Objectives: To evaluate Hy's law, DrILTox ALF Score, Robles-Diaz Model, and a new logistic regression model for predicting acute liver failure (ALF) in Chinese patients with drug-induced liver injury (DILI).
Methods: We conducted a retrospective study among 514 hospitalized DILI patients from 2011 to 2020. Logistic regression analysis was used to develop a predictive model for ALF. The sensitivity, specificity, positive predictive value (PPV), and negative predictive value (NPV) of these models were compared. Another 304 DILI patients were used for external validation.
Outcomes: Twenty-six of 514 DILI patients progressed to ALF. Among these models, Hy's law had 84.6% sensitivity, 59.8% specificity, 10.1% PPV, and 98.6% NPV. DrILTox ALF Score had 92.3% sensitivity, 51.8% specificity, 9.3% PPV, and 99.2% NPV, while Robles-Diaz Model had 50.0% sensitivity, 77.7% specificity, 10.7% PPV, and 96.7% NPV. The logistic regression model described as P = 1/(1+e(1.643 - 0.006* × TBIL (μmol/L) -- 1.302* × INR + 0.095* × ALB (g/L))) had 88.5% sensitivity, 73.1% specificity, 16.3% PPV, and 99.1% NPV at the cut-off of 0.04778 and kept 94.4% sensitivity, 66.8% specificity, 15.2% PPV, and 99.5% NPV in external validation.
Conclusions: The logistic regression model provided superior performance than Hy's law, DrILTox ALF Score, and Robles-Diaz Model for predicting DILI -related ALF.
Keywords: Acute liver failure; DrILTox ALF Score; Drug-induced liver injury; Hy’s law; Predictive model; Robles-Diaz Model.