Perioperative medicine & outcomes
Predicting surgical intensive care unit readmission with machine learning model: Bi-center training and validation
Authors' conclusion
Machine learning models offer greater accuracy and reliability compared to traditional regression methods when predicting SICU readmission.
Study context
- Population
- Adults admitted to the surgical ICU, Chang Gung Memorial Hospital (development: Kaohsiung, July 2020–December 2022; external validation: Linkou, January 2023–August 2023)development cohorts
- Participants
- 982 adults
- Comparison
- Gradient Boosting (GB) vs Published traditional logistic regression methods
- Outcome
- SICU readmission
Methodological detailsmachine learning, bi-centre, external validation
- Machine learning
- Bi-centre
- External validation
Performance
As reported in the abstract
Discrimination for SICU readmission (AUROC), internal validation
| Test or model | AUROC |
|---|---|
| GB | 0.82 (95% CI: 0.70-0.93) |
GB
AUROC0.82 (95% CI: 0.70-0.93)
Further findings
| Outcome | GB |
|---|---|
| Discrimination (AUROC) in the external validation cohort, GB vs three previously published traditional LR methodsExternal validation cohort (Linkou branch) | |
| AUROC | 0.80 |
- Discrimination (AUROC) in the external validation cohort, GB vs three previously published traditional Logistic Regression (LR) methods: Published LR method (first) 0.73 (95% CI 0.63–0.83) · AUROC 0.73
- Discrimination (AUROC) in the external validation cohort, GB vs three previously published traditional LR methods: Published LR method (second) 0.70 (95% CI 0.60–0.79) · AUROC 0.70
- Discrimination (AUROC) in the external validation cohort, GB vs three previously published traditional LR methods: Published LR method (third) 0.65 (95% CI 0.50–0.73) · AUROC 0.65
Important features for surgical intensive care unit (SICU) readmission
- 11 features significantly influenced SICU readmission; top four: central venous catheter usage days, pre-ICU stay duration, blood urea nitrogen, carbapenem usage days
Discrimination (AUROC) in the external validation cohort, GB vs three previously published traditional LR methods
- GB model surpassed the three previously published traditional LR methods
Original publication
Predicting surgical intensive care unit readmission with machine learning model: Bi-center training and validation. J Crit Care. Published 19 February 2026. doi:10.
Composed by AI from the published abstract and checked against it. How Briefs are made →