Logistic Regression: A Self-learning Text, Third Edition (Statistics in the Health Sciences)

Independent variables:

BIRTHWGT (in grams)

GENDER

DIARRHEA¼

1 if symptoms

present in

past month

0 otherwise

8

><

>:

DIARRHEA

 Exposure of interest

 Time-dependent variable

Infant Care Study Model

logit PðD¼ 1 jXÞ

¼b 0 þb 1 BIRTHWGT

þb 2 GENDERþb 3 DIARRHEA

Five GEE models presented, with
differentCi:

1. AR1 autoregressive


2. Exchangeable


3. Fixed


4. Independent


5. Independent (SLR)

The independent variables are BIRTHWGT

(the weight in grams at birth), GENDER (1¼

male, 2¼female), and DIARRHEA, a dichoto-

mous variable indicating whether the infant

had symptoms of diarrhea that month

(1¼yes, 0¼no). We shall consider DIAR-

RHEA as the main exposure of interest in this

analysis. Measurements for each subject were

obtained monthly for a 9-month period. The

variables BIRTHWGT and GENDER are time-

independent variables, as their values for a

given individual do not change month to

month. The variable DIARRHEA, however, is

a time-dependent variable.

The model for the study can be stated as shown

on the left.

Five GEE models are presented and compared,

the last of which is equivalent to a standard

logistic regression. The five models in terms

of their correlation structure (Ci) are as fol-

lows: (1) AR1 autoregressive, (2) exchangeable,

(3) fixed, (4) independent, and (5) independent

with model-based standard errors and scale

factor fixed at a value of 1 [i.e., a standard

logistic regression (SLR)]. After the output for

all five models is shown, a table is presented

that summarizes the results for the effect of the

variable DIARRHEA on the outcome. Addition-

ally, output from models using a stationary

4-dependent and a stationary 8-dependent cor-

relation structure is presented in the Practice

Exercises at the end of the chapter. A GEE

model using an unstructured correlation struc-

ture did not converge for the Infant Care data-

set using SAS version 9.2.

Presentation: II. Example 1: Infant Care Study 543