m¼Mþ1 [similar toð;Þandðþ;þÞcells in the 1:1 matching design] will
be ignored.
If we fix the numbermof exposed persons in each stratum, 1amaM,
then
Prðcases exposed=mexposed in a stratumÞ¼
mðORÞ
mðORÞþMmþ 1
where OR is the odds ratio representing the e¤ect of exposure. The result for
pair-matched design in Section 11.5 is a special case whereM¼m¼1.
For the strata, or matched sets with exactlymðm¼ 1 ; 2 ;...;MÞexposed
persons, let
n 1 ;m¼number of sets with an exposed case
n 0 ;m 1 ¼number of sets with an unexposed case
nm¼n 1 ;mþn 0 ;m 1
Then givennmfixed,n 1 ;mhasBðnm;pmÞ, where
pm¼
mðORÞ
mðORÞþMmþ 1
In the special case of 1:1 matching,M¼1, and the result is reduced to the
probabilitypof Section 11.5.1.
11.6.2 Estimation of the Odds Ratio
From the joint (conditional) likelihood function
LðORÞ¼
YM
m¼ 1
mðORÞ
mðORÞþMmþ 1
n 1 ;m 1
Mmþ 1
mðORÞþMmþ 1
n 0 ;m
one can obtain the estimate OR, but such a solution requires an iterative
procedure and a computer algorithm. We will return to this topic in the next
section.
Another simple method for estimating the odds ratio would be to treat a
matched set consisting of one case andMmatched controls as a stratum. We
then present the data from this stratum in the form of a 22 table (Table
11.10) and obtain the Mantel–Haenszel estimate for the odds ratio:
dORMH¼
P
ðad=ðMþ 1 ÞÞ
P
ðbc=ðMþ 1 ÞÞ
410 ANALYSIS OF SURVIVAL DATA