancer mortality. New England
Journal of Medicine. 299. 271-273.
Greer, S. (1974). Psychological aspects: Delay in the treatment of breast cancer.
Proceedings o f the Roval Society o f Medicine. 67.470-473.
Hedegaard, H. B., Davidson, A J., & Wright, R. A (1996). Factors associated with
screening mammography in low-income women. American Journal o f
Preventive Medicine. 12 (1), 51-56.
Heuser, L. S., Spratt, J. S., Kuhns, J. G., Chang, A F. C., Polk, H. C., & Buchanan,
J. B. (1984). The association o f pathologic and mammographie
characteristics o f primary human breast cancers with “slo V ’ and “fast”
growth rates and with axillary lymph node métastasés. Cancer. 53. 96-98.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
98
Hochbaum, G. (1958). Public participation in medical screening programs (DHEW
Publication No. 572, Public Health Service). Washington, DC: U. S.
Government Printing OfBce.
HoUingsbead, A. B. (1957>. Two factor index o f social position. New Ifaven, CT:
Hollingshead.
Holmes, F. F., & Heame, E., HI. (1981). Cancer stage-to-age relationsbip:
Inçlications for cancer screening in the elderly. Journal ofth e American
Geriatrics Society. 29 (2), 55-57.
Hunter, C. P., Redmond, C. E C , Chen, V. W., Austin, D. F., Greenberg, R. S., Correa,
P., Muss, H., Forman, M. R., Wesley, M. N., Blacklow, R. S., Kurman, R. J.,
Dignam, J. J., Edwards, B. K., Shapiro, S., & other members o f the
Black/White Cancer Survival Study Group. (1993). Breast Cancer: Factors
associated with stage at diagnosis in Black and White women. Journal ofthe
National Cancer Tnstitute. 85 (14), 1129-1137.
Ingram, D., Nottage, E., Ng, S., Sparrow, L., Roberts, A., & Willcox, D. (1989).
Obesity and breast disease: The role o f the female sex hormones. Cancer. 64.
1049-1053.
Janz, N. K., & Becker, M. H. (1984). The 33Health Belief Model: A decade later.
Health Education Quarterly. 11. 1-47.
lessor, R- (1987). Problem-Behavior Theory, psychosocial development, and
adolescent problem drinking. British Jotmial of Addiction. 82. 331-342.
Kaija, H., Matti, H., & Tapani, H. (1996). Use o f hospital services by breast cancer
patients by stage ofthe disease: implications on the costs of cancer control
Breast Cancer Research and Treatment. 37.237-241.
Kaplan, K. M., Weinberg, G. B., Small, A., & Herndon, J. L. (1991). Breast cancer
screening among relatives o f women with breast cancer. American Journal of
Public Health. 81 (9), 1174-1179.
Kelsey, J. L. (1993). Breast cancer epidemiology: Summary and future directions.
Epidemiologic Reviews. 15 (1), 256-263.
Kelsey, J. L., & Gammon, M. C. (1990). Epidemiology of breast cancer.
Epidemiologic Reviews. 12. 228-240.
Reproduced with permission ofthe copyright owner. Further reproduction prohibited without permission.
99
King, E. S., Resch, N., Rimer, B., Lerman, C., Boyce, A., & McGovem-Gorchov, P.
(1993). Breast cancer screening practices among retirement community
women. Preventive Medicine. 22 (1), 1-19.
Lane, D. S., Polednak, A. P., & Burg, M. A. (1992). Breast cancer screening
practices among users o f county-funded health centers vs women in the entire
community. American Journal of Public Health. 82 (2), 199-203.
Lerman, C., Rimer, B., Trock, B., Balshem, A., & Engstrom, P. F. (1990). Factors
associated with repeat adherence to breast cancer screening. Preventive
Medicine. 19, 279-290.
Lerman, C., & Schwartz, M. (1993). Adherence and psychological adjustment
among women at high risk for breast cancer. Breast Cancer Research and
Treatment 2 8 .145-155.
Longnecker, M. P., Bernstein, L., Paganini-Hill, A., Enger, S. M., Ross, R. K.
(1996). Risk foctors for in situ breast cancer. Cancer Epidemiology.
Binmarkers. and Prevention. 5, 961-965.
Longnecker, M. P., Paganini-Hill, A., Ross, R. K. (1995). Lifetime alcohol
consumption and breast cancer risk among postmenopausal women in Los
Angeles. Cancer Epidemiology. Binmarkers. and Prevention. 4, 721-725.
MacCallum, R. C. (1986). Specification searches in covariance structure modeling.
Psvchological Bulletin. 100. 107-120.
Marchant, D. J., & Sutton, S. M. (1990). Use of mammography - United States,
(1990). MMWR. 39 (36), 621-630.
Martin, L. M., Calle, E. E., Wingo, P. A., & Heath, C. W. (1996). Comparison of
mammography and Pap test use from the 1987 and 1992 National Health
