In matched pairs, or case-control, studies, conditional logistic regression is used to investigate the relationship between an outcome of being an event (case) or a nonevent (control) and a set of prognostic factors.
The following data are a subset of the data from the Los Angeles Study of the Endometrial Cancer Data in Breslow and Day (1980). There are 63 matched pairs, each consisting of a case of endometrial cancer (Outcome
=1) and a control (Outcome
=0). The case and corresponding control have the same ID
. Two prognostic factors are included: Gall
(an indicator variable for gall bladder disease) and Hyper
(an indicator variable for hypertension). The goal of the case-control analysis is to determine the relative risk for gall
bladder disease, controlling for the effect of hypertension.
data Data1; do ID=1 to 63; do Outcome = 1 to 0 by -1; input Gall Hyper @@; output; end; end; datalines; 0 0 0 0 0 0 0 0 0 1 0 1 0 0 1 0 1 0 0 1 0 1 0 0 1 0 0 0 1 1 0 1 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 1 1 0 0 1 1 0 1 0 1 0 0 1 0 1 0 0 0 0 1 1 0 0 1 1 0 0 0 1 0 1 0 0 0 0 1 1 0 1 0 1 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 1 1 0 0 1 0 0 0 1 1 0 0 0 0 1 0 0 0 1 0 0 0 1 0 0 0 1 0 0 0 0 0 0 1 1 1 1 0 0 0 1 0 1 0 0 0 1 0 1 0 1 0 1 0 1 0 0 0 0 0 0 0 1 1 0 0 0 0 1 0 0 0 0 1 0 0 0 0 0 0 0 1 1 0 0 0 1 0 0 0 0 0 0 0 1 0 1 0 0 0 0 0 1 0 1 0 1 0 0 0 1 0 0 1 0 0 0 0 0 0 0 1 1 1 0 0 0 0 0 0 0 0 0 1 1 0 0 1 0 1 0 0 1 0 0 1 0 0 0 ;
There are several ways to approach this problem with PROC LOGISTIC:
Specify the STRATA statement to perform a conditional logistic regression.
Specify EXACT and STRATA statements to perform an exact logistic regression on the original data set, if you believe the data set is too small or too sparse for the usual asymptotics to hold.
Transform each matched pair into a single observation, and then specify a PROC LOGISTIC statement on this transformed data without a STRATA statement; this also performs a conditional logistic regression and produces essentially the same results.
Specify an EXACT statement on the transformed data.
SAS statements and selected results for these four approaches are given in the remainder of this example.
In the following statements, PROC LOGISTIC is invoked with the ID
variable declared in the STRATA
statement to obtain the conditional logistic model estimates for a model containing Gall
as the only predictor variable:
proc logistic data=Data1; strata ID; model outcome(event='1')=Gall; run;
Results from the conditional logistic analysis are shown in Output 60.11.1. Note that there is no intercept term in the "Analysis of Maximum Likelihood Estimates" tables.
The odds ratio estimate for Gall
is 2.60, which is marginally significant (p = 0.0694) and which is an estimate of the relative risk for gall bladder disease. A 95% confidence interval for this relative
risk is (0.927, 7.293).
When you believe there are not enough data or that the data are too sparse, you can perform a stratified exact logistic regression. The following statements perform stratified exact logistic regressions on the original data set by specifying both the STRATA and EXACT statements:
proc logistic data=Data1 exactonly; strata ID; model outcome(event='1')=Gall; exact Gall / estimate=both; run;
Note that the score statistic in the "Conditional Exact Tests" table in Output 60.11.2 is identical to the score statistic in Output 60.11.1 from the conditional analysis. The exact odds ratio confidence interval is much wider than its conditional analysis counterpart,
but the parameter estimates are similar. The exact analysis confirms the marginal significance of Gall
as a predictor variable.
When each matched set consists of one event and one nonevent, the conditional likelihood is given by
where and are vectors representing the prognostic factors for the event and nonevent, respectively, of the ith matched set. This likelihood is identical to the likelihood of fitting a logistic regression model to a set of data with constant response, where the model contains no intercept term and has explanatory variables given by (Breslow, 1982).
To apply this method, the following DATA step transforms each matched pair into a single observation, where the variables
Gall
and Hyper
contain the differences between the corresponding values for the case and the control (case–control). The variable Outcome
, which will be used as the response variable in the logistic regression model, is given a constant value of 0 (which is the
Outcome
value for the control, although any constant, numeric or character, will suffice).
data Data2; set Data1; drop id1 gall1 hyper1; retain id1 gall1 hyper1 0; if (ID = id1) then do; Gall=gall1-Gall; Hyper=hyper1-Hyper; output; end; else do; id1=ID; gall1=Gall; hyper1=Hyper; end; run;
Note that there are 63 observations in the data set, one for each matched pair. Because the number of observations n is halved, statistics that depend on n such as R Square (see the Generalized Coefficient of Determination section) will be incorrect. The variable Outcome
has a constant value of 0.
In the following statements, PROC LOGISTIC is invoked with the NOINT option to obtain the conditional logistic model estimates. Because the option CLODDS=PL is specified, PROC LOGISTIC computes a 95% profile-likelihood confidence interval for the odds ratio for each predictor variable; note that profile-likelihood confidence intervals are not currently available when a STRATA statement is specified.
proc logistic data=Data2; model outcome=Gall / noint clodds=PL; run;
The results are not displayed here.
Sometimes the original data set in a matched-pairs study is too large for the exact methods to handle. In such cases it might be possible to use the transformed data set. The following statements perform exact logistic regressions on the transformed data set. The results are not displayed here.
proc logistic data=Data2 exactonly; model outcome=Gall / noint; exact Gall / estimate=both; run;