The Cost Equation for New Primary Care Models in Existing Frameworks (Part II)
TL;DR: traditional primary care costs $25 PMPM, new models cost somewhere around $50 — $75 PMPM. For insurers or employers to be interested they probably have to at least breakeven or demonstrate cost savings, so they need to demonstrate $25 — $50 PMPM of cost savings to break even, which is a large challenge. Part I of this goes through the cost structure differences; Part II outlines the challenges in demonstrating cost savings & some thoughts on moving forward.
How Does Primary Care Generate Cost Savings?
There are three primary ways I’ve seen that primary care models can save money for the overall healthcare system:
- Driving utilization patterns of downstream healthcare services to lower cost options
- Reducing existing mis-use of healthcare services
- Decreasing overall need to use healthcare services by keeping people healthier
All three of these levers present opportunities to demonstrate meaningful savings, but all three also present challenges as I’ve laid out in the chart below:
The potential of these savings buckets are huge. At Harken we demonstrated cost savings across the first two buckets and anecdotally thought we started showing evidence of the third bucket. Here’s a discussion of areas how these models could start showing cost reductions.
Cost Savings Opportunities Examples
Reducing Emergency Room / Urgent Care Spending — Bucket I / II
A number of attempts at proving out cost savings of primary care models (and most population health tools for that matter) turn to reducing ER / Urgent Care visits as a way to quantify success. The general thought is that having a relationship with your primary care doctor will lead you to call them first instead of showing up in an ER. Many people will show up to the ER today for things that can be handled elsewhere (i.e. in lower cost settings like primary care or urgent care) and your primary care doc would help direct you to those other less costly solutions.
This solution is one of the most tangible, immediate ways to demonstrate reduced utilization (and theoretically spend as a result, although most efforts stop short of showing it reduces spend). There is general agreement that our ERs are over utilized — although there is some disagreement on the extent, see this paper as an example.
While conceptually this makes sense, from a pure dollar perspective, reducing ER spend is not that meaningful. As the chart at left shows, ER spending is around $18 PMPM, which would be roughly 5% of total overall spend in a typical commercial population (this ER spend number seems high, but it comes from a data set that seems valid so lets go with it).
Even if a primary care model reduced ER spend by 25%, which would be a really meaningful reduction in ER visits, the savings generated from that would be only $4.50 PMPM ($18 x 0.25).
When you’re increasing primary care costs by $25 PMPM, this reduction in ER spend would only offset 18% of the increase in primary care costs. So, you still have another $20.50 PMPM of savings that you need to demonstrate just to break even. It’s a good start, but it’s just the beginning. It cannot be the only data these models hang their collective hats on.
This is the rub for these models — because primary care is so cheap in the existing industry framework, even demonstrating really meaningful savings on a tricky problem like ER overuse doesn’t even begin to cover the increase in costs. So lets look at a few other cost buckets to see where else savings can be derived.
Reducing Outpatient Specialist Spend — Mostly Bucket I / II
As the chart shown above demonstrates, $139.00 PMPM, or ~74% of total outpatient spend resides in the professional medical and facility buckets. This is where most of the spend going to specialists ends up.
This, for example, is where colonoscopy spend would be captured. At Harken we spent a few percent of our overall medical spend on colonoscopies. And a significant portion of those colonoscopies were done at academic medical centers, where they cost >$2,000 per procedure on average. Freestanding imaging centers on the other hand perform colonoscopies at <$1,000 on average. No difference in quality of care, but a 50% reduction in cost (this is why Anthem is refusing to pay for imaging at hospitals). If a few percent of spend alone is on colonoscopies, like it was at Harken, this would result in a reduction of overall medical spend by >1% just moving colonoscopies alone.
Another example is the number of tests that cardiologists and other specialists need to run. Some specialists require multiple visits, or multiple tests to handle patients where other specialists would do it all in a single visit without the barrage of tests. If you can perform the same services in fewer visits, costs are again significantly reduced.
