Acute cardiac care & resuscitation
Comparative analysis of machine learning vs. traditional modeling approaches for predicting in-hospital mortality after cardiac surgery: temporal and spatial external validation based on a nationwide cardiac surgery registry
Authors' conclusion
ML provided only marginal improvements over traditional modelling approaches in predicting cardiac surgery mortality with routine preoperative variables
Study context
- Population
- Adult cardiac surgery cases, Chinese Cardiac Surgery Registry, 2013–2018
- Participants
- 168,565 cases
- Comparison
- Machine learning vs logistic regression models
- Outcome
- Predictive performance (discrimination and calibration) for in-hospital mortality
Methodological detailstemporal and spatial external validation, nationwide registry, Temporal split: 2013-2017 training, 2018 testing; spatial split: 83 centres training, 22 testing
- Temporal and spatial external validation
- Nationwide registry
- Temporal split: 2013-2017 training, 2018 testing; spatial split: 83 centres training, 22 testing
Performance
As reported in the abstract
Discrimination (AUC) for in-hospital mortality
| Test or model | AUC, temporal testing set | AUC, spatial testing set |
|---|---|---|
| Machine learning (ML) | 0.797 (95% CI 0.779-0.815) | 0.732 (95% CI 0.710-0.754) |
| Logistic regression (LR) | 0.791 (95% CI 0.775-0.808) | 0.713 (95% CI 0.691-0.737) |
AUC, temporal testing set
Machine learning (ML)0.797 (95% CI 0.779-0.815)
Logistic regression (LR)0.791 (95% CI 0.775-0.808)
AUC, spatial testing set
Machine learning (ML)0.732 (95% CI 0.710-0.754)
Logistic regression (LR)0.713 (95% CI 0.691-0.737)
- ML vs LR: P = 0.12 — temporal testing set (n = 32 184)
- ML vs LR: P = 0.002 — spatial testing set (n = 28 323)
Further findings
Overall in-hospital mortality
- Overall in-hospital mortality
- 1.9%
Calibration
- Most ML and LR models were significantly miscalibrated
Feature selection methods
- Varying feature selection methods had relatively smaller effects on ML models
Original publication
Comparative analysis of machine learning vs. traditional modeling approaches for predicting in-hospital mortality after cardiac surgery: temporal and spatial external validation based on a nationwide cardiac surgery registry. Eur Heart J Qual Care Clin Outcomes. Published 22 May 2023. doi:10.
Summarised from the published abstract; every number is checked against it