Prediction of acetaminophen concentrations in overdose patients using a Bayesian pharmacokinetic model

Abstract

A pharmacokinetic program using population-based parameter estimates and a Bayesian forecasting model was retrospectively evaluated for predicting acetaminophen serum concentrations in overdose patients. Dynamic disposition factors known to affect acetaminophen disposition (emesis, activated charcoal, N-acetylcysteine, etc.) were included in the program. 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. Prediction of initial acetaminophen concentrations using only population-based parameter estimates resulted in a percent mean error (%ME) and percent mean absolute error (%MAE) of 9.3 and 42.2, respectively. Using only the initial concentration as feedback, the Bayesian forecasting model accurately predicted the second acetaminophen concentration (%ME = 4.0, %MAE = 23.6). The Bayesian model also accurately predicted all concentrations within 8 h of the ingestion (%ME = 10.6, %MAE = 24.0). The 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, respectively. Our data suggests that acetaminophen serum concentrations occurring within the first 4.5 h after ingestion can be reliably predicted by the set of population-based parameter estimates evaluated. Once a single acetaminophen concentration is available, the Bayesian forecasting model can accurately predict subsequent concentrations within the first 8 h after an acetaminophen ingestion.