Contrast-specific propensity scores for causal inference with multiple interventions.

Area of Science:

• Biostatistics
• Health Economics
• Pharmacoeconomics

Background:

• Estimating heterogeneous treatment effects (HTE) is crucial for personalized medicine.
• Existing propensity score methods are limited to binary or few treatment options.
• Non-experimental data analysis requires robust methods for causal inference.

Purpose of the Study:

• To develop and validate a novel propensity score-based method for HTE estimation with three or more treatment options.
• To address the limitations of existing methods in multi-treatment scenarios.
• To provide unbiased HTE estimates for clinical decision-making.

Main Methods:

• Development of a generalized propensity score framework for multi-treatment settings.
• Theoretical proof of unbiased estimation properties.
• Application to a real-world patient registry dataset (Singapore diabetic dyslipidemia).

Main Results:

• The proposed method yields unbiased HTE estimates for multiple treatment options.
• Demonstrated application in recommending treatments (Statins, fibrates, non-pharmacological) for diabetic dyslipidemia.
• Numerical studies show improved stability compared to multi-dimensional propensity score benchmarks.

Conclusions:

• The new method effectively extends propensity score analysis to complex treatment scenarios.
• It enables personalized treatment recommendations in multi-option settings.
• Offers a more stable and reliable approach for HTE estimation in non-experimental studies.