University of Southern California Dissertations and Theses

Predicting neuroradiographic abnormalities and death/disabled status occurring within 6 months in HIV-infected individuals with cryptococcal meningoencephalitis

(USC Thesis Other)

Predicting neuroradiographic abnormalities and death/disabled status occurring within 6 months in HIV-infected individuals with cryptococcal meningoencephalitis

PDF

Download

Open document

Flip pages

Download a page range

Download transcript

Copy asset link

Request this asset

Transcript (if available)

Content

PREDICTING NEURORADIOGRAPHIC ABNORMALITIES AND
DEATH/DISABLED STATUS OCCURRING WITHIN 6 MONTHS IN HIV-
INFECTED INDIVIDUALS WITH CRYPTOCOCCAL
MENINGOENCEPHALITIS

By

Wuchen Zhao

A Thesis Presented to the
FACULTY OF THE USC GRADUATE SCHOOL
UNIVERSITY OF SOUTHERN CALIFORNIA
In Partial Fulfillment of the
Requirements for the Degree
MASTER OF SCIENCE
(BIOSTATISTICS)

May 2014

Copyright 2014 Wuchen Zhao
ii

Acknowledgements

I would like to take this moment
to offer my sincerest appreciation and convey my utmost gratitude to
Dr. Christianne Lane
for allowing me to participate in this research and guiding me through the process.

I would also like to thank the chair of my thesis committee,
Dr. Wendy Mack,
along with other committee members,
Dr. Stanley Azen,
for their time, consideration, and insight.

iii

Table of Contents

Acknowledgements ......................................................................................................................... ii
List of Tables ...................................................................................................................................v
List of Figures ............................................................................................................................... vii
Abstract ........................................................................................................................................ viii
Chapter 1: Introduction ....................................................................................................................1
Chapter 2: Methods ..........................................................................................................................3
2.1 Study Design and Sample ..............................................................................................3
2.2 Evaluation of Study Outcomes ......................................................................................3
2.2.1 Neuroradiographic Assessment ............................................................................4
2.2.2 Death and Disability .............................................................................................4
2.3 Clinical Parameters ........................................................................................................4
2.3.1 CSF Cryptococcal Colony Forming Units (CFU) ................................................5
2.3.2 CSF Cryptococcal Antigen (CrAg) .......................................................................5
2.3.3 CSF Opening Pressure (OP) .................................................................................5
2.3.4 CSF Protein (Prot).................................................................................................5
2.3.5 CSF Glucose (Gluc) ..............................................................................................6
2.3.6 CSF Fungal Culture ..............................................................................................6
2.3.7 HIV Viral Load (HIV VL) ....................................................................................6
2.3.8 CD4 Cell Count.....................................................................................................7
2.3.9 Hemoglobin Level (Hgb) ......................................................................................7
2.3.10 Albumin Level (Alb)...........................................................................................7
2.4 Statistical Analysis .........................................................................................................7
Chapter 3: Results ..........................................................................................................................11
3.1 Neuroradiographic Abnormalities ...............................................................................12
3.2 Death/Disabled vs. Normal at 6 Months ......................................................................21
3.3 Goodness of Fit ............................................................................................................29
iv

3.4 Sensitivity and Specificity ...........................................................................................30
3.5 Comparison of Optimized Data and Original Data......................................................31
Chapter 4: Discussion ....................................................................................................................34
4.1 Small Sample Size .......................................................................................................34
4.2 Missing Data and Multiple Imputation ........................................................................36
Chapter 5: Conclusion....................................................................................................................39
References (Numerical) .................................................................................................................40
References (Alphabetical) ..............................................................................................................43

v

List of Tables

Table 1. Characteristics (Demographic Parameters) & Outcomes ............................................... 11
Table 2. Characteristics of Numeric Clinical Variables (N = 44)................................................. 12
Table 3. Characteristics of Categorical Clinical Variables (N = 44). ........................................... 12
Table 4. Comparison of Mean Values Of Continuous Clinical Variables among Participants with
Normal versus Abnormal In Neuroradiographic Findings ............................................ 13
Table 5. Categorical Variable Frequencies by Presence of Neuroradiographic Abnormalities ... 13
Table 6. Logistic Regression: Univariate Analysis of Continuous Variable Associations with
Neuroradiographc Abnormalities................................................................................... 14
Table 7. Logistic Regression: Univariate Analysis of Categorical Variable Associations with
Neuroradiographic Abnormalities ................................................................................. 15
Table 8. Summary of Stepwise Selection on Logistic Regression Analysis of Neuroradiographic
Abnormalities ................................................................................................................. 16
Table 9. Summary of Forward Selection on Neuroradiographic Abnormalities .......................... 16
Table 10. Results of Backward Selection on Neuroradiographic Abnormalities ......................... 17
Table 11. Candidate Models for Neuroradiographic Abnormalities ............................................ 19
Table 12. Final Multivariable Logistic Regression Model on Neuroradiographic Abnormalities 20
Table 13. Comparison of Mean Values of Continuous Clinical Variables among Participants with
Normal versus Death/Disability in Neurological Status at 6 Months ............................ 22
Table 14. Categorical Variable Frequencies by Presence of Death/Neurologic Disability 6
Months ........................................................................................................................... 22
vi

Table 15. Univariate Analysis of Continuous Variables (Logistic Regression) for Death/Disabled
vs. Normal at 6 months .................................................................................................. 23
Table 16. Logistic Regression on Death/Disabled vs. Normal at 6 Months ................................. 24
Table 17. Summary of Stepwise Selection on Death/Disabled vs. Normal at 6 Months ............. 26
Table 18. Summary of Forward Selection on Death/Disabled vs. Normal at 6 Months .............. 26
Table 19. Results of Backward Selection on Death/Disabled vs. Normal at 6 Months ............... 26
Table 20. Candidate Models for Death/Disabled vs. Normal at 6 Months ................................... 28
Table 21. Final model of Logistic Regression on Death/Disabled vs. Normal at 6 Months ........ 28
Table 22. Partition for the Hosmer and Lemeshow Test on Neuroradiographic Abnormalities .. 29
Table 23. Partition for the Hosmer and Lemeshow Test on Death/Disabled vs. Normal at 6
Months ........................................................................................................................... 29
Table 24. Final model of Logistic Regression on Neuroradiographic Abnormalities with Original
Data ................................................................................................................................ 32
Table 25. Final model of Logistic Regression on Death/Disabled vs. Normal at 6 Months with
Original Data .................................................................................................................. 33

vii

List of Figures

Figure 1. ROC Curve on Neuroradiographic Abnormalities ........................................................ 31
Figure 2. ROC Curve on Death/Disabled vs. Normal at 6 Months .............................................. 31
Figure 3. ROC Curve on Neuroradiographic Abnormalities, Original Data ................................ 32
Figure 4. ROC Curve on Death/Disabled vs. Normal at 6 Months, Original Data ...................... 33

viii

Abstract

The goal of this study is to assess if there are any clinical parameters that may predict abnormal
neuroradiographic findings in HIV-infected individuals who also have Cryptococcal
meningoencephalitis (CM). We investigated the relationships between clinical parameters and
neuroradiographic abnormalities measured at baseline, and the relationships between clinical
parameters and a combined outcome of death/disability occurring within 6 months of the
baseline. We selectively evaluated clinical parameters that are commonly used in diagnosing
and monitoring CM and are inexpensive and non-invasive compared to CT or MRI.

This study is a subset of a retrospective cohort study of 44 HIV-infected individuals with CM.
Cerebrospinal fluid and clinical parameters were collected at baseline, 14 days after baseline, and
6 months later. Our study revealed two statistically significant predictors of neuroradiographic
abnormalities: age and growth of C. neoformans in fungal culture measured at day 14 were
positively associated with neuroradiographic abnormalities. The interaction between HIV viral
load measured at baseline and fungal culture measured at day 14 was negatively associated with
neuroradiographic abnormalities; in persons who did not show any growth of C. neoformans in
fungal culture at day 14, the odds of being abnormal in neuroradiographic abnormalities decrease
by 0.7% (CI: 0.95-1.03), with each increase of 10,000 copies/mL in HIV viral load at baseline;
while in persons with growth of C. neoformans in fungal culture at day 14, the odds of being
abnormal in neuroradiographic abnormalities decrease by 7.6% (CI: 0.82-0.98), with each
increase of 10,000 copies/mL in HIV viral load at baseline.

ix

The interaction between HIV viral load measured at baseline and fungal culture measured at day
14 was positively associated with Death/Disabled (status) occurring within 6 months. In persons
who did not show any growth of C. neoformans in fungal culture at day 14, the odds of death or
disabled in 6 months decreases by 20.4% (CI: 0.62-1.03), with each increase of 10,000 units in
HIV viral load at baseline; while in persons with growth of C. neoformans in fungal culture at
day 14, the odds of death or disabled in 6 months increases by 17.9% (CI: 0.90-1.55), with each
increase of 10,000 units in HIV viral load at baseline.

Compared to CT or MRI, evaluating for clinical parameters such as CSF fungal culture and HIV
viral load are inexpensive and routinely performed in clinical practice. Predicting abnormal
neuroradiographic findings in individuals with cryptococcal meningoencephalitis using such
clinical parameters is a feasible first-step in deciding whether to perform CT or MRI, especially
in resource-limited settings. Our predictive model had 91.7% sensitivity and 82.4% specificity
for neuroradiographic abnormalities, and 76.5% sensitivity and 70.0% specificity for
death/disability after 6 months.

1

Chapter 1: Introduction

Cryptococcal meningoencephalitis (CM) is associated with a 20% mortality rate despite
appropriate antifungal therapy
1
.

In developed countries, most HIV-infected individuals with acute neurological symptoms will
undergo radiological brain evaluation. However, in resource-limited settings, brain imaging is
limited and considered a luxury, if it is available at all.

Few studies have specifically evaluated the utility of neuroimaging to assess the severity of
AIDS-associated cryptococcosis. Most radiological studies were completed before the
development and availability of Highly Active Anti-Retroviral Therapy (HAART); these older
studies often described abnormal cerebral imaging findings associated with CM
1
. Also, analysis
of the contribution of CT and/or MRI in detecting brain lesions and predicting mortality is
limited.

Since CT and MRI are expensive commodities in underdeveloped countries, we aimed to
evaluate the possibility of predicting neuroimaging abnormalities and mortality by Cryptococal
antigen titer (CrAg) or quantitative cultures, in addition to other baseline clinical characteristics
collected among newly-diagnosed CM individuals.

CrAg is widely used as a diagnostic tool for detecting cryptococcal meningitis. It can remain
detectable even months following successful therapy due to the possible continuous release of
2

capsular polysaccharide antigens from dead C. neoformans cells, which are slowly eliminated
from cerebrospinal fluid
2, 3
. Therefore, CrAg has limited utility for monitoring therapy and thus
it cannot be used as an index of cure. In contrast, cryptococcal quantitative cultures can indicate
treatment efficacy by determining the rate of cerebrospinal fluid sterilization and have greater
potential for determining clinical outcomes and mortality
4, 5
.

The goal of this study is to assess if there are clinical parameters commonly used in diagnosing
and monitoring CM that may predict the presence of abnormal neuroradiographic findings as
well as subsequent mortality/morbidity in persons co-infected with HIV and CM. Such
predictors would be cheaper and more efficient in predicting abnormal neuroradiography than
CT or MRI.

