Propensity score weighting methods for causal subgroup analysis with time-to-event outcomes

Area of Science:

• Biostatistics
• Epidemiology
• Health Services Research

Background:

• Evaluating intervention effects in specific patient subgroups is vital for comparative effectiveness research.
• Existing causal subgroup analysis methods have limitations for time-to-event outcomes.
• Propensity score weighting offers a promising approach for causal subgroup survival analysis.

Purpose of the Study:

• To investigate propensity score weighting methods for causal subgroup survival analysis.
• To introduce and estimate two novel causal estimands: subgroup marginal hazard ratio and subgroup restricted average causal effect.
• To compare the performance of different propensity score models and weighting schemes.

Main Methods:

• Developed propensity score weighting estimators for subgroup marginal hazard ratio and subgroup restricted average causal effect.
• Analytically established the link between subgroup covariate balance and bias for the restricted average causal effect.
• Conducted extensive simulations comparing logistic regression, random forests, LASSO, and GBM propensity score models with inverse probability weighting and overlap weighting.

Main Results:

• The logistic model incorporating subgroup-covariate interactions selected by LASSO consistently outperformed other propensity score models.
• Overlap weighting generally surpassed inverse probability weighting in terms of covariate balance, bias, and variance.
• The benefits of overlap weighting were most evident in small subgroups and situations with poor overlap.

Conclusions:

• Propensity score weighting, particularly with overlap weighting and carefully selected propensity score models, is effective for causal subgroup survival analysis.
• The findings provide robust methods for evaluating time-to-event outcomes in pre-specified patient subgroups.
• Applied methods to the "Comparing Options for Management: PAtient-centered REsults for Uterine Fibroids" study to assess myomectomy vs. hysterectomy effects on disease recurrence.