Black Holes, Stars & Member Engagement

Interview with Charith Peris, PhD, Data Scientist at Decision Point

Your background is in physics and astronomy, with a specific focus on using x-ray data to study black holes. How does that work relate to the work you do in healthcare and what similarities are there?

There are more similarities than you might think. First, what’s a black hole?  A black hole is a giant star that has collapsed on itself. What makes a black hole so interesting is that it possesses such an intense gravitational field that not even light can escape – so it’s basically invisible. The only way that we can observe a black hole is when other objects – like normal stars – interact with it.  When gas from a nearby star gets gobbled up by a black hole, it gets heated up and emits light, which we can observe. This interaction between the gases from the normal star and the black hole is what ultimately allows us to measure important characteristics of the black hole, like its mass and spin.

Obviously, at a glance the healthcare sphere is totally different. The similarity lies in the fact that from a distance every individual health plan member is an unknown – a black hole if you will.   Of course, we do have a lot more information about an individual member than we have of a black hole, such as his/her age, gender, race, language and so forth. But we don’t truly understand a person until they start interacting with other entities, whether it’s other people or a healthcare system. A member can undergo many interactions – visits to their doctor, hospital and ER visits, interactions with a health plan’s member services, visits to the pharmacy to fill prescriptions etc.  In the same way that I analyze the interactions of a black hole and a star in astrophysics, I analyze these member interactions to understand a member’s behavior. This enables me to get a good picture of that member’s history, and obtain great clues as to how that member will behave in the future.

What about the data?  Obviously, the data that goes into understanding black holes is different from healthcare data, but what about types of data and what the data represents?

Here’s another similarity.  In studying black holes, directly obtained data points are interesting, but not always completely representative of what will give us true understanding. Derived quantities (or calculated values) are crucial in providing true insight. For example, the spectrum  – that is the amount of light at different energies – of a black hole is a data point that can be directly obtained, but fitting this spectrum with a model and calculating its spin is much more interesting. The same is true for healthcare. For example, I may know where a member lives, and based on this draw some conclusions about their income level or employment status, but what’s more interesting is determining how far away they live from their doctor or their pharmacy.  Therefore, I can use a combination of directly obtainable and derived variables to quantify and test hypotheses on a member’s access to care.

Also, in healthcare, there are just too many data points to simply use one or two indicators of past behavior as an indicator of the future.  Just like a black hole system, an individual presents a complex network of parameters, and there’s a lot that goes into how a member chooses to live and seek health. Complex linear and non-linear models and machine-learning techniques are then needed to understand the member’s characteristics and interactions and make accurate predictions about his/her future behavior.

Other than the obvious, what is the most critical difference between black holes and healthcare member engagement?

The biggest difference – and this is what makes working at Decision Point so interesting – is the ability to change the behavior of the member.  With black holes, we’re simply observers and our goal is to understand and learn.  In healthcare, specifically the area of member engagement, simply understanding and observing is not enough.  Our goal is to use these observations, and our predictions based on those observations, to change behavior over time.  A member may be high risk for multiple admissions with complex conditions and a lack of engagement with the primary doctor. Our goal is to prevent those admissions.  A member may be predicted to be non-adherent with their medications. Our goal is to use that prediction and other observations to figure out how we can make them more consistently adherent.  How do we go about doing that?  What is the most cost effective way to do that?  These are questions that are unique to healthcare and what makes it an awesome sphere to work in.

 

Stay up to date

Keep in touch.  Give us your name and email and we’ll keep you in the loop.