Interview Surveys: Are we closing the gaps? American Journal o f Preventive
Medicine. 12 (2 \ 82-90.
Mayer-Oakes, S. A., Atchison, K. A., Matthias, R. E., De Jong, F. J., Lubben, J., &
Schweitzer, S. O. (1996). Mammography use in older women with regular
physicians: What are the predictors? American Journal o f Preventive
Medicine. 12 (1), 44-50.
McCarthy, B. D., Yood, M U., MacWfiUiam, C. H., & Lee, M. J. (1996). Screening
mammography use: The importance of a population perspective. American
Journal o f Preventive Medicine. 12 (2), 91-95.
Reproduced with permission ofthe copyright owner. Further reproduction prohibited without permission.
100
McGee, L., & Newcomb, M. D. (1992). General deviance syndrome: e^ganded
hierarchial evaluations at four ages from early adolescence to adulthood.
Journal of ConsuMng^ and Clinical Psvcholoev. 60(5), 766-776.
Mettler, F. A., Upton, A. C., Kelsey, C. A., Ashby, R. N., Rosenberg, R. D., &
Linver, M. N. (1996). Benefits versus risks from mammography. Cancer. 77
(5), 903-909.
Mohle-Boetani, J. C., Grosser, S., Whittemore, A. S., Malec, M., Kartç)ert, J. B., &
Pafifenbarger, R. S., Jr. (1988). Body size, reproductive fectors and breast
cancer survival Preventive Medicine. 17. 634-642.
Mutchler, J. E., & Burr, J. A. (1991). Racial differences in health and health care
service utilization in later Hfe: the effect of socioeconomic status. Journal o f
Health and Social Behavior. 32 (Dec.), 342-356.
National Cancer Institute Breast Cancer Screening Consortium. (1990). Screening
mammogr^hy: A missed clinical opportunity? JAMA. 264. 54-58.
Newcomb, M. D. (1990). What structural modeling techniques can tell us about
social support. In I. G. Sarason, B. R. Sarason, & G. R. Pierce (Eds.), Social
Support: An interactional view (pp. 26-63). New York: Wiley.
Newcomb, M. D. (1994). Drug use and intimate relationships among women and
men: Separating specific from general effects in prospective data using
structural equation models. Journal o f Consulting and Clinical Psvcholoev.
62(3), 463-476.
Newcomb, P. A., Olsen, S. J., Roberts, F. D., Storer, B. E., & Love, R. R. (1995).
Assessing breast self-examination. Preventive Medicine. 24. 255-258.
Owens, R- G., & Ashcroft, J. J. (1986). Breast cancer screening: The way ahead.
Journal of Psvchosocial Oncology. 4 (4), 15-26.
Patrick, D. L., Cheadle, A., Thompson, D. C , Diehr, P., Koepsell, T., & Kinne, S.
(1994). The validity of self-reported smoking: A review and meta-analysis.
American Journal o f Public Health. 84 (7), 1086-1093.
Pearlman, D. N., Rakowski, W., & Ehrich, B. (1996). Mammography, clinical breast
exam and pap testing: Correlates o f combined screening. American Journal o f
Preventive Medicine. 12 (1), 57-64.
Reproduced with permission ofthe copyright owner. Further reproduction prohibited without permission.
101
Pearlman, D. N., Rakowski, W., Ehrich, B., & Clark, M. A. (1996). Breast cancer
screening practices among Black, Hispanic and White women: reassessing
differences. American Journal o f Preventive Medicine. 12 (5), 327-337.
Peer, P. G. M., Verbeek, A. L. M., Straatman, H., Hendriks, J. H. C. L., & Holland,
R- (1996). Age-specific sensitivities o f mammographie screening for breast
cancer. Breast Cancer Research and Treatment. 38. 153-160.
Rakowski, W., Pearlman, D., Rimer, B. K., & Ehrich, B. (1995). Correlates o f
mammography among women with low and high socioeconomic resources.
Preventive Medicine. 24. 149-158.
Rimer, B. K., Trock, B., Engstrom, P. F,, Lerman, C., & King, E. (1991). Why do
some women get regular mammograms? American Journal of Preventive
Medicine. 7, 69-74.
Rimer, B. K., Conaway, M. R., Lyna, P. R., Rakowski, W., Woods-Powell, C. T.,
Tessaro, I., Yamall, K. S. H., & Barber, L. T. (1996). Cancer screening
practices among women in a community health center population. American
Journal o f Preventive Medicine. 12 (5), 351-357.
Rosenstock, I. M. (1974). The health belief model and preventive health behavior.
Health Education Monographs. 2, 354-386.
Rutledge, D. N., Hartmann, W. H , Kinman, P. O., & ^^Thfield, A. C. (1988).
E^qploration of fectors affecting mammography behaviors. Preventive
Medicine. 17, 412-422.
Saint-Germain, M. A., & Longman, A. J. (1993). Breast cancer screening among
older Hispanic women: Knowledge, attitudes and practices. Health Education
