5 big questions on innovation

One of the areas I am most active in is the application of machine learning to health care, specifically interpreting individual’s social media feeds and determining their health status. This sounds odd but social media provides a unique environment where people openly discuss their personal lives: how they feel, what they eat, their activities, etc. Social solves a big problem in health data in that it is immediate where most health data are not immediate and often take months to gather and process. At the same time hospitals and individual physicians are moving from fee for service to quality based programs that place emphasis on health outcomes, not how many procedures were performed. This tectonic shift in health care creates a need for information regarding the health of people around; let’s say a hospital, not just the people who come in the front door, but the ones living miles away.

To meet this challenge I developed a system that listens to social media posts within a certain geographically defined area and deconstructs the stream of posts to predict who displays signs of having diabetes. It works by looking for word patterns in the text of the post and then matching that information to the person’s profile information. In tests involving thousands of posts, it is 74% accurate. Some of the interesting patterns that emerge is that diabetics tend to have many friends on social media – over 1,900, but they don’t tend to status very often – less than 65 times in a year. They also tend to say really funny things regarding their disease. Actual tweet:

“Lets play a game called how many times will my relatives ask about my diabetes. #byyyyeeee”

This system could be helpful to health providers who are looking to engage with at risk populations as problems emerge, not just when patients end up in the ER.

The HITLAB is also active internationally and is part of (MOTECH), the groundbreaking mHealth initiative designed to increase the quantity and quality of pre- and post-natal care in Ghana.

MOTECH uses mobile phone technology to improve maternal and child health knowledge and health-seeking behavior in rural Ghana. The program’s Mobile Midwife Initiative provides pregnant women and new mothers with information on pregnancy and infant care, nutrition, malaria, maternal and childhood immunizations, and family planning, as well as reminders to seek timely health care. The initiative offers these services in either SMS or voice option, in multiple regional languages. MOTECH also helps community health workers identify women and newborns in their area who need healthcare services, while enabling these health workers to cut down on paperwork and increase accuracy by giving them the ability to enter patient data via their mobile phone.

Given the passage of the ACA and the increasing cultural focus on wellness, we are in a very supportive environment for our services. The major hurdle that remains is data. In the United States we just don’t have a standard format for health data or a central repository to keep it in. This is unlikely to change anytime soon so we use the data we have to fill in the gaps and create as accurate a picture of someone’s health as we can. Again, machine learning plays a big role here. 

Many of us at HITLAB have classical training in the sciences (MDs, nurses, psychologists, statisticians) so we tend to approach challenges from the scientific point of view. You won’t find anyone at the HITLAB who believes there is an unsolvable problem in health.

As I mentioned earlier, access to a standard set of data is one of our limitations. The market is rushing in to fill this gap as more people create and share health data through cell phones, wearables and medical devices. Not only are these data real-time, they capture aspects of health that no one has ever seen before in such quality and quantity. For example, a detailed record of an individual’s movements and physical activity, the actual locations of where that activity took place. 

When the HITLAB takes on a grant project or health study, the team always includes people from ‘outside’ healthcare. We include musicians, artists, HR people in solving some very deep technical health issues and it never ceases to amaze me how these people from varying backgrounds contribute so effectively to our work. This practice is actually codified in HITLAB procedures.

I’ve also seen research on what motivates people to change and develop healthy habits. Traditionally, healthcare looked at people with a disease such as type II diabetes and immediately focused on their need to lose weight, which seems logical but ignores the root cause of the disease. The root cause may be something very different, the person may be lonely or depressed. To directly attack the root cause researchers offered pet adoption to type II diabetics. This may seem unorthodox but what’s the first thing you need to do with a puppy? Chase it around and walk it. Perfect! I would expect to see great progress in the field of behavior change through similar methods as the one above.