Optimization-Based Stable Balancing Weights Versus Propensity Score Weighting for Samples With High Covariate...

Area of Science:

• Epidemiology
• Biostatistics
• Health Services Research

Background:

• Propensity score weighting (PSW) is a common method for covariate balance in observational studies.
• Stable balancing weights (SBW) offer an alternative approach to achieve covariate balance.
• Comparing the performance of SBW and PSW is crucial for selecting optimal weighting methods.

Purpose of the Study:

• To compare the performance of stable balancing weights (SBW) against propensity score weighting (PSW).
• To evaluate covariate balance and effective sample size (ESS) using both methods.
• To assess performance in cases of extreme covariate imbalance and substantial sample size discrepancy.

Main Methods:

• Utilized the Premier Healthcare Database for two applied cases involving surgical procedures and neurological procedures.
• Generated average treatment effects on the treated (ATT) weights.
• Implemented SBW using grid search and prespecified SMD tolerance techniques.
• Compared SBW and PSW on postweighting standardized mean differences (SMD), number of imbalanced covariates, and ESS.

Main Results:

• Both SBW techniques improved covariate balance.
• SBW methods achieved higher ESS for control groups compared to PSW.
• Sensitivity analyses with SBW and variable-specific SMD thresholds further increased ESS, outperforming PSW.
• All methods resulted in postweighting ESS lower than the original unweighted sample size.

Conclusions:

• Optimization-based SBW provides flexibility in prespecifying covariate balance goals.
• SBW resulted in superior postweighting covariate balance and larger ESS compared to PSW in applied analyses.
• SBW is a promising alternative to PSW for enhancing causal inference from observational data.