Research on machine learning model-assisted screening of high-risk tumor surgery patients and the effectiveness of pre-admission management

GONG Yuhong; SHANG Nan; HE Yasha; ZHANG Shaoguo; WANG Wentian; YANG Zhuang; YANG Lulu; LIU Yanmei; GAO Hongmei; ZHAO Jingyuan; ZHOU Peibin

doi:10.12173/j.issn.1004-5511.202503188

Abstract：Objective This study aimed to identify high-risk factors associated with high resource consumption during hospitalization among tumor patients using machine learning(ML)models,and to screen high-risk patients for pre-admission management to evaluate its impact on length of stay and hospitalization costs.Methods The date were retrospectively collected from inpatients treated at First Hospital of Shanxi Medical University between April 2021 and October 2022.Lasso regression was applied for feature selection,and multiple ML models were constructed to identify key factors influencing length of stay and hospitalization costs.SHAP values were used to interpret model results.Based on the modeling results,a risk factor screening and pre-admission management program was established between November and December 2022.The program was implemented from January to June 2023,and data were collected from patients who received or did not receive pre-admission management.Propensity score matching(PSM)was used to control baseline differences between groups,and causal forest(CF)analysis was performed to assess the effects of pre-admission intervention on hospitalization duration and costs.Results A total of 5,211 patients were included.The CatBoost model had the best results among the nine ML models constructed,and the prediction results showed that tumor type,preoperative waiting time,and age were the main high-risk factors affecting hospital resource utilization.Based on the model,698 patients received pre-admission management.After PSM,563 matched pairs were analyzed.CF analysis showed that the pre-admission group had a mean reduction in length of stay of 3.004[95％CI(-3.334 to-2.675)]days compared with the control group and a mean reduction in hospitalization costs of 1,473.124[95％CI(-2,166.09 to-780.16)]CNY,indicating that pre-admission management effectively improved hospital resource utilization efficiency.Conclusion Implementing pre-admission management for high-risk patients identified by ML models can significantly reduce length of stay and hospitalization costs for tumor patients,and may positively contribute to optimizing hospital workflows and resource allocation.