Predicting Ketamine Exposure in Pediatric Status Epilepticus: A Pharmacokinetic-Machine Learning Approach.

Area of Science:

• Pharmacology
• Machine Learning
• Pediatric Neurology

Background:

• Status epilepticus is a critical neurological emergency.
• Ketamine and levetiracetam show promise for treating status epilepticus.
• Accurate early drug exposure estimation is vital for pediatric dosing but challenging due to sampling constraints.

Purpose of the Study:

• To develop a framework using machine learning and simulated pharmacokinetic data.
• To predict early ketamine exposure in pediatric status epilepticus with limited sampling.
• To overcome logistical and ethical challenges of intensive sampling in pediatric emergency settings.

Main Methods:

• Simulated 100,000 concentration-time profiles using a published population pharmacokinetic model.
• Trained machine learning algorithms (LASSO, RF, KNN, GBM) to predict early exposure (AUC0-2h) using two samples per individual.
• Validated machine learning models internally.

Main Results:

• Ensemble models (GBM and RF) demonstrated superior performance without overfitting.
• RF achieved a root mean square error of 0.183 mg h/L and mean absolute error of 0.103 mg h/L.
• GBM achieved a root mean square error of 0.185 mg h/L and mean absolute error of 0.102 mg h/L.

Conclusions:

• Population pharmacokinetic-informed machine learning models can accurately predict early ketamine exposure.
• This framework is feasible for pediatric status epilepticus where sampling is difficult.
• The study highlights a novel approach for optimizing drug dosing in critical pediatric care.