Consumer wearable devices for evaluation of heart rate control using digoxin versus beta-blockers: the RATE-AF randomized trial

Abstract

Consumer-grade wearable technology has the potential to support clinical research and patient management. Here, we report results from the RATE-AF trial wearables study, which was designed to compare heart rate in older, multimorbid patients with permanent atrial fibrillation and heart failure who were randomized to treatment with either digoxin or beta-blockers. Heart rate (n = 143,379,796) and physical activity (n = 23,704,307) intervals were obtained from 53 participants (mean age 75.6 years (s.d. 8.4), 40% women) using a wrist-worn wearable linked to a smartphone for 20 weeks. Heart rates in participants treated with digoxin versus beta-blockers were not significantly different (regression coefficient 1.22 (95% confidence interval (CI) -2.82 to 5.27; P = 0.55); adjusted 0.66 (95% CI -3.45 to 4.77; P = 0.75)). No difference in heart rate was observed between the two groups of patients after accounting for physical activity (P = 0.74) or patients with high activity levels (≥30,000 steps per week; P = 0.97). Using a convolutional neural network designed to account for missing data, we found that wearable device data could predict New York Heart Association functional class 5 months after baseline assessment similarly to standard clinical measures of electrocardiographic heart rate and 6-minute walk test (F1 score 0.56 (95% CI 0.41 to 0.70) versus 0.55 (95% CI 0.41 to 0.68); P = 0.88 for comparison). The results of this study indicate that digoxin and beta-blockers have equivalent effects on heart rate in atrial fibrillation at rest and on exertion, and suggest that dynamic monitoring of individuals with arrhythmia using wearable technology could be an alternative to in-person assessment. ClinicalTrials.gov identifier: NCT02391337 .

Fig. 1. RATE-AF wearables substudy flowchart.

Flowchart for the wearables study enrollment.

Fig. 2. Wearable device data for measurement of heart rate and physical activity.

a, Examples of data capture for heart rate (red lines) and step count (green bars) using a wrist-worn wearable and smartphone over a single 24-h period for two individual patients with AF and heart failure. b, Correlations between daytime 10-s intervals of heart rate and physical activity for 50 patients who remained in AF at each visit. Light blue columns indicate the range of positive and negative correlations between heart rate and physical activity, with medians indicated by dark blue bars (correlation <0.19, very weak; 0.20–0.59, weak to moderate).

Fig. 3. Heart rate data from the wearable device.

The start of the wearables study was preceded by a period of dose adjustment and stabilization (mean of 30 weeks from randomization). a, Mean (solid line) and s.d. (shaded area) in heart rate over the 20-week period of follow-up in patients randomized to treatment with digoxin (brown) or beta-blockers (blue). b, Individual patient heart rate trajectories over the 20-week period of follow-up in patients randomized to treatment with digoxin (brown) and beta-blockers (blue). Bold lines indicate the fitted generalized linear model curves with corresponding 95% CI (shaded). No significant difference was demonstrated between the two groups using a generalized linear model with random-effects to account for repeated measurements (unadjusted P = 0.55; adjusted P = 0.75; after accounting for physical activity P = 0.74).

Fig. 4. Prediction of clinical progress with a neural network based on wearable data.

F1 scores combining precision and recall of each model are presented along with 95% CI for the prediction of NYHA functional class at the end of follow-up (mean 5 months); an F1 score of 0.35 (dashed line) is equivalent to chance. Derived from wearable sensor data from n = 41 individual patients.

Extended Data Fig. 1. Overview of the RATE-AF randomized trial wearable study.

mcg = micrograms; mg = milligrams; RATE-AF = RAte control Therapy Evaluation in permanent Atrial Fibrillation.

Extended Data Fig. 2. Neural network architecture.

Top panel: At time-point 1 (t1), 6-minute walk distance (6MWD), electrocardiogram (ECG) heart rate (HR), age, sex and body mass index are taken at the closest trial appointment to the start of the wearables sub-study. S1 denotes start of sensor data collection. S2 denotes sensor data used for validation. Time-point 2 (t2) is the future for New York Heart Association (NYHA) functional class prediction. Lower panel: The convolutional neural network (CNN) architecture uses dilated convolution and max-pool layers to capture long-term structure. Global average pooling is used to reduce spatial extent to fixed-length vector representation. Dropout and L2 penalty are applied to avoid overfitting to training set and label smoothing is used for faster learning. The final layer is binary sigmoid for data discrimination task. The penultimate layer of 32 units is used as latent space features for regression modeling.