Explaining Individualized Treatment Rules: Integrating LIME and SHAP With Xgboost in Precision Medicine.

Area of Science:

• Computational biology
• Biostatistics
• Precision medicine

Background:

• Precision medicine requires interpretable predictive models for individualized treatments.
• Extreme gradient boosting (XGBoost) offers high predictive accuracy but lacks interpretability.
• Understanding feature influence is crucial for identifying patient subgroups and biomarkers.

Purpose of the Study:

• To develop an interpretable framework for estimating individualized treatment rules (ITR) using XGBoost.
• To assess treatment effect heterogeneity using a global permutation test.
• To enhance model interpretability using Local Interpretable Model-agnostic Explanations (LIME) and SHAPley Additive exPlanations (SHAP).

Main Methods:

• Developed a permutation-based pipeline within an XGBoost framework for ITR estimation.
• Incorporated LIME and SHAP for model-agnostic interpretability at global and individual levels.
• Validated the approach through simulations and real-world clinical trial data analysis.

Main Results:

• The permutation-based pipeline successfully detected empirical signals of treatment effect heterogeneity.
• LIME and SHAP provided valuable exploratory insights into feature contributions to ITR.
• The integrated framework demonstrated improved understanding of complex predictive models in clinical settings.

Conclusions:

• The proposed framework enhances the interpretability of XGBoost models for precision medicine.
• It enables robust identification of treatment effect heterogeneity, guiding individualized treatment strategies.
• Combining permutation tests with LIME and SHAP offers a powerful approach for advancing precision medicine research.