Heart failure & cardiomyopathiesCohort study
Machine Learning–Driven Risk Prediction Model in Transthyretin Amyloid Cardiomyopathy
- Multicentre cohort study (prognostic model development and validation)
- internal-external cross-validation
Authors' conclusion
An ML-based time-to-event prediction model demonstrated promising predictive performance and showed improved discrimination compared with established staging systems in patients with ATTR-CM, supporting its potential for individualized prognostication.
Study context
- Population
- Patients with confirmed ATTR-CM in the Swiss Cardiac Amyloidosis Registry (February 2018 to November 2025) and the Medical University of Vienna Cardiac Amyloidosis Registry (July 2014 to February 2025)
- Participants
- 850 patients · 3 cohorts
- Exposure
- Random survival forest machine learning (ML) model vs National Amyloidosis Centre and Mayo Clinic staging systems
- Bern-Swissn = 352
- Other-Swissn = 260
- Viennan = 238
- Outcome
- Composite of:
- All-cause mortality
- Hospitalization for HF
Main finding
As reported in the abstract
Discrimination of the random survival forest model in held-out cohorts
| Group | Harrell C-index | 3-year AUC |
|---|---|---|
| Bern-Swiss | 0.74 (95% CI, 0.69-0.78) | 0.73 (95% CI, 0.66-0.80) |
| Other-Swiss | 0.77 (95% CI, 0.69-0.83) | 0.80 (95% CI, 0.70-0.88) |
| Vienna | 0.72 (95% CI, 0.67-0.77) | 0.75 (95% CI, 0.67-0.83) |
Observational data: associations, not effects.
Further findings
Discrimination vs National Amyloidosis Centre and Mayo Clinic staging systems
- Harrell concordance indices higher with the ML model
- 2% to 10%
- 3-year AUC improvements across cohorts and scoring systems
- 3% to 19%
Calibration
- Generally acceptable across cohorts and time points
Patients receiving disease-modifying therapy (subgroup)
- Model performance remained satisfactory
Original publication
Machine Learning–Driven Risk Prediction Model in Transthyretin Amyloid Cardiomyopathy. JAMA Cardiol. Published 28 August 2026. doi:10.1001/jamacardio.2026.3496
Summarised from the published abstract; every number is checked against it