PREDICTION OF ACETAMINOPHEN CONCENTRATIONS IN OVERDOSE PATIENTS USINGA BAYESIAN PHARMACOKINETIC MODEL

A pharmacokinetic program using population-based parameter estimates a nd a Bayesian forecasting model was retrospectively evaluated for pred icting acetaminophen serum concentrations in overdose patients. Dynami c disposition factors known to affect acetaminophen disposition (emesi s, activated charcoal, N-acetylcysteine, etc.) were included in the pr ogram. Twenty six patients who reported an acetaminophen ingestion of at least 70 mg/kg within 24 h of presentation to the hospital and had at least one measured acetaminophen concentration were included. Predi ction of initial acetaminophen concentrations using only population-ba sed parameter estimates resulted in a percent mean error (%ME) and per cent mean absolute error (%MAE) of 9.3 and 42.2, respectively. Using o nly the initial concentration as feedback, the Bayesian forecasting mo del accurately predicted the second acetaminophen concentration (%ME = 4.0, %MAE = 23.6). The Bayesian model also accurately predicted all c oncentrations within 8 h of the ingestion (%ME = 10.6, %MAE = 24.0). T he prediction of concentrations between 2 to 4 h and 4 to 4.5 h after ingestion with only population-based parameter estimates resulted in % ME of 17.0 and 13.2, respectively, and %MAE of 36.5 and 35.1, respecti vely. Our data suggests that acetaminophen serum concentrations occurr ing within the first 4.5 h after ingestion can be reliably predicted b y the set of population-based parameter estimates evaluated. Once a si ngle acetaminophen concentration is available, the Bayesian forecastin g model can accurately predict subsequent concentrations within the fi rst 8 h after an acetaminophen ingestion.