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

Phevos Paschalidis; Vasiliki Stoumpou; Lisa Everest; Yu Ma; Talhat Azemi; Jawad Haider; Steven Zweibel; Eleftherios M. Protopapas

cs.LGDec 9, 2025

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

Authors: Phevos Paschalidis, Vasiliki Stoumpou, Lisa Everest, Yu Ma, Talhat Azemi, Jawad Haider, Steven Zweibel, Eleftherios M. Protopapas, +6 more

Organizations: J. Paulson School of Engineering and Applied Research, Harvard University, Cambridge, MA, USA. · Operations Research Center, Massachusetts Institute of Technology, Cambridge, MA, USA. · Heart & Vascular Institute, Hartford HealthCare, Hartford, CT, USA. · Athens Heart Surgery Institute, Iaso Children’s Hospital, Athens, Greece. · Hartford HealthCare Research Institute, Hartford HealthCare, Hartford, CT, USA. · Section of Cardiovascular Medicine, Yale School of Medicine, New Haven, CT, USA.

Abstract

Transcatheter Aortic Valve Replacement (TAVR) has emerged as a prominent, minimally invasive treatment 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 a topic of ongoing debate within the medical community. 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, combining U.S. and Greek patient populations, that integrates data from three distinct sources (patient demographics, computed tomography scans, echocardiograms) while harmonizing the different encoding processes specific to each country's record system. We propose leaf-level analysis to leverage the heterogeneity of the patient populations and avoid benchmarking against uncertain counterfactual risk estimates. The final prescriptive model shows a reduction in PPI rates of 26% and 16% compared to the current standard of care in our internal U.S. population and external, Greek validation set, respectively. To the best of our knowledge, this work represents the first unified, personalized prescription strategy for THV selection in TAVR.

Figures & tables

Figure 1 : Inclusion criteria detailing all criterions used to define the final participating cohort

Valve	No. Patients	Description	Pacemaker Rate (No.)
Edwards Sapien	72	Edwards Sapien	13.63% (178)
Edwards Sapien 3	1017
Edwards Sapien 3 Ultra	217
Medtronic Evolut Pro	166	Medtronic Evolut	15.22% (72)
Medtronic Evolut Pro Plus	307	Medtronic Evolut	15.22% (72)

Table 1 : The number of patients in the internal cohort that received each type of valve, as well as the pacemaker rate amongst the Edwards Sapien and Medtronic Evolut populations.

Valve	No. Patients	Description	Pacemaker Rate (No.)
Edwards Sapien 3	118	Edwards Sapien	3.39% (4)
Medtronic Evolut Pro	33	Medtronic Evolut	18.18% (6)

Table 2 : The number of patients in the external validation set that received each type of valve, as well as the pacemaker rate amongst the Edwards Sapien and Medtronic Evolut populations.

Figure 2 : 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.

	Node	Estimated Sapien Pacemaker Rate (%)	Estimated Evolut Pacemaker Rate (%)	Historical Sapien Pacemaker Rate (%)	Historical Evolut Pacemaker Rate (%)
Train N=890	4	6.07	17.36	2.5	22.73
	5	28.41	15.14	28.57	13.04
	6	35.36	20.44	37.88	26.67
	8	14.32	6.88	13.43	3.45
	10	15.40	10.88	15.62	10.53
	11	7.12	17.96	6.91	16.38

Table 3 : Performance of Counterfactual Estimation: The estimated pacemaker rate of Edwards Sapien and Medtronic Evolut at each node among the train and test set as conjectured by the corresponding counterfactual estimator compared to the historical pacemaker rate amongst Edwards Sapien and Medtronic Evolut patients at the node.

Figure 3 : 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 4 : 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.

