Some challenges linking medication adherence to medical cost offsets include the potential of reverse causality, omitted variable bias, how adherence is measured between studies, and the outdated nature of some of the most-cited literature, said Ben Urick, PharmD, PhD, principal health outcomes researcher at Prime Therapeutics.
At AMCP 2023, Ben Urick, PharmD, PhD, principal health outcomes researcher at Prime Therapeutics, explains some of the various challenges with existing literature associating medication adherence improvements with medical cost offsets.
What are some challenges with the current literature associating adherence improvements with medical cost offsets?
Current literature looking at the relationship between improvements in medication adherence and medical cost offsets has several aspects that I think we should be aware of if we're thinking about using that literature for modeling medical cost offsets from existing adherence improvement programs.
First, from a design perspective, some of the studies used, for example, a single-period design where you're looking at adherence in the same year that you're looking at changes in medical spending. One challenge with that is there's a possibility of what's called reverse causality. A great example is, you would assume that taking statins as the prescriber intended—which is to say, being adherent—would eventually reduce your health care spending. But if, for example, you are starting on a statin medication, you're not adherent to it, you have a heart attack, the cardiologist says to you, "You need to keep taking this statin." Well, that's a pretty big wake up call for you, and you probably will be adherent for at least a period of time after that.
Now, that's an instance where an actual health care event causes adherence versus the other way around. Using the same time period that you're looking at for adherence measurement as medical spending, again has that threat of reverse causality.
Another piece is what's called omitted variable bias, which is a big broad term from [epidemiology] and econometrics. But really what that means is a lot of the studies don't necessarily control for common factors that influence adherence. One of the most common factors is what's called healthy adherer bias. So healthy adherer bias is this concept that people who are more likely to practice healthy behaviors are more likely to be adherent, and those healthy behaviors can independently influence your health care spending.
For example, if you are somebody who takes your statin all the time, you might also exercise regularly. If exercising regularly, independent of your statin adherence, has a positive effect on your health care spending, you would expect that that health care spending would be lower if you exercise. So, trying to account for that healthy adherer effect can be pretty important, and a lot of the current literature does not. So, that's kind of from a design perspective.
Current literature is also quite old. So, for example, there's a study by Sokol et al that was published in 2005 that used data from the end of the '90s. That study still gets cited almost 100 times a year, and people like it because it shows a strong relationship between improvements in adherence and reductions in medical spending. But again, that's data from the end of the '90s, which is a completely different landscape—from a health care spending perspective, from a medication perspective—than we have today.
Other literature—for example, when Medicare looks at this, they look at 3 studies for the National Impact Report—all of those studies use data from the mid-2000s to make their estimates, again which is a pretty different time period, which these studies again are quite old.
The third piece of this is, the way in which adherence is actually measured within the current literature is different than the way adherence is measured as a part of adherence support programs for most health plans and most PBMs [pharmacy benefit managers]. Most health plans and most PBMs structure their adherence measurement around the Medicare Part D Stars Ratings measures, which use specifications endorsed by the Pharmacy Quality Alliance. And that works great from a health plan perspective. From the Part D Stars perspective, these are sort of the measures to which you are being held to account if you are if you're a Part D participating plan.
But the challenge is, when trying to take those measures and implement this as a part of a study, epidemiologists and economists and others doing this work make changes to these measures to better fit the study design and to, for example, make a person who definitely has hyperlipidemia but isn't taking a statin, that person could be considered nonadherent when the Part D Stars measure would not even have them in the measure. So, those types of different decisions can greatly affect the relationship that you observe between adherence and medical spending. You need to be very aware of how the existing literature is actually measuring adherence if you're going to use that to model medical cost offsets for a program that is based, for example, on Medicare Part D Stars.