Raltegravir Plasma Exposure: A Machine Learning-Based Model for its Prediction Using Limited Sampling Strategy.

Area of Science:

• Pharmacology
• Machine Learning
• Computational Biology

Background:

• Limited sampling strategies (LSS) are crucial for estimating drug exposure (AUC).
• Machine learning (ML) shows promise in developing accurate LSS, outperforming traditional methods.
• Predicting raltegravir (RAL) exposure is essential for optimizing patient treatment.

Purpose of the Study:

• To develop and validate an ML-based LSS for predicting raltegravir (RAL) exposure.
• To evaluate the performance of different ML algorithms for LSS.
• To assess the utility of the developed LSS in various pharmacokinetic contexts.

Main Methods:

• Four ML algorithms (XGBoost, Random Forest, GLMNet, SVM) were trained on simulated pharmacokinetic profiles.
• Pharmacokinetic data were generated using Monte Carlo simulation from a population pharmacokinetic (POPPK) model.
• Model performance was evaluated using root mean square error (RMSE) on training, test, and external validation datasets.

Main Results:

• XGBoost model using concentrations at 0.5, 2, and 4 hours demonstrated superior predictive performance.
• The model achieved low bias and RMSE in test sets (0.8%/8.7%) and independent simulations (1.9%/14.3%).
• Performance showed a decrease with real patient data (5.0%/24.1%), indicating potential extrapolation challenges.

Conclusions:

• An ML-based LSS using three sampling points accurately estimates raltegravir AUC₀-₁₂.
• This validated approach offers a valuable tool for pharmacokinetic and PK/PD studies.
• The method reduces the need for intensive sampling in clinical settings.