Can ML flag high-risk AS+HFpEF after TAVR?

In 326 TAVR patients, an 8-feature SVM reached validation AUC 0.756 for 12-month MACCE, explained with SHAP.

Source

Multimodal Visualization and Explainable Machine Learning-Driven Markers Enable Early Identification and Prognosis Prediction for Symptomatic Aortic Stenosis and Heart Failure With Preserved Ejection Fraction After Transcatheter Aortic Valve Replacement: Multicenter Cohort Study

Wang J, Zhu J, Li H, et al. · Journal of medical Internet research · 2025

doi.org/10.2196/70587 Read the full paper ↗17 citationscc by

Study at a glance

Design: Cohort — Multicenter retrospective TAVR cohort with train/validation ML risk models
N: N=326 · 195 derivation + 131 external validation patients
Population: Older patients with severe aortic stenosis and HFpEF undergoing TAVR
Outcome: 12-month major adverse cardiovascular and cerebrovascular events predicted by explainable ML

Structured fields used in claim comparison tables when every cited study has a complete layer.

What they did

Selected features via LASSO+Boruta, trained five ML models on 195 patients, validated on 131 from other hospitals, and interpreted with SHAP.

What they found

SVM performed best (AUC 0.756); features centered on age, NT-proBNP, and metabolic lipid-glucose indices; good calibration/net benefit claimed vs alternatives.

The limits

What it doesn't show

Prospective deployment benefit, or psychology-primary mechanisms (cardiac cohort framed for aging risk markers).

Key terms

TAVR: Transcatheter aortic valve replacement.
HFpEF: Heart failure with preserved ejection fraction.
MACCE: Major adverse cardiovascular and cerebrovascular events.
SVM: Support vector machine classifier/regressor.
SHAP: Shapley-value explanations of model predictions.
TyG index: Triglyceride-glucose index of insulin resistance.

Flashcards

1 / 8

Research intelligence for this paper

See its role on concept claims, tensions it is part of, placement history, and related discoveries.

Open paper intelligence

Quiz yourself

1 / 5

Validation AUC for SVM?

Common questions

326 (195+131).

Best AUC?

SVM 0.756 validation.

Features?

8 clinical/metabolic markers.

Explainability?

SHAP values.