Valvular & structural heart disease
Parsimonious machine learning model to predict 1-year mortality and procedural futility after transcatheter aortic valve replacement
Authors' conclusion
This externally validated ML model, using 13 routinely available variables, significantly outperforms existing risk scores in predicting 1-year mortality post-TAVR. Its simplicity and generalisability support its potential use in real-world clinical decision-making to identify patients at high risk of procedural futility.
Study context
- Population
- Patients undergoing transcatheter aortic valve replacement (TAVR)discovery (training) cohort and independent external validation cohort · Discovery (training) cohort 1,025 · External validation cohort 270
- Predictors
- ML model using 13 preprocedural variables (age, atrial fibrillation, creatinine, haemoglobin, pulmonary function, frailty markers (Katz Index, poor mobility) and tricuspid regurgitation), selected from 52 clinical and echocardiographic variables
- Predicted outcome
- 1-year all-cause mortality
Methodological detailsretrospective cohort, external validation in an independent cohort, multiobjective evolutionary algorithm for feature selection and model tuning
- Retrospective cohort
- External validation in an independent cohort
- Multiobjective evolutionary algorithm for feature selection and model tuning
Model performance
As reported in the abstract
AUC
ML modeldiscovery cohort
0.81
ML modelexternal validation cohort
0.84
EuroSCORE IIdiscovery cohort
0.61
EuroSCORE IIexternal validation cohort
0.70
FRANCE-2discovery cohort
0.52
FRANCE-2external validation cohort
0.58
TAVI2-SCOREdiscovery cohort
0.56
TAVI2-SCOREexternal validation cohort
0.64
Further findings
| Outcome | NRI (95% CI) |
|---|---|
| Net reclassification improvement (NRI), test setML model vs FRANCE-2 | 0.62 (0.49–0.75) |
| Net reclassification improvement (NRI), test setML model vs TAVI2-SCORE | 0.36 (0.14–0.61) |
Calibration
- Calibration plots showed strong agreement between predicted and observed risks
Misclassification analysis
- Most errors were clustered near the decision threshold, with no evidence of systematic bias
Subgroup and robustness performance
- Performance was consistent across subgroups and robust to temporal and institutional variation
Original publication
Parsimonious machine learning model to predict 1-year mortality and procedural futility after transcatheter aortic valve replacement. Heart. Published 25 July 2025. doi:10.
Composed by AI from the published abstract and checked against it. How Briefs are made →