Machine learning to guide infant voriconazole dosing

In a retrospective study of immunocompromised infants under 24 months receiving oral voriconazole, researchers integrated population pharmacokinetic (PopPK) modeling with machine learning (ML) to predict steady-state trough concentrations. The XGBoost model, incorporating PopPK-derived clearance and volume alongside clinical variables, demonstrated strong predictive performance internally and in external validation. Clearance, weight, and select labs were the strongest drivers of model accuracy, suggesting the potential to guide initial dosing before therapeutic drug monitoring (TDM) results are available. Despite limitations such as single-center design and lack of CYP2C19 genotyping, findings support further prospective evaluation for clinical decision support.

Shen L, Hu M, Xu X, et al. Precision dosing of voriconazole in immunocompromised children under 2 years: integrated machine learning and population pharmacokinetic modeling. Front Pharmacol. 2025;16:1671652. Published 2025 Sep 15. doi:10.3389/fphar.2025.1671652