4 questions every value-based care company must answer

I’ve spoken with lots of value-based care leaders this year who are grappling with the increasing sophistication of value-based care customers, usually insurance plans and employers. There’s lots of variability, but selling in is getting harder. Ultimately, this is a good thing – a lot of what’s been pitched hasn’t delivered. Now we’re in a stage where customers are trying to screen out solutions that appear to have lots of evidence but don’t actually work while keeping the door open for solutions that don’t have a lot of evidence yet but are likely to work.

With that in mind, here are four questions value-based solutions should be able to answer (and which customers should be asking).

Do you save the system more money than you cost?

This may seem obvious, but it’s not intuitive for most people who have lived in traditional provider organizations. The customary framing of the value equation – quality divided by costs – suggests that if you improve quality a lot and increase costs a little, you’re in a good position. But you’re not. You’re in a quite bad position.

This is because plans and employers themselves are so sensitive to cost. If an employer’s insurance premiums go up an extra 5%, they’ve got to find that money from foregoing employee raises or increasing prices, both of which put them in a bind versus their competitors. Same with insurance plans. If their plan costs 5% more on the exchange than a competitor, they’re going to lose lots of members.

We can debate if our healthcare system should support initiatives that increase costs a little and quality a lot, but that’s a debate about how other people should spend their money. For now, we should just recognize and accept the constraints that employers and plans face.

If you’re a smaller company, you probably don’t really know if you’re saving the system money because you haven’t done a claims analysis. For now, just do a thoughtful back-of-the-envelope. For example, say you serve 1,000 patients, and you charge your customer $1,000 per patient per year. During that year, that set of patients would have had 200 hospital admissions, but your solution will prevent 20 of them. Let’s say the average admission for this group costs $15,000. You’ve just told me you cost your customer $1,000,000 and you save them $300,000.

In this example, it doesn’t matter if your back-of-the-envelope is off by 25%. You already know you’re not saving the system money. If your math has to get super squishy to pass this test, it’s time to go back to the drawing board on your solution. And if you do pass the test, make a plan to get an analytic answer based on claims data as soon as you can.

What measures matter?

In theory, this should be easy based on what I wrote above – total claims cost, including any fees charged. In practice, total relevant claims cost is almost impossible to calculate for any solution because the risk pool is unique (more on this later), and only a subset of claims costs is relevant. If you’re a solution for chronic back pain, you’re not going to impact labor and delivery in a meaningful way.

If claims costs are what matters and you can’t track your impact on them, what do you do? You pick measures that are as close as possible to claims costs while still being reliable. We think about this in three buckets at Ryse Health.

• Utilization measures (e.g., diabetes-related pharmacy claims cost per patient month, emergency department visits per 1,000 patients) – these are both large areas of spend we’re directly impacting. We’re not held accountable for someone’s HIV meds, but we should be held accountable for their diabetes drugs because we’re the ones prescribing them. Similarly, our population have frequent ED visits prior to Ryse for both hypo- and hyperglycemia. Our work is designed to prevent that, and we want to get credit for that.
• Clinical outcome measures (e.g., A1c, blood pressure) – these are core clinical measures that our customers see having a direct impact on inpatient admissions, which they want to see reduced. Because health is complex and most patients of ours have multiple comorbidities outside our scope, it’s hard to draw a straight line on admissions. But our customers are willing to put real dollars on these because they see them as important. If you can use HEDIS measures rather than custom measures, you’ll also find an easier reception.
• Clinical process measures (e.g., diabetic retina exam, foot exam) – what we really care about here is preventing blindness and amputations, both of which are awful for patients and costly for the system. But the incidence of these outcomes is so low already, it’s tough to reliably estimate our impact on them. Process measures are usually quite suspect because they’re two degrees removed from the utilization tied to them. However, in these cases there’s a tight logic line.

Measures that matter are always a work in progress. I’d be surprised if I gave you the exact same answer on specific measures two years from now. But the intent of finding measures that can be tracked reliably and have a strong connection to claims costs is essential.

How are you doing on each step of the impact equation?

People have different versions of this equation, but here’s how I think of it.

Impact = target population X

% of target population engaged initially X

% of initially engaged population with sustained engagement X

average impact per user with sustained engagement

All three of these steps are important. Unfortunately, the easiest way to solve any one step is to undermine the others. If you lower all the barriers to signing up for your solution, you’re likely to get a lot of signups from users who don’t really want what you’re providing and drop out quickly. If you’re asking patients to do hard things to improve their health, you’re probably going to drive away some who view it as too hard.

So be honest with yourself here. And be honest with your customers. I’ve seen companies that tout high engagement rates and great outcomes per engaged patient while slyly shifting the definition of engaged when making customer pitches, effectively omitting the middle step of sustaining engagement. Don’t do that. That’s bad juju.

How much does your risk pool look like a risk pool your customer cares about?

You may have heard the term “adverse selection”, when an organization (typically an insurer) has a population that is higher risk/higher cost/less healthy than the general population or other larger comparison group. The reverse of this is called “positive selection”, when an organization has a lower risk/lower cost/healthier population than the larger comparison group.

Selection bias is generally much more powerful than interventions. Take this study that “showed” 45% lower costs from One Medical than legacy primary care. Was this some miracle intervention? No. Sicker individuals actively use primary care and seldom switch doctors. People using One Medical were all switching doctors and were therefore much healthier on average.

Almost every healthcare intervention benefits from positive selection. Why? Because individuals availing themselves of an intervention are definitionally more engaged in their care than the population that is not. And few customers care about the subpopulation that is ready to download an app and spend hours a week on their care. Chances are that subpopulation would have been fine anyway.

With that said, you don’t have to mirror a payer’s entire population with a condition or meeting a cost threshold to be meaningful. For example, at Ryse, we like to compare ourselves to members with diabetes being seen by traditional endocrinology. Since we and traditional endos both get most patients from PCP referrals, there’s minimal bias. And that population is still large, sick, and expensive.

In conclusion

Value-based care is hard. It’s tempting to use magical math to demonstrate impact or “fake it until you make it”. But that’s a tough strategy with increasingly sophisticated customers. If you hone your answers to these 4 questions early, you’ll be built for the long run.