Detailed
Outline
I. Overview(page 392)
Focus:
Basics of matching
Model for matched data
Control of confounding and interaction
Examples
II. Basic features of matching(pages 392–394)
A. Study design procedure: Select referent group to
be constrained so as to be comparable to index
group on one or more factors:
i. Case-control study (our focus):
referent¼controls, index¼cases
ii. Follow-up study: referent¼unexposed,
index¼exposed
B. Category matching: If case-control study, find,
for each case, one or more controls in the same
combined set of categories of matching factors
C. Types of matching: 1-to-1,R-to-1, other
D. To match or not to match:
i. Advantage: Can gain efficiency/precision
ii. Disadvantages: Costly to find matches and
might lose information discarding controls
iii. Safest strategy: Match on strong risk factors
expected to be confounders
iv. Validity not a reason for matching: Can get
valid answer even when not matching
III. Matched analyses using stratification(pages
394–397)
A. Strata are matched sets, e.g., if 4-to-1 matching,
each stratum contains five observations
B. Special case: 1-to-1 matching: four possible
forms of strata:
i. Both case and control are exposed (Wpairs)
ii. Only case is exposed (Xpairs)
iii. Only control is exposed (Ypairs)
iv. Neither case nor control is exposed (Zpairs)
C. Two equivalent analysis procedures for 1-to-1
matching:
i. Mantel–Haenszel (MH): Use MH test on all
strata and compute MOR estimate of OR
ii. McNemar approach: Group data by pairs
(W,X,Y, andZas in B above). Use
McNemar’s chi-square statistic (X–Y)^2 /
(XþY) for test andX/Yfor estimate of OR
D. R-to-1 matching: Use MH test statistic and MORDetailed Outline 415