3

Chapter 2: Methods

2.1 Study Design and Sample

This study is a retrospective cohort study. Data were collected via retrospective chart review.
The data from 44 HIV-infected persons represent a subset of Dr. Robert Larsen’s study
6
on
persons co-infected with HIV and CM (A5225) whose cerebrospinal fluid (CSF) was collected
for quantitative culture studies. Subject eligibility criteria for the subset included: (1) age 18+
years, (2) HIV-positive, (3) newly diagnosed with Cryptococcal Meningoencephalitis (CM), (4)
culture-proven CM, (5) followed-up at LAC-USC Medical Center facilities (hospital +/-clinics)
for at least 6 months post-treatment or until death, and (6) had brain imaging data – CT and/or
MRI. Non-HIV individuals, incarcerated individuals, and transplanted individuals were
excluded from the study. Of the 44 eligible participants with a CSF sample, 43 (97.7%)
completed a CT and/or MRI of the brain to assess neuroradiographic abnormalities. Twenty-
nine of 44 participants (65.9%) completed the follow-up survey 6 months after the subjects were
first recruited in the study.

The data from these 44 participants were collected from 2005 to 2011.

2.2 Evaluation of Study Outcomes

The criteria and methods on ascertainment of neuroradiographic abnormalities and death or
disability status were defined and assessed by Dr. Karen Michelle U. Guiang.

4

2.2.1 Neuroradiographic Assessment
Several criteria were used to judge whether abnormalities were present or absent on brain images
from CT/MRI obtained at the baseline (End point #1). Neuroradiographic abnormalities were
defined to be present if any of the following were seen on the images: (1) Atropy, (2) Dilated
Virchow-Robin spaces, (3) Pseudocysts, (4) Intracerebral masses, (5) Hydrocephalus, (6)
Radiological meningitis. In the absence of all of these conditions, the image was otherwise
considered to be normal.

2.2.2 Death and Disability
Neurological status at 6 months following the baseline visit (End point #2) was defined as
normal unless the individual died within 6 months of the CM diagnosis or demonstrated any
persistent neurologic complaints including: (1) Altered mental status, (2) Headaches, (3)
Seizures, (4) CNS deficits, (5) Gait disturbances.

2.3 Clinical Parameters

Cryptococcal colony forming units (CFU) were measured at the baseline (denoted as Day 0) and
also 14 days after the baseline (Day 14)
7
. Cryptococcal antigen (CrAg)
8, 9
, Opening Pressure
(OP), CSF protein (Prot), and CSF glucose (Gluc) were also measured twice at days 0 and 14.
HIV viral load
10
, CD4 cell count, and hemoglobin and albumin levels were measured only once
at the baseline visit.

5

2.3.1 CSF Cryptococcal Colony Forming Units (CFU)
The number of CFUs was measured in each participant’s cerebrospinal fluid (CSF) at days 0 and
14. The reduction or growth in CSF CFUs measured at days 0 and 14 was found to be
exponential, thus the rate of decrease or increase of CSF CFU was transformed as log CFU when
analyzing its association with the study outcomes
11
.

2.3.2 CSF Cryptococcal Antigen (CrAg)
The CrAg titer was measured by a latex agglutination assay (Immuno-Mycologics, Inc., Norman,
Okla.). Serum and CSF specimens were sent overnight to a central laboratory on wet ice, and
titers were measured within 24 to 48 h of collection. Samples were then placed in batches and
stored frozen at −70°C until they were thawed and the titers were measured at the conclusion of
the study by latex agglutination assay and enzyme-linked immunoabsorption assay (Meridian,
Cincinnati, Ohio)
12
. In this study, CSF CrAg was measured at days 0 and 14.

2.3.3 CSF Opening Pressure (OP)
Opening pressure was measured with the individual in a reclining (lateral decubitus) position
during the initial lumbar puncture
13
. In this study, OP mm CSF was measured at days 0 and 14.

2.3.4 CSF Protein (Prot)
The amount of protein in the cerebrospinal fluid (CSF) is one of the most sensitive indicators of
pathology within the central nervous system
14
. In this study, CSF protein was measured at days 0
and 14.

6

2.3.5 CSF Glucose (Gluc)
CSF glucose or glycorrhachia is a measurement used to determine the levels of glucose in
cerebrospinal fluid
14-16
. As a general rule, CSF glucose is about two thirds of the serum glucose
measured during the preceding two to four hours in a normal adult. This ratio decreases with
increasing serum glucose levels
14
. Normal glucose levels do not rule out infection, because up to
50 percent of individuals who have bacterial meningitis will have normal CSF glucose levels
17
.
There is no pathologic process that causes CSF glucose levels to be elevated
14
. In this study, CSF
glucose was measured at days 0 and 14.

2.3.6 CSF Fungal Culture
CSF fungal culture was measured at 14 days after the baseline using a culture of cerebrospinal
fluid. This test is used to help diagnose suspected central nervous system infection caused by
cryptococcal meningitis
18-22
. This type of fungal infection causes inflammation of the membrane
surrounding the brain and spinal cord. Negative or no growth in cryptococcal meningitis on CSF
fungal culture is considered to be the normal result
23
.

2.3.7 HIV Viral Load (HIV VL)
HIV viral load tests are reported as the number of HIV copies in a milliliter (copies/mL) of
blood and is a measure of the degree of active infection
24
. HIV viral load was measured at the
baseline of this study.

7

2.3.8 CD4 Cell Count
HIV disease is associated with profound immunosuppression, usually occurring at CD4 counts
<100 cells/μl
11
. The World Health Organization recommends cryptococcal antigen screening in
HIV-infected persons with CD4<100 cells/μL
25
. Typically, if the HIV viral load is high, the
CD4 count will be low—making individuals more vulnerable to opportunistic infections
26
. CD4
cell count was measured at the baseline of this study.

2.3.9 Hemoglobin Level (Hgb)
Hemoglobin is a protein in red blood cells that carries oxygen. The normal range for
hemoglobin for men and women are 13.5 to 17.5 grams per deciliter (135 to 175 grams per liter)
and 12.0 to 15.5 grams per deciliter (120 to 155 grams per liter) respectively
27
. Hemoglobin
level was measured at the baseline of this study.

2.3.10 Albumin Level (Alb)
Albumin is a protein made by the liver. Serum albumin test was performed to measure the
amount of this protein in the clear liquid portion of the blood. The results of albumin test for
adults can be considered as normal if the albumin level is between 3.5-5.5 g/dL (35-55 g/L)
28
.
Albumin level was measured at the baseline of this study.

2.4 Statistical Analysis

Student’s t-tests were conducted as exploratory and descriptive analyses comparing participants
with normal versus abnormal neuroradiographic findings on continuous variables (Table 4).
Univariate logistic regression analyses were conducted for each continuous and categorical
8

clinical parameter to assess the univariate association of clinical parameters with the presence of
neuroradiographic abnormalities (Table 7). For the CM-related and other clinical variables
measured at days 0 and 14, we also assessed the association of 14-day change in these levels
with neuroradiographic abnormalities. The same methods were used for analysis of
death/neurologic disability occurring within 6 months of CM (Table 13 and
All candidate clinical parameters and potential two way interaction effects were entered into the
multivariate logistic model and eliminated based on lack of significance. We used forward
selection, backward selection, and stepwise selection methods to build candidate models and
used a relaxed significance level of 0.3 for entering and removing effects, rather than a
traditional level of 0.05.

Table 16).

We evaluated the log-linearity of association between clinical parameters and the two study
outcomes. The reduction or growth in CSF CFUs measured at days 0 and 14 was found to be
exponential, thus the rate of decrease or increase of CSF CFU was transformed as log CFU when
analyzing its association with the study outcomes.

For neuroradiographic abnormalities, clinical parameters with data available on more than 25
participants were analyzed. Independent variables with a relaxed p-value <0.3 in univariate
logistic regression or on Student’s t-test were considered as candidate clinical parameters for the
stepwise multivariable logistic regression model. Since the sample size for the outcome of
9

death/disabled vs. normal at 6 months was only 29, we used the p-value of <0.3 in univariate
logistic regression or on Student’s t-test to select candidate clinical parameters.

Due to the small sample size, the high rate of missing data, with missing values occurring for
different clinical parameters scattered among different individuals, we used a multiple
imputation strategy for candidate predictors to optimize the dataset. SAS PROC MI was used to
create multiply imputed (5 imputations) datasets. A Markov chain Monte Carlo (MCMC)
method
29
that assumes multivariate normality was used to impute missing values for continuous
variables assuming that the data set showed an arbitrary missing pattern
30
. The 5 complete data
sets were analyzed using PROC LOGISTIC and the statistical results were combined using
PROC MIANALYZE.

To identify and validate independent predictors and potential two way interaction effects for the
two study outcomes, forward selection, backward selection, and stepwise selection algorithms
were conducted. In the forward selection procedure, candidate clinical parameters were entered
into the model according to priority based on the corresponding significance level. In the
backward selection procedure, all candidate clinical parameters were entered into the model and
were eliminated based on lack of significance. The stepwise selection combines the elements of
the previous two selection methods: candidate clinical parameters were entered into the model
according to priority based on the corresponding significance level and were eliminated based on
lack of significance. The clinical parameters were included or eliminated from the model at a
relaxed significance level of 0.3.

10

Since the logistic models should be used with a minimum of 10 outcome events per predictor
variable
31
, we built final models for both neuroradiographic abnormalities and death/disability at
6 months using a significance level of 0.05, selecting variables from the candidate forward,
backward and stepwise selection models. The overall goodness of fit for the final models was
evaluated with Pearson and Hosmer and Lemeshow goodness-of-fit tests, which are used
frequently for logistic regression in risk prediction models
32
. Pearson’s chi-squared test examines
the sum of the squared differences between the observed and expected number of cases per
covariate pattern divided by its standard error. For the Hosmer-Lemeshow goodness-of-fit test,
the observations were sorted in increasing order of their estimated event probability and then
divided into groups, the statistic was obtained by calculating the Pearson chi-square statistic from
the contingency table of observed versus expected frequencies. ROC curves plotted true positive
rates versus false positive rates for each prospective cutoff to indicate the relation of sensitivity
and specificity for the predictive models
33
.

All analyses were performed using the Statistical Analysis System (version 9.3, SAS Institute
Inc., Cary, NC).
11

Chapter 3: Results

Table 1 shows the participant characteristics and distributions of the two outcomes. The study
sample consists of 44 participants, with 37 males (84%) and 7 females (16%). At baseline, the
participants’ age ranged from 21 to 63 years of age (mean=43.0 years, standard deviation=9.4
years), one participant’s age was unidentified. Table 2 and Table 3 display the characteristics of
numeric and categorical clinical variable predictors respectively.

All 44 participants were included for statistical analyses, 43 of whom contributed to the analysis
of neuroradiographic abnormalities and 29 to the analysis of death/disability outcome.

Table 1. Characteristics (Demographic Parameters) & Outcomes
Characteristics

N Mean ± SD
Age 43 43.0 ± 9.4

1 Missing

N (%)
Sex

F 7 (15.91)
M 37 (84.09)
Outcome

N (%)
Neuroradiographic Abnormalities at Baseline

Normal 18 (41.86)

Abnormal 25 (58.14)

Death/Disabled vs. Normal at 6 Months

Normal 11 (37.93)
Dead/Disabled 18 (62.07)
Note. 1 value was missing in neuroradiographic abnormalities, 15 values were missing in
death/disabled vs. normal at 6 months

12

Table 2. Characteristics of Numeric Clinical Variables (N = 44).
Day 0 Day 14
Variable N Mean ± SD N Mean ± SD
Cryptococcal Antigen (mg/kg) 40 9.5 ± 3.3 26 6.9 ± 2.4
Protein (mg/100 mL) 41 106.1 ± 109.8 28 59.7 ± 21.3
Glucose (mg/100 mL) 40 39.0 ± 20.3 28 45.3 ± 17.2
HIV Viral Load (per 10,000 copies/mL) 33 21.0 ± 38.6

CD4 (cells/mm
3
) 42 32.0 ± 31.4

Hemoglobin (g/dL) 44 11.8 ± 2.0

Albumin (g/dL) 42 3.2 ± 0.6

Table 3. Characteristics of Categorical Clinical Variables (N = 44).
Variable N (%)
Cryptococcal Antigen (Day 0) (mg/kg) <128 6 (15.00)
128-256 5 (12.50)
>256 29 (72.50)
Cryptococcal Antigen (Day 14) (mg/kg) <128 8 (30.77)
128-256 13 (50.00)
>256 5 (19.23)
Fungal Culture (Day 0) No growth 0 (0.00)
C. neoformans 43 (100.00)
Fungal Culture (Day 14) No growth 20 (47.62)
C. neoformans 22 (52.38)
Fungal Blood Culture (Day 0) No growth 2 (18.18)
C. neoformans 9 (81.82)
Fungal Blood Culture (Day 14) No growth 5 (83.33)
C. neoformans 1 (16.67)

3.1 Neuroradiographic Abnormalities

A total of 43 individuals had data on neuroradiographic abnormalities; 18 (41.9%) individuals
were normal and the remaining 25 individuals (58.1%) demonstrated at least one
neuroradiographic abnormality.

