Abstract
BACKGROUND: Elective surgeries for older adults are increasing. Machine learning could enhance risk assessment, influencing surgical planning and postoperative care. Preoperative cognitive assessment may facilitate early detection and management of postoperative delirium (POD). OBJECTIVE: This study aims to assess machine learning models' predictive ability for POD, focusing on the added predictive value of the neuropsychological assessments before elective surgery. METHODS: This retrospective cohort study analyzed data from the multicenter PAWEL (Patient safety, Efficiency and Life quality in elective surgery) and PAWEL-R (risk) studies, encompassing older patients (≥70 y) undergoing elective surgeries from July 2017 to April 2019. A total of 1624 patients (52.3% male, N=850; age: mean 77.9, SD 4.9 years) were included, with a POD diagnosis made before discharge. Sociodemographic, clinical, surgical, and neuropsychological features were collected pre- and intraoperatively by care providers. Machine learning models' performance was evaluated using the area under the receiver operating characteristic curve (AUC), with permutation testing for significance, and Shapley Additive Explanations to identify effective neuropsychological assessments. RESULTS: Predicting POD before surgery with a random forest model achieved an AUC of 0.760. Incorporating all pre- and intraoperative features into the model yielded a slightly higher AUC of 0.783, with no statistically significant difference observed (P=.24). While cognitive factors alone were not strong predictors (AUC=0.617), specific tests within neuropsychological assessments, such as the Montreal Cognitive Assessment and Trail Making Tests, showed high feature attribution and played a crucial role in further enhancing prediction before surgery. CONCLUSIONS: Preoperative risk prediction for POD can increase risk awareness in presurgical assessment and improve perioperative management in older patients at a high risk for delirium.