Adaptive Paracetamol Exposure Risk Classification with Cost-Aware Information Acquisition

Paracetamol is widely used worldwide, yet substantial inter-individual pharmacokinetic variability can lead to markedly different exposure following similar dosing. This paper presents a computational framework for adaptive, cost-aware classification of elevated paracetamol exposure in resource-limited settings. A virtual population of 10,000 patients incorporating age, sex, body weight, serum creatinine, and creatinine clearance (CrCl) was generated, and 48-hour concentration profiles were simulated using a one-compartment pharmacokinetic model. Toxic Exposure Time (TET) above 20 mg/L was used as the model-defined exposure endpoint. Sensitivity analysis identified clearance as the dominant driver of TET (PRCC = −0.573, p < 0.001). Information ablation showed modest discrimination from freely observable variables (AUC = 0.712), while inclusion of CrCl increased AUC to 0.731. The full model achieved AUC = 0.728. A two-stage adaptive acquisition strategy, using additional CrCl information only for uncertain cases, achieved AUC = 0.735 while reducing testing expenditure by 88% relative to universal testing. However, sensitivity remained low, indicating that the available feature set is insufficient for reliable identification of elevated exposure. The results demonstrate both the potential and limitations of cost-aware adaptive information acquisition for pharmacokinetic risk classification and motivate future validation with additional clinically informative biomarkers.