POSTOPERATIVE DELIRIUM prediction using machine learning showed promising performance in patients undergoing cardiac surgery with cardiopulmonary bypass, according to a prospective cohort study of 729 adults.
The research also identified distinct patient subgroups with markedly different delirium burdens, highlighting the clinical heterogeneity of this common postoperative complication.
Researchers analysed perioperative data from 729 adults undergoing cardiac surgery with cardiopulmonary bypass and developed seven machine learning models to predict postoperative delirium.
During the study observation period, 262 patients developed postoperative delirium, representing 35.9% of the cohort.
The predictive models incorporated preoperative, intraoperative, and early postoperative variables, with postoperative factors restricted to measurements obtained before the first documented delirium episode.
In an independent test set, discrimination across the seven models ranged from 0.718–0.861.
Among the evaluated approaches, random forest demonstrated the strongest overall performance.
Random forest additionally showed the greatest net clinical benefit across a broad range of predicted risk thresholds.
Beyond patient-level prediction, investigators performed unsupervised hierarchical clustering to explore clinical heterogeneity within the cohort.
Three data driven patient subgroups emerged, each with substantially different rates of postoperative delirium.
The low incidence subgroup recorded postoperative delirium in 6.4% of patients, compared with 30.3% in the intermediate incidence subgroup and 100.0% in the highest incidence subgroup.
These subgroups displayed distinct perioperative characteristics across several domains, including inflammatory markers, comorbidity burden, cardiovascular stress indicators, frailty, psychological vulnerability, educational attainment, and surgical complexity.
Patients in higher burden subgroups generally exhibited greater inflammatory activity, higher levels of cardiovascular injury markers, and increased prevalence of frailty and hypertension.
The analysis also examined associations between postoperative delirium and healthcare utilisation.
Patients with postoperative delirium experienced longer intensive care unit stays and prolonged hospitalisation compared with those who did not develop delirium.
Multivariable analyses further showed that postoperative delirium, diabetes, male sex, and more complex surgical procedures were independently associated with longer hospital stays.
The findings suggest that postoperative delirium reflects a broad perioperative risk profile shaped by baseline vulnerability, operative stress, and early postoperative instability rather than any single factor.
The researchers concluded that machine learning models integrating information from across the perioperative pathway may offer a useful approach for predicting postoperative delirium after cardiac surgery with cardiopulmonary bypass.
Peng Q et al. Machine-learning prediction of postoperative delirium and characterization of clinical heterogeneity after cardiac surgery with cardiopulmonary bypass: a prospective observational cohort study. Sci Rep. 2026;DOI:10.1038/s41598-026-72612-w.
Featured image: sudok1 on Adobe Stock
Each article is made available under the terms of the Creative Commons Attribution-Non Commercial 4.0 License.
