Abstract Body (Do not enter title and authors here): Background
Ambulatory ECG (AECG) enables heart rhythm monitoring during daily activities, with a focus on arrythmia detection. Detection and quantification of sleep and activity patterns during monitoring may provide insights into lifestyle and aid in contextualizing arrhythmia events.
Aims
We developed an AI algorithm to classify sleep, activity, and inactivity periods using a novel AECG patch with embedded accelerometry. We assessed algorithm performance and compared it to FDA-cleared actigraphy and consumer devices, which have shown 93-99% sensitivity (SE) and 39-54% specificity (SP) in classifying sleep vs. polysomnography (PSG).
Methods
We conducted a prospective study of 81 participants at 3 sites who wore the Zio^® monitor AECG patch (iRhythm Technologies, San Francisco, CA) and Actigraph wGT3X (Pensacola, FL) simultaneously for 14 days. Sleep disordered breathing was excluded. Participants underwent a 6-minute walk test at the beginning of wear and in-clinic overnight PSG sleep testing at 7±3 days.
Data were split into training (n=40) and validation (n=41) sets. Feature and model selection utilized five-fold cross-validation on the training set, focusing on total activity and body angle. The final machine model was trained on selected features using the entire training set. In validation, SE and SP for sleep were assessed based on 1-minute epochs vs. PSG and also vs. a 24-hour reference (combined PSG and actigraphy-wake labeling). SE and SP for activity were assessed vs. actigraphy over 8 minutes sampled per subject (4min walk test, 4min PSG wake periods).
Results
The study population was diverse (age 43±14 years, 57% female, 64% White, 25% Black, 20% Hispanic). In the validation set, average sleep and wake times were 5.9 and 1.2 hours, respectively, during PSG. SE and SP were 88.8% and 54.0%, respectively, in sleep detection vs. PSG, or 88.8% SE and 95.6% SP vs. the 24-hour reference (Table 1). SE and SP for activity detection were 97.0% and 100%, respectively.
Conclusion
Assessment of sleep and activity during AECG is feasible, with performance comparable to FDA-cleared actigraphy. This feature offers insights into patient wellness patterns, highlighting its potential for personalized healthcare monitoring.