Model Prediction of Infection Risk After Glucocorticoid Treatment for Severe Drug-Induced Liver Injury: Globulin is a Core Predictive Factor

Jun Ling,1,* Yiwen Xv,1,* Sa Lv,1 Xiaogang Hao,2 Weiwei Chen,3 Dongze Li,1 Wanshu Liu,1 Zhengsheng Zou,1 Bing Zhu,1 Shaoli You1 1Hepatology Department, The Fifth Medical Center of Chinese PLA General Hospital, Beijing, People’s Republic of China; 2Inpatient and Medical Record Management Department, The Fifth Medical Center of Chinese PLA General Hospital, Beijing, People’s Republic of China; 3Infectious Disease Department, The Fifth Medical Center of Chinese PLA General Hospital, Beijing, People’s Republic of China*These authors contributed equally to this workCorrespondence: Bing Zhu, The Fifth Medical Center of Chinese PLA General Hospital, No. 100, Xisi Huanzhong Road, Beijing, 10039, People’s Republic of China, Tel +86 138 1163 3607, Email [email protected] Shaoli You, The Fifth Medical Center of Chinese PLA General Hospital, No. 100, Xisi Huanzhong Road, Beijing, 10039, People’s Republic of China, Tel +86 136 5118 9002, Email [email protected]: Glucocorticoids are frequently administered in cases of severe drug-induced liver injury (DILI) to promote patient recovery and shorten hospitalization duration. However, their use is associated with an increased risk of infection. This study developed a predictive model for infection after glucocorticoid therapy in patients with DILI.Patients and Methods: We retrospectively analyzed patients with severe DILI treated with glucocorticoids at the Fifth Medical Center of the Chinese People’s Liberation Army between 2017 and 2024. We constructed and interpreted eight machine learning models: random forest, support vector machine, generalized linear model, gradient boosting machine, least absolute shrinkage and selection operator, XGBoost, K-nearest neighbor classification, and artificial neural network. Decision curve analysis, calibration curves, receiver operating characteristics (ROC), and Shapley Additive Explanations model scores were used to interpret the optimal model.Results: Among the eight models, the gradient boosting machine showed the best results (area under the ROC curve: 0.981 and 0.928 for the validation and test sets, respectively) and had the smallest residuals. Decision curve analysis and calibration curves confirmed the model’s strong clinical prediction performance. Globulin (GLO) was a key variable in the models, with significantly low levels in infected patients compared with those in the control group (p < 0.001). Patients with pre-treatment GLO levels below 20 g/L had a higher infection rate (41.1%), while those with post-treatment GLO levels below 21.5 g/L exhibited an even greater infection rate (82.3%).Conclusion: Our early warning model for the prediction of infection is valuable for guiding hormonal therapy for severe DILI. Monitoring changes in GLO levels may provide a simple and effective clinical monitoring tool for preventing infection development.Keywords: drug safety, hepatoprotective, liver damage, patient recovery