LISTEN TO THE PODCAST
An interview with Dr. Scott Halpern of the University of Pennsylvania on innovation in testing what works to change healthcare outcomes.
What’s driving innovation in testing and developing interventions to improve healthcare?
Dr. Halpern: We need to improve knowledge about what works and what doesn’t. To change health-related behaviors, we need to improve the evidence around the highest impact and most cost-effective interventions. We must engage those who are in positions to implement interventions to use the evidence wisely so that they’re choosing the best interventions to make a difference
It’s really important when it comes to serious illnesses. We know a lot about how people die in this country. We understand what their last months and years are like and how these experiences impact their friends and family. Yet, we know very little about how to actually improve on the current state of affairs.
Why? The problem is that most interventions haven’t been rigorously tested to determine whether they actually achieve their goals and interventions. The few that have been rigorously tested either haven’t worked very well, or have only shown promise in narrow settings.
What’s the problem with current practices?
Dr. Halpern: First and foremost, it’s a failure of academics who are developing interventions to effectively partner with health systems, insurers, and others who determine healthcare for the majority of Americans. By partnering with these organizations to rigorously test interventions, academics could better achieve their mission of improving the current state of healthcare.
It’s also a failure of health systems and insurers to learn what works and what doesn’t. Too often, organizations are overcome by the very understandable impulse to just do something. Yet a more prudent response might be to take a step back and take the time necessary to evaluate what works in a particular setting and then implement what works best.
How are you getting buy-in to changing the way we test and learn?
Dr. Halpern: We’ve been very fortunate to collaborate with some really innovative and effective health systems, employers, and insurers around the country.
Three principles of innovation drive that partnership. One is that we map our programs to health priorities that health systems, insurers, and employers face, and we are adaptable in how we apply tools of behavioral change to reach consumers.
Second, we know that these organizations are feeling pressured to show evidence of change quickly–whether or not that evidence is beyond a shadow of a doubt.
The days of proposing a 6-year randomized control trial before we’ve got any data are changing. We now can propose much shorter time horizons for seeing change. It really speaks to the virtues of pragmatic testing designs. We can get answers faster.
The third thing that we’ve learned is to go beyond randomizing at the level of an organization. Before, we might have worked at the hospital level. Now, we’ve innovated with what are called “stepped-wedge” designs. Instead of randomly assigning some hospitals to get the intervention while others don’t, all the hospitals may get the intervention, but we randomly assign the time at which they adopt it. We allow those units to serve as their own controls. We compare those who already were randomly assigned to those who are still waiting.
That’s generated a ton more buy-in than the older way of doing things where some people are going to be left out entirely.
Is emerging technology changing the way we test and learn in healthcare?
Dr. Halpern: There’s no question. There are a variety of advances on the technological front which are greatly expediting our ability to not only conduct research but to affect change.
On the research side, we have better ways of predicting who is likely to benefit from an intervention and to target our interventions according to people’s underlying potential benefit using machine learning algorithms. We also use natural language processing to go above and beyond what we can learn about individuals simply through discrete data fields.
We are seeing new technologies for monitoring vital signs and measuring symptoms remotely. We can feed those data back into a machine learning algorithm. In the future, we may be able to process those data and present it to nurses, doctors, and other clinicians in real-time. That has the potential to allow them to better serve patients before that small problem becomes a big one.
The Holy Grail is if we can use insights from behavioral change science to deliver interventions through smartphones and other technologies.