Ordinary wearables can flag signs of diabetes, according to new Cardiogram study

New Orleans, LA - Feb 7th, 2018

An interview with Cardiogram co-founders Johnson Hsieh (JH) and Brandon Ballinger (BB).

What's the news?

JH: A new N=14,011 study shows heart rate sensors you’re wearing already—like the Apple Watch, Android Wear, Garmin, or Fitbits—can detect early signs of diabetes. Researchers at Cardiogram and UCSF validated the accuracy of DeepHeart, a deep neural network, in distinguishing between people with and without diabetes, achieving 85% accuracy on a large data set which included 200 million heart rate and step count measurements.

Where and when is this research on wearables and diabetes being presented?

BB: The paper was accepted to the Thirty-Second AAAI Conference on Artificial Intelligence (AAAI-18) in New Orleans. Avesh Singh (Software Engineer, Cardiogram) and Johnson Hsieh (Co-Founder, Cardiogram) will present the research in a session titled "DeepHeart: Semi-Supervised Sequence Learning for Cardiovascular Risk Prediction", at 9am CST on Wednesday, February 7.

Why does it matter?

BB: According to the CDC, more than 100 million U.S. adults are now living with prediabetes or diabetes. 1 in 4 of those with diabetes are undiagnosed and, even worse, 88.4% of people with prediabetes don’t realize they have it [1]. While there have been many attempts to build special-purpose glucose-sensing hardware to detect diabetes, this is the first large-scale study showing that ordinary heart rate sensors—when paired with an artificial intelligence-based algorithm—can identify early signs of diabetes. By detecting diabetes earlier, we can help people live longer and healthier lives.

What other health conditions can DeepHeart detect?

JH: Cardiogram and UCSF have previously validated DeepHeart's accuracy at detecting hypertension (chronic high blood pressure), sleep apnea, and atrial fibrillation. These studies were presented at the American Heart Association's annual scientific sessions in November 2017 [5] and Heart Rhythm Society in May 2017 [6]. These health conditions are important since 19% of people with hypertension, 80% of people with sleep apnea, and roughly 20% of people with atrial fibrillation remain undiagnosed [2-4].

Who's behind this new study?

BB: The study is a collaboration between UC San Francisco's Health eHeart Study (https://health-eheartstudy.org) and digital health startup Cardiogram (https://cardiogr.am). Cardiogram was founded in 2016 by ex-Google tech leads Johnson Hsieh and Brandon Ballinger, is backed by Andreessen-Horowitz's Bio Fund, was named the best iPhone app of 2016 by iMore, and launched for Android in October 2017. The Health eHeart Study is an ambitious, online study to end heart disease which has enrolled more than 100,000 participants worldwide.

How was DeepHeart trained to detect diabetes? What were the challenges?

JH: 14,011 users of Cardiogram for Apple Watch and Android Wear were recruited into UCSF’s Health eHeart Study. Then, 33,628 person-weeks of health sensor data was used to train a deep neural network by presenting it with samples from people with and without diabetes, hypertension, sleep apnea, atrial fibrillation, and high cholesterol.

Typical deep learning algorithms are data-hungry, requiring millions of labeled examples, but in medicine, each label represents a human life at risk—for example, a person who recently suffered a heart attack or experienced an abnormal heart rhythm. To solve this challenge, researchers applied two semi-supervised deep learning techniques (“unsupervised sequence pretraining” and “weakly-supervised heuristic pretraining”) which made use of both labeled and unlabeled heart rate data to improve accuracy.

The final deep neural network contained 564,227 neural network weights and both convolutional and recurrent layers.

How, exactly, was accuracy measured?

JH: DeepHeart’s accuracy was measured using 12,790 person-weeks of data from a separate set of participants (whose data was not used during training). The final c-statistic was 0.8451 (85% accuracy), where 0.5 corresponds to random performance and 1.0 corresponds to a perfect classifier.

