Heart Podcast Predicting 1-year futility of TAVI procedures using machine learning
Oct 21, 2025
Dr. Mehdi Eskandari, an imaging cardiologist at King's College Hospital, shares his expertise on predicting TAVI procedure outcomes using machine learning. He dives into the limitations of current models and highlights how his research utilizes 13 key pre-procedural variables to enhance predictive accuracy. Topics include the impact of frailty, lung function, and more on one-year futility rates. Dr. Eskandari also discusses future directions, including the potential of advanced imaging, as he emphasizes the need for prospective validation and a deeper integration of data science.
AI Snips
Chapters
Transcript
Episode notes
TAVI Eligibility Has Broadened
- TAVI moved from only high-risk/inoperable older patients to class I indication for those 70+ with suitable anatomy.
- This broadened eligibility drives the need to refine selection to avoid futile procedures.
Nonlinear Risks Need Nonlinear Models
- Logistic regression assumes linear, additive risk and misses complex interactions between variables.
- Machine learning captures non-linear relationships and hidden thresholds, improving prediction accuracy.
Guard Against Overfitting
- Use cross-validation, model simplicity, and external validation to avoid overfitting machine learning models.
- Combine these safeguards when developing clinical prediction tools to ensure real-world robustness.