13

Table 4. Comparison of Mean Values Of Continuous Clinical Variables among Participants with
Normal versus Abnormal In Neuroradiographic Findings
Normal Abnormal
Variable Mean ± SD (N) Mean ± SD (N) t p-value
Age (years) 38.3 ± 10.7 (18) 46.0 ± 6.5 (24) -2.72 0.01
Cryptococcal Antigen (Day 0) 9.5 ± 3.6 (17) 9.4 ± 3.2 (23) 0.09 0.93
Cryptococcal Antigen (Day 14) 7.1 ± 2.0 (12) 6.7 ± 2.8 (14) 0.38 0.70
Protein (Day 0) (mg/100 mL) 80.4 ± 65.6 (18) 129.7 ± 134.8 (22) -1.51 0.14
Protein (Day 14) (mg/100 mL) 59.5 ± 23.8 (14) 59.9 ± 19.3 (14) -0.05 0.96
Glucose (Day 0) (mg/100 mL) 38.8 ± 25.1 (18) 39.2 ± 15.8 (22) -0.07 0.95
Glucose (Day 14) (mg/100 mL) 40.6 ± 16.4 (14) 49.9 ± 17.2 (14) -1.46 0.16
HIV Viral Load (Day 0) (per
10,000 copies/mL) 31.8 ± 58.7 (12) 12.0 ± 14.6 (20) 1.15 0.27
CD4 (Day 0) (cells/mm3) 34.6 ± 20.0 (17) 30.1 ± 33.5 (24) 0.44 0.66
Hemoglobin (Day 0) (g/dL) 11.4 ± 2.0 (18) 12.3 ± 2.0 (25) -1.44 0.16
Albumin (Day 0) (g/dL) 3.1 ± 0.7 (17) 3.3 ± 0.5 (24) -1.07 0.29
P-values from Student’s t-test
Continuous variables identified for further analysis are highlighted in Bold.

Table 5. Categorical Variable Frequencies by Presence of Neuroradiographic Abnormalities
Normal Abnormal

Variable N (%) N (%)
p-value
Sex
0.11
Female 5 (27.78) 2 (8.00)

Male 13 (72.22) 23 (92.00)

Cryptococcal Antigen (Day 0)

0.83

<128 3 (17.65) 3 (13.04)

128-256 2 (11.76) 3 (13.04)

>256 12 (70.59) 17 (73.91)

Cryptococcal Antigen (Day 14)

0.83

<128 3 (25.00) 5 (35.71)

128-256 7 (58.33) 6 (42.86)

>256 2 (16.67) 3 (21.43)

Fungal Culture (Day 14)

0.03

No growth 12 (70.59) 8 (33.33)

C. neoformans 5 (29.41) 16 (66.67)

p-value from Fisher’s Exact test

14

Table 4 presents the results of Student's t-test comparing the levels of continuous clinical
variables between participants with and without neuroradiographic abnormalities. No violation
of normality was detected and no significant association was found between neuroradiographic
abnormalities and clinical variables. Comparisons on categorical clinical variables with Fisher’s
Exact test
34
are summarized in Table 5. Only the association of CSF fungal culture at day 14
with neuroradiographic abnormalities was statistically significant (p=0.03). Table 6 and Table 7
show the results from univariate logistic regression evaluating associations of continuous and
categorical clinical variables with the presence of neuroradiographic abnormalities.

Table 6. Logistic Regression: Univariate Analysis of Continuous Variable Associations with
Neuroradiographc Abnormalities
Variable
Odds
Ratio
p-
value
Confidence
Interval
Age (years) 1.12 0.01 (1.02, 1.22)
Protein (Day 0) (mg/100 mL) 1.01 0.21 (1.00, 1.01)
Protein (Day 14) (mg/100 mL) 1.00 0.96 (0.97, 1.04)
Glucose (Day 0) (mg/100 mL) 1.00 0.94 (0.97, 1.03)
Glucose (Day 14) (mg/100 mL) 1.04 0.16 (0.99, 1.09)
HIV Viral Load (Day 0) (per 10,000
copies/mL) 0.98 0.22 (0.96, 1.01)
CD4 (Day 0) (cells/mm
3
) 1.00 0.65 (0.98, 1.02)
Hemoglobin (Day 0) (g/dL) 1.27 0.16 (0.91, 1.78)
Albumin (Day 0) (g/dL) 1.87 0.29 (0.59, 5.96)
Variables selected for further analysis are highlighted in Bold.

Based on these initial exploratory analyses, only the fungal culture measurement after 14 days
showed a significant association (at p<0.05) with neuroradiographic abnormalities (OR=4.8,
p=0.02, 95% CI=1.25-18.42). Individuals with growth of C. neoformans in fungal culture were
4.8 times more likely to show a neuroradiographic abnormality compared to persons with no
fungal growth.

15

We performed logistic regression to test the association of the change in CSF protein and CSF
glucose from day 0 to day 14 on neuroradiographic abnormalities; no significant associations
were found (p=0.34 for change in CSF protein and p=0.67 for change in CSF glucose).

Table 7. Logistic Regression: Univariate Analysis of Categorical Variable Associations with
Neuroradiographic Abnormalities
Variable Odds Ratio p-value Confidence Interval
Sex
Female 1.00
Male (vs. Female) 4.42 0.10 (0.75, 26.10)

Cryptococcal Antigen (Day 0)

<128 1.00

128 - 256 (vs. <128) 1.50 0.74 (0.14, 16.54)

>256 (vs. <128) 1.42 0.70 (0.24, 8.26)

Cryptococcal Antigen (Day 14)

<128 1.00

128 - 256 (vs. <128) 0.51 0.47 (0.09, 3.11)

>256 (vs. <128) 0.90 0.93 (0.09, 8.90)

Fungal Culture (Day 14)

No growth 1.00
C. neoformans 4.80 0.02 (1.25, 18.42)
Clinical parameters selected for further analysis are highlighted in Bold.

We selected 6 variables for further analysis, which met the criteria of: (1) more than 25
individuals had data available, (2) the p-value in univariate logistic regression or the p-value in
two sample t-test was <0.3. These candidate variables for the multivariable analyses are noted in
bold in Table 6 and Table 7, including one categorical variable (fungal culture (Day 14)) and five
continuous variables (CSF protein (Day 0), CSF glucose (Day 14), HIV viral load at baseline,
hemoglobin level (Day 0), and albumin level (Day 0)). Age and sex were also selected as
candidate clinical parameters for further analysis.
16

Table 8. Summary of Stepwise Selection on Logistic Regression Analysis of Neuroradiographic
Abnormalities
Step Effect Number
In
OR Score
Chi-
Square
Wald
Chi-
Square
Pr > Chi
Sq
Entered Removed
1 Age (years) 1 1.23 7.77 < 0.01
2 Albumin (Day 0)
(g/dL)
2 7.13 3.87 0.05
3 HIV Viral Load (Day
0) (per 10,000
copies/mL)
3 0.98 4.39 0.04
4 Fungal Culture (Day
14)
4 11.05 4.30 0.04
5 HIV Viral Load (Day 0)
(per 10,000
copies/mL)* Fungal
Culture (Day 14)
5 0.95 2.36 0.12
6 Glucose (Day 14)
(mg/100 mL)
6 1.04 1.61 0.20
Bold Indicate a statistically significant effect at α = 0.05 2-tailed hypothesis test.

Table 9. Summary of Forward Selection on Neuroradiographic Abnormalities
Step Effect Entered Number
In
OR Score Chi-
Square
Pr >
ChiSq
1 Age (years) 1 1.23 7.77 < 0.01
2 Albumin (Day 0) (g/dL) 2 7.13 3.87 0.05
3 HIV Viral Load (Day 0) (per 10,000
copies/mL)
3 0.98 4.39 0.04
4 Fungal Culture (Day 14) 4 11.05 4.30 0.04
5 HIV Viral Load (Day 0) (per 10,000
copies/mL)* Fungal Culture (Day 14)
5 0.95 2.36 0.12
6 Glucose (Day 14) (mg/100 mL) 6 1.04 1.61 0.20
Bold Indicate a statistically significant effect at α = 0.05 2-tailed hypothesis test.

Missing data for four of the five continuous variables, CSF protein (Day 0), CSF glucose (Day
14), HIV viral load (Day 0), and albumin level (Day 0), were imputed using multiple imputation.
There were 3 missing values imputed for CSF protein (Day 0), and 16, 11, and 2 missing values
were imputed for CSF glucose (Day 14), HIV viral load (Day 0), and albumin level (Day 0)
respectively. We did not use multiple imputation method for fungal culture (Day 0) because it is
17

a categorical variable and there are only two missing values. A total of 41 subjects were included
in further multivariable analyses.

Table 10. Results of Backward Selection on Neuroradiographic Abnormalities
Step Effect Entered Number
In
Score Chi-
Square
Pr >
ChiSq
1 Age (years) * Albumin (Day 0) (g/dL) 33 < 0.01 > 0.99
2 Glucose (Day 14) (mg/100 mL) *Sex 32 < 0.01 0.99
3 Age (years) * Protein (Day 0) (mg/100 mL) 31 < 0.01 0.99
4 HIV Viral Load (Day 0) (per 10,000 copies/mL) *
Hemoglobin (Day 0) (g/dL)
30 < 0.01 0.99
5 Age (years) * Hemoglobin (Day 0) (g/dL) 29 < 0.01 0.96
6 Fungal Culture (Day 14)*Sex 28 < 0.01 0.98
7 Glucose (Day 14) (mg/100 mL) * HIV Viral Load
(Day 0) (per 10,000 copies/mL)
27 < 0.01 0.95
8 Hemoglobin (Day 0) (g/dL) * Fungal Culture (Day 14) 26 < 0.01 0.97
9 Albumin (Day 0) (g/dL) * Fungal Culture (Day 14) 25 < 0.01 0.93
10 Age (years) * Fungal Culture (Day 14) 24 < 0.01 0.92
11 Age (years) * HIV Viral Load (Day 0) (per 10,000
copies/mL)
23 < 0.01 0.96
12 Age (years) *Sex 22 < 0.01 0.93
13 Hemoglobin (Day 0) (g/dL) * Albumin (Day 0) (g/dL) 21 < 0.01 0.97
14 Protein (Day 0) (mg/100 mL) * HIV Viral Load (Day
0) (per 10,000 copies/mL)
20 0.01 0.92
15 Glucose (Day 14) (mg/100 mL) * Albumin (Day 0)
(g/dL)
19 0.02 0.89
16 HIV Viral Load (Day 0) (per 10,000 copies/mL) *
Fungal Culture (Day 14)
18 < 0.01 0.94
17 Protein (Day 0) (mg/100 mL) *Sex 17 0.02 0.88
18 Glucose (Day 14) (mg/100 mL) * Hemoglobin (Day 0)
(g/dL)
16 < 0.01 0.95
19 Protein (Day 0) (mg/100 mL) * Albumin (Day 0)
(g/dL)
15 0.02 0.90
20 HIV Viral Load (Day 0) (per 10,000 copies/mL) *
Albumin (Day 0) (g/dL)
14 0.13 0.72
21 Glucose (Day 14) (mg/100 mL)* Fungal Culture (Day
14)
13 0.29 0.59
22 Age* Glucose (Day 14) (mg/100 mL) 12 0.38 0.54

18

All candidate clinical parameters and potential two way interaction effects were entered into the
multivariate logistic model and eliminated based on lack of significance. We used forward
selection, backward selection, stepwise selection methods to build candidate models, using a
more relaxed significance level of 0.3 for entering and removing effects, rather than a traditional
level of 0.05.

