Machine Learning‐Based Prediction of Life‐Threatening Complications During Hemodialysis in Hospitalized Patients With Poor General Conditions

Background Patients undergoing hemodialysis (HD) face a significantly elevated risk of cardiovascular mortality, with sudden events during treatment posing a critical threat to survival. These risks are particularly pronounced in high‐risk populations, such as patients recovering from cardiovascular surgery or those being treated for sepsis. Therefore, the development of effective preventive strategies is essential for improving patient outcomes. This study aimed to develop a machine learning model that uses pretreatment patient characteristics to predict sudden adverse events during HD and within 24 h after treatment in high‐risk inpatients at acute care hospitals. Methods His retrospective study analyzed data from 739 patients who underwent HD at Hirosaki University Hospital between 2018 and 2021. Sudden events were defined as fatal arrhythmia, refractory intradialytic hypotension, or respiratory arrest. A logistic regression model was constructed using backward stepwise selection from 51 patient characteristics (demographic data, clinical parameters, laboratory data, and HD‐related information). Results Among the 739 patients, 17 (2.3%) experienced sudden events. The model identified 23 pre‐HD covariates and achieved an area under the receiver operating characteristic curve (AUC) of 0.889. Key covariates included emergency hospitalization (present in 71% of patients with sudden events), recent surgery (76%), shorter HD history, elevated pre‐HD heart rate, lower serum albumin levels, and higher C‐reactive protein concentrations. Conclusions Our model enables the early identification of high‐risk inpatients receiving hemodialysis using pre‐dialysis data, thereby supporting timely clinical interventions, optimized resource allocation, and improved patient safety.