Theoretically, primary care providers can help patients navigate where to use these services — i.e. they can help the member to decide to get the colonoscopy at a imaging center instead of hospital — thus reducing overall costs. If successful, this cost savings can be shown relatively quickly — referral patterns can be changed immediately.
Let’s assume that 10% of outpatient professional medical and facility spend can be cut by influencing referral patterns. Given the colonoscopy example alone could probably cut spend by >1%, this doesn’t seem outlandish. Reducing a PMPM spend of $139.00 by 10% would thus result in $13.90 of cost savings generated by the primary care model, covering off on almost half of the increase in cost almost immediately. These are the cost buckets where savings will need to be demonstrated in order to prove out these models — if that reduction could go from 10% to 20%, all of a sudden the increase in primary care costs would be paid for. Anything above 20% and these models start to work miracles.
In-Patient Spend — Bucket III
In-patient spend starts to capture where the long-term opportunity for these primary care models lies — from preventing the major events that end up getting someone admitted to the hospital.
The challenge with inpatient savings is that it’s not immediate. I think most would agree that most inpatient visits are necessary —when someone is getting admitted to the hospital it is usually for a good reason. Thus, there isn’t an immediate reduction that should come from improving efficiency in the current system like there is with outpatient. The reduction in inpatient visits will come over the longer term where the health of a population is being better managed, resulting in less overall illness.
There may be some instances where there are short term benefits from these primary care models — for example, a heart failure patient who stays on their meds and doesn’t get readmitted because they had a primary care doc that coordinated their transition in care could generate some cost savings. But the real savings generated by these models here should be longer term savings that aren’t generated in the first few months, or even years, of offering a primary care intensive model. The savings will show up in later years, assuming the model still exists at that point in time.
In the commercial population, inpatient medical spend is $65.00 PMPM per the data above. So if a model like this can eventually reduce admissions by 20%, that will eventually generate a cost savings of $13.00 PMPM. The key question is how long it takes to generate that savings, and what are the savings in year 1… 0%? 5%? Regardless, it’s not as big as the outpatient bucket, particularly in the first year. If it’s only 5% in year 1, which still would be a notable reduction in inpatient spend, that would reduce the PMPM savings to only $3.25. That makes offsetting the primary care cost increase much more difficult.
So… What Could Savings Be?
Adding up these total potential savings estimates from ER spend, outpatient spend, and inpatient spend, you get to a total potential savings of $31.40 PMPM for a typical commercial population. Thus, if this model increases primary care spend by $25.00 PMPM, it is theoretically able to offset that number and generate $6.40 PMPM of cost savings on top of that. Does that cost savings estimate seem right? Let’s look at a report PWC put together to try to estimate a similar number.
The image at left is from a report PWC authored that highlighted potential cost savings from a reduction in inpatient admissions as well as ER spending. By their calculations, $1.2 million could be saved annually in these two buckets. To put that in perspective, that savings is only $10.08 PMPM ($1.21 million / 10,000 members / 12 months = $10.08). They exclude any savings generated from reducing outpatient costs, but it gets to the same general ballpark of savings for inpatient admission reduction as well as ER reductions. So it seems like these estimates are directionally accurate. Note that if you read the report — which is worth perusing — it takes a different approach to calculating the cost of these primary care teams, which I think underestimates costs as I laid out in Part I of this.
At Harken, we set out to prove these kinds of cost savings across the buckets above. While anecdotally we were able to start demonstrating a few areas of cost savings, we encountered a lot of obstacles along the way as well. Here are some examples of those issues in action:
Challenges With Demonstrating Cost Savings
Network Issues
Downstream network issues are a large problem for these primary care models that gets in the way of demonstrating tangible cost savings. As shown above, there’s $139.00 PMPM of outpatient spend that could be impacted by these models through better and more efficient use of the network — with the example of reducing colonoscopy spend by driving them from high cost hospital owned outpatient facilities to lower cost imaging centers.
