A machine learning algorithm using wearable watch data detected CAD with 80% sensitivity, 84% specificity, and AUC of 0.82 in the test group.
Does a machine learning algorithm utilizing ECG and PPG data from a wearable watch accurately detect coronary artery disease compared to healthy controls?
A machine learning algorithm utilizing ECG and PPG data from a smart wearable watch demonstrated good diagnostic accuracy (AUC 0.82) for detecting coronary artery disease.
Tasa de eventos absoluta: 0% vs 0%
Abstract Background Wearable devices have become increasingly important in the early detection of coronary artery disease (CAD). However, the application of machine learning to harness data from these devices for early CAD detection has been limited. Purpose This study aimed to develop and validate a machine learning algorithm for early CAD detection utilizing data from one already marketed smart wearable watch (ECG version). Methods We conducted a study involving patients undergoing coronary angiography to train and validate our machine learning model for early CAD detection based on participants' watch data. From August 2024 to January 2025, 152 patients and 148 healthy individuals from one local hospital were consecutively enrolled into this study. The cohort was divided into a training group (106 patients and 103 healthy participants), and a test group (46 patients and 45 healthy individuals) to facilitate the development and validation of the algorithm. For each participant, more than 60 features were extracted from electrocardiogram (ECG) and photoplethysmography (PPG) data obtained from the watch. Logistic Regression was selected as our core algorithm due to its accuracy, efficiency and interpretability. Sensitivity, specificity, and area under the receiver operating characteristic curve (AUC) were calculated to validate our trained algorithm. Results Key population characteristics of the study cohort revealed: 92.1% (140/152) of CAD patients were aged ≥50 vs 85.8% (127/148) controls 40 years; elevated BMI (≥25 kg/m²) prevalence was nearly doubled in CAD group (46.7% vs 24.3%). Our CAD detection algorithm achieved a sensitivity of 0.80 (95% confidence interval (CI), 0.74-0.86), a specificity of 0.84 (95% CI, 0.78-0.89) and an AUC of 0.82(95% CI, 0.74-0.89) in the test group. Notably, subgroup analysis of the test group revealed differential sensitivity across CAD subtypes: 83.3% (20/24) for stable angina, 50% (2/4) for STEMI, and 83.3% (15/18) for other CAD types. The top five pivotal features of the model were identified: T-wave width variability, QRS complex duration, S-wave width variability, percentage of NN intervals 50ms, and high-frequency heart rate variability. Physiological patterns emerged where CAD patients exhibited significantly higher values in the first four parameters, while showing reduced high-frequency heart rate variability. Conclusions Our results indicate that a machine learning algorithm based on wearable watch can effectively assist in CAD detection, which would offer a promising opportunity to enhance early detection of CAD among the general population.Summary figure
Xu et al. (Sat,) reported a other. A machine learning algorithm using wearable watch data detected CAD with 80% sensitivity, 84% specificity, and AUC of 0.82 in the test group.