Table 8 presents the results of the stepwise logistic regression analyses for neuroradiographic
abnormalities. The variables independently associated with the presence of a neuroradiographic
abnormality in order of importance were older age, elevated albumin level (Day 0), lower HIV
viral load (Day 0), growth of C. neoformans in fungal culture (Day 14), negative interaction
between elevated HIV viral load (Day 0) and growth of C. neoformans in fungal culture (Day
14), and elevated CSF glucose (Day 14). Results of forward and backward procedures are listed
in Table 9 and Table 10 respectively. Table 11 displays the parameter estimates, standard errors
and p-values from stepwise, forward selection and backward selection models for comparison.
We also tried multiple combinations of the possible predictors in the model, refining the
candidate models and retaining only variables achieving a 0.05 level of significance.

The results for the final multivariable logistic model to predict neuroradiographic abnormalities
are summarized in Table 12. Age, Fungal Culture (Day 14), and HIV viral load (Day 0) were
selected in the final model, and a significant interaction effect between HIV viral load (Day 0)
and Fungal Culture (Day 14) was detected. A total of 41 observations were used in this model,
among which 17 (41.5%) individuals were normal and 24 individuals (58.5%) demonstrated at
least one neuroradiographic abnormality.
19

Model:
Logit (Radiographic Abnormalities) = -7.27 + 0.17 * Age + 3.22 * Fungal Culture (Day 14) –
0.007 * HIV viral load (Day 0) – 0.07 * HIV viral load (Day 0) * Fungal Culture (Day 14)
Note: HIV viral load (Day 0) equal to values multiply by 10,000

Table 11. Candidate Models for Neuroradiographic Abnormalities
Stepwise
Selection
Forward
Selection
Backward
Selection
Estimate (SE) Estimate (SE) Estimate (SE)
Intercept -17.10 ( 7.11 ) -17.10 ( 7.11 ) -642.1 ( 351.7 )
p = 0.02 p = 0.02 p = 0.07
Age (years) 0.21 ( 0.08 ) 0.21 ( 0.08 ) 2.47 ( 1.45 )
p < 0.01 p < 0.01 p = 0.09
Fungal Culture (Day 14) 2.40 ( 1.25 ) 2.40 ( 1.25 ) -2.61 ( 13.35 )
p = 0.05 p = 0.05 p = 0.85
Glucose (Day 14) (mg/100 mL) 0.04 ( 0.03 ) 0.04 ( 0.03 ) -0.94 ( 0.64 )
p = 0.22 p = 0.22 p = 0.14
HIV Viral Load (Day 0) (per 10,000
copies/mL)
-0.02 ( 0.02 ) -0.02 ( 0.02 ) -2.73 ( 1.61 )
p = 0.36 p = 0.36 p = 0.09
HIV Viral Load (Day 0) (per 10,000
copies/mL)*Fungal Culture (Day 14)
-0.05 ( 0.04 ) -0.05 ( 0.04 ) . ( . )
p = 0.19 p = 0.19 p = .
Albumin (Day 0) (g/dL) 1.96 ( 1.23 ) 1.96 ( 1.23 ) 84.53 ( 47.91 )
p = 0.11 p = 0.11 p = 0.08
Sex 34.31 ( 26.80 )
p = 0.20
Protein (Day 0) (mg/100 mL) 3.07 ( 1.67 )
p = 0.07
Protein (Day 0) (mg/100 mL)*Fungal
Culture (Day 14)
0.41 ( 0.27 )
p = 0.13
Protein (Day 0) (mg/100 mL)*Glucose
(Day 14) (mg/100 mL)
0.03 ( 0.01 )
p = 0.08
HIV Viral Load (Day 0) (per 10,000
copies/mL)*Sex
2.42 ( 1.48 )
p = 0.10
Hemoglobin (Day 0) (g/dL) 23.09 ( 12.56 )
p = 0.07
Protein (Day 0) (mg/100 mL)*
Hemoglobin (Day 0) (g/dL)
-0.37 ( 0.20 )
p = 0.070
Bold Indicate a statistically significant effect at α = 0.05 2-tailed hypothesis test.

20

The odds of having at least one neuroradiographic abnormality increase by 18.5%, with each
increase of 1 year in age (p < 0.01, CI: 1.05-1.34). The association between fungal culture (Day
14) and the outcome neuroradiographic abnormalities is significant (p=0.01). The interaction
between HIV viral load measured at baseline and fungal culture measured at day 14 was
negatively associated with neuroradiographic abnormalities: among those whose HIV viral load
(Day 0) are lower than 448,000 copies/mL, individuals with growth of C. neoformans in fungal
culture are more likely to be abnormal in neuroradiographic abnormalities than those whose C.
neoformans level did not grow; among those whose HIV viral load (Day 0) are higher than
448,000 copies/mL, individuals with growth of C. neoformans in fungal culture are less likely to
be abnormal in neuroradiographic abnormalities than those whose C. neoformans level did not
grow.

Table 12. Final Multivariable Logistic Regression Model on Neuroradiographic Abnormalities
Parameter
Names
Fixed
Intercept
Age
(yesrs)
Fungal
Culture
(Day 14)
HIV Viral Load
(Day 0) (per
10,000 copies/mL)
HIV Viral Load (Day 0)
(per 10,000 copies/mL) *
Fungal Culture (Day 14)
Parameter
Estimates
-7.27
P < 0.01
0.17
P < 0.01
3.22
P= 0.01
-0.01
P= 0.73
-0.07
P= 0.04
Bold Indicate a statistically significant effect at α = 0.05 2-tailed hypothesis test.

Although the main effect of HIV viral load (Day 0) is not significant (Chi-square = 0.12, p=0.73),
it was kept in the final model because the interaction effect of HIV viral load (Day 0) and fungal
culture (Day 14) is significant (Chi-square = 4.06, p=0.04). For individuals with growth of C.
neoformans in fungal culture at day 14, the odds of having at least one neuroradiographic
abnormality decrease by 7.6% (1-exp (-0.0071-0.0718) = 7.6%, CI: 0.82-0.98), with each
increase of 10,000 copies/mL in HIV viral load at baseline; for those who didn’t show any
growth of C. neoformans in fungal culture at day 14, the odds of being abnormal in
21

neuroradiographic abnormalities decrease by 0.7% (1-exp (-0.0071) = 0.7%, CI: 0.95-1.03), with
each increase of 10,000 copies/mL in HIV viral load at baseline.

The overall goodness of fit by Pearson test and Hosmer and Lemeshow goodness-of-fit test using
10 groups were performed for the final model
32
. No significant deviations between expected
(model predicted) and observed outcomes were detected through either Pearson test (Chi-square
(36 df) = 36.22, p=0.46) or Hosmer and Lemeshow test (Chi-square (8 df) = 9.15, p=0.33).

3.2 Death/Disabled vs. Normal at 6 Months

Only 29 of the 44 subjects had a record of death/disabled vs. normal at 6 months after the
baseline; 18 individuals were classified as dead or disabled (62.1%).

Table 13 presents the results of Student's t-test comparing the levels of continuous clinical
variables between participant status for death/neurologic disability. No violation of normality
was detected and no significant association was found between death/disabled status at 6 months
and clinical variables. Comparisons on categorical clinical variables with Fisher’s Exact test
34

are summarized in Table 14; no significant associations were found. Univariate logistic
regression analyses were conducted; Table 15 and
All candidate clinical parameters and potential two way interaction effects were entered into the
multivariate logistic model and eliminated based on lack of significance. We used forward
selection, backward selection, and stepwise selection methods to build candidate models and
used a relaxed significance level of 0.3 for entering and removing effects, rather than a
traditional level of 0.05.
22

Table 16 show the univariate associations of continuous and categorical clinical parameters with
neurological status at 6 months respectively.

Table 13. Comparison of Mean Values of Continuous Clinical Variables among Participants with
Normal versus Death/Disability in Neurological Status at 6 Months
Normal Death/Disabled
Variable Mean ± SD (N) Mean ± SD (N) t p-value
Age (years) 45.09 ± 5.07 (11) 40.00 ± 9.68 (17) 1.82 0.08
Cryptococcal Antigen (Day 0) 8.56 ± 3.09 (9) 9.47 ± 3.56 (17) -0.65 0.52
Cryptococcal Antigen (Day 14) 6.43 ± 2.88 (7) 7.00 ± 2.61 (11) -0.44 0.67
Protein (Day 0) (mg/100 mL) 126.91 ± 134.25 (11) 66.40 ± 33.73 (15) 1.46 0.17
Protein (Day 14) (mg/100 mL) 58.75 ± 25.14 (8) 52.55 ± 18.70 (11) 0.62 0.54
Glucose (Day 0) (mg/100 mL) 47.64 ± 23.66 (11) 41.29 ± 14.33 (14) 0.83 0.41
Glucose (Day 14) (mg/100 mL) 47.00 ± 13.26 (8) 44.64 ± 17.21 (11) 0.32 0.75
HIV Viral Load (Day 0) (per
10,000 copies/mL) 16.91 ± 32.44 (11) 10.91 ± 21.25 (11) 0.51 0.61
CD4 (Day 0) (cells/mm3) 32.00 ± 29.12 (10) 24.47 ± 18.44 (17) 0.83 0.42
Hemoglobin (Day 0) (g/dL) 12.17 ± 1.66 (11) 11.91 ± 2.06 (18) 0.36 0.72
Albumin (Day 0) (g/dL) 3.36 ± 0.36 (11) 3.22 ± 0.53 (18) 0.78 0.44
P-values from Student’s t-test
Continuous variables identified for further analysis are highlighted in Bold.

Table 14. Categorical Variable Frequencies by Presence of Death/Neurologic Disability 6
Months
Normal Dead/Disabled

Variable N (%) N (%)
p-value
Sex 0.62

Female 2 (18.18) 2 (11.11)

Male 9 (81.82) 16 (88.89)

Cryptococcal Antigen (Day 0)

0.62

<128 3 (33.33) 2 (11.76)

128-256 0 (0.00) 3 (17.65)

>256 6 (66.67) 12 (70.59)

Cryptococcal Antigen (Day 14)

0.21
23

<128 4 (57.14) 2 (18.18)

128-256 2 (28.57) 6 (54.55)

>256 1 (14.29) 3 (27.27)

Fungal Culture (Day 14)

1.00

No growth 6 (60.00) 9 (52.94)

C. neoformans 4 (40.00) 8 (47.06)

p-value from Fisher’s Exact test
Based on these exploratory analyses, no significant association between potential predictors and
death/disabled vs. normal at 6 months was detected. We performed logistic regression to test the
association of the change in CSF protein and CSF glucose (from day 0 to day 14); no significant
associations were found (p=0.39 for CSF protein and p=0.75 for CSF glucose).

