Prediction of acetaminophen-induced hepatotoxicity in acetylcysteine-treated patients using routine admission biomarkers

Study objective. Some acetaminophen-overdose patients develop hepatotoxicity despite acetylcysteine treatment. Established tools struggle to prospectively identify this cohort. We developed a model using only routine admission biomarkers to identify acetylcysteine-treated patients at highest risk, and compared it with the current benchmark, the alanine aminotransferase x acetaminophen product (ALTxAPAP). Methods. Retrospective cohort of all acetaminophen overdose admissions (ICD-10 T39.1) to three UK hospitals (2008-2024) with alanine aminotransferase (ALT) >1000U/L at admission. We fitted elastic-net logistic models stratified by presentation ALT. The outcome was peak ALT >1,000U/L. Performance was assessed on a 25% held-out test set and benchmarked against ALTxAPAP. Results. Of 4,705 admissions, 119 (2.5%) developed hepatotoxicity. The model used seven routine blood tests, four per stratum: acetaminophen, sodium, potassium and lymphocyte count where presentation ALT was <50U/L; ALT, bilirubin, alkaline phosphatase and lymphocyte count where it was 51-1,000U/L. In the test set (n=1,175) it achieved an area under the curve of 0.93 (95% CI 0.89-0.97) versus 0.82 (0.72-0.91) for ALTxAPAP (paired difference 0.11; 95% CI 0.01-0.22; p=0.03), with higher specificity and a higher positive likelihood ratio at every matched sensitivity. Matched to current ALTxAPAP >1,500 practice (sensitivity 89.7%), specificity was 82.5% versus 62.6% and the positive likelihood ratio 5.1 versus 2.4, more than halving false-positive escalations (171 versus 365 per 1,000 patients). Conclusion. A stratified model using only routine admission biomarkers identifies acetylcysteine-treated patients at highest residual hepatotoxicity risk, outperforming the ALTxAPAP rule across decision thresholds, supporting selection for intensified therapy.