Machine Learning: A New Approach for Dose Individualization.

Area of Science:

• Precision Medicine
• Computational Biology
• Pharmacometrics

Background:

• Machine learning (ML) excels at analyzing complex data, offering potential for personalized medicine.
• Individualized dosing using ML is an emerging field requiring systematic evaluation.
• Existing research has not fully explored ML's capabilities for optimizing drug dosages.

Purpose of the Study:

• To systematically review study designs and modeling details of ML applications in individualized drug dosing.
• To summarize current practices, identify limitations, and propose improvements for ML in pharmacometrics.
• To assess the quality and risk of bias in studies applying ML for dose individualization.

Main Methods:

• Conducted a systematic review of studies using ML for individualized drug dosing.
• Analyzed study populations, predictive targets, data sources, ML algorithms, and feature selection.
• Evaluated predictive performance and used the Prediction model Risk of Bias Assessment Tool (PROBAST) for quality assessment.

Main Results:

• ML is applicable for both a priori and a posteriori dose selection, optimization, and therapeutic drug monitoring.
• Studies predominantly focus on narrow therapeutic index drugs like immunosuppressants and anti-infectives.
• Limited attention is given to special populations (e.g., children), and most studies exhibit poor methodological quality and high risk of bias.

Conclusions:

• The clinical implementation of ML for dose individualization is hindered by a lack of external validation and clinical utility assessment.
• Improvements in study design, methodological rigor, and validation are crucial for translating ML models into clinical practice.
• Further research is needed to enhance the clinical relevance and applicability of ML-driven individualized dosing strategies.