We applied a multiple imputation strategy for continuous variables with data available for more
than 25 participants using PROC MI in SAS, including numeric variables CSF protein (Day 0
and 14), CSF glucose (Day 0 and 14), HIV viral load (Day 0), CD4 cell count (Day 0),
hemoglobin level (Day 0), and albumin level (Day 0). Age and fungal culture (Day 14) were also
selected as candidate clinical parameters for further analysis due to potential interaction effects.
We did not use multiple imputation methods for fungal culture (Day 0) because it is a categorical
variable and there are only two missing values. A total of 27 subjects were included in further
multivariable analyses.

Table 15. Univariate Analysis of Continuous Variables (Logistic Regression) for Death/Disabled
vs. Normal at 6 months
Variable
Odds
Ratio
p-
value
Confidence
Interval
Age (years) 0.92 0.13 (0.82, 1.03)
Protein (Day 0) (mg/100 mL) 0.98 0.16 (0.96, 1.01)
Protein (Day 14) (mg/100 mL) 0.99 0.52 (0.94, 1.03)
Glucose (Day 0) (mg/100 mL) 0.98 0.41 (0.94, 1.03)
24

Glucose (Day 14) (mg/100 mL) 0.99 0.73 (0.93, 1.05)
HIV Viral Load (Day 0) (per 10,000
copies/mL) 0.99 0.60 (0.96, 1.03)
CD4 (Day 0) (cells/mm3) 0.99 0.41 (0.95, 1.02)
Hemoglobin (Day 0) (Day 0) (g/dL) 0.93 0.71 (0.62, 1.39)
Albumin (Day 0) (Day 0) (g/dL) 0.51 0.43 (0.10, 2.72)
Clinical parameters selected for further analysis are highlighted in Bold.

All candidate clinical parameters and potential two way interaction effects were entered into the
multivariate logistic model and eliminated based on lack of significance. We used forward
selection, backward selection, and stepwise selection methods to build candidate models and
used a relaxed significance level of 0.3 for entering and removing effects, rather than a
traditional level of 0.05.

Table 16. Logistic Regression on Death/Disabled vs. Normal at 6 Months
Variables Odds Ratio p-value Confidence Interval
Sex
Female 1.00
Male 1.78 0.60 (0.21, 14.86)

Cryptococcal Antigen (Day 0)

<128 1.00

128 - 256 (vs. <128) >99.99 0.96 (<0.01, >99.99)

>256 (vs. <128) 3.00 0.29 (0.39, 23.07)

Cryptococcal Antigen (Day 14)

<128 1.00

128 - 256 (vs. <128) 6.00 0.13 (0.58, 61.84)

>256 (vs. <128) 6.00 0.21 (0.35,101.57)

Fungal Culture (Day 14)

No growth 1.00
C. neoformans 1.33 0.72 (0.27, 6.50)
Clinical parameters selected for further analysis are highlighted in Bold.

25

Table 17 presents the results of the stepwise logistic regression analyses. The dependent variable
was death/disabled vs. normal at 6 months; the independent variables were all candidate clinical
parameters associated with death/disabled vs. normal at 6 months and are highlighted in Table 15
and
All candidate clinical parameters and potential two way interaction effects were entered into the
multivariate logistic model and eliminated based on lack of significance. We used forward
selection, backward selection, and stepwise selection methods to build candidate models and
used a relaxed significance level of 0.3 for entering and removing effects, rather than a
traditional level of 0.05.

Table 16. According to the results in Table 17, the variables independently associated with the
risk of death or disability in order of importance were lower CSF protein (Day 0) and lower
albumin level (Day 0). Logistic regression analyses were also conducted using forward and
backward procedures. During the forward selection procedure, candidate clinical parameters
were entered into the model according to priority based on the corresponding significance.
During the backward selection procedure, all candidate clinical parameters and potential two
way interaction effects were entered into the model and were eliminated based on lack of
significance. Results of forward and backward procedures are listed in Table 18 and Table 19
respectively. Table 20 displays the parameter estimates, standard errors and p-values for
comparison for the three variable selection procedures. The result of forward selection is the
same as stepwise selection; 7 effects remained using the backward selection procedure.

26

We also tried multiple combinations of the possible predictors in the model. We refined the
candidate models, keeping only variables achieving a 0.05 level of significance. The results for
the final logistic model to predict death/disabled vs. normal at 6 months is summarized in Table
21. There are only 27 observations that were used in this model, among which 10 (37.0%)
individuals were normal and the remaining 17 (63.0%) had died or were disabled.

Table 17. Summary of Stepwise Selection on Death/Disabled vs. Normal at 6 Months
Step Effect Number
In
OR Score Chi-
Square
Wald Chi-
Square
Pr >
ChiSq
Entered Removed
1 Protein (Day 0)
(mg/100 mL)
1 0.99 1.44 0.23
2 Albumin (Day 0)
(g/dL)
2 0.34 1.33 0.25

Model:
Logit (Death/Disabled vs. Normal at 6 months) = 1.58 - 1.84 * Fungal Culture (Day 14) - 0.23 *
HIV Viral Load (Day 0) + 0.39 * Fungal Culture (Day 14) * HIV Viral Load (Day 0)
Note: HIV viral load (Day 0) equal to values multiply by 10,000
Table 18. Summary of Forward Selection on Death/Disabled vs. Normal at 6 Months
Step Effect Entered Number In OR Score Chi-Square Pr > ChiSq
1 Protein (Day 0) (mg/100 mL) 1 0.99 1.44 0.23
2 Albumin (Day 0) (g/dL) 2 0.34 1.33 0.25

Table 19. Results of Backward Selection on Death/Disabled vs. Normal at 6 Months
St
ep
Effect Entered Number
In
Score Chi-
Square
Pr > Chi
Sq
1 Glucose (Day 14) (mg/100 mL)* HIV Viral Load
(Day 0) (per 10,000 copies/mL)
20 < 0.01 0.98
2 Glucose (Day 14) (mg/100 mL)* Hemoglobin
(Day 0) (g/dL)
19 < 0.01 0.95
3 Hemoglobin (Day 0) (g/dL)* Albumin (Day 0)
(g/dL)
18 < 0.01 0.98
4 Protein (Day 0) (mg/100 mL)* Hemoglobin (Day
0) (g/dL)
17 < 0.01 0.96
27

5 Glucose (Day 14) (mg/100 mL)* Albumin (Day 0)
(g/dL)
16 0.01 0.94
6 Glucose (Day 14) (mg/100 mL)* Fungal Culture
(Day 14)
15 0.03 0.85
7 Albumin (Day 0) (g/dL)* Fungal Culture (Day 14) 14 < 0.01 0.97
8 HIV Viral Load (Day 0) (per 10,000 copies/mL)*
Albumin (Day 0) (g/dL)
13 0.02 0.90
9 HIV Viral Load (Day 0) (per 10,000 copies/mL)*
Fungal Culture (Day 14)
12 0.08 0.78
10 Hemoglobin (Day 0) (g/dL)* Fungal Culture (Day
14)
11 0.25 0.62
11 Protein (Day 0) (mg/100 mL)* Fungal Culture
(Day 14)
10 0.09 0.77
12 HIV Viral Load (Day 0) (per 10,000 copies/mL)*
Hemoglobin (Day 0) (g/dL)
9 0.64 0.42
13 Hemoglobin (Day 0) (g/dL) 8 0.10 0.75
14 Fungal Culture (Day 14) 7 0.58 0.45

In this model, the interaction between fungal culture (Day 14) and HIV viral load (Day 0) is
statistically significant (Chi-square = 4.29, p=0.04). For the individuals with growth of C.
neoformans in fungal culture at day 14, the odds of death or disabled in 6 months increases by
17.9% (1-exp (-0.2278+0.3923) = 17.9%, CI: 0.90-1.55), with each increase of 10,000
copies/mL in HIV viral load at baseline; for the individuals who did not show any growth of C.
neoformans in fungal culture at day 14, the odds of death or disabled in 6 months decreases by
20.4% (1-exp (-0.2278) = 20.4%, CI: 0.62-1.03), with each increase of 10,000 copies/mL in HIV
viral load at baseline. The interaction between HIV viral load measured at baseline and fungal
culture measured at day 14 was positively associated with Death/Disabled (status) occurring
within 6 months: among those whose HIV viral load (Day 0) are lower than 47,000 copies/mL,
individuals with growth of C. neoformans in fungal culture are less likely to be death or disabled
after 6 months of the baseline than those whose C. neoformans level did not grow; among those
whose HIV viral load (Day 0) are higher than 47,000 copies/mL, individuals with growth of C.
28

neoformans in fungal culture are more likely to be death or disabled after 6 months of the
baseline than those whose C. neoformans level did not grow.

Table 20. Candidate Models for Death/Disabled vs. Normal at 6 Months
Stepwise
Selection
Forward
Selection
Backward
Selection
Estimate (SE) Estimate (SE) Estimate (SE)
Intercept 4.65 ( 3.33 ) 4.65 ( 3.33 ) 30.94 ( 17.29 )
p = 0.16 p = 0.16 p = 0.07
Protein (Day 0) (mg/100 mL) -0.01 ( < 0.01 ) -0.01 ( < 0.01 ) -0.48 ( 0.30 )
p = 0.21 p = 0.21 p = 0.11
Albumin (Day 0) (g/dL) -1.07 ( 0.95 ) -1.07 ( 0.95 ) -4.63 ( 3.43 )
p = 0.26 p = 0.26 p = 0.18
Glucose (Day 14) (mg/100 mL) -0.24 ( 0.14 )
p = 0.09
Protein (Day 0) (mg/100 mL)*
Glucose (Day 14) (mg/100 mL)
< 0.01 ( < 0.01 )
p = 0.09
HIV Viral Load (Day 0) (per
10,000 copies/mL)
-0.28 ( 0.15 )
p = 0.06
Protein (Day 0) (mg/100 mL)*
HIV Viral Load (Day 0) (per
10,000 copies/mL)
0.01 ( < 0.01 )
p = 0.07
Protein (Day 0) (mg/100 mL)*
Albumin (Day 0) (g/dL)
0.06 ( 0.05 )
p = 0.28

Table 21. Final model of Logistic Regression on Death/Disabled vs. Normal at 6 Months
Parameter
Names
Fixed
Intercept
Fungal
Culture
(Day 14)
HIV Viral Load (Day 0)
(per 10,000 copies/mL)
Fungal Culture (Day 14) *
HIV Viral Load (Day 0) (per
10,000 copies/mL)
Parameter
Estimates
1.58
P= 0.09
-1.84
P= 0.15
-0.23
P= 0.08
0.39
P= 0.04
Bold Indicate a statistically significant effect at α = 0.05 2-tailed hypothesis test.
The overall goodness of fit by Pearson test and Hosmer and Lemeshow goodness-of-fit tests
using 9 groups were performed for the final model. Significant deviation between expected
(model predicted) and observed outcomes was not detected through either the Pearson test (Chi-
square (DF=22) = 18.41, p=0.68) or the Hosmer and Lemeshow test (Chi-square (DF=7) = 4.63,
p=0.71).
29

3.3 Goodness of Fit

Table 22. Partition for the Hosmer and Lemeshow Test on Neuroradiographic Abnormalities
Group Total RadiographicChanges = 1 RadiographicChanges = 0
Observed Expected Observed Expected
1 4 0 0.15 4 3.85
2 4 1 0.73 3 3.27
3 4 1 0.96 3 3.04
4 4 0 1.38 4 2.62
5 4 4 2.17 0 1.83
6 4 2 2.82 2 1.18
7 4 4 3.32 0 0.68
8 4 3 3.68 1 0.32
9 4 4 3.82 0 0.18
10 5 5 4.97 0 0.03

For the analysis on neuroradiographic abnormalities, even though we did not detecte any
significant deviation between expected and observed outcomes using Hosmer and Lemeshow
test (Chi-square = 9.15, p=0.33), the result is not valid in that the cell counts of neither the
observed values nor the expected values within each group were more than 5
32
. For the analysis
on death/disabled vs. normal at 6 months, the results of Hosmer and Lemeshow test (Chi-square
= 4.63, p=0.71) is not valid for the same reason. Table 22 and Table 23 display the results of
Hosmer and Lemeshow test on neuroradiographic abnormalities and death/disabled vs. normal at
6 months.

