External Comparator Studies: Performance of Four Missing Data-Handling Approaches, Stratified by Four Different...

Area of Science:

• Causal inference
• Observational studies
• Missing data analysis

Background:

• External comparator (EC) studies are susceptible to bias from missing data and unmeasured confounding.
• Existing research has quantified these effects, but a broader evaluation of missing data-handling approaches was needed.
• Knowledge gaps exist in understanding the performance of various statistical methods for addressing missing data in EC studies.

Purpose of the Study:

• To investigate the performance of four distinct missing data-handling strategies.
• To assess these strategies across four different marginal estimators: ATU, ATE, ATT, and ATO.
• To provide clarity on minimizing bias in EC studies using propensity score weighting with missing data.

Main Methods:

• An extensive simulation study was conducted using two real-world EC case studies.
• Four missing data-handling approaches were assessed: within-cohort MI, across-cohort MI, mixed MI, and covariate omission.
• Propensity score weighting was employed as the causal inference method, with missingness simulated only in the EC cohort.

Main Results:

• Within-cohort multiple imputation (MI) and the average treatment effect of the untreated (ATU) estimator showed the best performance in bias mitigation.
• The strategy of omitting prognostic factors (covariates) due to high missingness yielded the worst results.
• Performance varied across different marginal estimators and missing data-handling techniques.

Conclusions:

• The study clarifies the performance of different missing data strategies for marginal estimators in EC studies.
• Findings aid researchers in selecting appropriate statistical approaches to minimize bias when using propensity score weighting.
• Recommendations include considering bias estimation and correction steps for more robust EC study results.