Note: An SAS program would include these instructions:PROC PHREG DATA = CANCER;
MODEL WEEKS*RELAPSE(0) = DRUG TESTEE;
TESTEE = DRUG*WEEKS;
where WEEKS is the variable name for duration time, RELAPSE is the vari-
able name for survival status, ‘‘0’’ is the coding for censoring, DRUG is the 0/1
group indicator (i.e.,X 1 ), and TESTEE is the newly created variable (i.e.,X 2 ).
11.5 PAIR-MATCHED CASE–CONTROL STUDIES
Case–control studies have been perhaps the most popular form of research
design in epidemiology. They generally can be carried out in a much shorter
period of time than cohort studies and are cost-e¤ective. As a technique for
controlling the e¤ects of confounders, randomization and stratification are
possible solutions at the design stage, and statistical adjustments can be made
at the analysis stage. Statistical adjustments are done using regression methods,
such as the logistic regression described in Chapter 9. Stratification is more
often introduced at the analysis stage, and methods such as the Mantel–
Haenszel method (Chapters 1 and 6) are available to complete the task.
Stratification can also be introduced at the design stage; its advantage is that
one can avoid ine‰ciencies resulting from having some strata with a gross
imbalance of cases and controls. A popular form of stratified design occurs
when each case ismatched individuallywith one or more controls chosen to
have similar characteristics (i.e., values of confounding or matching variables).
Matched designs have several advantages. They make it possible to control for
confounding variables that are di‰cult to measure directly and therefore di‰-
cult to adjust at the analysis stage. For example, subjects can be matched using
area of residence so as to control for environmental exposure. Matching also
provides more adequate control of confounding than can adjustment in analy-
sis using regression because matching does not need specific assumptions as to
functional form, which may be needed in regression models. Of course, match-
ing also has disadvantages. Matches for subjects with unusual characteristics
are hard to find. In addition, when cases and controls are matched on a certain
specific characteristic, the influence of that characterisric on a disease can no
longer be studied. Finally, a sample of matched cases and controls is not usu-
ally representative of any specific population, which may reduce our ability to
generalize analysis results.
One-to-one matching is a cost-e¤ective design and is perhaps the most pop-
ular form used in practice. It is conceptually easy and usually leads to a simple
analysis.
PAIR-MATCHED CASE–CONTROL STUDIES 405