Utilizing stratified generalized propensity score matching to approximate blocked randomized designs with multiple...

Area of Science:

• Causal Inference
• Biostatistics
• Epidemiology

Background:

• Propensity score matching (PSM) is limited to binary treatments.
• Generalized propensity score matching (GPSM) extends PSM to multi-level treatments but typically emulates only completely randomized trial (CRT) designs.
• Emulating more efficient blocked randomized trial designs is desirable for causal inference.

Purpose of the Study:

• To develop a new generalized propensity score matching (GPSM) estimator.
• To incorporate stratifying variables into GPSM to emulate blocked randomized trial designs.
• To enable more efficient retrospective causal inference analyses.

Main Methods:

• Developed a novel GPSM estimator incorporating relevant stratifying variables.
• Extended existing variance estimation methods for GPSM to accommodate stratification.
• Applied the method to analyze the effect of household size on adult systolic blood pressure.
• Conducted a simulation study to assess performance against non-stratified GPSM.

Main Results:

• The proposed method successfully incorporates stratifying variables into GPSM.
• The new estimator allows for the emulation of stratified randomized trial designs.
• Simulation results demonstrate the performance of the stratified GPSM approach.

Conclusions:

• The developed method enhances GPSM by enabling emulation of stratified randomized trials.
• This provides a more efficient approach for causal inference with multi-level treatments.
• The method allows retrospective analyses to better align with prospective stratified experimental designs.