	Node	Prescribed Treatment	No. Implanted Edwards Sapiens	Historical Sapien Pacemaker Rate (%)	No. Implanted Medtronic Evoluts	Historical Evolut Pacemaker Rate (%)
Total N=1779	4	Edwards Sapien	104	14.42	41	29.27
	5	Medtronic Evolut	133	24.81	54	20.37
	6	Medtronic Evolut	142	35.92	29	17.24
	8	Medtronic Evolut	129	9.3	49	4.08
	10	Medtronic Evolut	65	9.23	40	5.0
	11	Edwards Sapien	733	8.32	260	15.38

Table 4 : Historical Outcome Performance: The size and PPI rate of the historical Edwards Sapien and Medtronic Evolut patient implants in (i) the combined train and test set, (ii) the train set, and (iii) the test set for each node, as well as the treatment prescribed by the OPT.

	Historical Pacemaker Rate	Observed Pacemaker Rate under OPT	Percent Improvement
Total	14.05%	10.32%	26.54
Train	13.6%	8.87%	34.74
Test	14.51%	11.39%	21.5

Table 5 : Pacemaker Rate Improvement: The historical pacemaker rate of the total patient population, the train set, and the test set compared to the pacemaker rate observed under the OPT’s prescription strategy.

	Actual Pacemaker Rate	Observed Pacemaker Rate under OPT	Percent Improvement
Train	13.6%	9.29%	31.7
Test	14.51%	12.05%	16.92

Table 6 : Pacemaker Rate Improvement (CE): The historical pacemaker rate of the training and tests set compared to the pacemaker rate of the policy prescribed by the OPT as evaluated by the counterfactual estimators.

Figure 5 : 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.

	Node	Prescribed Treatment	No. Implanted Edwards Sapiens	Historical Sapien Pacemaker Rate (%)	No. Implanted Medtronic Evoluts	Historical Evolut Pacemaker Rate (%)
Greek N=151	4	Edwards Sapien	5	0.00	0	0.00
	5	Medtronic Evolut	5	0.00	4	0.00
	6	Medtronic Evolut	4	25.00	3	3.33
	8	Medtronic Evolut	12	0.00	2	0.00
	10	Medtronic Evolut	8	0.00	3	0.00
	11	Edwards Sapien	84	3.57	21	19.05

Table 7 : External Cohort Performance: The size and PPI rate of the actual Edwards Sapien and Medtronic Evolut patient implants in the external validation set as well as the treatment prescribed by the OPT.

	Actual Pacemaker Rate	Observed Pacemaker Rate under OPT	Percent Improvement
Greek	6.62%	5.52%	16.62

Table 8 : External Cohort Improvement: The historical pacemaker rate of the training, and the test set as well as the pacemaker rate of the policy prescribed by the OPT as evaluated by the counterfactual estimators.

Fin1	Fin2	OPT1	OPT2	OPT3	OPT4	OPT5	OPT6	OPT7	OPT8	OPT9	OPT10
0.81	0.72	0.75	0.72	0.85	0.71	0.75	0.74	0.74	0.72	0.78	0.74

Table 9 : Concordance between our chosen model and the twelve other high-performing OPTs, including the other finalists, Fin1 and Fin2.

Fin1	Fin2	OPT1	OPT2	OPT3	OPT4	OPT5	OPT6	OPT7	OPT8	OPT9	OPT10
16.36%	16.96%	6.33%	5.49%	6.01%	3.52%	16.46%	11.35%	9.14%	16.86%	10.39%	10.8%

Table 10 : Estimated improvement (as compared to historical policy) of other trees on their respective test sets. Recall that the estimated improvement of our chosen model was 21.5%.

Figure 6 : Visualization of the online algorithm user interface

Appendix figures & tables19 assets

Supplementary material from the paper’s appendix.

Appendix

Characteristic	Train, No. (%)	Test, No. (%)
Sex
Female	435 (48.88%)	416 (46.79%)
Male	455 (51.12%)	473 (53.21%)
Weight (kg), median (IQR)	77.1 (66.0-88.98)	77.0 (66.0-90.9)
Porcelain Aorta
False	875 (98.31%)	877 (98.65%)

Appendix

Table A1 : A summary of the train and test patient cohorts. For quantitative variables, the median and inter-quartile range of both sets is presented. For categorical variables, the number and percentage of patients assigned to each category.

