Atrial Fibrillation History Feature

The Atrial Fibrillation History Feature is an over-the-counter software application designed for individuals with a pre-existing AFib diagnosis to retrospectively track their AFib burden using photoplethysmography data from an Apple Watch. It employs a convolutional neural network to analyze pulse rhythms, providing users with estimated AFib burden percentages and lifestyle data visualizations to enhance their understanding and management of their condition.

The Atrial Fibrillation History Feature is an over-the-counter (OTC) software-only mobile medical application designed for users aged 22 and older who have a prior diagnosis of atrial fibrillation (AFib). It operates in conjunction with compatible Apple Watch models (Series 4, 5, SE) and the Health app on an iPhone, utilizing the watch's photoplethysmography (PPG) sensor, which employs green LED lights and light-sensitive photodiodes, to detect changes in blood flow at the wrist and thereby measure beat-to-beat intervals from pulse rate data. The feature opportunistically analyzes this data to identify episodes of irregular heart rhythms suggestive of AFib, subsequently providing users with a retrospective estimate of their AFib burden, defined as the amount of time spent in AFib during periods of Apple Watch wear. The core of the AFib History Feature is its rhythm classification algorithm, which is based on a convolutional neural network (machine learning) trained on an extensive dataset comprising over 2500 subjects and more than 3 million pulse rate recordings, and has been specifically optimized for individuals with a pre-existing AFib diagnosis. In addition to tracking AFib burden over time—presenting estimates weekly, by day of the week over the past 42 days, and by time of day in 4-hour segments over the past 42 days—the feature incorporates lifestyle data visualizations to help users understand potential correlations between their habits and AFib occurrences. Clinical validation, involving 413 participants (with 280 contributing to the primary endpoint analysis), demonstrated the feature's accuracy in estimating weekly AFib burden when compared to a reference electrocardiogram (ECG) patch, yielding Bland-Altman limits of agreement of -11.4% and 12.8%, an average difference of 0.67%, and showing that 92.9% of subjects had paired weekly AFib burden differences within ±5%. The classification algorithm itself achieved 92.6% sensitivity and 98.8% specificity in this clinical validation study.

AI Regulatory Concerns

Powered by

Jev reads the full FDA letter and returns calibrated concern probabilities per regulatory category. These are AI estimates, not regulatory advice.

• Q1. How does the opportunistic analysis of pulse rate data account for periods when the Apple Watch is not worn or when data quality is insufficient for analysis, and how might this impact the accuracy of the retrospective AFib burden estimate?; What specific lifestyle factors are tracked and visualized alongside AFib burden, and what is the evidence supporting their clinical relevance in influencing AFib burden in the target population?; Given the algorithm's optimization for users with a prior AFib diagnosis, what is its performance (sensitivity and specificity) in differentiating AFib from other common arrhythmias or ectopic beats in this specific population?; How does the feature's 'Day of Week' and 'Time of Day' AFib burden estimation account for variability in watch wear time across different days or times, and how is this communicated to the user?; What were the criteria for including/excluding subjects from the primary endpoint analysis (280 out of 413 enrolled) in the clinical validation study, and how might this selection have influenced the reported Bland-Altman limits of agreement?