What if we could predict which expectant mothers are going to have a preterm birth? Or which members are going to end up in the hospital if they don’t take their medications? What if we could predict which arthritis or cancer medications will help cure a patient or best manage their disease?
What if we could answer these and our other most pressing ‘what if’ care management questions with precision and accuracy? And what if we could not only predict the future – how populations will behave – but we could actively change behaviors, optimize them and their related clinical outcomes? All based on understanding cause and effect relationships.
This is the essence of our appoach to machine learning – answering these questions – and ultimately answering What will happen if I do this?
A quick definition of machine learning is a set of algorithms that can learn and make data-driven predictions or decisions, without relying on rules-based programming.
I’d like to illustrate GNS Healthcare's approach to machine learning and its applicability to care management by using the graphic you see below – it’s a pyramid that shows the hierarchy of data analytics capabilities from the most basic to the most advanced. This was published by Tom Davenport, a professor at Babson College and a N.Y. Times best-selling author of Competing on Analytics.
At the most basic are standard reports that tell us What happened? For example, what is the cost of care for diabetics in my Medicare population? And how much did it rise in the last year? The next level is statistical analysis, which drills down and answers Why is this happening? In our example, statistical analysis tells us why the cost of diabetics in my Medicare population increased in the last year.
The next level is forecasting, which extrapolates to answer the question What if these trends continue? If the cost of diabetes care continues to increase next year and the year after, what will the cost be over time? And then predictive modeling goes beyond simple forecasting to predict the future and answer What will happen next? In our example, predictive modeling can tell us where the next big driver of costs will be across the book of business in care management.
And now as we approach the very top of the pyramid, things get interesting. At the very top is optimization and inference, supported by advances in machine learning. Can we go further than just predicting the future? Can we predict future ‘what ifs’ in response to action? Can we answer the cause and effect question What will happen if I do this?
In our example, machine learning answers the question, what will happen if I deliver one specific care management intervention to my diabetic population versus another? What behaviors can I expect and what will be the clinical outcome and financial impact? What is the best future path and how can I optimize my decisions? Machine learning answers these crucial questions and enables not only an ability to predict the future, but an opportunity to change it.
Bruce Church is Chief Mathematics Officer of GNS Healthcare.