Your browser is out-of-date!

Update your browser to view this website correctly. Update my browser now


How machine-learning driven interventions can build value in health care

One of the main goals of efforts to improve the quality of health care in the U.S. is to reduce variation in care, where some patients are treated optimally while others miss out on the best of medicine. This is a key driver behind the focus on population health, which in recent decades has spawned the standards of care used in practice throughout clinics and hospitals today. Measurable improvements have been made, but in 2015, we can do better, argues Colin Hill, CEO, chairman and co-founder of GNS Healthcare. “Population is the total of individuals that make up a whole, but where is the individual in population?” Hill asked attendees at Institute 2015, the annual gathering of America’s Health Insurance Plans. “I’ll tell you where, the individual is missing,” he said. Hill cited some data on the efficacy of standard treatments for well-known conditions.

  • 40% of asthma patients don’t respond to any of the FDA-approved drugs available to them.
  • 50% of arthritis patients don’t respond to any of the FDA-approved drugs available to them.
  • 75% of cancer patients don’t respond to any of the FDA-approved drugs available to them.

Similar figures can be seen for other types of interventions: devices, care management programs, surgery and more, Hill says. “Half of the interventions that we apply today in this country don’t work for the patients they’re intended for. So I ask you: ‘Is standard of care failing us?’” It is certainly costing us, Hill argues, amounting to $500 billion in annual spending on interventions that aren’t helping people get better. Hill calls for a smarter system that capitalizes on the wealth of data available and novel machine learning capabilities to optimize treatment, reduce waste and improve outcomes by predicting risks for preterm birth, hospitalization, treatment efficacy and more. Machine learning involves employing algorithms to learn and make data-driven predictions or decisions, Hill says. To illustrate, he walks through layers of analysis, beginning with the most basic types of questions:

  • What happened? In a health care setting, this question might involve an insurer looking at spending on a subset of diabetes patients to find that costs are rising.
  • Why did this happen? Digging into data, the health plan can identify basic drivers of the higher costs.
  • What if these trends continue? Based on existing trajectories, data can be used to predict future spending.
  • What will happen next? This level of analysis might involve looking at who else is at risk of developing diabetes and then cost scenarios based on those assumptions.

There is value in all these questions, but Hill argues the next step – machine learning — is where the real opportunity lies. “Machine learning can do more than just predict the future, because what’s the point of predicting the future if you can’t change it?” Breakthroughs in inference and understanding of cause-and-effect mechanisms allow health insurers to try to modify scenarios. Consider a set of patients and everything that is known about them – age, demographics, health history, all the factors needed to understand a patient’s risk by way of predictive analytics. Take that approach and then build in machine learning capability that incorporates intervention models. That’s where it’s possible to try manipulating the trajectory, modifying risk. “This allows you to predict many future what-ifs in response to actions, and then select the best future path,” Hill says. “That can now be used to get to the holy grail of what we want in care management: Individualized ROI.” The result is insight that shows the way to invest in care – and often, machine-learning guided interventions do not align with established population health principles. But they deliver millions of dollars in savings by driving interventions that will have the greatest effect. “Member by member, intervention by intervention, we can now determine where we are getting the most bang for our buck. And it’s not just about money, it’s about health outcomes,” Hill says. “Isn’t this what we’ve dreamed of? The ability to deliver much more personalized interventions, much more personalized care and not break the bank doing it. In fact, [we can] to do it for less.” “We now have the ability to break the rules of population health, to think differently, to act differently, to drive to different outcomes.”