# Novel integrative models to predict the severity of inflammation and fibrosis in patients with drug-induced liver injury

## TL;DR

This study develops new models to predict liver inflammation and fibrosis severity in drug-induced liver injury patients, outperforming existing non-invasive methods.

## Contribution

The backward stepwise regression model is shown to be more effective than current non-invasive biomarkers for predicting DILI severity.

## Key findings

- The backward stepwise regression model achieved an AUC of 0.856 for predicting ≥G2 inflammation.
- The model also achieved an AUC of 0.889 for predicting ≥S2 fibrosis, outperforming existing indices.
- Calibration curves and decision curve analysis confirmed the model's high effectiveness in predicting DILI severity.

## Abstract

Drug-induced liver injury (DILI) is becoming a worldwide emerging problem. However, few studies focus on the diagnostic performance of non-invasive markers in DILI. This study aims to develop novel integrative models to identify DILI-associated liver inflammation and fibrosis, and compare the predictive values with previously developed indexes.

A total of 72 DILI patients diagnosed as DILI through liver biopsy were enrolled in this study. Patients were divided into absent-mild (S0–S1, G0–G1) group and moderate–severe (S2–S4, G2–G4) group based on the histological severity of inflammation and fibrosis. We used the area under the receiver operating characteristics curves (AUC) to test the model performances. Backward stepwise regression, best subset and logistic regression models were employed for feature selection and model building. Prediction models were presented with nomogram and evaluated by AUC, Brier score, calibration curves and decision curve analysis (DCA).

For diagnosing moderate–severe inflammation and fibrosis, we calculated the AUC of gamma-glutamyl transpeptidase-to-platelet ratio (GPR), aspartate aminotransferase-to-platelet ratio index (APRI), fibrosis-4 index (FIB-4) and fibrosis-5 index (FIB-5), which were 0.708 and 0.676, 0.778 and 0.667, 0.822 and 0.742, 0.831 and 0.808, respectively. Then, backward stepwise regression, best subset and logistic regression models were conducted for predicting significant liver inflammation and fibrosis. For the prediction of ≥G2 inflammation grade, the AUC was 0.856, 0.822, 0.755, and for the prediction of ≥S2 fibrosis grade, the AUC was 0.889, 0.889, 0.826. Through Brier score, calibration curves and DCA, it was further demonstrated that backward stepwise regression model was highly effective to predict both moderate–severe inflammation and fibrosis for DILI.

The backward stepwise regression model we proposed in this study is more suitable than the existing non-invasive biomarkers and can be conveniently used in the individualized diagnosis of DILI-related liver inflammation and fibrosis.