How people have behaved in the past can tell us a lot about how they will act in the future.
This statement holds true in medicine. As scores of data — from medical, behavioral, socioeconomic and even psychographic sources — can attest, medical histories can say a lot about how someone will react to a recommended course of treatment.
But because these data are spread across all those aforementioned categories, insight into future behavior isn’t so clear. After all, there are vast stores of data to collect, standardize, validate and link before you can even begin the process of utilizing it all. It’s no surprise, then, that traditional treatments have generally focused on the most acute or costly symptoms first, regardless of a patient’s background or temperament.
My colleagues at OptumRx® weren’t satisfied with a cookie-cutter approach to treatment plans, so they came up with something better. They use predictive analytics to power a strategy we call Next Best Action — a framework that uses these many different types of data for effective decision making at the individual patient level.
A crystal ball for patient behavior
This article tells the story of how the Next Best Action strategy guides treatment plans for Jack and Jill, two patients who are suffering from similar symptoms: lower back pain and signs of COPD. What is novel about this approach is that it pairs up traditional treatment paradigms with principles from behavioral economics — as in, the study of human behavior and the choices that we make.
One of the key machine learning models within the Next Best Action engine is an estimate of a consumer’s “Propensity to Engage.” It’s a metric that informs us about an individual’s likelihood to perform an action. For example, if I live in Minnesota and search online for Minnesota Twins box scores every morning in the summer, read the sports section daily and have a history of attending Twins games in person, somebody with access to that information could infer that other people with similar attributes/behaviors might have a high propensity to purchase tickets to a Twins game.
Similarly, my neighbor Bryan may do the same things each morning, but he is 18 years old and moved with his family last year from Detroit. He follows both teams because they are competitors in the division but he hasn’t had a strong history of purchasing tickets. In this case, the additional data indicate that people like Bryan (with similar attributes/behaviors) may not have a high propensity to purchase tickets to a Twins game.
In order to gauge people’s preferences accurately and estimate this type of propensity, you need access to a lot of data from many different sources, you need to bring it all together in one place and you need to apply analytic models to it to find these patterns. To learn more, you can click here.
Given that example, let’s now think about the implication for health care. Offering one-size-fits-all treatments can have serious consequences from a health and financial perspective, because some patients won’t adhere to the plan. Personalization changes the dynamic. Recommendation engines like Next Best Action simply weren’t feasible until now because of the depth and breadth of the data you need to connect. We can now tie together vastly different types of data — geographic, psychographic, demographic, clinical, etc. — and we can do it at scale.
Because of that scale, we’ve now opened the door for machine-learning models to identify patterns in patient profiles and behavior. That evolution means we can accurately anticipate which types of treatment a patient is more likely to adhere to, which in turn can lead to better outcomes and lower costs.
The power of prediction
You may be familiar with the saying, “Don’t let perfect be the enemy of good.” If we can tailor the treatment plan in a way that aligns with the person’s lifestyle and outlook, we are more likely to put him or her on the path to a better, less costly outcome.
Health care providers are treating people, not symptoms — and that means there are multiple factors that should be considered when developing care plans. It’s not as straightforward as ranking the diagnoses by cost and crossing ailments off the top of the list one by one. Knowing each patient’s propensity to engage allows care providers to craft realistic and achievable plans to improve their patients’ health. Thanks to advances in data and analytics, this new approach is already making a difference for the Jacks and Jills of the world.
About the author:
Steve Griffiths, PhD, MS
Senior Vice President, Chief Operating Officer
Optum Enterprise Analytics
Steve Griffiths has more than 20 years’ experience in health analytics management, and currently heads up the Optum Enterprise Analytics organization. His main focus is driving growth and innovation through Optum products and services. Steve has a master’s degree in biostatistics from the University of Washington and a PhD in health services research, policy and administration from the University of Minnesota.