Abstract: Objective To construct a risk prediction model for relapse of chronic drug-induced liver injury (DILI) and explore the correlation between DILI relapse risk and liver fibrosis. Methods We retrospectively collected the clinical data of 1138 patients with chronic DILI hospitalized from January, 2017 to January, 2022, including 154 patients with and 984 without DILI relapse. Based on the results of univariable and multivariable logistic regression analyses, a risk prediction model for DILI relapse was constructed, evaluated for its discrimination and calibration using AUC value and Hosmer-Lemeshow test, and verified with a 200 times 5, 10 and 20 folds cross validation method. Spearman correlation analysis was used to evaluate the correlation between the new model and liver fibrosis, and its diagnostic efficiency for liver fibrosis was assessed by comparison with APRI and FIB-4 using ROC curve. Results The proportions of patients with S0, S1, S2, S3 and S4 liver fibrosis were 1.9%, 13.1%, 42.2%, 27.9% and 14.9% in the relapse group, respectively, as compared with 8.9%, 43.5%, 26.1%, 17.1% and 4.4% in the non- relapse group, respectively, showing severer liver fibrosis in patients with than those without DILI relapse. Multivariable logistic regression analysis identified LSM≥13.7 kPa (OR=4.35, 95%CI: 2.61-7.25, P<0.001), CHE <2500 U/L (OR=5.17, 95%CI: 2.13-12.53, P<0.001), CHE of 2500-5000 U/L (OR=4.07, 95%CI: 2.75-6.01, P<0.001), and AST >2×ULN (OR=2.29, 95%CI: 1.38-3.80, P=0.001) as risk factors for relapse of chronic DILI. The ACLS model constructed based on these non-invasive indicators had an AUC value of 0.803 (95%CI: 0.78-0.83). The results of Hosmer-Lemeshow goodness of fit test (χ2=7.73, P=0.46) and the cross validation tests (average AUC of 0.803) all showed a good stability of the model. Spearman correlation analysis showed that ACLS score was positively correlated with the severity of liver fibrosis (rho=0.530, P<0.001). At the optimal cut-off value of 3 points for diagnosing moderate liver fibrosis, the ACLS model had an AUC value of 0.78 (with specificity of 72.7% and sensitivity of 73.3% ), demonstrating a better efficacy than that of APRI and FIB-4 (P<0.001). At the cut-off value of 6 for severe liver fibrosis, the diagnostic efficacy of the model (AUC=0.83; specificity 75.7%, sensitivity 72.7% ) was still better than that of APRI (P<0.001) but comparable with that of FIB-4 (P=0.38). Conclusion The patients at high risks of chronic DILI relapse have severer liver fibrosis and should be followed up regularly for timely aggressive treatment.

Key words: chronic drug-induced liver injury; relapse; liver fibrosis; risk factors; prediction model