Wearable Monitoring and Interpretable Machine Learning Can Objectively Track Progression in Patients during Cardiac Rehabilitation

Abstract

Cardiovascular diseases (CVD) are often characterized by their multifactorial complexity. This makes remote monitoring and ambulatory cardiac rehabilitation (CR) therapy challenging. Current wearable multimodal devices enable remote monitoring. Machine learning (ML) and artificial intelligence (AI) can help in tackling multifaceted datasets. However, for clinical acceptance, easy interpretability of the AI models is crucial. The goal of the present study was to investigate whether a multi-parameter sensor could be used during a standardized activity test to interpret functional capacity in the longitudinal follow-up of CR patients. A total of 129 patients were followed for 3 months during CR using 6-min walking tests (6MWT) equipped with a wearable ECG and accelerometer device. Functional capacity was assessed based on 6MWT distance (6MWD). Linear and nonlinear interpretable models were explored to predict 6MWD. The t-distributed stochastic neighboring embedding (t-SNE) technique was exploited to embed and visualize high dimensional data. The performance of support vector machine (SVM) models, combining different features and using different kernel types, to predict functional capacity was evaluated. The SVM model, using chronotropic response and effort as input features, showed a mean absolute error of 42.8 m (±36.8 m). The 3D-maps derived using the t-SNE technique visualized the relationship between sensor-derived biomarkers and functional capacity, which enables tracking of the evolution of patients throughout the CR program. The current study showed that wearable monitoring combined with interpretable ML can objectively track clinical progression in a CR population. These results pave the road towards ambulatory CR.

Keywords: cardiac rehabilitation; machine learning; patient progression monitoring; physical fitness assessment; wearable sensor.

Figure 1

Added variable plot for the whole model. The adjusted whole model visualizes the fit of two independent variables (effort and chronotropic response) against the dependent variable. 6MWD, 6-min walking distance. R-squared: 0.661, p-value < 0.001.

Figure 2

6MWD prediction based on accelerometer derived effort and chronotropic response.

Figure 3

Error distribution in 6MWD prediction. Mean average error is equal to 42.5 m (±35.5 m).

Figure 4

(a) t-distributed stochastic neighboring embedding (t-SNE) projection, colors represent rehabilitation time (session number) (upper left graph); (b) t-SNE projection, colors represent distance. Note that distance has not been used as input feature in the model, but it has only been used to color the map (upper right graph); (c) t-SNE projection, colors represent chronotropic response (bottom left graph); (d) t-SNE projection, colors represent effort during the last 2 min of the 6MWT (bottom right graph). The axes do not represent distance or time but are the results of a composite projection of features in a 3D space.

Figure 5

(a) Tracking all patients during the rehabilitation program (top left); (b) tracking of rehabilitation for patient 38. The 6MWD has increased from about 600 m (green) to more than 700 m (yellow) between the first and last session of the rehabilitation program (top right); (c) tracking of patient number 24. This patient shows a small difference in the 6MWD between the start and the end of the rehabilitation program (bottom left); (d) tracking of patient number 15. This patient shows an increase in 6MWD between baseline and the second 6MWT, stable afterwards (bottom right).