A Propensity-score-based Fine Stratification Approach for Confounding Adjustment When Exposure Is Infrequent

Area of Science:

• Epidemiology
• Biostatistics
• Health Services Research

Background:

• Propensity-score matching can lack precision with infrequent exposures due to discarding many unexposed individuals.
• The performance of propensity-score stratification in such scenarios remains underexplored.

Purpose of the Study:

• To compare the relative performance of propensity-score matching and stratification methods for confounding adjustment in observational studies.
• To evaluate these methods under varying exposure prevalences and outcome risks.

Main Methods:

• Compared propensity-score matching, cohort-based stratification, exposed-group-based stratification, and standardized mortality ratio (SMR) weighting.
• Utilized an empirical example of first-trimester statin exposure and congenital malformations.
• Conducted simulations with 1,000 cohorts (n=20,000) across diverse exposure prevalences (0.5%-10%) and outcome risks (3.5%-10%).

Main Results:

• Propensity-score stratification (exposed group) showed greater precision than matching and SMR weighting for low exposure prevalence (<5%).
• In simulations, propensity-score stratification (exposed group) yielded smaller relative bias than the cohort approach.
• Empirical example indicated greater covariate imbalance with cohort-based stratification compared to exposed-group stratification and matching.

Conclusions:

• Propensity-score stratification, particularly using fine strata based on the exposed group's propensity scores, is optimal for infrequent exposures (<5%).
• For more common exposures, all evaluated propensity-score-based confounding adjustment methods demonstrated comparable performance.