Valvular & structural heart disease
Machine Learning Identification of Modifiable Predictors of Patient Outcomes After Transcatheter Aortic Valve Replacement
Authors' conclusion
This study identified four potentially modifiable predictors of excellent outcome after TAVR
Study context
- Population
- Patients undergoing transcatheter aortic valve replacement (TAVR), 21 hospitals, January 2016–December 2021
- Participants
- 8,332 TAVR cases
- Outcome
- Excellent outcome: composite of no major complications (in-hospital or at 30 days), post-TAVR length of stay ≤1 day, discharge to home, no readmission and alive at 30 days
- Follow-up
- 30 days
Methodological detailsmachine learning (random forest), 21 hospitals, Recursive feature elimination with cross-validation and Shapley Additive Explanation feature importance; 57 candidate characteristics
- Machine learning (random forest)
- 21 hospitals
- Recursive feature elimination with cross-validation and Shapley Additive Explanation feature importance; 57 candidate characteristics
Main finding
As reported in the abstract
Final random forest model performance (29 predictors)
- AUC
- 0.77
- Sensitivity
- 0.67
- Specificity
- 0.73
- Predictors retained: 15 patient characteristics and 14 care process components
- 29
- Potentially modifiable predictors: type of anaesthesia, direct movement to stepdown unit post-TAVR, time between catheterisation and TAVR, preprocedural length of stay
- 4
Observational data: associations, not effects.
Original publication
Machine Learning Identification of Modifiable Predictors of Patient Outcomes After Transcatheter Aortic Valve Replacement. JACC Adv. Published 16 July 2024. doi:10.
Summarised from the published abstract; every number is checked against it