Stratified analysis:
Each matched pair is a stratum
or
Pool exchangeable matched setsE DD DEP
RSQ
DWithout pooling!McNemar’s
table
MRRd ¼PþQ
PþR
MORd ¼Q
Rw^2 MH¼ðQRÞ^2
QþRMOR andd w^2
MHuse discordant
pairs information
MRRd uses discordant and concordant
pairs informationThe fourth type describes a “concordant pair”
in which both the exposed and the unexposed
do not have the disease. AssumeSpairs of this
type.The analysis of data from a matched pair fol-
low-up study can then proceed using a strati-
fied analysis in which each matched pair is
a separate stratum or the number of strata is
reduced by pooling exchangeable matched
sets.If pooling is not used, then, as with case-
control matching, the data can be rearranged
into a McNemar-type table as shown at the left.
From this table, a Mantel–Haenszel risk ratio
can be computed as (PþQ)/(PþR). Also, a
Mantel–Haenszel odds ratio is computed as Q/R.Furthermore, a Mantel–Haenszel test of asso-
ciation between exposure and disease that
controls for the matching is given by the chi-
square statistic (QR)^2 /(QþR), which has
one degree of freedom under the null hypothe-
sis of noE–Dassociation.In the formulas described above, both the
Mantel–Haenszel test and odds ratio estimate
involve only the discordant pair information in
the McNemar table. However, the Mantel–
Haenszel risk ratio formula involves the
concordant diseased pairs in addition to the
discordant pairs.As an example, consider a pair-matched follow-
up study with 4,830 matched pairs designed to
assess whether vasectomy is a risk factor for
myocardial infarction. The exposure variable
of interest is vasectomy status (VS: 0¼no,
1 ¼yes), the disease is myocardial infarction
(MI: 0¼no, 1¼yes), and the matching vari-
ables are AGE and YEAR (i.e., calendar year of
follow-up).EXAMPLE
Pair-matched follow-up study 4,830
matched pairs
E¼VSð 0 ¼no; 1 ¼yesÞ
D¼MIð 0 ¼no; 1 ¼yesÞMatching variables: AGE and YEARPresentation: VIII. Analysis of Matched Follow-up Data 411