Arrhythmias & electrophysiologyDiagnostic or prognostic study
Machine Learning-Based Prediction of Death and Hospitalization in Patients With Implantable Cardioverter Defibrillators
- Prognostic model development and external validation study
- machine learning (random forest, RF-SLAM)
- nationwide training cohort
- external validation in a separate nonveteran clinical registry
Median follow-up 3.3 years (training) and 3.6 years (validation)
Authors' conclusion
The RF-SLAM algorithm accurately predicted all-cause mortality and death/HF hospitalization at 3 months and 1 year during the first 5 years of device implant, demonstrating good internal and external validity.
Study context
- Population
- Patients with implantable cardioverter-defibrillators (ICDs) with and without cardiac resynchronization therapy (CRT), first 5 years after device implant
- Participants
- 12,043 vs 1,394
- Comparison
- Training vs Validation
- Trainingn = 12,043
- Validationn = 1,394
- Outcome
- All-cause mortality and composite of death/HF hospitalization
- Follow-up
- 3 months and 1 year
Performance
As reported in the abstract
Discrimination of the most accurate models (baseline demographics + time-varying ICD data)
AUC, death at 3 months
0.91 (95% CI: 0.87-0.94)
AUC, death at 1 year
0.80 (95% CI: 0.78-0.82)
AUC, death/heart failure (HF) hospitalization at 3 months
0.81 (95% CI: 0.79-0.83)
AUC, death/HF hospitalization at 1 year
0.71 (95% CI: 0.70-0.72)
Further findings
External validation
- High discrimination and good calibration in the validation cohort
Predictor importance
- Time-varying physiologic ICD data, especially daily physical activity, had substantial importance in predicting outcomes
Original publication
Machine Learning-Based Prediction of Death and Hospitalization in Patients With Implantable Cardioverter Defibrillators. J Am Coll Cardiol. Published 20 November 2024. doi:10.1016/j.jacc.2024.09.006
Summarised from the published abstract; every number is checked against it