statistics for case-control study

[E. Bernardo]

Dear Everyone,

We used case-control design for our study, where each case is matched to two controls.  The variable under consideration is binary, presence or absence of a disease.  May I know what statistical test is available in SPSS that is appropriate for our data.  For example, for a five matched subjects, our dataset looks like below:

Match#   Case     Control1    Control2

1             yes        No             No

2             No         yes            No

3             No        No             Yes

4             yes        yes              yes

5             No        yes              yes

Your inputs are highly solicited.

Thank you.

Eins

[Bruce Weaver]

One option is conditional logistic regression.  But as there is no conditional logistic procedure, you would have to fit that model using COXREG.  Search the archives of this list for an example posted by Marta.  And see this note:

  http://www-01.ibm.com/support/docview.wss?uid=swg21477360

Another way to take into account the dependencies between matched cases & controls is by using generalized estimating equations (GEE).  You can fit that type of model via GENLIN.  Here's a modification of syntax I used for 3:1 matching (i.e., 3 controls for each case).  I had one dichotomous predictor variable (X).  Because it was dichotomous, I could treat it as either continuous (i.e., WITH X) or categorical  (BY X).  I chose the former because the output is a bit tidier that way.

GENLIN Case_Ctl (REFERENCE=FIRST) WITH X
  /MODEL X INTERCEPT=YES
     DISTRIBUTION=BINOMIAL LINK=LOGIT
  /REPEATED SUBJECT=MatchGrp SORT=YES
     CORRTYPE=UNSTRUCTURED ADJUSTCORR=YES
     COVB=ROBUST MAXITERATIONS=100 PCONVERGE=1e-006(ABSOLUTE) UPDATECORR=1
  /MISSING CLASSMISSING=EXCLUDE
  /PRINT CPS DESCRIPTIVES MODELINFO FIT SUMMARY SOLUTION (EXPONENTIATED).

Variable Case_Ctl = 1 for cases, 0 for Controls
Variable MatchGrp:  each case and their matched controls have the same value
Data layout:  One row per person

By the way, I had previously analyzed these same data using conditional logistic regression (with Stata).  As shown below, the results from the two analyses were very similar.  

Comparison of Conditional Logistic Regression & GEE
(View in fixed font, such as Courier)
----------------------------------------------------------
Outcome      | Odds Ratio   [95% Conf. Interval]   METHOD
----------------------------------------------------------
maj_surg  gp |   1.599      1.064       2.403         1
             |   1.606      1.072       2.406         2
----------------------------------------------------------
all_maj   gp |   1.626      1.126       2.346         1
             |   1.631      1.128       2.358         2
----------------------------------------------------------
blood     gp |   0.755      0.536       1.065         1
             |   0.839      0.616       1.142         2
----------------------------------------------------------
maj_oth   gp |   1.579      0.734       3.340         1
             |   1.585      0.735       3.419         2
----------------------------------------------------------

Method 1:  Conditional logistic regression (Stata)
Method 2:  GEE (via GENLIN in SPSS)

HTH.

Eins Bernardo wrote

Dear Everyone,

We used case-control design for our study, where each case is matched to two controls.  The variable under consideration is binary, presence or absence of a disease.  May I know what statistical test is available in SPSS that is appropriate for our data.  For example, for a five matched subjects, our dataset looks like below:

Match#   Case     Control1    Control2
1             yes        No             No
2             No         yes            No
3             No        No             Yes
4             yes        yes              yes
5             No        yes              yes


Your inputs are highly solicited.

Thank you.
Eins