Characteristic	Medtronic Evolut	Edwards Sapien
Sex, No. (%)
Female	275 (58.14%)	576 (44.1%)
Male	198 (41.86%)	730 (55.9%)
Weight (kg), median (IQR)	75.0 (64.0-86.36)	78.0 (66.5-91.0)
Porcelain Aorta, No. (%)
False	466 (98.52%)	1286 (98.47%)

Appendix

Table A2 : A summary of the Medtronic Evolut and Edwards Sapien patient cohorts. For quantitative variables, the median and inter-quartile range of both sets is presented. For categorical variables, the number and percentage of patients assigned to each category.

Characteristic	Hospital A	Hospital B
Age, median (IQR)	82.0 (76.0-87.0)	81.0 (74.3-85.0)
Sex, No. (%)
Female	851 (47.84%)	71 (40.8%)
Male	928 (52.16%)	103 (59.2%)
Weight (kg), median (IQR)	77.0 (66.0-90.0)	78.0 (68.0-89.0)
Conduction Defect, No. (%)

Appendix

Table A3 : A comparison between the patient cohorts of U.S. Hopsital and Greek Hospital (used for external validation) with respect to the patient demographics and the variables used by the proposed OPT. For quantitative variables, the median and inter-quartile range of both sets is presented. For categorical variables, the number and percentage of patients assigned to each category.

Characteristic	% Missing
Sex	0.0
Weight (kg)	0.06
Porcelain Aorta	0.0
Prior Peripheral Arterial Disease	0.0
Transient Ischemic Attacks	0.0
Prior Myocardial Infarction	0.17

Appendix

Table A4 : A summary of the missingness for each feature present in our dataset.

Figure B1 : Calibration Curves for Training Set Outcome Estimators. We calibrate the train set outcome estimators for Edwards Sapien and Medtronic Evolut on patients in the testing cohort who received Sapien and Evolut, respectively. We leave out 2.5% outliers on either side of the predicted probability.

Figure B2 : Calibration Curves for Testing Set Outcome Estimators. We calibrate the test set outcome estimators for Edwards Sapien and Medtronic Evolut on patients in the training cohort who received Sapien and Evolut, respectively. We leave out 2.5% outliers on either side of the predicted probability.

	Propensity	Outcome- Medtronic Evolut	Outcome- Edwards Sapien
Train Estimator	0.6149	0.4674	0.5942
Test Estimator	0.6823	0.6288	0.5970

Appendix

Table B1 : AUC Performance: The AUC-ROC of the propensity and outcome models used to estimate the counterfactual pacemaker rate. The estimations across the train and test sets are largely equivalent/close to each other.

Figure B3 : Evolut Feature Importance: Feature importances of the Medtronic Evolut valve outcome estimators. Importance score are averaged over the associated train and test set models.

Age Range	Cohort Size	Percent Improvement (CE)
$<75$	195	18.78
$76-80$	195	19.89
$81-85$	216	20.13
$86-90$	193	9.77
$91+$	95	12.55

Appendix

Table B2 : Subgroup Analysis. A summary of the estimate pacemaker rate improvement of the proposed policy as evaluated by the counterfactual estimator for patients of different age ranges. We observe consistent improvement, with the model offering increased benefit to younger patients.

Imputation Method	Percent Improvement	Percent Improvement (CE)
Mean	21.5	16.92
Random Forest	21.58	16.57
Conditional Mean	21.50	16.92

Appendix

Table C1 : Imputation Method Comparison. A summary of the estimate pacemaker rate improvement of the proposed policy for different imputation strategies. For each method, we train an imputation model on the training set and apply it to the test set before evaluating our model using both the node analysis technique and counterfactual estimation (CE). Thus, we investigate the model’s robustness to test sets of varying data quality.

Counterfactual Estimation Method	Percent Improvement	Percent Improvement (CE, 1 model)	Percent Improvement (CE, 2 models)
Single Model	$9.04\pm 0.37$	$11.31\pm 0.52$	$7.79\pm 0.58$
Two Models	$8.61\pm 0.42$	$11.10\pm 0.59$	$7.52\pm 0.56$

Appendix

Table C2 : Average performance and standard error of trees trained on 20 training test splits under different counterfactual estimation approaches, as evaluated by the node analysis and each counterfactual estimation approach.

