Analysis of matched case-control data with incomplete strata: applying longitudinal approaches.

Area of Science:

• Biostatistics
• Epidemiology
• Statistical Modeling

Background:

• Matched case-control data share similarities with longitudinal data but employ a retrospective sampling scheme.
• Conditional logistic regression (CLR) analysis can suffer efficiency losses with incomplete covariate data or strata with identical covariate values, particularly in small sample sizes.
• Retrospective models for longitudinal data are explored as alternatives for analyzing matched case-control data.

Purpose of the Study:

• To compare the statistical properties of matched case-control data analyses using conditional likelihood and generalized estimating equations (GEE).
• To evaluate the efficiency and consistency of these methods under various simulation scenarios.

Main Methods:

• Simulations were conducted to compare conditional likelihood and GEE methods.
• Scenarios included one-to-one and one-to-two matching designs with varying strata sizes, complete and incomplete strata, and dichotomous and normal exposures.

Main Results:

• Both CLR and GEE methods produced consistent estimates with proper coverage for binary and continuous exposures.
• GEE yielded estimates with smaller standard errors compared to CLR, indicating greater efficiency.
• Efficiency losses for CLR were more pronounced with incomplete matched sets and smaller strata sizes.

Conclusions:

• GEE is a more efficient analytical approach for matched case-control data than CLR, especially when dealing with incomplete data or small strata.
• Increasing the number of controls within strata can improve the efficiency of CLR.
• The magnitude of the association influences the relative efficiency losses observed.