Quarterly. 20 (4), 539-553.
SAS Institute, Inc. (1996). SAS Statistical Package (version 6.12) [ConQ)uter
software]. Cary, NC: Author.
Satariano, W. A., Belle, S. EL, & Swanson, G. M. (1986). The severity o f breast
cancer at diagnosis: A conqiarison o f age and extent o f disease in Black and
White women. American Journal o f Public Health. 76 (7), 779-782.
Seidman, H., Gelb, S. K., Silverberg, E., LeVerda, N., & Lubera, J. A. (1987).
Survival ogerience in the Breast Cancer Detection Demonstration Project.
CA- A Cancer Journal for Clinicians. 37.258-290.
Reproduced with permission ofthe copyright owner. Further reproduction prohibited without permission.
102
Sobell, L. C., Toneatto, T., & Sobell, M. B. (1990). A procedure for reducing errors
in reports of life events. Journal o f Psychosomatic Research. 34. 163-170.
Stein, J. A , Newcomb, M. D., & Bentler, P. M. (1994). Psychosocial correlates and
predictors o f AIDS risk behaviors, abortion, and drug use among a
community sample o f young adult women. Health Psychology. 13_(4), 308-
318.
Strecher, V. J., Becker, M. H , Clark, N. M., & Prasada-Rao, P. (1989). Using
patients' descrÿtions o f alcohol consmrçtion, diet, medication conçjliance
and cigarette smoking: The validity o f self-reports in research and practice.
Tmimal o f General Internal Medicine. 4, 160-66.
Sutton, S., Bidder, G., Sancho-Aldridge, J., & Saidi, G. (1994). Prospective study of
predictors of attendance for breast screening in inner London. Journal o f
F.pidemioloev & Community Health. 48 (1), 65-73.
Taplin, S. H., Anderman, C., Grothaus, L., Curry, S., & Montano, D. (1994). Using
physician correspondence and postcard reminders to promote mammography
use. American Journal o f Public Health. 84 (4), 571-574.
Taylor, S. E. (1991). Health Psychology (2nd ed., pp. 66-70). New York: McGraw
HilL
Toneatto, T., Sobell, L. C., & Sobell, M. B. (1992). Predictors o f alcohol abusers’
inconsistent self-reports o f their drinking and life events. Alcoholism:
Clinical and Experimental Research. 16 (3), 542-546.
Ungemack, J. A . (1994). Patterns of personal health practice: Men and women in the
United States. American Journal o f Preventive Medicine. 10 (1), 38-44.
Velicer, W. F., Prochaska, J. O., Rossi, J. S., & Snow, M. G. (1992). Assessing
outcome in smoking cessation studies. Psychological Bulletin. 111. 23-41.
Vernon, S. W., Laville, E. A , & Jackson, G. L. (1990). Participation in breast
screening programs: A review. Social Science and Medicine. 30. 1107-1118.
Vernon, S. W., Tilley, B. C., Neale, A V., & Steinfeldt, L. (1985). Ethnicity,
survival and delay in seeking treatment for symptoms o f breast cancer.
Cancer. 55. 1563-1571.
Reproduced with permission ofthe copyright owner. Further reproduction prohibited without permission.
103
Vogel, V. G., Bondy, M., Halabi, S., Lord, J., & Laville, E. A. (1993). The Texas
breast screening project; Part H. Demographics, risk profiles, and health
practices of participants. Southern Medical JoumaL 86 (4), 391-396.
Weke, J. W., & Russell, M. (1993). Influence of socially desirable responding in a
study of stress and substance abuse. Alcoholism: Clinical and Experimental
Research. 17 (4), 758-761.
Wertheimer, M. D., Costanza, M. E., Dodson, T. F., D’Orsi, C., Pastides, H., &
Zapka, J. G. (1986). Increasing the effort toward breast cancer detection.
JAMA. 255. 1311-1315.
Zaloznik, A. J. (1995). Breast cancer stage at diagnosis: Caucasians versus Afro-
Americans. Breast Cancer Research and Treatment 3 4 .195-198.
Zumoff B., & Dasgupta, I. (1983). Relationship between body weight and the
incidence of positive axillary nodes at mastectomy for breast cancer. Journal
o f Surgical Oncology. 22.217-220.
Reproduced with permission ofthe copyright owner. Further reproduction prohibited without permission.
IMAGE EVALUATION
TEST TARGET (Q A -3 )
/
'/r,
%
1 . 0
l.i
1.25
L à mit 2 . 8
y o “ = =
u IM
îf ^
! r BAO
1 .4
I I M
1 2.2
2.0
1 . 8
1 . 6
1 5 0 m m
V
V
/
o
y
A
y i P P L I E D ^ IIV 1 4 G E . I n c
1653 E ast Main S treet
• Roctiester, NY 14609 USA
Phone: 716/482-0300
-------- Fax: 716/288-5989
O ' 9 9 3 . A p p l i e d I m a g o . I n c . , A il R i g h t s R e s e r v e d
Reproduced with permission ofthe copyright owner. Further reproduction prohibited without permission.

