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

hypertensive subject with a 220 cholesterol count isexp(0.1960)¼1.2166. The
estimated odds ratio for CAT¼1 vs. CAT¼0 for a nonhypertensive subject with a
220 cholesterol count isexp(2.5278)¼12.5262. The table titled “Contrast Results”
gives the chi-square test statistic (53.16) andp-value (<0.0001) for the likelihood ratio
test on the two interaction terms.

C. Events/Trials Format

The Evans County datasetevans.datcontains individual level data. Each observation
represents an individual subject. PROC LOGISTIC and PROC GENMOD also accom-
modate summarized binomial data in which each observation contains a count of the
number of events and trials for a particular pattern of covariates. The dataset EVANS2
summarizes the 609 observations of the EVANS data into eight observations, where
each observation contains a count of the number of events and trials for a particular
pattern of covariates. The dataset contains five variables described below:
CASES – number of coronary heart disease cases
TOTAL – number of subjects at risk in the stratum
CAT – serum catecholamine level (1¼high, 0¼normal)
AGEGRP – dichotomized age variable (1¼age55, 0¼age<55)
ECG – electrocardiogram abnormality (1¼abnormal, 0¼normal)

The code to produce the dataset is shown next. The dataset is small enough that it can
be easily entered manually.

DATA EVANS2;

INPUT CASES TOTAL CAT AGEGRP ECG;

CARDS;

17 274 0 0 0

15 122 0 1 0

7 59001

5 32011

1 8100

9 39110

3 17101

14 58 1 1 1

;

To run a logistic regression on the summarized data EVANS2, the response is put into
anEVENTS/TRIALSform for either PROC LOGISTIC or PROC GENMOD. The
model is stated as follows:

logit PðCHD¼ 1 jXÞ¼b 0 þb 1 CATþb 2 AGEGRPþb 3 ECG

The code to run the model in PROC LOGISTIC using the dataset EVANS2 is:

PROC LOGISTIC DATA¼EVANS2;

MODEL CASES/TOTAL¼CAT AGEGRP ECG;

RUN;

608 Appendix: Computer Programs for Logistic Regression