A new study out from health startup Cardiogram and the University of California San Francisco (UCSF) suggests wearables like the Apple Watch, Fitbit and others are able to accurately detect common but serious conditions like hypertension and sleep apnea.
The USCF team and Cardiogram trained the AI-based algorithms to detect the two conditions based on the principles that patients with low heart rates are more likely to be diagnosed with high blood pressure also known as hypertension., while those with a certain beat-to-beat rate variability are very likely wot be diagnosed with sleep apnea. At the same time, 80 percent of people with sleep apnea - a disorder that disrupts your breathing during sleep - don't even realize they have it.
Cardiogram and UCSF's study on heart health is the third major study of deep learning in medicine, following Google Brain's results on diabetic eye disease in December of 2016 and a Stanford study on skin cancer this January.
Of the 6,115 Cardiogram users in the study, 2,230 had hypertension and 1,016 had sleep apnea. This new study shows the Watch can detect sleep apnea with a 90 percent accuracy and hypertension with an 82 percent accuracy. DeepHeart was trained on 57,675 person-weeks of data, sifting through a grand total of 140 million heart rate measurements and 63 million step counts.
Though the study was conducted through the Apple Watch, Cardiogram co-founder and study lead Johnson Hsieh believes it shows promise for most wearables that incorporate a heart-rate sensor as "they basically all have the same technology built inside", he told TechCrunch.
Rather than saying a specific wearable can detect health problems like hypertension and sleep apnea, it says majority can as long as they're paired up with a powerful AI agent like DeepHeart. It's incredibly exciting to see that the potential of wearables to let us know about risky health conditions - non-invasively - is alive and well, but this also requires the companies in charge of these platforms to deliver on software and AI as good as DeepHeart.