Towards Optimal Valve Prescription for Transcatheter Aortic Valve Replacement (TAVR) Surgery: A Machine Learning Approach

TL;DR

This work introduces a data-driven prescriptive framework to optimize transcatheter heart valve (THV) selection in TAVR with the aim of reducing permanent pacemaker implantation (PPI). It fuses a multimodal US-Greece dataset (demographics, CT, echocardiography) and applies counterfactual, doubly robust estimation with Optimal Policy Trees to yield an interpretable leaf-based policy. The proposed method achieves substantial PPI reductions in both internal (26%) and external (16%) cohorts and demonstrates generalizability via an external Greek validation and a publicly available decision tool. The study provides actionable clinical insights, supports proactive valve planning, and highlights limitations including single-center training and the need for randomized validation. Overall, it delivers the first unified, interpretable, personalized THV prescription strategy with demonstrated potential operational and patient-level impact.

Abstract

Transcatheter Aortic Valve Replacement (TAVR) has emerged as a minimally invasive treatment option for patients with severe aortic stenosis, a life-threatening cardiovascular condition. Multiple transcatheter heart valves (THV) have been approved for use in TAVR, but current guidelines regarding valve type prescription remain an active topic of debate. We propose a data-driven clinical support tool to identify the optimal valve type with the objective of minimizing the risk of permanent pacemaker implantation (PPI), a predominant postoperative complication. We synthesize a novel dataset that combines U.S. and Greek patient populations and integrates three distinct data sources (patient demographics, computed tomography scans, echocardiograms) while harmonizing differences in each country's record system. We introduce a leaf-level analysis to leverage population heterogeneity and avoid benchmarking against uncertain counterfactual risk estimates. The final prescriptive model shows a reduction in PPI rates of 26% and 16% compared with the current standard of care in our internal U.S. population and external Greek validation cohort, respectively. To the best of our knowledge, this work represents the first unified, personalized prescription strategy for THV selection in TAVR.

Figures (11)

• Figure 1: Proposed Prescriptive Policy: Our proposed prescriptive policy as a decision tree. The policy splits the patients on a variety of variables extracted from computed tomography scans and echocardiograms to arrive at a partition of six patient groups represented by nodes 4, 5, 6, 8, 10, and 11 and prescribes each group one of Edwards Sapien or Medtronic Evolut.
• Figure 2: Sapien Feature Importance: Feature importances of the Edwards Sapien valve outcome estimators. Importance scores are averaged over the associated train and test set models. Note that the absolute importance of any one variable has no interpretation. Instead, it is the relative difference in feature importances that offer insight into which variables are used by the model.
• Figure 3: The standardized distributions of patients prescribed the Edwards Sapien and Medtronic Evolut platforms with tails cut-off after three standard deviations for each of the ten most important features for the propensity estimators. Feature importance scores are averaged over the train and test set models.
• Figure 4: The percent improvement in pacemaker rate induced by our proposed policy over 1000 bootstrap test sets as evaluated by our node analysis technique and by the counterfactual estimators. The means are denoted by white lines. The 95% confidence intervals of the techniques are (0.036, 0.377) and (0.0646, 0.275) respectively.
• Figure 5: Visualization of the online algorithm user interface