Imagine a world where your daily health habits are not just tracked, but also accurately linked to your medical well-being. That's the exciting prospect Empirical Health is exploring with their groundbreaking research.
The Power of Wearables: Unlocking Medical Insights
Presented at the prestigious Timeseries for Health workshop at NeurIPS 2025, Empirical Health has developed a foundation model that predicts blood test results and diagnoses from wearable data. This is a significant step forward in the field of health technology, bridging the gap between consumer wearables and medical actionability.
But here's where it gets controversial: Empirical's model, JETS, has achieved an impressive 87% accuracy in detecting high blood pressure. It also successfully identified other conditions like atrial flutter (70% accuracy), ME/CFS (81% accuracy), and sick sinus syndrome (87% accuracy). These results are a testament to the potential of AI in healthcare.
JETS, or Joint Embedding for Timeseries, was trained on an extensive dataset of 3 million person-days of wearable data from various devices like Apple Watch, Fitbit, Pixel Watch, and Samsung Galaxy Watch. It utilizes the JEPA architecture proposed by Yann LeCun, a renowned figure in the AI community.
This research study touches on several critical themes:
Blood Testing Meets Wearables: Many wearables now offer lab testing features, but the link between wearable data and blood tests has been largely unexplored. Empirical Health's study is a pioneering effort in this domain, showing how AI can connect the dots between these two sources of health information.
Beyond LLMs: The Future of AI in Healthcare: While large language models have achieved remarkable success, the question arises: what's next? Empirical Health suggests that the next frontier in AI is physiological ground truth, which wearables can provide. This study paves the way for developing health superintelligence based on physiological data.
Extracting Meaningful Insights from Wearables: The JETS system employs a unique training method with twin encoders. One encoder sees the full sequence of data, while the other sees only a portion, approximately 30%. By learning to align their latent representations, the system focuses on understanding the meaning behind the data, rather than surface details.
Empirical Health's mission is ambitious yet crucial: to prevent one million heart attacks. Their program starts by assessing over 100 biomarkers, modeling an individual's risk of heart disease, and then creating a personalized plan to mitigate that risk. Founded by an ex-Kaiser doctor and an ex-Google machine learning expert, Empirical Health has already impacted the lives of over 100,000 people.
This research study is a significant step towards achieving Empirical Health's mission. It showcases the potential of AI to revolutionize healthcare, providing actionable insights from wearable data. But what are your thoughts? Do you see this as a promising development or a potential cause for concern? We'd love to hear your opinions in the comments below!