Linked assets

University of Southern California Dissertations and Theses

Conceptually similar

PDF

Child sexual abuse in a sample of male and female Hispanic and White nonclinical adolescents: Extending the reliability and validity of the Trauma Symptom Inventory (TSI)

PDF

Attachment as a predictor of self -monitoring and sexual behavior

PDF

Cognitive and non-cognitive factors as predictors of retention among academically at-risk college students: A structural equation modelling approach

PDF

Gender differences in motivation for sexual intercourse: Implications for risky sexual behavior and substance use in a university and community sample

PDF

A cross-cultural comparison of marital power and dyadic adjustment among American, Indo-American, and East Indian dual-career and single-career couples

PDF

Cognitive predictors of intention -behavior consistency in safer sex practices

PDF

A correlational study of feminist/womanist identity development and depression in women

PDF

An investigation of the psychological well-being of unaccompanied Taiwanese minors/parachute kids in the United States

PDF

A comparison of values among counseling psychologists characterized as scientist-practitioners, practitioners, or scientists

PDF

Concordant and discordant drug use in intimate relationships: A longitudinal study

PDF

Impact of language and culture on a neuropsychological screening battery for Hispanics

PDF

Counselor Trainees' Perceptions Of And Attitudes Toward Sexual Harassment In Relation To Feminist Beliefs

PDF

Development and validation of the Cooper Quality of Imagery Scale: A measure of vividness of sporting mental imagery

PDF

Attachment style, interpersonal guilt, parental alcoholism, parental divorce and eating disordered symptomatology in college women

PDF

"It's a person": An ethnographic analysis of gender and power in coed softball

PDF

Integration of science and practice: A collective case study of scientist -practitioner programs in counseling psychology

PDF

A Comparison Of Female Inmates With And Without Histories Of Prostitution On Selected Psychosocial Variables

PDF

Comparison of novel oncologic phase-I clinical trial designs incorporating time to trial completion

PDF

Intergenerational conflict, family functioning, and acculturation experienced by Asian American community college students

PDF

A daily diary approach to compare the accuracy of depressed and nondepressed participants' estimation of positive and negative mood: A test of the depressive realism hypothesis

Asset Metadata

Creator Clark, Linda Joyce (author)

Core Title Behavioral and demographic predictors of breast cancer stage at diagnosis

Degree Doctor of Philosophy

Degree Program Education (Counseling Psychology)

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

Tag health sciences, oncology,OAI-PMH Harvest,psychology, psychobiology,women's studies

Language English

Contributor Digitized by ProQuest (provenance)

Advisor Newcomb, Michael (committee chair), [illegible] (committee member), Goodyear, Rodney (committee member)

Permanent Link (DOI) https://doi.org/10.25549/usctheses-c17-425863

Unique identifier UC11353734

Identifier 9931870.pdf (filename),usctheses-c17-425863 (legacy record id)

Legacy Identifier 9931870.pdf

Dmrecord 425863

Document Type Dissertation

Rights Clark, Linda Joyce

Type texts

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

Access Conditions The author retains rights to his/her dissertation, thesis or other graduate work according to U.S. copyright law. Electronic access is being provided by the USC Libraries in agreement with the au...

Repository Name University of Southern California Digital Library

Repository Location USC Digital Library, University of Southern California, University Park Campus, Los Angeles, California 90089, USA