Table 23. Partition for the Hosmer and Lemeshow Test on Death/Disabled vs. Normal at 6
Months
Group Total DeathDisabledvsNormalat6M = 1 DeathDisabledvsNormalat6M = 0
Observed Expected Observed Expected
1 3 0 0.16 3 2.84
2 4 2 1.45 2 2.55
30

3 3 2 1.41 1 1.59
4 3 2 1.92 1 1.08
5 3 1 2.23 2 0.77
6 3 2 2.42 1 0.58
7 3 3 2.53 0 0.47
8 3 3 2.89 0 0.11
9 2 2 2 0 0

3.4 Sensitivity and Specificity

The predictive model for neuroradiographic abnormalities contained three predictors and one
interaction term. The ROC curve indicating the relation of sensitivity and specificity is shown in
Figure 1. From the ROC curve we can say the model we found predicts the results of
neuroradiographic abnormalities very well. The ROC curve is above the chance discrimination
line and the area between the ROC curve and the chance discrimination line (the straight line
between the origin and the point where both sensitivity and 1-specificity are equal to 1) is large;
the area under the ROC is equal to 0.892.

Based on the ROC curve, we used 0.44 as the cut point for the predicted outcome because this is
the best tradeoff between the results of true positive and true negative. Individuals with predicted
values larger than 0.44 are categorized as neuroradiographically abnormal and the rest are
categorized as normal. Using this cutpoint, the sensitivity is equal to 91.7% and specificity is
equal to 82.4%.
Figure 2 displays the ROC curve for the predictive model of death/disabled vs. normal at 6
months. According to the ROC curve, the predictive model, which includes two predictors and
one interaction term, predicts the results of death/disabled vs. normal at 6 months very well. The
31

area under the ROC is 0.844 and the area between the ROC curve and the chance discrimination
line is large also.

Using 0.69 as the cut point for death/disabled vs. normal at 6 months, the corresponding
sensitivity is 76.5% and specificity is 70.0%.

Figure 1. ROC Curve on Neuroradiographic
Abnormalities

Figure 2. ROC Curve on Death/Disabled vs.
Normal at 6 Months

3.5 Comparison of Optimized Data and Original Data

Using the predictive models from Table 12 and Table 21, we performed logistic regression on
the original data (not imputing missing data) and compared the results between the optimized
data (imputing missing data) and original data. The results for analyses of neuroradiographic
abnormalities and death/disabled vs. normal at 6 months are listed in Table 24 and Table 25
respectively. Only the effect of age on neuroradiographic abnormalities (Chi-square=5.60,
p=0.02) and the effect of fungal culture (Day 14) on neuroradiographic abnormalities were
32

significant. The odds of being abnormal in neuroradiographic abnormalities increases by 18.6%
with each year of age (Chi-square=5.60, p=0.02); individuals with growth of C. neoformans in
fungal culture are 26.7 times more likely to have neuroradiographic abnormalities (Chi-
square=4.08, p=0.04).

For neuroradiographic abnormalities, the multiple imputation method increases the number of
observations used from 29 to 41, the R-square of the optimized data decreases from 0.474 to
0.433 and the p-value of goodness-of-fit test drops from 0.964 to 0.330.

For death/disabled vs. normal at 6 months using the original data, only 20 individuals were used,
among which 10 were classified as dead or disabled and the other 10 individuals were normal.
Compared with the results with original data, the R-square increases from 0.271 to 0.349 after
applying multiple imputation, and the p-value of goodness-of-fit test increases from 0.434 to
0.705.

Table 24. Final model of Logistic Regression on Neuroradiographic Abnormalities with Original
Data
Parameter
Names
Fixed
Intercept
Age
(years)
Fungal
Culture
(Day 14)
HIV Viral Load
(Day 0) (per
10,000 copies/mL)
HIV Viral Load (Day 0)
(per 10,000 copies/mL) *
Fungal Culture (Day 14)
Parameter
Estimates
-7.23
P= 0.02
0.17
P= 0.02
3.28
P= 0.04
-0.01
P= 0.65
-0.02
P= 0.67
Bold Indicate a statistically significant effect at α = 0.05 2-tailed hypothesis test.

Figure 3. ROC Curve on Neuroradiographic
Abnormalities, Original Data
33

Figure 4. ROC Curve on Death/Disabled vs.
Normal at 6 Months, Original Data

Table 25. Final model of Logistic Regression on Death/Disabled vs. Normal at 6 Months with
Original Data
Parameter
Names
Fixed
Intercept
Fungal Culture
(Day 14)
HIV Viral Load
(Day 0) (per
10,000 copies/mL)
Fungal Culture (Day 14) *
HIV Viral Load (Day 0)
(per 10,000 copies/mL)
Parameter
Estimates
0.91
P= 0.39
-1.16
P= 0.39
-0.18
P= 0.18
0.29
P= 0.12
Bold Indicate a statistically significant effect at α = 0.05 2-tailed hypothesis test.

34

Chapter 4: Discussion

We identified older age and an interaction of HIV viral load (Day 0) and fungal culture (Day 14)
to be associated with the presence of neuroradiographic abnormalities in HIV-infected
individuals who were co-infected with Cryptococcal meningoencephalitis. For individuals with
growth of C. neoformans in fungal culture (Day 14), elevated HIV viral load (Day 0) was found
to be associated with abnormal in death/disabled vs. normal at 6 months after the baseline; for
individuals with no growth of C. neoformans in fungal culture (Day 14), lowered HIV viral load
(Day 0) was found to be associated with abnormal in death/disabled vs. normal. We did not find
the effect of cryptococal antigen titer to be significant on either neuroradiographic abnormalities
or death/disabled at 6 months.

4.1 Small Sample Size

In addition to the analyses above, we still have interest in some predictors if their odds ratios
between each level are large or small enough, even though the estimates are not significant (p-
value≥0.05), since our sample size is relatively small, and the proportion of missing values is
relatively high without multiple imputation.

For neuroradiographic abnormalities, we found that individuals with cryptococcal antigen units
between 128 and 256 at Day 0 are 1.50 times as likely to have abnormal neuroradiographic
abnormalities as those whose cryptococcal antigen units are less than 128 (p=0.74, CI: 0.14-
16.54), and the odds ratio between cryptococcal antigen units above 256 and below 128 is 1.42
(p=0.70, CI: 0.24-8.26). Individuals with cryptococcal antigen units between 128 and 256 at
35

Day 14 are 0.49 times less likely to have abnormal neuroradiographic abnormalities as those
whose cryptococcal antigen units are less than 128 (p=0.47, CI: 0.09-3.11), and the odds ratio
between cryptococcal antigen units above 256 and below 128 is 0.90 (p=0.93, CI: 0.09-8.90).

For death/disabled vs. normal at 6 months, we found that individuals with cryptococcal antigen
units between 128 and 256 at Day 0 are more than 99.99 times as likely to be found as death or
disabled in 6 months as those whose cryptococcal antigen units are less than 128 (p=0.96, CI:
0.01-99.99); the cell count of the individuals with normal outcomes and with cryptococcal
antigen units between 128 and 256 was 0. The odds ratio between cryptococcal antigen units
above 256 and below 128 is 3 (p=0.29, CI: 0.39-23.07). he odds ratio for death or disabled in 6
months of individuals with cryptococcal antigen units between 128 and 256 at Day 14 was 6
compared to those whose cryptococcal antigen units were less than128 (p=0.13, CI: 0.58-61.84);
the odds ratio between cryptococcal antigen units above 256 and below 128 is 6 (p=0.21, CI:
0.35-101.57).

Even though these odds ratios were relatively pronounced in comparison to other potential
predictors, we still excluded these two variables during modeling, because some cell counts
between cryptococcal antigen (Day 0 and 14) and neuroradiographic abnormalities were very
small, which would lead to unstable estimates in logistic regression. In larger samples, these
variables could be demonstrated to be of importance in prediction of neuroradiographic
abnormalities and death/disability in this patient population.

36

4.2 Missing Data and Multiple Imputation

In epidemiological and clinical research, missing data is a common issue during data
manipulation and data analysis. The potential of missing data to undermine the validity of
research results has often been overlooked in the medical literature
35, 36
. Simply omitting the
missing values is a usual strategy for estimating potential effects when missing values are rare.
Missing data can compromise the integrity of the study by affecting reliability, validity and
generalizability. The validity of conclusions drawn from incomplete data may be questionable as
well as the reliability of the original study design. If the proportion of incomplete cases is high, it
can limit the generalizability of the analyses because they represent only a subset of the study
population
37, 38
. In this study, among the 21 predictors, missing values were higher than 30% for
10 variables; missing values were less than 10% for only 11 variables. In this case, only
analyzing cases with available data on each variable would most certainly reduce statistical
power in this already very small sample size. A complete case approach would also waste the
rest valuable data of omitted subjects, and yield biased estimates of association.

A usual approach of dealing with missing data is single imputation. If we replace missing values
with a sample mean or mode, than we would be able to use complete case analysis methods.
However, this imputation approach will not only reduce the variability of each variable which
has been filled in with imputed data, but weaken the covariance and correlation estimates in the
data, as relationships between variables were not considered in the imputation process. If we
replace missing values with predicted values using a regression approach, we could use the
information from observed data, but the model fit and correlation estimates would be
overestimated, and the variance would be reduced.
37

In this study, we used multiple imputation to handle missing values, filling in missing values
with imputed values. The multiple imputation procedure assumes that the missing data are
missing at random (MAR), that is, the probability that an observation is missing may depend on
the observed data but not the missing values. Multiply imputed data sets were created by
repeatedly filling in the missing values with imputed values, and then the complete imputed data
were analyzed by performing logistic regression within each complete dataset. The last step
pooled the regression results over the multiply imputed datasets into one estimate. The
variability of estimates will be more accurate with multiple imputations for each missing value
because both the variability due to sampling and variability due to imputation are considered.
On the other hand, multiple imputation requires more statistical work to generate the vectors for
missing values, needs more space to store the data and requires more time on analyzing and
summarizing the dataset. However, since the sample size is small (n=44) and the statistical
methods are supported by SAS, the disadvantages of multiple imputation was not be a big issue
in this study.

During the process of multiple imputation in SAS, PROC MI was performed to create multiply
imputed datasets. A Markov chain Monte Carlo (MCMC) method
29
that assumes multivariate
normality was used to impute missing values for continuous variables, assuming the data set
showed an arbitrary missing pattern. The complete datasets were analyzed using PROC
LOGISTIC and the statistical results were combined using PROC MIANALYZE.

Since the multiple imputation approach is based on measured values, using multiple imputation
on variables with a high proportion of missing values may lead to bias and will affect the validity
38

of the optimized dataset. Thus, in this study, only the missing values of clinical parameters with
data measured for at least 25 participants were filled with imputed values.

39

Chapter 5: Conclusion

The goal of this study was to assess use of commonly measured clinical parameters for
prediction of abnormal neuroradiographic findings among HIV-infected individuals who were
co-infected with CM. We found two significant predictors of neuroradiographic abnormalities,
age and fungal culture (Day 14), along with one interaction terms between HIV viral load (Day 0)
and fungal culture (Day 14). We found one interaction term between HIV viral load (Day 0) and
fungal culture (Day 14) of death/disabled (status) after 6 months of the baseline.

