Machine-Learning-Based Prediction of Exercise Intolerance of Patients With Heart Failure Using Pragmatic Submaximal...

Area of Science:

• Cardiology
• Exercise Physiology
• Machine Learning in Medicine

Background:

• Low peak oxygen uptake (V̇O₂) is a critical prognostic indicator in heart failure (HF).
• Measuring peak V̇O₂ can be challenging if maximal exercise capacity is not achieved.
• Accurate prediction of low V̇O₂ is essential for risk stratification in HF patients.

Purpose of the Study:

• To develop and validate a machine learning (ML) model for predicting low peak V̇O₂ in HF patients.
• To identify key clinical and exercise parameters predictive of low peak V̇O₂.
• To enable reliable V̇O₂ estimation when maximal testing is not feasible.

Main Methods:

• Retrospective analysis of 343 chronic HF patients with LVEF <50%.
• Utilized 33 variables from laboratory, echocardiographic, and submaximal exercise data.
• Employed machine learning, including support vector machines, for model development and validation.

Main Results:

• Identified 5 key predictors: age, B-type natriuretic peptide, left ventricular end-diastolic diameter, resting V̇O₂, and V̇O₂ at RER 1.00.
• The optimized ML model achieved 85% accuracy, an F1 score of 0.81, and an AUC of 0.94.
• The model demonstrated high predictive performance on the independent testing dataset.

Conclusions:

• Machine learning models can accurately predict low peak V̇O₂ in HF patients.
• Readily available clinical and exercise parameters are sufficient for this prediction.
• This approach facilitates prognostic assessment in HF when maximal cardiopulmonary exercise testing is limited.