Acetaminophen overdose remains a leading cause of acute liver injury. Some patients develop significant hepatotoxicity despite receiving standard antidotal therapy with acetylcysteine (NAC). Existing prognostic approaches can fail to prospectively identify those at highest residual risk. This study aimed to develop and validate a prognostic model that uses only routine admission biomarkers to identify NAC-treated patients at increased risk of peak ALT >1,000 U/L, and to compare its performance with the established alanine aminotransferase × acetaminophen product (ALTxAPAP) benchmark.
The investigators performed a retrospective cohort study of all acetaminophen overdose admissions (ICD-10 T39.1) to three UK hospitals between 2008 and 2024. The source reports a full dataset of 4,705 admissions. The prespecified outcome was development of hepatotoxicity defined as peak ALT >1,000 U/L. Overall, 119 of 4,705 admissions (2.5%) met this outcome.
The dataset was partitioned to allow internal validation: 25% of admissions were held out as a test set (n=1,175) for independent assessment of model performance.
Models were fitted using elastic-net logistic regression and were stratified by presentation ALT to account for differing baseline liver enzyme levels at admission. The final approach used seven routine blood tests across the two strata.
The investigators report that these variables were derived from routinely obtained laboratory tests at admission; no additional biomarkers or point-of-care tests were required for the model as presented in the source.
Model performance was assessed in the 25% held-out test set (n=1,175). On this independent test set the stratified elastic-net model achieved an area under the receiver operating characteristic curve (AUC) of 0.93 (95% CI 0.89–0.97).
By contrast, the ALTxAPAP benchmark achieved an AUC of 0.82 (95% CI 0.72–0.91) on the same test set. The reported paired AUC difference between the new model and ALTxAPAP was 0.11 (95% CI 0.01–0.22; p=0.03), indicating statistically superior discrimination for the stratified routine-biomarker model.
The authors benchmarked the new model against the ALTxAPAP product across clinically relevant decision thresholds. They report that the routine-biomarker model demonstrated higher specificity and a higher positive likelihood ratio at every matched sensitivity evaluated.
When matched to current ALTxAPAP practice using a threshold corresponding to ALTxAPAP >1,500 (reported sensitivity 89.7%), the stratified model achieved specificity of 82.5% versus 62.6% for ALTxAPAP, and a positive likelihood ratio of 5.1 versus 2.4. This improvement translated to fewer false-positive escalations, reported in the source as 171 versus 365 per 1,000 patients.
Within the limitations of a retrospective, preprint analysis, the findings suggest a pragmatic approach to risk stratification: a model using only routine admission biomarkers can identify NAC-treated patients at higher residual risk of hepatotoxicity, with improved specificity compared with the ALTxAPAP product. The authors propose this could support selection of patients for intensified monitoring or additional therapeutic interventions to mitigate progression to severe liver injury.
Implementation would rely on routinely available laboratory parameters and stratification by presentation ALT, potentially fitting into existing clinical pathways without the need for specialised assays.
The source is a preprint and has not undergone peer review; the authors explicitly state this caveat. The abstract does not provide detailed prospective external validation beyond the held-out test set, and it does not report implementation or impact data. The full text notes competing interests for some authors: a patent on a single-biomarker liver toxicity point-of-care test held by one author, industry ties related to a macrophage cell therapy product and consulting; other authors declared no competing interests. Ethical approvals were obtained from the Edinburgh Medical School Research Ethics Committee and local data governance approvals were in place. Funding sources declared included the Medical Research Council and the Chief Scientist Office.
In a multicentre retrospective cohort of acetaminophen overdose admissions, a stratified elastic-net model using routine admission biomarkers demonstrated superior discrimination for predicting peak ALT >1,000 U/L in acetylcysteine-treated patients compared with the ALTxAPAP benchmark. The model achieved an AUC of 0.93 in a held-out test set and reduced false-positive escalations at matched sensitivity. As reported in the source, these results support further evaluation of routine-biomarker risk stratification to guide intensified therapy decisions, but findings should be interpreted in light of preprint status and the need for external prospective validation.