devote

We pair existing apple watch users with research partners to build the largest labeled dataset of heart rate monitor data that fosters insight in to user health status and condition.

Comment

Heart rate and features derived from wearable monitors have been proven to predict cardiovascular risk, but these features can likely predict and provide insight beyond cardiovascular risk. Yet labeled dataset availability is the bottleneck for future model development from this wearable data. In parallel, apps for social good have been proven to gain extremely large user bases, engagement, and participation.

What we do

We present a framework to build large datasets of labeled data, where users feel good about participating knowing that they are sharing their labeled, but anonymized data with impactful research institutions like the Stanford Center for Digital Health and the MIT Media Lab. We collect heart rate in real time and allow users to build corresponding datasets of labels with timestamps that are compilations of of selected labels, natural language, and audio narratives.

Users are able to engage with the community around them, see what other users are sharing about their day to day health and personal experience. A user can see how this may correlate with particular heart rate patterns or measures of heart rate variability.

Why we are better

The largest study to date using apple watch based prediction of basic health conditions like sleep apnea, hypertension, and diabetes had 14,000 users. Apple sold 3.1 million apple watches last year. If we engage and retain only 1 percent of those new customers, we will already double the user base that the next closest competitor engaged with to establish their simple cardiovascular only model.

Growth model

This framework can generate revenue through some early basic paths and has significant opportunity for growth and expansion. 1. All users are given access to the ‘freemium’ app, they can share, observe other anonymized users, and monitor their own current condition. 2. But for a few extra dollars a month, users can monitor family, elderly, and children, share that information with health care providers, and get predictive insight in to various health metrics as the model begins to develop insight with a large user base adding to the health narrative alongside their heart rate.

Where we are going

We present a skeleton of the iOS app, a proven background in data visualization, modeling, and prediction, and experience in developing public private partnerships with research institutions. We need to develop a functioning iOS front end, get on to the app store, engage a number of early that we’ve already made contact with, and begin building the user base through promotion on various media outlets (wired, sf chronicle, medium).

We went in to this thinking we could leverage our skillset to build a powerful neural net to predict previous unpredictable health conditions from basic wearable health monitoring. We realized there is far more growth to be had by shooting for that end goal, but building the path to get there by resolving existing bottlenecks for labeled data.

We’re devoted to data, users are devoted to their health, and partners are devoted to research.

Built With

  • adobe-ui-dev
  • containerized-analytic-pipelines
  • python/r
  • tensorflow-model-dev
Share this project:

Updates