Apple’s Wearables Can Now Spot Health Problems with AI
Apple researchers use 2.5B hours of Watch data to train an AI that predicts health issues from how users move, sleep, and live—more than just biometrics.
image for illustrative purpose
In a major advancement for digital health technologies, researchers have unveiled a foundation model that leverages behavioral data from wearable devices to enhance the prediction of various medical conditions. The system, named the Wearable Behavior Model (WBM), diverges from traditional models that prioritize immediate sensor readings by analyzing longer-term patterns in user behavior.
The study, titled "Beyond Sensor Data: Foundation Models of Behavioral Data from Wearables Improve Health Predictions," outlines how this machine learning framework processes metrics such as movement, rest cycles, heart rate variability, and daily activity. These behavioral signals are derived from Apple Watch data using embedded algorithms.
Unlike previous models that focus solely on momentary biometric feedback, WBM identifies temporal trends that may signal underlying health issues. The model proved more effective at diagnosing both constant health markers, such as medication usage (e.g., beta blockers), and short-term concerns like respiratory infections or sleep disruptions.
In cases like pregnancy detection, WBM, when paired with biometric inputs, reached an accuracy rate of up to 92 per cent. This hybrid method significantly outperforms systems that rely exclusively on sensor outputs.
The foundation model was trained using data from Apple’s Heart and Movement Study. This ongoing initiative includes health information voluntarily shared by over 160,000 users via Apple Watch and iPhone. The dataset consists of more than 2.5 billion recorded hours, making it one of the most extensive behavioral health datasets to date.
Researchers evaluated the model across 57 distinct predictive tasks. Using a time-series approach, the system monitors behavioral changes spanning several days or weeks, allowing it to detect conditions that develop progressively.
While there is no confirmation yet on whether this model will feature in public Apple Watch updates, the research confirms that existing consumer wearables have the computational capacity to support advanced AI health tools.