Figure D1 : Alternative Candidate Model 1: An alternative high performing OPT developed during model training.

	Node	Prescribed Treatment	No. Implanted Edwards Sapiens	Historical Sapien Pacemaker Rate (%)	No. Implanted Medtronic Evoluts	Historical Evolut Pacemaker Rate (%)
Total N=1779	3	Edwards Sapien	155	18.06%	56	23.21%
	4	Medtronic Evolut	224	31.7%	68	22.06%
	5	Edwards Sapien	927	8.52%	349	12.61%
Train N=890	3	Edwards Sapien	78	20.51%	22	27.27%
	4	Medtronic Evolut	106	36.79%	32	25.0%
	5	Edwards Sapien	485	8.04%	167	10.78%

Appendix

Table D1 : Historical Outcome Performance: The size and PPI rate of the historical Edwards Sapien and Medtronic Evolut patient implants in (i) the combined train and test set, (ii) the train set, and (iii) the test set for each node, as well as the treatment prescribed by Alternative Candidate Model 1 (see Figure D1 ).

	Actual Pacemaker Rate	Observed Pacemaker Rate under OPT	Percent Improvement
Total	14.05%	11.88%	15.49
Train	14.16%	12.07%	14.73
Test	13.95%	11.67%	16.36

Appendix

Table D2 : Pacemaker Rate Improvement: The historical pacemaker rate of the total patient population, the train set, and the test set compared to the pacemaker rate observed under the policy prescribed by Alternative Candidate Model 1 (see Figure D1 ).

	Actual Pacemaker Rate	Observed Pacemaker Rate under OPT	Percent Improvement
Train	14.16%	11.71%	17.28
Test	13.95%	11.17%	19.9

Appendix

Table D3 : Pacemaker Rate Improvement (CE): The historical pacemaker rate of the training and test sets compared to the pacemaker rate observed under the policy prescribed by Alternative Candidate Model 1 (see Figure D1 ) as evaluated by the counterfactual estimators.

Figure D2 : Alternative Candidate Model 2: An alternative high performing OPT developed during model training.

	Node	Prescribed Treatment	No. Implanted Edwards Sapiens	Historical Sapien Pacemaker Rate (%)	No. Implanted Medtronic Evoluts	Historical Evolut Pacemaker Rate (%)
Total N=1779	3	Edwards Sapien	86	11.63%	21	23.81%
	5	Medtronic Evolut	189	33.33%	66	13.64%
	6	Edwards Sapien	104	25.0%	37	37.84%
	7	Edwards Sapien	927	8.52%	349	12.61%
Train N=890	3	Edwards Sapien	43	16.28%	10	30.0%
	5	Medtronic Evolut	95	40.0%	34	11.76%

Appendix

Table D4 : Historical Outcome Performance: The size and PPI rate of the historical Edwards Sapien and Medtronic Evolut patient implants in (i) the combined train and test set, (ii) the train set, and (iii) the test set for each node, as well as the treatment prescribed by Alternative Candidate Model 2 (see Figure D2 ).

	Actual Pacemaker Rate	Observed Pacemaker Rate under OPT	Percent Improvement
Total	14.05%	10.75%	23.52
Train	13.82%	9.62%	30.43
Test	14.29%	11.86%	16.96

Appendix

Table D5 : Pacemaker Rate Improvement: The historical pacemaker rate of the total patient population, the train set, and the test set compared to the pacemaker rate observed under the policy prescribed by Alternative Candidate Model 2 (see Figure D2 ).

	Actual Pacemaker Rate	Observed Pacemaker Rate under OPT	Percent Improvement
Train	13.82%	10.08%	27.06
Test	14.29%	12.43%	13.02

Appendix

Table D6 : Pacemaker Rate Improvement (CE): The historical pacemaker rate of the training and test sets compared to the pacemaker rate observed under the policy prescribed by Alternative Candidate Model 2 (see Figure D2 ) as evaluated by the counterfactual estimators.

