Perioperative medicine & outcomes
Leveraging data-driven machine learning: From explainable risk prediction to hierarchical clustering-based subtypes of postoperative delirium in a prospective non-cardiac surgery cohort
Authors' conclusion
Our study validated the utility of ML models, particularly RF, in predicting POD and identified three novel data-driven subtypes with distinct clinical characteristics.
Study context
- Population
- Patients who had non-cardiac surgery
- Participants
- 1,106 patients
- Comparison
- Six machine learning (ML) algorithms for postoperative delirium (POD) prediction; data-driven POD subtypes compared
- Primary endpoint
- POD prediction performance of ML models
Methodological detailsprediction model development and clustering, machine learning, least absolute shrinkage and selection operator (LASSO) feature selection, SHAP explainability, external validation, hierarchical clustering, External validation in another large tertiary hospital
- Prediction model development and clustering
- Machine learning
- LASSO feature selection
- SHAP explainability
- External validation
- Hierarchical clustering
- External validation in another large tertiary hospital
Primary result
As reported in the abstract
POD prediction, best model (RF)
Random forest (RF) performed best among six ML models; AUC
0.85
95% CI 0.78–0.91
Number of POD risk factors identified
14
Further findings
Days
| Outcome | Subtype 1 | Subtype 3 |
|---|---|---|
| Postoperative hospital stay by POD subtype | 21.5 (IQR 19–28) | 5 (IQR 4–8) |
- Postoperative hospital stay by POD subtype — Subtype 1 longest hospitalization
- p < 0.001
12-month survival by POD subtype
- Significant survival differences: Subtype 2 > Subtype 3 > Subtype 1 (Kaplan-Meier)
- P < 0.001
- Subtype 2 optimal survival (log-rank)
- P < 0.001
Original publication
Leveraging data-driven machine learning: From explainable risk prediction to hierarchical clustering-based subtypes of postoperative delirium in a prospective non-cardiac surgery cohort. J Clin Anesth. Published 18 September 2025. doi:10.
Composed by AI from the published abstract and checked against it. How Briefs are made →