Subgroup identification-based model selection to improve the predictive performance of individualized dosing.

Area of Science:

• Pharmacometrics
• Pediatric Pharmacology
• Computational Biology

Background:

• Model-informed precision dosing (MIPD) currently relies on a single population pharmacokinetic model.
• Optimizing individualized dosing requires improved predictive performance for specific patient subgroups.

Purpose of the Study:

• To develop and evaluate a subgroup identification-based model selection approach for precision dosing.
• To enhance the predictive performance of vancomycin dosing in neonates and infants.

Main Methods:

• Utilized a training and test dataset of vancomycin concentrations in neonates/infants.
• Calculated population predictions from published pharmacokinetic models.
• Employed clustering and genetic algorithms for subgroup identification and model selection.
• Developed classification trees to predict the best-performing model for individual patients.

Main Results:

• The single best-performing model showed limited predictive performance (P20: 26.2-42.6%) in the test dataset.
• The proposed subgroup identification approach achieved improved predictive performance (P20: 44.1-45.5%).
• Percentage of predictions within 60% (P60) was comparable between approaches.

Conclusions:

• Subgroup identification-based model selection holds potential to improve precision dosing accuracy.
• This approach offers a promising alternative to the single best-fit model strategy in pediatric vancomycin therapy.
• Further validation is warranted to confirm the clinical utility of this novel method.