Development and validation of interpretable machine learning models for predicting AKI risk in patients treated with...

Area of Science:

• Oncology
• Nephrology
• Immunotherapy
• Machine Learning

Background:

• Anti-programmed cell death protein 1 (PD-1)/programmed cell death ligand 1 (PD-L1) immunotherapy has transformed cancer treatment but can cause acute kidney injury (AKI).
• Early AKI prediction is crucial for managing treatment interruptions and improving patient outcomes.
• Existing research lacks robust AKI risk prediction models for patients on PD-1/PD-L1 inhibitors.

Purpose of the Study:

• To develop and validate an interpretable machine learning (ML) model for early AKI prediction.
• To identify patients at high risk of developing AKI during PD-1/PD-L1 inhibitor therapy.
• To facilitate personalized therapeutic strategies and improve patient prognosis.

Main Methods:

• Retrospective cohort design utilizing data from January 2018 to January 2024.
• Evaluation of nine ML models, with Gradient Boosting Machine (GBM) selected for optimal performance.
• Application of SHapley Additive exPlanations (SHAP) for feature importance and model interpretability.

Main Results:

• The GBM model demonstrated strong predictive performance with an AUC of 0.850 in the validation set and 0.795 in the test set.
• Calibration curves confirmed good agreement between predicted and observed AKI risks.
• An interpretable GBM model with 13 features was developed, and a web-based calculator is available for clinical use.

Conclusions:

• An interpretable ML model was successfully developed and validated for predicting AKI in patients receiving PD-1/PD-L1 inhibitors.
• The model enables early identification of high-risk individuals, supporting personalized treatment adjustments.
• This tool can assist clinicians in proactive AKI management within cancer immunotherapy settings.