Doubly robust adaptive LASSO for effect modifier discovery.

Area of Science:

• Causal inference
• Biostatistics
• Epidemiology

Background:

• Effect modification, where treatment effects vary by pre-treatment variables, is crucial but difficult to assess.
• Identifying these effect modifiers is essential for accurate interpretation of treatment effects in observational studies.

Purpose of the Study:

• To propose and evaluate a novel two-stage procedure for the automatic selection of effect modifying variables.
• To enhance the accuracy of causal inference by identifying key variables that modify treatment effects within Marginal Structural Models (MSMs).

Main Methods:

• A two-stage procedure is presented to automatically select effect modifying variables.
• The method utilizes nuisance quantities: conditional outcome expectation and propensity score within an MSM framework.
• Performance is assessed via simulation studies and application to real-world pregnancy data.

Main Results:

• The proposed method demonstrates effective automatic selection of effect modifying variables.
• Simulation studies confirm the performance and reliability of the two-stage procedure.
• Application to pregnancy data illustrates the practical utility in estimating counterfactual outcomes, such as birth weight differences.

Conclusions:

• The developed two-stage procedure offers a tractable and effective approach for identifying effect modifiers in MSMs.
• This method facilitates more precise causal effect estimation, particularly in complex observational health data.
• The study highlights the importance of effect modification assessment in understanding treatment impacts on health outcomes.