Conditional Logistic Regression

Conditional logistic regression estimates odds ratios in matched case-control, case-crossover or binary panel data by conditioning on each matched set, which removes set-specific intercepts.

조건부 로지스틱 회귀는 짝지은 집합마다 조건을 걸어 집합별 절편을 없애고, 짝지은 사례-대조·사례-교차·이진 패널 자료에서 오즈비를 추정하는 방법입니다.

I have a 1:3 matched case-control study; how do I estimate exposure odds ratios while respecting the matching?

1:3으로 짝지은 사례-대조 연구인데, 짝짓기를 반영해서 노출의 오즈비를 어떻게 추정하나요?

Conditional logistic regression uses the likelihood of which members of each matched set are cases given how many cases the set contains, so the matching factors drop out of the model. The resulting odds ratios compare members within the same set, which is the correct analysis for matched case-control and fixed-effects binary panel data.

When to use Conditional Logistic Regression

• Cases are individually matched to one or more controls (1:1, 1:M or M:N).
• A case-crossover design compares hazard and control periods within the same person.
• Binary panel data need a fixed-effects (Chamberlain) logit.

When to use something else

• The data are not matched or grouped; use ordinary logistic regression.
• The effect of a set-level variable (such as a matching factor) is of interest; use a mixed-effects logistic model.

Assumptions and requirements

• Observations within a set are exchangeable apart from the covariates.
• Covariate effects are the same across matched sets.
• Sets with only cases or only controls carry no information and are dropped.

What the analysis reports

• Coefficients and odds ratios with Wald confidence intervals
• Number of informative sets and rows used, sets dropped
• Likelihood-ratio test against the null model
• Exact conditional likelihood or Breslow/Efron approximation

How to report it

In conditional logistic regression on [number] matched sets, the odds ratio for [exposure] was [value], 95% CI [[value], [value]], p = [value].

Related methods

• Binary Logistic Regression
• McNemar Test
• Stratified Cox Regression

Run Conditional Logistic Regression on Quorant