Machine learning models outperform traditional scoring systems for predicting post-TAVR mortality riskMachine Learning Models Show Better Accuracy for TAVR Mortality Risk

Frontiers in Medicine Published July 15, 2026 DOI ↗ Editorial oversight: Dr. Amelia Tan, PhD · Internal Medicine & Chronic Disease

AI-generated summary of the cited source, checked by automated accuracy review. How we work

Key Takeaway

Note that machine learning models provide superior C-index scores for predicting mortality after TAVR than traditional tools.

This meta-analysis evaluated the predictive accuracy of various mortality risk models in patients undergoing transcatheter aortic valve replacement (TAVR). The analysis included 36 studies and a total population of 272,390 patients to compare traditional scoring systems against machine learning models using the concordance index (C-index).

Key findings indicate that machine learning models achieved a C-index of 0.705 (95% CI: 0.677 to 0.733). In comparison, established tools showed lower predictive accuracy: STS risk model had a C-index of 0.648 (95% CI: 0.622 to 0.674), ACC TVT risk model reached 0.632 (95% CI: 0.616 to 0.648), and EuroSCORE II was 0.615 (95% CI: 0.588 to 0.643). Other models, including EuroSCORE I (0.610) and OBSERVANT score (0.594), also showed lower C-indices than machine learning counterparts.

The authors note that there are currently not enough subjects available to develop even more accurate models. While machine learning models appear more effective for predicting mortality risk after TAVR compared to traditional scoring tools, the evidence is limited by the current data volume. These findings suggest a shift toward advanced computational models in clinical risk stratification.

How this fits prior evidence

This meta-analysis addresses a gap in risk stratification tools for patients undergoing transcatheter aortic valve replacement (TAVR). It builds upon existing knowledge regarding TAVR outcomes, such as the finding that TAVR is associated with higher 10-year mortality compared to surgery in intermediate-risk aortic stenosis. While previous evidence established the clinical outcomes of TAVR, this study specifically evaluates the predictive accuracy of different scoring systems for post-procedure mortality.

Researchers analyzed data from over 272,000 patients who underwent a procedure called transcatheter aortic valve replacement (TAVR). The goal was to see which mathematical models best predict the risk of death following this specific heart surgery.

The study compared several established scoring systems, such as EuroSCORE I and EuroSCORE II, against more modern machine learning models. The results showed that machine learning models had a higher concordance index, which is a measure of how well a model predicts an outcome. Specifically, these advanced models scored 0.705 compared to lower scores for traditional methods like the STS risk model (0.648) or the France II score (0.578).

While machine learning appears more effective at predicting mortality than standard tools, it is important to note that this is a large-scale review of existing data. The study notes that there are still not enough subjects available to create even more precise models. Patients and doctors should view these findings as an indication of improved prediction accuracy rather than a definitive change in clinical practice.

What this means for you:

Machine learning models currently show higher accuracy in predicting TAVR mortality risk than traditional scoring tools.

Common questions

What is TAVR and who does this study help?

TAVR stands for transcatheter aortic valve replacement. This study looked at 272,390 patients who received this specific heart procedure. The findings help doctors understand which mathematical tools are most accurate when predicting the risk of death for these patients after their surgery.

How do machine learning models compare to traditional scores?

Machine learning models showed a higher concordance index of 0.705 for predicting mortality. In comparison, established tools like the STS risk model had a score of 0.648, and the France II score was 0.578. This suggests machine learning is currently more accurate than these traditional scoring systems.

Are there any limitations to these findings?

The study notes that there are still not enough subjects available to develop even more accurate models at this time. While machine learning shows better results than current tools, the data is used to compare prediction accuracy rather than to provide a guaranteed medical outcome.

Study Details

Study typeMeta analysis

EvidenceLevel 1

PublishedJul 2026

View Original Abstract ↓

BackgroundTranscatheter aortic valve replacement (TAVR) is increasingly used due to the rising incidence of aortic stenosis (AS). Early identification of mortality risk after TAVR is challenging. Although various prediction models have been developed, no systematic review has evaluated their effectiveness in predicting mortality risk. Therefore, this study aimed to systematically evaluate the performance of models for early prediction of mortality risk after TAVR, so as to provide evidence-based support for the future development or updating of risk assessment tools.MethodsDatabases (PubMed, Web of Science, Embase, and Cochrane Library) were systematically searched for studies on tools for predicting the risk of mortality after TAVR, up to June 2024. PROBAST was used to assess the risk of bias in the included studies. A subgroup analysis was conducted based on different time points.ResultsThis systematic review included 36 studies with 272,390 patients receiving TAVR and 6 major scoring tools encompassing 23 new machine learning models. The meta-analysis showed that the concordance index (C-index) was 0.610 (95% CI: 0.588–0.632) for European System for Cardiac Operative Risk Evaluation I (EuroSCORE I), 0.615 (95% CI: 0.588–0.643) for EuroSCORE II, 0.578 (95% CI: 0.531–0.625) for French Aortic National CoreValve and Edwards II (France II), 0.594 (95% CI: 0.554–0.633) for the OBSERVANT score, 0.648 (95% CI: 0.622–0.674) for the Society of Thoracic Surgeons (STS) risk model, 0.632 (95% CI: 0.616–0.648) for the American College of Cardiology Transcatheter Valve Therapy (ACC TVT) risk model, and 0.705 (95% CI: 0.677–0.733) for summarized machine learning models.ConclusionDetermining the predictive performance of current established risk assessment tools for predicting the risk of modality after TAVR is challenging. Machine learning models seem to be more effective. Therefore, future research should include more subjects to develop more accurate models.Systematic Review Registrationhttps://www.crd.york.ac.uk/PROSPERO/, identifier CRD42023485237.

Machine learning models outperform traditional scoring systems for predicting post-TAVR mortality riskMachine Learning Models Show Better Accuracy for TAVR Mortality Risk

How this fits prior evidence

Common questions

Study Details

Meta-analysis finds systemic lupus erythematosus elevates coronary artery disease risk

People with lupus face higher heart disease risk but genetics show no direct link

Clinical research that matters. Delivered to your inbox.

Machine learning models outperform traditional scoring systems for predicting post-TAVR mortality riskMachine Learning Models Show Better Accuracy for TAVR Mortality Risk

How this fits prior evidence

Common questions

More on Aortic Stenosis

Study Details

Meta-analysis finds systemic lupus erythematosus elevates coronary artery disease risk

People with lupus face higher heart disease risk but genetics show no direct link

Clinical research that matters. Delivered to your inbox.

Related in Cardiology

From Other Specialties