Within-subject exposure dependency in case-crossover studies

Area of Science:

• Epidemiology
• Biostatistics
• Public Health Research

Background:

• The case-crossover design is a specialized epidemiological method used to investigate the effects of transient exposures on acute events.
• This design relies on within-subject comparisons, contrasting exposure status at the time of an event with exposure status during control periods.
• Conditional logistic regression is a common analytical approach for case-crossover studies, but its assumptions regarding exposure patterns require careful consideration.

Purpose of the Study:

• To evaluate the impact of within-subject dependence of exposures across successive time intervals in case-crossover designs.
• To identify conditions under which the conditional logistic model yields biased estimates in this design.
• To explore alternative analytical strategies for case-crossover data that ensure unbiased estimation.

Main Methods:

• Theoretical analysis of the conditional logistic model applied to case-crossover data with dependent exposures.
• Investigation of the properties of the Mantel-Haenszel estimator under exposure stationarity.
• Derivation of maximum likelihood methods for case-crossover analysis, drawing parallels with cohort study models.

Main Results:

• Conditional logistic model estimates in case-crossover studies are shown to be biased when within-subject exposure dependence exists.
• This bias is eliminated if the distribution of exposures across the event time and control periods is exchangeable.
• The Mantel-Haenszel estimator provides approximately unbiased odds ratio estimates when exposures are stationary.

Conclusions:

• Standard conditional logistic regression may produce biased results in case-crossover studies due to exposure dependence.
• Exchangeability of exposure distributions is a key condition for unbiased estimation using conditional logistic models.
• Maximum likelihood methods, potentially adapted from cohort models, offer a robust alternative for analyzing case-crossover designs.