Corresponding matched sets are called
exchangeable.
C. Example (continued):
60 exchangeable smoker matched pairs.
40 exchangeable nonsmoker matched pairs.
D. Recommendation:
Identify and pool exchangeable matched
sets.
Carry out stratified analysis or logistic
regression using pooled strata.
Consider using unconditional ML
estimation (but conditional ML estimation
always gives unbiased estimates).
E. Reason for pooling: Treating exchangeable
matched sets as separate strata is artificial.
VIII. Analysis of Matched Follow-up Data(pages
409–413)
A. In follow-up studies, unexposed subjects are
selected to have same distribution on matching
variables as exposed subjects.
B. In follow-up studies, frequency matching rather
than individual matching is typically used
because of practical convenience and to obtain
larger sample size.
C. Model same as for matched case-control studies
except dummy variables defined by exposed/
unexposed matched sets:
logit PðXÞ¼aþbEþ~g 1 iV 1 iþ~g 2 jV 2 j
þE~dkWk:D. Analysis if frequency matching used: Consider
unconditional ML estimation when number of
strata is small, although conditional ML
estimation will always give unbiased answers.
E. Analysis if pair-matching is used and no pooling
is done: Use McNemar approach that considers
concordant and discordant pairs (P,Q,R, andS)
and computes
MRRd ¼ðPþQÞ=ðPþRÞ;MORd ¼Q=R, and
w^2 MH¼ðQRÞ^2 =ðQþRÞ.
F. Example: Pair-matched follow-up study with
4,830 matched pairs,E¼VS (vasectomy
status),D¼MI (myocardial infarction status),
match on AGE and YEAR (of follow-up);
P¼0,Q¼20,R¼16,S¼4790.
MRRd ¼ 1 : 25 ¼MORd;w^2 MH¼ 0 : 44 ðN:S:Þ:418 11. Analysis of Matched Data Using Logistic Regression