Sensors are an essential part of a wearable Health Tech device, creating the data that is used to provide meaningful results for the wearer. The real work, however, is done by sophisticated algorithms that can extract meaning from noisy data sets, singling out the important trends and ignoring the random static. As we have repeatedly written here, the future of the wearable Health Tech industry lies in algorithms that can produce reliable, actionable information. And we need to prove the accuracy of the results. As the market matures, we are seeing more and more scientific studies that try to evaluate how reliable this information really is.

That’s why we’re particularly interested in a recent study by Cardiogram and the University of California San Francisco Cardiology Department. The project is called mRhythm, and has more than 6,000 subjects enrolled. The goal was to create an algorithm that could detect the most common form of irregular heartbeats: atrial fibrillation (AFib). This condition is a leading cause of strokes, but many people are not aware that they have this condition, going undiagnosed and untreated. Cardiogram trained a deep neural network to read Apple Watch heart rate data and identify incidents of AFib. Once the system was trained to identify the condition, they enlisted 51 AFib patients who were schedule to undergo “cardioversion,” in which their hearts are converted back to normal sinus rhythm. Analyzing data from their Apple Watches before and after the return to normal rhythm, the system was able to correctly identify AFib data sets¬†with 98% sensititivity and 90% specificity.

AFib is often intermittent, and it can be difficult to “catch” an episode when it occurs. With ongoing monitoring by a consumer wearable — like the Apple Watch — could mean that that individuals are watched around the clock, increasing the chances of identifying an AFib or other arrhythmia event. Early detection can lead to early treatment, which in turn can lead to better outcomes and reduced overall healthcare costs.