Applying interpretable machine learning workflow to evaluate exposure-response relationships for large-molecule...

Area of Science:

• Pharmacometrics
• Machine Learning in Drug Development
• Oncology Drug Efficacy

Background:

• Logistic Regression (LR) and Cox Proportional Hazard (CoxPH) models are standard for evaluating exposure-response (E-R) relationships in oncology drug development.
• The application of machine learning (ML) models for E-R assessment remains underexplored.

Purpose of the Study:

• To develop and validate a machine learning workflow for assessing E-R relationships in oncology drugs.
• To compare the performance of traditional LR/CoxPH models with tree-based XGBoost (XGB) models in E-R evaluation.
• To provide an interpretable ML approach for informing key dosing decisions.

Main Methods:

• Developed a workflow to train regularized LR/CoxPH and XGB models.
• Derived odds ratios for best overall response and hazard ratios for overall survival.
• Evaluated E-R relationships across exposure quantiles using clinical trial datasets.

Main Results:

• The E-R conclusions from LR/CoxPH and XGB models were largely consistent.
• Findings aligned well with historical pharmacometric analyses.
• The interpretable ML workflow demonstrated reliable E-R assessment.

Conclusions:

• The developed interpretable ML workflow offers a promising alternative for assessing E-R relationships.
• This approach can significantly impact and inform critical dosing decisions in drug development.
• Consistent results between ML and traditional models validate the utility of this new workflow.