• The 1-minute sit-to-stand test and 6-minute walk test are key tools for evaluating exercise capacity in COPD patients. • A smartphone-based 1-minute sit-to-stand test predicts the 6-minute walk distance. • Combines inertial motion unit sensor data with machine learning for accurate estimation. • Enables remote COPD management using smartphone sensors and advanced algorithms. Regular assessment of exercise tolerance is essential for managing COPD, nevertheless, the standard 6-Minute Walk Test (6MWT) is difficult to perform outside clinical settings. This study aimed to develop and validate a smartphone-based digital 1-Minute Sit-to-Stand Test (1MSTST) to estimate 6-Minute Walking Distance (6MWD) by integrating data from the phone's Inertial Motion Unit (IMU) with advanced machine learning algorithms, offering a convenient alternative for remote functional assessment. The enrolled COPD patients completed the smartphone-based digital 1MSTST and 6MWT with a minimum 15-minute rest period between the two tests. Accelerometer and gyroscope data were recorded by a smartphone throughout the 1MSTST. Systolic and Diastolic Blood Pressure (SBP, DBP), Heart Rate (HR) and Pulse Oxygen Saturation (SpO 2 ) were measured before and after the 1MSTST and 6MWT. The authors used a smartphone-based dataset of 66 subjects and algorithms such as Extreme Gradient Boosting (XGBoost), Light Gradient Boosting Machine (LightGBM), Support Vector Machine (SVM), Linear Regression (LR) and Random Forest (RF) to estimate the prediction of digital 1MSTST to 6MWD. The correlation between the features extracted from digital 1MSTST and the 6MWD was analyzed. A total of 66 patients with stable COPD were enrolled to build the predictive model for 6MWD. The change of HR and SBP after 1MSTST was higher than that of 6MWT (paired t -test, ΔHR: p < 0.0001, ΔSBP: p < 0.0001) with no significant difference in the change of DBP and SpO 2 (paired t -test, ΔDBP: p = 0.974, ΔSpO 2 : p = 0.072). Pearson correlation analysis identified the top seven features that were strongly correlated with 6MWD. The RF machine learning model had the best performance to predict the 6MWD (R-square = 0.86, MAE = 24.3). Bland-Altman analysis indicates that the RF model provides a bias of -0.61 ± 31.04 and that the limits of agreement (95%, 1.96 SD) range from -60.22 to 61.45. The smartphone-based digital 1MSTST, combined with machine learning, can accurately estimate 6MWD. The significance of this study lies in proposing a novel assessment paradigm that may serve as a practical tool for remote monitoring of exercise capacity in COPD management.
Xie et al. (Thu,) studied this question.