Continuous Learning in Model-Informed Precision Dosing: A Case Study in Pediatric Dosing of Vancomycin.

Area of Science:

• Pharmacology
• Clinical Pharmacy
• Computational Biology

Background:

• Model-informed precision dosing (MIPD) uses pharmacokinetic (PK) models for individualized drug regimens.
• Selecting appropriate PK models is crucial for accurate dosing and avoiding errors.
• Continuous learning, updating models with new data, is a proposed method to refine MIPD.

Purpose of the Study:

• To investigate the benefits of a continuous learning approach for pediatric vancomycin MIPD.
• To evaluate the performance of existing PK models and newly developed continuous learning models.
• To determine optimal sample sizes for continuous learning models in this setting.

Main Methods:

• Evaluated five existing PK models in 273 pediatric ICU patients.
• Developed and tested two simple PK models using a continuous learning approach with 50-350 patients.
• Assessed prediction error reduction and the impact of sample size on model performance.
• Utilized routine clinical data, including trough samples, to estimate simulated area under the curve (AUC).

Main Results:

• Previously published models performed adequately.
• Continuous learning models reduced prediction error by 2-13% compared to existing models.
• Sample sizes of at least 200 patients were sufficient for capturing vancomycin variability.
• Sparsely sampled routine data enabled reasonable AUC estimation.

Conclusions:

• Continuous learning offers a viable approach for enhancing pediatric vancomycin MIPD.
• This method can improve dose selection accuracy and potentially optimize therapeutic outcomes.
• Foundational data supports the clinical implementation of automated continuous learning for MIPD.