Risk Factors for Postoperative Delirium in Nonintensive Care Unit Patients: Machine Learning Approach.

Postoperative delirium is a frequent and serious complication lacking effective prediction tools for general ward patients. This study aimed to identify postoperative delirium risk factors in non-ICU patients using machine learning. A retrospective analysis of 85,884 surgical patients (2017-2022) from a tertiary hospital was conducted. Postoperative delirium, diagnosed from nursing records, was predicted using 53 variables, with LightGBM showing the best performance. Key risk factors included a higher comorbidity count, age, drain count, sodium levels, and lower albumin levels. Age was the primary predictor in most surgical specialties, while ICU transfer was a key factor in neurosurgery. This AI-based model using electronic health record data offers a foundation for improved postoperative delirium prediction and enhanced patient care.