Perioperative medicine & outcomes
Development and validation of a machine learning model to predict postoperative delirium using a nationwide database: A retrospective, observational study
Authors' conclusion
Using extensive Diagnostic Procedure Combination data, we successfully created and validated a machine learning model for predicting postoperative delirium.
Study context
- Population
- Patients aged ≥65 years who underwent GA for surgery, Japanese Diagnosis Procedure Combination inpatient data
- Participants
- 557,990 patients
- Outcome
- Postoperative delirium (newly prescribed antipsychotic drugs or corresponding insurance claim code after the date of surgery)
Methodological detailsmachine learning prediction model development and validation, 10-fold cross-validation, temporal validation, nationwide database
- Machine learning prediction model development and validation
- 10-fold cross-validation
- Temporal validation
- Nationwide database
Main finding
As reported in the abstract
Predictive performance for postoperative delirium (internal validation)
| Group | AUC | Recall | Precision |
|---|---|---|---|
| Light-gradient boosting machine | 0.826 (95% CI 0.822-0.829) | 0.124 (95% CI 0.119-0.129) | 0.659 (95% CI 0.641-0.677) |
Light-gradient boosting machine
AUC0.826 (95% CI 0.822-0.829)
Recall0.124 (95% CI 0.119-0.129)
Precision0.659 (95% CI 0.641-0.677)
Observational data: associations, not effects.
Further findings
Temporal validation performance (2019.01-2019.10)
- AUC, temporal validation
- 0.815 [95% CI: 0.811-0.818]
- Specificity at a sensitivity of 0.80
- 0.672 (95% CI: 0.670-0.674])
- NPV at a sensitivity of 0.80
- 0.975 (95% CI: 0.974-0.975)
- PPV at a sensitivity of 0.80
- 0.176 (95% CI: 0.176-0.179)
- Performance sustained in the temporal validation
Original publication
Development and validation of a machine learning model to predict postoperative delirium using a nationwide database: A retrospective, observational study. J Clin Anesth. Published 27 April 2024. doi:10.
Composed by AI from the published abstract and checked against it. How Briefs are made →