Methodological Techniques Used in Machine Learning to Support Individualized Drug Dosing Regimens Based on...

Area of Science:

• Pharmacokinetics and Drug Development
• Computational Biology and Bioinformatics
• Artificial Intelligence in Medicine

Background:

• Individualized drug dosing optimizes therapeutic outcomes for drugs with high inter-individual variability.
• Population pharmacokinetic (PPK) modeling is standard but labor-intensive and requires expertise.
• Machine learning (ML) offers a promising alternative for personalized drug dosing.

Purpose of the Study:

• To conduct a scoping review of ML-based pharmacokinetic modeling for dose optimization.
• To examine the methodologies and predictive performance of ML models in this field.
• To identify gaps in current ML applications for individualized drug dosing.

Main Methods:

• Systematic search of five databases up to May 2025.
• Inclusion of studies comparing ML and PPK model predictions for drug concentrations or parameters.
• Exclusion of non-English studies, reviews, protocols, and studies not using ML for individualized dosing.

Main Results:

• Fifty-eight studies were included, with boosting, tree-based, instance-based, and regression models being common ML approaches.
• 31% of studies integrated ML with PPK models; others used standalone ML models.
• ML models demonstrated comparable or superior predictive accuracy to PPK models, particularly for drugs with high pharmacokinetic variability.

Conclusions:

• Substantial heterogeneity exists in ML modeling approaches, feature selection, and evaluation methods.
• Standardization in reporting and methodology is essential for ML model reproducibility.
• Enhanced standardization will improve the clinical applicability of ML models in individualized drug dosing.