Evaluating Preclinical-to-Human Bioavailability Classification Using Smooth ROC Curve Estimation

Area of Science:

• Pharmacokinetics
• Drug Development
• Biostatistics

Background:

• Human oral bioavailability is crucial for drug development, but animal models often show poor predictive accuracy.
• Receiver Operating Characteristic (ROC) analysis is used to assess animal model predictability, but empirical curves can be discontinuous and difficult to interpret, especially with small sample sizes.

Purpose of the Study:

• To enhance the interpretability and accuracy of ROC analysis for predicting human oral bioavailability from animal data.
• To apply a smooth ROC estimation method using Bernstein polynomials to a multispecies pharmacokinetic dataset.

Main Methods:

• Utilized a curated multispecies dataset previously analyzed with empirical ROC techniques.
• Applied a smooth ROC estimation method based on Bernstein polynomials.
• Compared the performance of the Bernstein estimator against traditional empirical ROC curves.

Main Results:

• The Bernstein estimator generated smoother and more visually coherent ROC curves compared to the empirical approach.
• A slight underestimation of the Area Under the Curve (AUC) was observed for species with limited sample sizes using the Bernstein estimator.
• Demonstrated a trade-off between curve smoothness and discriminatory power.

Conclusions:

• Smooth ROC estimation using Bernstein polynomials offers practical advantages for translational pharmacokinetic modeling.
• Nonparametric smoothing techniques can improve the interpretation of predictive performance from animal models in drug development.
• Further research may explore optimizing smoothing parameters to balance interpretability and discriminatory power.