Connected Health Predictive Analytics: A Long Road Ahead


We’re spending a lot of time at the Center for Connected Health (CCH) these days thinking about and experimenting with algorithms.  It’s part of our general interest in micro-segmenting the population and creating unique, engaging health messaging for each individual that will keep them on the path to better health.  Healthrageous is working fast and furious on this as well.  Of course, we’re not the only ones.  A number of other labs and firms are on the same journey.  The vision is compelling.

However, today when you get health related messages from your insurer or another source, they are typically public health focused.  Stop smoking!  Get your mammogram! Get your flu shot!  These three messages illustrate the challenge. I’ve been the recipient of all of them recently.  I’ve never smoked, clearly do not need a mammogram and was vaccinated for influenza in early October.

I always thought our friends on the consumer web side were doing better.  The first time you experience Amazon’s or Netflix’s recommendation engines, they tend to raise eyebrows.  Over time, the experience is less salient.  And let’s face it, it’s got to be easier to guess which type of movie I might want to watch or a book that might interest me than to predict what a really engaging health-related message might be.

At CCH we’re in the middle of an interesting trial funded by the McKesson Foundation, where we collect three types of data (a measure of readiness to change, ongoing activity data and location data) and use an algorithm to generate motivational messages based on these variables.  It’s ongoing now, so I don’t know how it will turn out, but we’re excited about the possibilities. Still, it’s only three variables and only one (activity level) is continuous. My instinct is that we have a long journey ahead of us.

I got some confirmatory evidence when I read a front-page article in the Boston Globe last week.  A Harvard professor picked up a sad fact about Google’s algorithms.  If one searches for an African American sounding name such as Trevon, Rasheed or Tamika, you are more likely to get an ad offering you an opportunity to run a background check on that individual than if you put in a more Caucasian sounding name such as Brad, Cody or Amy.

Imagine if we goofed like this in our ‘highly customized’ health care messaging.  What a disaster.  We’d lose patients and consumers by the legion.

Health messaging is complex. It needs to be highly individualized, motivational, caring and of course accurate.

Indeed we have a long way to go. Although our friends in consumer, retail and financial services industries have not quite conquered this complex challenge (witness the Google gaff noted above), they are way ahead of where we are in healthcare. So, I hope some of those bright algorithm scientists can be convinced to turn their attention to connected health. We need the talent. The problem is intellectually challenging enough and the rewards will be great.