Development and validation of a neural network survival prediction model for ischemic heart disease
Authors' conclusion
More complex and feature-rich machine learning models can better predict all-cause mortality in ischemic heart disease and may be used by clinicians and patients to inform and guide treatment and management.
Study context
- Population
- Patients with coronary angiography-confirmed coronary atherosclerosis, 2006–2016, Danish EHR data; Icelandic EHR data for external testtest set n = 5000
- Participants
- 39,746 patients
- Comparison
- PMHnet vs GRACE2.0 risk score and less feature-rich neural network models
- Outcome
- All-cause mortality prediction (tdAUC, Brier score)
- Follow-up
- Until 2019 or up to 5 years
Methodological detailsneural network-based discrete-time survival models, hold-out test set, external validation (Icelandic EHR data), SHAP analysis
- Neural network-based discrete-time survival models
- Hold-out test set
- External validation (Icelandic EHR data)
- SHAP analysis
Performance
As reported in the abstract
tdAUC of PMHnet on test set
| Test or model | Value | Time-dependent AUC (tdAUC) at six months (case count = 196) | tdAUC at one year (cc = 261) | tdAUC at three years (cc = 395) | tdAUC at five years (cc = 763) |
|---|---|---|---|---|---|
| PMHnet | 0.88 (CI 0.86–0.90) | 0.88 [ 0.86-0.90] | 0.88 [0.86-0.90] | 0.84 [0.82-0.86] | 0.82 [0.80-0.84] |
PMHnet
Value0.88 (CI 0.86–0.90)
Time-dependent AUC (tdAUC) at six months (case count = 196)0.88 [ 0.86-0.90]
tdAUC at one year (cc = 261)0.88 [0.86-0.90]
tdAUC at three years (cc = 395)0.84 [0.82-0.86]
tdAUC at five years (cc = 763)0.82 [0.80-0.84]
Further findings
External validation (Icelandic data)
- PMHnet showed similar performance in the Icelandic data
Original publication
Development and validation of a neural network survival prediction model for ischemic heart disease. Cardiovasc Diabetol. Published 28 January 2026. doi:10.
Composed by AI from the published abstract and checked against it. How Briefs are made →