External Pharmacokinetic Model Evaluation of Caffeine in Critically Ill Neonates With Heart Disease Using Data...

While most population pharmacokinetic (PopPK) models undergo internal validation, few pediatric PopPK models undergo external evaluation. We aimed to externally evaluate a previously developed PopPK model of caffeine in neonates with congenital heart disease using data collected from a pragmatic platform trial.

We prospectively enrolled neonates with congenital heart disease who received caffeine per standard of care for prostaglandin-associated apnea and collected opportunistic or scavenged pharmacokinetic samples in a pragmatic platform trial. The predictive performance of the PopPK model was evaluated using model prediction error (PE), mean prediction error (MPE), mean absolute prediction error (MAPE), proportion of predicted values within 20% (F20) and 30% (F30) of observed values, prediction-corrected visual predictive checks, and normalized prediction distribution errors. Dosing simulations were conducted to match exposures seen in preterm infants.

Fourteen neonates contributed 37 pharmacokinetic samples to the validation cohort. The model demonstrated strong quantitative predictive performance (PE -0.18 mg/L, MPE -1.3%, and MAPE 13.3%, F20 79%, and F30 88%), meeting predefined thresholds. Graphical diagnostics supported model accuracy. Dosing simulations revealed lower caffeine exposures in neonates with congenital heart disease, particularly after cardiopulmonary bypass, than in preterm neonates without heart disease.

External evaluation of a pediatric PopPK model using data from a pragmatic platform trial was successful. Our external evaluation supports broader adoption of pragmatic platform trials for pediatric model development and evaluation. These studies can inform starting doses for future trials, which cannot necessarily be extrapolated from other populations due to variability in drug disposition.