Predicting Heart Rate at the Anaerobic Threshold Using a Machine Learning Model Based on a Large-Scale Population
Atsuko Nakayama1,2, Tomoharu Iwata3,4, Hiroki Sakuma3,4
1Department of Cardiovascular Medicine, Sakakibara Heart Institute, Tokyo 183-0003, Japan.
Journal of Clinical Medicine
|January 11, 2025
Summary
A new machine learning model accurately predicts target heart rate at anaerobic threshold (AT-HR) using non-exercise clinical data. This approach offers a more precise method for exercise prescription in cardiovascular disease patients than traditional formulas.
Area of Science:
- Cardiology
- Exercise Physiology
- Machine Learning in Medicine
Background:
- Determining target heart rate at the anaerobic threshold (AT-HR) is crucial for effective exercise prescription in cardiovascular disease (CVD) patients.
- Cardiopulmonary exercise testing (CPET) is the standard method for AT-HR determination but can be resource-intensive.
- Developing alternative methods to predict AT-HR from readily available clinical features is needed.
Purpose of the Study:
- To develop and validate a machine learning (ML) model for predicting AT-HR using only non-exercise clinical features.
- To compare the accuracy of the ML model against established guideline-recommended equations for AT-HR prediction.
- To identify key clinical features that significantly contribute to AT-HR prediction.
Main Methods:
- Utilized a dataset of 8228 participants (healthy and CVD patients) from 21,482 CPET cases.
- Employed a gradient boosting ML model trained on 78 clinical features (e.g., demographics, vitals, blood tests, echocardiography).
- Evaluated prediction accuracy using Mean Absolute Error (MAE) and compared ML model results with Karvonen and simpler formulas.
Main Results:
- The ML model achieved a significantly lower MAE (7.7 ± 0.2 bpm) compared to guideline equations (e.g., Karvonen: 34.5 ± 0.3 bpm, 11.9 ± 0.2 bpm; simpler formulas: 15.9 ± 0.3 bpm, 9.7 ± 0.2 bpm).
- Key predictors for AT-HR included resting heart rate, age, N-terminal pro-brain natriuretic peptide (NT-proBNP), resting systolic blood pressure, hsCRP, CVD diagnosis, and beta-blocker use.
- High prediction accuracy was maintained using the top 10-20 features, indicating feature efficiency.
Conclusions:
- An accurate ML-based prediction model for AT-HR from non-exercise clinical features has been successfully developed.
- This model can potentially simplify and enhance exercise prescription for cardiac rehabilitation.
- The study identified novel determinants of AT-HR, such as NT-proBNP and hsCRP, offering new insights into CVD pathophysiology.