Compared to CT or MRI, evaluating for clinical parameters such as cerebrospinal fluid fungal
culture and HIV viral load are inexpensive and routinely performed in clinical practice.
Predicting abnormal neuroradiographic findings in cryptococcal meningoencephalitis individuals
with clinical parameters is a feasible first step in deciding whether to perform CT or MRI for
these individuals, especially in resource-limited countries.

40

References (Numerical)

1. Caroline Charlier, Francoise Dromer, Christophe Leveque, et al. (2008). Cryptococcal
Neuroradiological Lesions Correlate with Severity during Cryptococcal
Meningoencephalitis in HIV-Positive Patients in the HAART Era. PLoS ONE, 3(4):
e1950. doi:10.1371/journal.pone.0001950

2. J., C. (2003). Opportunistic infections of the CNS in patients with AIDS: diagnosis and
management. CNS Drugs, 869-87.

3. Mitchell, A. P. (2003). Updated View of Cryptococcus neoformans Mating Type and
Virulence. INFECTION AND IMMUNITY, 4829–4830. doi:10.1128/IAI.71.9.4829–
4830.2003

4. Brouwer AE, Teparrukkul P, Rajanuwong A, et al. (2010). Cerebrospinal fluid HIV-1
viral load during treatment of cryptococcal Meningitis. Journal of Acquired Immune
Deficiency Syndromes, 668-9. doi:10.1097/QAI.0b013e3181ba489a.

5. Bicanic T, Brouwer AE, Meintjes G, et al. (2009). Relationship of cerebrospinal fluid
pressure, fungal burden and outcome in patients with cryptococcal meningitis undergoing
serial lumbar punctures. AIDS (London, England), 701-6.
doi:10.1097/QAD.0b013e32832605fe.

6. High-Dose Fluconazole for the Treatment of Cryptococcal Meningitis in HIV-Infected
Individuals - Tabular View - ClinicalTrials.gov. (2013, 12 2). Retrieved from
ClinicalTrials.gov: http://clinicaltrials.gov/ct2/show/record/NCT00885703

7. Koshi G, Anandi V, Shastry JC, et al. (1989). Coagglutination (COA) test for the rapid
diagnosis of cryptococcal meningitis. Journal of Medical Microbiology, 189-94.

8. Antinori S, Radice A, Galimberti L, et al. (2005). The role of cryptococcal antigen assay
in diagnosis and monitoring of cryptococcal meningitis. Journal of Clinical Microbiology,
5828-9. doi:10.1128/JCM.43.11.5828-5829.2005

9. Powderly WG, Cloud GA, Dismukes WE, et al. (1994). Measurement of cryptococcal
antigen in serum and cerebrospinal fluid: value in the management of AIDS-associated
cryptococcal meningitis. Clinical Infectious Diseases, 789-92.

10. Phillips AN, Staszewski S, Weber R, et al. (2001). HIV viral load response to
antiretroviral therapy according to the baseline CD4 cell count and viral load. The Journal
of the American Medical Association, 2560-7.

11. Tihana Bicanic, Thomas S. Harrison. (2004). Cryptococcal meningitis. British Medical
Bulletin, 99-118. doi:10.1093/bmb/ldh043

41

12. Robert A. Larsen, Peter G. Pappas, John Perfect, et al. (2005). Phase I Evaluation of the
Safety and Pharmacokinetics of Murine-Derived Anticryptococcal Antibody 18B7 in
Subjects with Treated Cryptococcal Meningitis. Antimicrob Agents Chemother, 49(3):
952–958. doi:10.1128/AAC.49.3.952-958.2005

13. Michael S. Saag, Richard J. Graybill, Robert A. Larsen, et al. (2000). Practice Guidelines
for the Management of Cryptococcal Disease. Clinical Infectious Diseases (CID), 30 (4):
710-718. doi:10.1086/313757

14. Dean A. Seehusen, M.D., Mark M. Reeves, M.D., and Demitri A. Fomin, M.D. (2003).
Cerebrospinal Fluid Analysis. American Family Physician, 1103-1109.

15. CSF glucose test: MedlinePlus Medical Encyclopedia. (2011, 6 18). Retrieved from
National Institutes of Health:
http://www.nlm.nih.gov/medlineplus/ency/article/003633.htm

16. Mohammadi M, Mohebbi MR, Naderi F. (2003). CSF Glucose Concentrations in Infants
with Febrile Convulsions and the Possible Effect of Acetaminophen. Indian Pediatr,
1183-6.

17. Dougherty JM, Roth RM. (1986). Cerebral spinal fluid. Emergency Medicine Clinics of
North America, 281-97.

18. Thomson RB, Bertram H. (2001). Laboratory diagnosis of central nervous system
infections. . Infectious Disease Clinics of North America, 1047-1071.

19. Jenny-Avital ER, Abadi M. (2002). Immune reconstitution cryptococcosis after initiation
of successful highly active antiretroviral therapy. . Clinical Infectious Diseases (CID),
e128-e133.

20. Saag MS, Graybill RJ, Larsen RA, et al. (2000). Practice guidelines for the management
of cryptococcal disease. Infectious Diseases Society of America. . Clinical Infectious
Diseases (CID), 710-718.

21. Dunbar SA, Eason RA, Musher DM, et al. (1998). Microscopic examination and broth
culture of cerebrospinal fluid in diagnosis of meningitis. . Journal of Clinical
Microbiology (JCM), 1617-1620.

22. AA, P. (1998). Infections of the nervous system. . Neurologic Clinics, 419-447.

23. Cerebrospinal Fluid, Fungus Culture. (n.d.). Retrieved from The University of Michigan:
https://www.pathology.med.umich.edu/apps/handbook/details.php?testID=254

24. HIV Viral Load Testing: How It's Done, Results, and More. (2012, 6 16). Retrieved from
WebMD: http://www.webmd.com/hiv-aids/hiv-viral-load-what-you-need-to-know

42

25. Rapid advice: Diagnosis, prevention and management of cryptococcal disease in HIV-
infected adults, adolescents and children. (2011, 12). Retrieved from World Health
Organization: http://www.who.int/hiv/pub/cryptococcal_disease2011/en/

26. Viral Load. (2009, 8 6). Retrieved from AIDS.gov: http://www.aids.gov/hiv-aids-
basics/just-diagnosed-with-hiv-aids/understand-your-test-results/viral-load/

27. Hemoglobin test: Results. (2011, 4 9). Retrieved from Mayo Clinic:
http://www.mayoclinic.com/health/hemoglobin-test/MY00529/DSECTION=results

28. Lab Tests & Results of Albumin measurement, serum. (n.d.). Retrieved from Medical
University of South Carolina: http://www.muschealth.com/lab/content.aspx?id=150007

29. Schafer, J. (1997). Analysis of Incomplete Multivariate Data (1st ed.). Chapman and
Hall/CRC.

30. Yuan, Y. C. (2000). Multiple Imputation for Missing Data: Concepts and New
Development. SUGI 25, (pp. Paper 267-25). Indianapolis, IN.

31. Peduzzi P, Concato J, Kemper E, et al. (1996). A simulation study of the number of
events per variable in logistic regression analysis. Journal of Clinical Epidemiology,
49:1373-9. doi:10.1016/S0895-4356(96)00236-3

32. Kellie J. Archer, Stanley Lemeshow. (2006). Goodness-of-fit test for a logistic regression
model fitted using survey sample data. The Stata Journal, Volume 6 Number 1: pp. 97-
105.

33. Andrew G. Stead, Karen G. MacDonald. (1997). Constructing ROC Curves with the SAS.
SUGI 22. San Diego, California.

34. Severino, R. (2000). PROC FREQ: It’s More Than Counts. SUGI 25, (pp. Paper 69-25).
Indianapolis, IN .

35. Jonathan A C Sterne, Ian R White, John B Carlin. (2009). Multiple imputation for
missing data in epidemiological and clinical research: potential and pitfalls. BMJ, 338.
doi:10.1136/bmj.b2393

36. Wood AM, White IR, Thompson SG. (2004). Are missing outcome data adequately
handled? A review of published randomized controlled trials in major medical journals.
Clinical Trials, 368-76.

37. Theresa Schwartz, Rachel Zeig-Owens. (2012). Knowledge (of your missing data) is
power: handling missing. SAS Global Forum, (pp. 319-2012).

38. McKnight, P. E., McKnight, K. M., Sidani, S. and Figueredo, A. J. (2007). Missing Data:
A Gentle Introduction. . New York: The Guilford Press.

43

References (Alphabetical)

AA, P. (1998). Infections of the nervous system. . Neurologic Clinics, 419-447.
Andrew G. Stead, Karen G. MacDonald. (1997). Constructing ROC Curves with the SAS. SUGI
22. San Diego, California.
Antinori S, Radice A, Galimberti L, et al. . (2005). The role of cryptococcal antigen assay in
diagnosis and monitoring of cryptococcal meningitis. Journal of Clinical Microbiology,
5828-9. doi:10.1128/JCM.43.11.5828-5829.2005
Bicanic T, Brouwer AE, Meintjes G, et al. . (2009). Relationship of cerebrospinal fluid pressure,
fungal burden and outcome in patients with cryptococcal meningitis undergoing serial
lumbar punctures. AIDS (London, England), 701-6.
doi:10.1097/QAD.0b013e32832605fe.
Brouwer AE, Teparrukkul P, Rajanuwong A, et al. . (2010). Cerebrospinal fluid HIV-1 viral load
during treatment of cryptococcal Meningitis. Journal of Acquired Immune Deficiency
Syndromes, 668-9. doi:10.1097/QAI.0b013e3181ba489a.
Caroline Charlier, Francoise Dromer, Christophe Leveque, et al. . (2008). Cryptococcal
Neuroradiological Lesions Correlate with Severity during Cryptococcal
Meningoencephalitis in HIV-Positive Patients in the HAART Era. PLoS ONE, 3(4):
e1950. doi:10.1371/journal.pone.0001950
Caroline Charlier, Francoise Dromer, Christophe Leveque, Loıc Chartier, Yves-Sebastien
Cordoliani. (2008). Cryptococcal Neuroradiological Lesions Correlate with Severity
during Cryptococcal Meningoencephalitis in HIV-Positive Patients in the HAART Era.
PLoS ONE, 3(4): e1950. doi:10.1371/journal.pone.0001950
Cerebrospinal fluid (CSF) analysis - Meningitis. CSF protein, glucose, gram stain, cultures, wbc,
leukocytes. (2013, 3 28). Retrieved from GlobalRPh:
http://www.globalrph.com/cerebrospinal_fluid.htm
Cerebrospinal Fluid, Fungus Culture. (n.d.). Retrieved from The University of Michigan:
https://www.pathology.med.umich.edu/apps/handbook/details.php?testID=254
CSF glucose test: MedlinePlus Medical Encyclopedia. (2011, 6 18). Retrieved from National
Institutes of Health: http://www.nlm.nih.gov/medlineplus/ency/article/003633.htm
CSF total protein: MedlinePlus Medical Encyclopedia. (2011, 4 30). Retrieved from National
Institutes of Health: http://www.nlm.nih.gov/medlineplus/ency/article/003628.htm
44

