Predicting prolonged dalbavancin exposure using machine learning: a validated strategy for individualized redosing.

Area of Science:

• Pharmacology
• Infectious Diseases
• Machine Learning

Background:

• Dalbavancin, a lipoglycopeptide, is used for complex Gram-positive infections but faces challenges due to pharmacokinetic variability and MIC heterogeneity.
• Optimizing dalbavancin dosing is crucial for effective treatment, especially with its long-acting profile.

Purpose of the Study:

• To develop and validate machine learning (ML) models for predicting dalbavancin plasma concentrations against pharmacokinetic/pharmacodynamic targets.
• To enable early, individualized redosing decisions for dalbavancin therapy.

Main Methods:

• Trained ML models (Support Vector Machine) on simulated pharmacokinetic profiles.
• Validated models using independent simulated datasets and real-world cohorts (Limoges and Nantes University Hospitals).
• Input features included patient demographics, creatinine clearance, MIC, and a single pre-dose plasma concentration.

Main Results:

• ML models achieved high accuracy (>88%) and sensitivity (>90%) across validation settings.
• Clinical validation demonstrated accuracy approaching 95% with no false negatives observed.
• ML models outperformed traditional Bayesian estimation in accuracy and sensitivity for dalbavancin dosing.

Conclusions:

• A machine learning approach provides a pragmatic strategy for model-informed precision dosing of dalbavancin.
• This method supports early, individualized redosing decisions, complementing Bayesian forecasting and reducing serial sampling.