Interpretable Machine Learning Techniques for Causal Inference Using Balancing Scores as Meta-features.

Area of Science:

• Causal Inference
• Machine Learning
• Medical Interventions

Background:

• Estimating individual causal effects is crucial for informed decision-making, particularly in medical interventions.
• Observational data presents challenges in accurately determining treatment efficacy.
• Existing methods may lack interpretability, hindering clinical application.

Purpose of the Study:

• To develop an interpretable and accurate algorithm for estimating causal effects from observational data.
• To enhance the understanding of medical intervention effectiveness through interpretable causal inference.
• To apply the proposed algorithm to analyze the impact of t-PA therapy in stroke patients.

Main Methods:

• Combining multiple predictor outputs using an interpretable model (e.g., linear predictor, if-then rules).
• Utilizing interpretable predictors and balancing scores as meta-features for causal inference.
• Adapting machine learning algorithms for calculating balancing scores to ensure accuracy.

Main Results:

• The proposed scheme demonstrated slightly lower cross-validation AUC compared to original machine learning schemes.
• The algorithm provided interpretability, revealing that t-PA therapy is effective for severe stroke patients.
• Analysis was conducted on a real-world dataset of 64,609 stroke patients with 362 variables.

Conclusions:

• The proposed algorithm successfully balances accuracy and interpretability in causal effect estimation.
• Interpretability derived from the algorithm offers valuable insights into treatment effectiveness for specific patient subgroups.
• The findings support the targeted application of t-PA therapy for severe stroke cases.