Model of baseline clinicopathological features predicts non-resolution of drug-induced liver injury at 6 months.

Introduction: Chronicity in drug-induced liver injury (DILI) is assessed at 12 months, leading to a large time gap from its initial presentation. In this study, we developed a model that could predict biochemical non-resolution in DILI (DILI-NR) patients at 6 months using baseline clinicopathological data.

Patients and methods: Cases of DILI with liver biopsies were enrolled between January 2016 and December 2021. BSEP, MDR3, and MRP2 were assessed immunohistochemically. DILI-NR was considered a biochemical non-resolution 6 months after the onset of DILI. A separate cohort of 126 patients was taken as a validation cohort.

Results: DILI-NR was noted in 59/407 patients (14.5%). DILI-NR patients had significantly higher body mass index, lower hemoglobin, more severe disease at the presentation, autoantibody positivity, higher IgG, association with co-morbidities, and were more aged. Pathologically, DILI-NR had increased ductular reaction, duct damage, duct loss, ductular bile plugs, and autoimmune hepatitis-like morphology along with lesser expression of canalicular transporters. On multivariate logistic regression (LR) analysis and XGBoost analysis, BMI, hemoglobin, presence of autoantibodies, disease severity at baseline, and lower expression of any one transporter were associated with DILI-NR (AUROC = 0.92). After calibrating the model on the test cohort, the LR model showed AUROC of 0.89 with an accuracy of 87.3% and precision of 91.5%, confirming the effectiveness of the model.

Conclusion: The model encompassing hemoglobin, BMI, presence of autoantibodies, disease severity, and reduced expression of canalicular proteins at baseline predicts the biochemical non-resolution of DILI at six months.