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

Area Under the Curve

Test Result Variable(s): Predicted probability

Area

.745

Conditional Logistic Regression

SPSS does not perform conditional logistic regression except in thespecial casein
which there is only one case per stratum, with one or more controls. The SPSS
survival analysis procedure COXREG can be used to obtain coefficient estimates
equivalent to running a conditional logistic regression.

Recall that the MI dataset contains information on 39 cases diagnosed with myocar-
dial infarction, each of which is matched with two controls. Thus, it meets the
criterion of one case per stratum. Atimevariable must be created in the data, coded
to indicate that all cases had the event at the same time, and all controls were
censored at a later time. This variable has already been included in the SPSS dataset
mi.sav. The variable has the value 1 for all cases and the value 2 for all controls.

The first step is to open the SPSS dataset,mi.sav, into the Data Editor window. The
corresponding command syntax is:

GET

FILE¼‘C:\mi.sav’.

To run the equivalent of a conditional logistic regression analysis, select Analyze!
Survival!Cox Regression from the drop-down menus to reach the dialog box to
specify the model. Select SURVTIME from the variable list and enter it into the Time
box. The Status box identifies the variable that indicates whether the subject had an
event or was censored. For this dataset, select and enter MI into the Status box. The
value of the variable that indicates that the event has occurred (i.e., that the subject is
a case) must also be defined. This is done by clicking on the Define Event button and
entering the value “1” in the new dialog box. Next, select and enter the covariates of
interest (i.e., SMK, SBP, ECG) into the Covariate box. Finally, select and enter the
variable which defines the strata in the Strata box. For the MI dataset, the variable is
called MATCH. Click on OK to run the model.

The corresponding syntax, with the default specifications regarding the modeling
process follows:

COXREG

survtime /STATUS¼mi(1) /STRATA¼match

/METHOD¼ENTER smk sbp ecg

/CRITERIA¼PIN(.05) POUT(.10) ITERATE(20).

642 Appendix: Computer Programs for Logistic Regression