How well do digital health interventions work? A diabetes study hints at where the space is headed
Does digital health work? Years of investment and excitement have been followed by a period of intense scrutiny and skepticism as to whether digital health solutions deliver results.
Last year, the Peterson Health Technology Institute, which runs independent evaluations of digital health technologies, examined a series of digital diabetes tools and found that the overwhelming majority of these solutions deliver small incremental benefits and that their relative cost increases rather than decreases healthcare spending for an adopting population. A later Institute report about digital solutions for hypertension control yielded similar findings.
A key component of Oscar’s model is that great member engagement leads to better outcomes and higher retention. So how do we feel about this? Let’s look at our diabetes digital health data.
In short: digital health does work - it is possible to engage sicker members to get them better. But, the Peterson Institute is right: digital health interventions today are too high-cost, and that is a problem.
Analyzing the numbers
First, here’s a recap of the numbers from the Peterson diabetes study. The study quotes $64 per activated member per month (PMPM) in cost to the employer for digital health diabetes programs. The programs typically come with actual free supplies to the member (like diabetes strips), which the study estimates at roughly $14-15 PMPM. So the actual program cost charged by the digital health vendors is around $49 PMPM.
On the benefits side, are the diabetes programs able to reduce members’ hbA1c? That is a measure of blood sugar (glucose) levels, and the goal for most adults should be a hbA1c of 7% or less.
The Peterson study estimates a HbA1c reduction for activated members of 0.23% to 0.6%-points. The paper assumes that a 1% reduction in HbA1c yields 1.7% in cost savings across all payer types. That lets it estimate the dollar savings impact at $6-15 PMPM. Here is the problem: on average, digital health diabetes programs cost $49 PMPM - and while they deliver impact, those cost savings are around $6-15 PMPM. In other words, digital health costs a lot more than it delivers.
What does Oscar’s data suggest? We used one of these solutions for three years as part of our Diabetes Care program. Below is the impact on HbA1c for our various cohorts, depending on their initial HbA1c.
You can actually see an impact, which is great news: hbA1c levels for unmanaged diabetics come down. Member satisfaction of this program also tends to be high. So let’s mark this insight: digital health works, and members like it.
Looking closely, things get more complicated. For one, look at the reductions in HbA1c by various HbA1c cohorts. The overall impact for all activated members is quite similar to the numbers in the Peterson study. But this splits in a big way: the impact is large for initially very uncontrolled members, and small-to-nonexistent for previously controlled members.
For Oscar, nearly two-thirds of the members who use the solution have a HbA1c below 8%, where you don’t see much impact. That means that a third of the members in these solutions would have to save enough to pay for a program that costs most employers $64 PMPM. That’s a very tall order. (Note that we also saw a large influx of members joining the control group which made it more challenging to achieve savings.) This punctuates problem number one with digital health solutions: they are simply too expensive as currently priced and structured.
Connecting back to Oscar’s mission
How do the actual economics work for Oscar? Using a difference-in-difference actuarial analysis on our own data of diabetics using the solution vs. a control group, we actually see a $22-25 PMPM savings impact for activated members, or $8-$11 PMPM savings once our average of $14 PMPM in supply costs are subtracted.
This again aligns closely with the savings projected in the Peterson study (estimated incremental savings of roughly $9 PMPM), and it would just be too little for an employer investing $64 PMPM. (We should note that we did not pay more than the savings delivered, because of how we had set up this program.)
But here is the other problem with diabetes solutions, and perhaps with digital health more broadly. The first chart above shows another strange phenomenon: virtually all impact happens instantly, within the first three months after activation. If you’re driving members’ behavior change, how come you don’t see this play out over a much longer period of time?
That puzzled us at Oscar for a while, until we interviewed, surveyed and studied our diabetic members: in our Affordable Care Act population, we have a high degree of diabetic members not “activated” previously. Most of the members we speak to say they were not measuring their blood sugar regularly. So, while the bump in improvement you're seeing here is real, it might be much more driven by the fact that any kind of member activation in a largely unactivated group is useful to get members to bring down their blood sugar levels.
That sets up the real question for digital health: if engaging and managing your members works, then how can you do it at scale, and in a way that’s cost-effective? And how can you generate lots of touch points that matter to members, but do so more automatically, while still emphatically? Here comes the obligatory plug for AI: language models certainly enable more personalized and more frequent interactions with members (as we have shown with our campaign builder messaging to members based on individualized reasons for why they visited emergency rooms, for example).
Digital health works, but it needs to be personalized and low-cost. Only then can it actually bring down healthcare costs.