Explore similar work

TAVI-TEC: An AI-Based Tool for Procedural Planning of Transcatheter Aortic Valve Implantation
Jul 31, 2026Alessandra Zerillo, Stefano Cannata, Diego Bellavia +4Coronary Angiography Ai-Based
Department of Research, IRCCS-ISMETT, Palermo, Italy. · Department for the Treatment and Study of Cardiothoracic Diseases and Cardiothoracic Transplantation, IRCCS-ISMETT, Palermo, Italy. · DICOM Vision, Dalmine, Italy. +3
Robust and Explainable Bicuspid Aortic Valve Diagnosis Using Stacked Ensembles on Echocardiography
May 13, 2026Christos Chrysanthos Nikolaidis, Vasileios Sachpekidis, Nikolas Moustakidis +2Echocardiography Ensemble
Department of Electrical and Computer Engineering, Democritus University of Thrace, Xanthi, 67100, Greece · Department of Cardiology, Papageorgiou Hospital, Thessaloniki, Greece · Department of Informatics, Aristotle University of Thessaloniki, Thessaloniki, Greece +1
TAVR-VLM: Risk-Conditioned Causal Grounding for Hallucination-Resistant Report Generation
Jun 25, 2026Zhixiang Lu, Xiwei Liu, Sifan Song +5Medical Vision-Language Models Hallucination Mitigation
Xi’an Jiaotong-Liverpool University · University of Liverpool · Mohamed bin Zayed University of Artificial Intelligence +2

Jul 31, 2026, cs.CVJ/K move · Enter open · S save

TAVI-TEC: An AI-Based Tool for Procedural Planning of Transcatheter Aortic Valve Implantation

Alessandra Zerillo, Stefano Cannata, Diego Bellavia, Daniele Ciriello+3

Department of Research, IRCCS-ISMETT, Palermo, Italy. · Department for the Treatment and Study of Cardiothoracic Diseases and Cardiothoracic Transplantation, IRCCS-ISMETT, Palermo, Italy. · DICOM Vision, Dalmine, Italy.+3

Coronary Angiography Ai-Based

Open paper AskPDF arXiv

Computed tomography angiography (CTA) is crucial for preprocedural TAVI planning, providing the anatomical information required for prosthesis sizing and vascular access assessment. As the volume of TAVI procedure increases, improving efficiency and standardizing annotations is becoming essential in clinical practice. This study presents TAVI-TEC, a fully automated artificial intelligence-based framework integrated into a web based DICOM viewer for routine preoperative TAVI planning. Pre-procedural CTA scans from patients undergoing TAVI with SAPIEN 3 Ultra (S3U) prostheses were processed using a fully automated pipeline. Deep learning-based segmentation of cardiovascular structures, calcification detection, centerline extraction, landmark identification, and annular plane definition was implemented to quantify key annular and aortic root measurements and color-coded maps of lumen reduction and vessel diameter for vascular access. A multilayer perceptron classifier was trained to predict prosthesis size prior to the TAVI procedure. Results revealed that TAVI-TEC enabled pre-procedural measurements in approximately 2-6 min. Strong agreement with clinician-derived measurements was observed for annular area (coefficient of concordance, CCC = 0.934; interclass correlation coefficient, ICC = 0.935; R^2 = 0.881) and perimeter (CCC = 0.909; ICC = 0.909; R^2 = 0.854). The valve-size prediction model achieved 82% overall accuracy, with most misclassifications occurring between adjacent prosthesis sizes. Though further multicenter validation and extension to additional measurements and valve platforms are required, the TAVI-TEC methodology may reduce operator variability in pre-TAVI measurements and streamline the preoperative workflows of the Heart Team for decision-making.

Abstract

Explore similar work

TAVI-TEC: An AI-Based Tool for Procedural Planning of Transcatheter Aortic Valve Implantation

Robust and Explainable Bicuspid Aortic Valve Diagnosis Using Stacked Ensembles on Echocardiography

TAVR-VLM: Risk-Conditioned Causal Grounding for Hallucination-Resistant Report Generation