Dean A. Seehusen, M.D., Mark M. Reeves, M.D., and Demitri A. Fomin, M.D. (2003).
Cerebrospinal Fluid Analysis. American Family Physician, 1103-1109.
Dougherty JM, Roth RM. (1986). Cerebral spinal fluid. Emergency Medicine Clinics of North
America, 281-97.
Dunbar SA, Eason RA, Musher DM, et al. (1998). Microscopic examination and broth culture of
cerebrospinal fluid in diagnosis of meningitis. . Journal of Clinical Microbiology (JCM),
1617-1620.
Ernest S. Shtatland, Emily Cain, Mary B. Barton. (2001). The Perils of Stepwise Logistic
Regression and How to Escape Them Using Information Criteria and the Output Delivery
System. SUGI 26, (pp. Paper 222-26). Long Beach, CA .
Fishman, R. A. (1992). Cerebrospinal fluid in diseases of the nervous system. 2nd edition.
Philadelphia: Saunders.
Hemoglobin test: Results. (2011, 4 9). Retrieved from Mayo Clinic:
http://www.mayoclinic.com/health/hemoglobin-test/MY00529/DSECTION=results
Hemoglobin: MedlinePlus Medical Encyclopedia. (2013, 3 22). Retrieved from National
Institutes of Health: http://www.nlm.nih.gov/medlineplus/ency/article/003645.htm
High-Dose Fluconazole for the Treatment of Cryptococcal Meningitis in HIV-Infected
Individuals - Tabular View - ClinicalTrials.gov. (2013, 12 2). Retrieved from
ClinicalTrials.gov: http://clinicaltrials.gov/ct2/show/record/NCT00885703
HIV Viral Load Testing: How It's Done, Results, and More. (2012, 6 16). Retrieved from
WebMD: http://www.webmd.com/hiv-aids/hiv-viral-load-what-you-need-to-know
HIV Viral Load: The Test. (2012, 11 30). Retrieved from Lab Tests Online:
http://labtestsonline.org/understanding/analytes/viral-load/tab/test
J., C. (2003). Opportunistic infections of the CNS in patients with AIDS: diagnosis and
management. CNS Drugs, 869-87.
Jenny-Avital ER, Abadi M. (2002). Immune reconstitution cryptococcosis after initiation of
successful highly active antiretroviral therapy. . Clinical Infectious Diseases (CID), e128-
e133.
John R. Perfect, William E. Dismukes, Francoise Dromer, et al. (2010). Clinical Practice
Guidelines for the Management of Cryptococcal Disease: 2010 Update by the Infectious
Diseases Society of America. Clinical Infectious Diseases (CID), 50 (3): 291-322.
doi:10.1086/649858
45

Jonathan A C Sterne, Ian R White, John B Carlin. (2009). Multiple imputation for missing data
in epidemiological and clinical research: potential and pitfalls. BMJ, 338.
doi:10.1136/bmj.b2393
Kellie J. Archer, Stanley Lemeshow. (2006). Goodness-of-fit test for a logistic regression model
fitted using survey sample data. The Stata Journal, Volume 6 Number 1: pp. 97-105.
Kellie J. Archer, Stanley Lemeshow. (2006). Goodness-of-fit test for a logistic regression model
fitted using survey sample data. The Stata Journal, 6, Number 1, pp. 97–105.
Koshi G, Anandi V, Shastry JC, et al. . (1989). Coagglutination (COA) test for the rapid
diagnosis of cryptococcal meningitis. Journal of Medical Microbiology, 189-94.
Lab Tests & Results of Albumin measurement, serum. (n.d.). Retrieved from Medical University
of South Carolina: http://www.muschealth.com/lab/content.aspx?id=150007
McKnight, P. E., McKnight, K. M., Sidani, S. and Figueredo, A. J. (2007). Missing Data: A
Gentle Introduction. . New York: The Guilford Press.
Michael S. Saag, Richard J. Graybill, Robert A. Larsen, et al. (2000). Practice Guidelines for the
Management of Cryptococcal Disease. Clinical Infectious Diseases (CID), 30 (4): 710-
718. doi:10.1086/313757
Mitchell, A. P. (2003). Updated View of Cryptococcus neoformans Mating Type and Virulence.
INFECTION AND IMMUNITY, 4829–4830. doi:10.1128/IAI.71.9.4829–4830.2003
Mohammadi M, Mohebbi MR, Naderi F. (2003). CSF Glucose Concentrations in Infants with
Febrile Convulsions and the Possible Effect of Acetaminophen. Indian Pediatr, 1183-6.
Peduzzi P, Concato J, Kemper E, et al. (1996). A simulation study of the number of events per
variable in logistic regression analysis. Journal of Clinical Epidemiology, 49:1373-9.
doi:10.1016/S0895-4356(96)00236-3
Phillips AN, Staszewski S, Weber R, et al. . (2001). HIV viral load response to antiretroviral
therapy according to the baseline CD4 cell count and viral load. The Journal of the
American Medical Association, 2560-7.
Powderly WG, Cloud GA, Dismukes WE, et al. . (1994). Measurement of cryptococcal antigen
in serum and cerebrospinal fluid: value in the management of AIDS-associated
cryptococcal meningitis. Clinical Infectious Diseases, 789-92.
Rapid advice: Diagnosis, prevention and management of cryptococcal disease in HIV-infected
adults, adolescents and children. (2011, 12). Retrieved from World Health Organization:
http://www.who.int/hiv/pub/cryptococcal_disease2011/en/
46

Robert A. Larsen, Peter G. Pappas, John Perfect, et al. (2005). Phase I Evaluation of the Safety
and Pharmacokinetics of Murine-Derived Anticryptococcal Antibody 18B7 in Subjects
with Treated Cryptococcal Meningitis. Antimicrob Agents Chemother, 49(3): 952–958.
doi:10.1128/AAC.49.3.952-958.2005
Saag MS, Graybill RJ, Larsen RA, et al. (2000). Practice guidelines for the management of
cryptococcal disease. Infectious Diseases Society of America. . Clinical Infectious
Diseases (CID), 710-718.
Schafer, J. (1997). Analysis of Incomplete Multivariate Data (1st ed.). Chapman and Hall/CRC.
Severino, R. (2000). PROC FREQ: It’s More Than Counts. SUGI 25, (pp. Paper 69-25).
Indianapolis, IN .
Theresa Schwartz, Rachel Zeig-Owens. (2012). Knowledge (of your missing data) is power:
handling missing . SAS Global Forum, (pp. 319-2012).
Thomson RB, Bertram H. (2001). Laboratory diagnosis of central nervous system infections. .
Infectious Disease Clinics of North America, 1047-1071.
Tihana Bicanic, Thomas S. Harrison. (2004). Cryptococcal meningitis. British Medical Bulletin,
99-118. doi:10.1093/bmb/ldh043
Viral Load. (2009, 8 6). Retrieved from AIDS.gov: http://www.aids.gov/hiv-aids-basics/just-
diagnosed-with-hiv-aids/understand-your-test-results/viral-load/
Volunteer for the NIAID Cryptococcosis Study. (2011, 12 20). Retrieved from National Institutes
of Health: http://www.niaid.nih.gov/volunteer/crypto/Pages/default.aspx
Wood AM, White IR, Thompson SG. (2004). Are missing outcome data adequately handled? A
review of published randomized controlled trials in major medical journals. Clinical
Trials, 368-76.
Yuan, Y. C. (2000). Multiple Imputation for Missing Data: Concepts and New Development.
SUGI 25, (pp. Paper 267-25). Indianapolis, IN.

Abstract (if available)

Abstract The goal of this study is to assess if there are any clinical parameters that may predict abnormal neuroradiographic findings in HIV‐infected individuals who also have Cryptococcal meningoencephalitis (CM). We investigated the relationships between clinical parameters and neuroradiographic abnormalities measured at baseline, and the relationships between clinical parameters and a combined outcome of death/disability occurring within 6 months of the baseline. We selectively evaluated clinical parameters that are commonly used in diagnosing and monitoring CM and are inexpensive and non‐invasive compared to CT or MRI. ❧ This study is a subset of a retrospective cohort study of 44 HIV‐infected individuals with CM. Cerebrospinal fluid and clinical parameters were collected at baseline, 14 days after baseline, and 6 months later. Our study revealed two statistically significant predictors of neuroradiographic abnormalities: age and growth of C. neoformans in fungal culture measured at day 14 were positively associated with neuroradiographic abnormalities. The interaction between HIV viral load measured at baseline and fungal culture measured at day 14 was negatively associated with neuroradiographic abnormalities

Linked assets

University of Southern California Dissertations and Theses

Conceptually similar

PDF

Hepatitis C virus quasispecies in HIV/HCV co-infected women

PDF

Comparative study of laparoscopy vs. laparotomy for ovarian mass removal

PDF

The relationship of resistin, leptin, adiponectin, and ghrelin with bone mineral density in healthy postmenopausal women: longitudinal analysis

PDF

Insulin’s effect on lactate levels in extremely low birth weight neonates. a multi-center, observational study

PDF

The association between self-reported physical activity and cognition in elderly clinical trial participants

PDF

Effect of soy isoflavones on anthropometric and metabolic measurements in postmenopausal women

PDF

Pharmacogenetic association studies and the impact of population substructure in the women's interagency HIV study

PDF

Marginalization in acculturation is related to objectively measured physical activity in Latina adolecents

PDF

The association of asthma and asthma-related medications on subclinical atherosclerosis: a cross-sectional analysis of four randomized clinical trials

PDF

Evaluation of factors influencing Los Angeles Tiered-Dispatch System’s improvement on bystander CPR rate and inter reliability between electronic patient care report (ePCR) and 911 call review on...

PDF

Cross-sectional association of blood pressure, antihypertensive medications, MRI volumetric measures and cognitive function scores in an aging population

PDF

Evaluating the associations between the baseline and other exposure variables with the longitudinal trajectory when responses are measured with error

PDF

Effect of estradiol on circulating levels of inflammatory cytokines in postmenopausal women

PDF

An assessment of impact of early local progression on subsequent risk for the treatment failure in adolescent and young adult patients with non-metastatic osteosarcoma

PDF

Need for tissue plasminogen activator for central venous catheter malfunction and its association with occurrence of vVenous thromboembolism

PDF

Development and validation of survey instrument designed for cervical cancer screening in Malawi, and other low resource settings

PDF

Effects of post-menopausal hormone therapy on arterial stiffness in the ELITE trial

PDF

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

PDF

The evaluation of the long-term effectiveness of zero/low fluoroscopy workflow in ablation procedures for the treatment of paroxysmal and persistent atrial fibrillation

PDF

Associations of ambient air pollution exposures with perceived stress in the MADRES cohort

Asset Metadata

Creator Zhao, Wuchen (author)

Core Title Predicting neuroradiographic abnormalities and death/disabled status occurring within 6 months in HIV-infected individuals with cryptococcal meningoencephalitis

School Keck School of Medicine

Degree Master of Science

Degree Program Biostatistics

Publication Date 02/18/2014

Defense Date 02/18/2014

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

Tag C. neoformans,CM,CrAg,cryptococal antigen titer,cryptococcal meningoencephalitis,CSF fungal culture,CT,HIV viral load,MRI,neurological status at 6 months,neuroradiographic abnormality,OAI-PMH Harvest

Format application/pdf (imt)

Language English

Contributor Electronically uploaded by the author (provenance)

Advisor Mack, Wendy Jean (committee chair), Azen, Stanley P. (committee member), Lane, Christianne (committee member)

Creator Email wuchenzh@usc.edu,zhaowuchen@gmail.com

Permanent Link (DOI) https://doi.org/10.25549/usctheses-c3-363206

Unique identifier UC11296378

Identifier etd-ZhaoWuchen-2253.pdf (filename),usctheses-c3-363206 (legacy record id)

Legacy Identifier etd-ZhaoWuchen-2253.pdf

Dmrecord 363206

Document Type Thesis

Format application/pdf (imt)

Rights Zhao, Wuchen

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 a...

Repository Name University of Southern California Digital Library

Repository Location USC Digital Library, University of Southern California, University Park Campus MC 2810, 3434 South Grand Avenue, 2nd Floor, Los Angeles, California 90089-2810, USA