Why can diabetes be detected from heart rate and step count data?

JH: Your heart is connected with your pancreas via the autonomic nervous system. As people develop the early stages of diabetes, their pattern of heart rate variability shifts. In 2015, the Framingham Heart Study showed that high resting heart rate and low heart rate variability predicts who will develop diabetes over a 12-year period (https://academic.oup.com/jcem/article/100/6/2443/2829673). In 2005, the ARIC study showed that heart rate variability declines faster in diabetics than non-diabetics over a 9-year period (http://care.diabetesjournals.org/content/28/3/668.short).

Does Cardiogram have any user statistics to share?

BB: Yes. More than a quarter of a million people use Cardiogram actively, and 73% of them open the app on a daily basis. In an abstract presented at the American Heart Association's scientific sessions [7], Cardiogram's user retention was 5x higher than the first five ResearchKit apps, exceeding even Twitter or Instagram.

What types of wearables does Cardiogram work with, and where can users download it?

BB: Cardiogram is currently compatible with Apple Watch (all versions) and any Android Wear watch with a heart rate sensor, including the Huawei Watch, LG Watch Sport, LG Watch Urbane, Moto 360, New Balance RunIQ, Polar M600, Montblanc Summit, Misfit Vapor, and Mobvoi Ticwatch S&E. Cardiogram can be downloaded from the App Store or Play Store at the below links:

Where can I download the paper?

JH: “DeepHeart: Semi Supervised Sequence Learning for Cardiovascular Risk Prediction” is available here: https://www.dropbox.com/s/wk4ll59uafixk1j/deepheart_aaai_2018.pdf?dl=0

What’s next?

BB: While these research results are promising, ultimately, Cardiogram’s goal is to save lives in the real world. In 2018, you’ll see us launch new features to incorporate DeepHeart directly within the Cardiogram app for iOS and Android. Stay tuned!

References

 [1] https://www.cdc.gov/media/releases/2017/p0718-diabetes-report.html

 [2] https://www.cdc.gov/features/undiagnosed-hypertension/index.html

 [3] https://aasm.org/resources/pdf/sleep-apnea-economic-crisis.pdf

 [4] http://www.revespcardiol.org/en/prevalence-of-undiagnosed-atrial-fibrillation/articulo/90207387/

 [5] http://circ.ahajournals.org/content/136/Suppl_1/A21042

 [6] http://www.abstractsonline.com/pp8/#!/4227/presentation/11303

 [7] https://circ.ahajournals.org/content/136/Suppl_1/A21029

About Cardiogram

Cardiogram's mission is to reinvent preventive medicine with consumer wearables like Android Wear, Apple Watch, Garmin, and Fitbit.

More than 250,000 people use Cardiogram for Apple Watch or Android Wear each month, generating more than 30 billion heart rate measurements with which we've trained DeepHeart, an AI-based algorithm to assess cardiovascular risk. In a sequence of studies with UCSF Cardiology, DeepHeart has shown high accuracy at detecting atrial fibrillation, hypertension, sleep apnea, and diabetes.

Cardiogram was founded in 2016 by ex-Google tech leads Johnson Hsieh and Brandon Ballinger, is backed by Andreessen-Horowitz's Bio Fund, was named the best iPhone app of 2016 by iMore, and launched for Android in October 2017.

Cardiogram is available for both iOS and Android here: App Store (iOS): https://itunes.apple.com/us/app/cardiogram/id1000017994?ls=1&mt=8 Play Store (Android): https://play.google.com/store/apps/details?id=com.cardiogram.v1



Brandon Ballinger

Brandon Ballinger

Co-Founder
brandon@cardiogr.am
@bballinger

Leo Schwartz

Media Contact
leo.schwartz@onupbeat.com
Cardiogram
cardiogr.am

Social Links
Twitter
Facebook

Located in San Francisco, CA, United States

Use these media items in your story

Images

Previous Coverage