Imagine a world where your smartwatch doesn't just track your steps, but actually predicts your risk for serious health conditions like high blood pressure. That future might be closer than you think, thanks to groundbreaking research from Empirical Health.
Presented at the prestigious Timeseries for Health workshop at NeurIPS 2025, the leading artificial intelligence conference, Empirical Health unveiled a new wearable foundation model with some astonishing capabilities. We all know that our daily habits significantly impact our health, but pinpointing the exact effect of any single action is incredibly difficult. Blood tests, considered the medical 'gold standard,' only provide snapshots every few months. Consumer wearables, on the other hand, offer a continuous stream of data, but often lack the clinical validation to be truly actionable... until now.
What if we could use the rich data from wearables to predict the kind of detailed health information we get from blood tests? That's exactly what Empirical Health set out to achieve.
Their research study introduces a novel foundation model designed to predict blood test results and diagnose conditions directly from wearable data. The implications are huge.
Empirical's model, cleverly named JETS (Joint Embedding for Timeseries), achieved an impressive 87% accuracy in detecting high blood pressure. But it doesn't stop there. JETS also demonstrated significant accuracy in identifying other conditions, including atrial flutter (70%), ME/CFS (81%), and sick sinus syndrome (87%). JETS was trained on a massive dataset of 3 million person-days of wearable data, encompassing a wide range of devices like Apple Watch, Fitbit, Pixel Watch, and Samsung Galaxy Watch. This dataset included 63 independent timeseries, providing a comprehensive view of each individual's health metrics. The model leverages the JEPA architecture, a cutting-edge approach pioneered by Yann LeCun, formerly Meta's Chief Scientist.
And this is the part most people miss: the potential applications are vast.
This research touches upon several crucial themes that are reshaping the future of healthcare:
Blood Testing Meets Wearables: While some wearables have recently incorporated lab testing features, very few attempt to directly link wearable data with blood test results. Empirical Health's study is one of the first published research efforts demonstrating how artificial intelligence can bridge this gap. Imagine getting proactive alerts about potential health issues based on the combined analysis of your wearable data and predicted blood test results! This could lead to earlier diagnoses and more effective interventions.
AI Beyond LLMs: Large language models (LLMs) have taken the world by storm, but some experts believe we're reaching a saturation point with text-based training data. The next frontier in AI lies in leveraging physiological ground truth - the kind of rich, real-time data generated by wearables. This study offers a glimpse into how we can use this data to create a 'health superintelligence' capable of personalized health insights and preventative care. But here's where it gets controversial... Could this also lead to concerns about data privacy and algorithmic bias?
Extracting Meaningful Signal from Wearables: The system employs twin encoders, where one encoder processes the full data sequence, while the other sees only approximately 30%. This forces the model to align their latent representations without simply reconstructing raw signals. This encourages the model to learn the underlying meaning of the wearable data, rather than just focusing on superficial details. Think of it like learning the core concepts of a book, rather than just memorizing individual words. This ability to extract meaningful signal is crucial for developing robust and reliable health predictions.
About Empirical Health
Empirical Health's ambitious mission is to prevent one million heart attacks. Their program analyzes over 100 biomarkers to model your risk of heart disease and creates a personalized plan to mitigate that risk.
Founded by an ex-Kaiser doctor and an ex-Google machine learning tech lead, and backed by Y Combinator S23, Empirical Health has already helped more than 100,000 people take control of their health.
What do you think about using AI to predict health conditions from wearable data? Do you find the potential benefits outweigh the potential risks? Share your thoughts in the comments below!
