Abstract: Objective To construct prediction models of hospitalization risk and high medical cost for patients with hepatitis B cirrhosis, in order to improve the scientific nature of patient management and clinical decision-making. Methods The study samples were collected from patients diagnosed as hepatitis B cirrhosis in an infectious disease hospital in Guangzhou. The data were divided into a training set (70%) and a validation set (30%) by random sampling. In view of the class imbalance, SMOTE method was used to balance the training set, and the prediction models of hospitalization risk and high medical cost in the following year were established by the random forest algorithm combined with logistic regression. The models were then validated with class-balanced and validation datasets to evaluate its predictive effectiveness. Results This study included 7 022 patients with hepatitis B cirrhosis, of whom 602 (8.57%) were hospitalization in the following year, and 179 (2.55%) had high medical expenses in the following year. Random forest algorithm and logistic regression prediction models showed that hospitalization, abnormal total protein, and low albumin in the current year were risk factors for hospitalization and high cost in the following year (all OR 95% CI >1). Glutamic-pyruvic transaminase and entecavir use were protective factors for hospitalization and high costs in the following year (both OR 95% CI <1). In the class-balanced dataset, the AUC for the following year hospitalization risk prediction model was 0.944, and the AUC for the high medical cost prediction model was 0.962. In the validation dataset, the AUC of the following year hospitalization risk prediction model was 0.787, and the AUC of the high medical cost prediction model was 0.857, indicating good predictive performance. Conclusion The predictive models constructed in this study showed good performance in predicting the risk of hospitalization and high medical expenses in the following year of patients with hepatitis B cirrhosis, which are of great value for optimizing patient management, reducing medical costs, and improving the quality of medical services.

Key words: Hepatitis B cirrhosis, Random forest, Hospitalization risk, High medical cost