Collaborative-controlled LASSO for constructing propensity score-based estimators in high-dimensional data.

Area of Science:

• Causal inference
• Observational studies
• High-dimensional data analysis

Background:

• Propensity score methods are vital for causal inference in observational studies.
• Model selection for propensity scores in high-dimensional settings is challenging and often overlooks the causal parameter of interest.
• Traditional methods focus on treatment mechanism fit, potentially biasing causal effect estimates.

Purpose of the Study:

• To introduce a novel collaborative model selection approach for propensity score estimation using LASSO in high-dimensional settings.
• To improve causal inference by integrating information about the causal parameter of interest into propensity score model selection.
• To minimize the bias-variance tradeoff in estimated treatment effects.

Main Methods:

• Developed a novel approach for collaborative model selection with LASSO for propensity score estimation.
• Designed quasi-experiments using electronic healthcare data with simulated potential outcomes.
• Evaluated the performance of collaborative minimum loss-based estimation against competing estimators.

Main Results:

• Collaborative minimum loss-based estimation significantly outperformed competing estimators in point estimation and confidence interval coverage.
• The propensity score models selected via this collaborative method enhanced other propensity score-based estimators.
• The approach demonstrated substantive improvements in both point estimation and confidence interval coverage when applied broadly.

Conclusions:

• Collaborative minimum loss-based estimation offers a superior strategy for propensity score model selection in high-dimensional observational studies.
• This method enhances the accuracy and reliability of causal effect estimation.
• The collaborative approach can be broadly applied to improve various propensity score-based causal inference methods.