However, the primary care companies don’t ultimately own the downstream network. That network is typically owned the insurers they partner with, and ultimately the primary care practices are bound by the network contracts already in place. In many cases, these contracts prevent the primary care providers from seeing any cost data, so they are unable to actually know the differences in price of treatments at various places. (If you want to read more on this topic there’s a good article from a Kaiser Health News reporter trying to navigate cost data here).
This is one of the reasons why you’ll see primary care models publishing their numbers in terms of utilization data as opposed to cost data. The primary care practice can impact utilization more directly, but total cost is the result of utilization x price, and the practice doesn’t have control over — or any visibility into — the price side of the equation. The opacity behind provider contract rates is really problematic — the contracts between insurers and providers usually make it so that the insurer can’t share pricing info with providers.
Provider networks in general are a messy problem that make the theoretical benefits of these primary care models harder to achieve. A lot of the benefits from the primary care model are derived from the primary care physician being the ‘quarterback’ of an individuals care and directing any downstream care that is required. That is why many of the primary care companies invest heavily in building relationships with specialists outside of the primary care clinic who understand the value in this approach and are willing to work in that manner.
The issue is that most specialists in any given insurance network haven’t agreed to this approach and have their own economic interests in mind. Those economic interests make it really challenging to demonstrate a reduction in the oupatient facility and medical spend (all spend for that matter).
Practical Data Challenges
While a lot of savings opportunities make sense in theory, many of them get lost in mundane data issues that often go overlooked. In theory, it’s really easy to say, ‘we’re going to find the people who visit an ER too often and intervene to get them out the ER’. Or, ‘we’re going to identify everyone who is currently inpatient and call them to facilitate transitions of care as they leave the hospital setting’. Ok, both of those make sense intellectually — but how do you operationalize that?
It is really challenging practically to create an end-to-end intervention — meaning going from identifying a population of individuals that you want to perform an intervention on, to performing said intervention, to tracking the impact of said intervention — that actually works on a everyday basis.
It turns out it isn’t that easy to identify in real time who is in the ER too often or get a list of everyone who is currently inpatient. Nor is it to get any clean data set in healthcare for that matter. There are countless vendors that have a solution like being a part of a health information exchange to theoretically get around these issues. And those all sound great in theory, and should work in the long run. But in the short term, they just don’t really work all that well unfortunately, and so you have a lot of data issues. There is a reason why data folks spend the majority of their time cleaning data (google ‘data janitor’ for stuff like this).
And then even if you do come up with a method for identification, figuring out how to get that list into the workflow of the clinical team so that they can perform a defined intervention based off of that data is no small task either. We learned at Harken that there isn’t a great source to provide real-time data on which of our members were inpatient at any given time. When we had our clinical teams perform outreach phone calls to those members based on our list of who was inpatient at any given point in time, it was a mess. This results in really poor scenarios playing out — i.e. clinical teams checking in on a member who has passed away.
Then lastly, there’s the problem of actually demonstrating and tracking the results of these interventions. Again, conceptually, it’s really easy to envision how one might do this. The challenge is in the reality of getting an entire organization to track the results of an intervention in the same way. You try to do this and quickly gain an appreciation for how EMRs have become such a difficult to navigate mess with boxes to click everywhere. Tracking these interventions in a way that you can demonstrate results becomes nightmare-ish very quickly.
Short Term vs. Long Term Cost Allocation
This is really just a practical issue for these models to grapple with when figuring out the financial impacts. Actuaries at insurance companies price a commercial population for a one year time period. There is no meaningful concept of retention year over year in a population. That means that savings that are generated over a multi-year period essentially don’t count in the financial models of a commercial insurer. Yes, theoretically they could take into account an increasing amount of savings each year, and that would be one way to capture the benefit of cost savings.
These models look artificially worse because they do not fit into the standard year long actuarial model. The costs associated with these primary care models are relatively easily accounted for and baked into the model. However, the long term savings generated are harder to account for, as there is no ‘long term’ in a year long model. Realistically it means that for any cost savings to matter for these models in the current environment, the cost savings are going to need to be demonstrated in year 1. It might be a nice intellectual exercise to show future cost savings, but that’s not going to change how products are filed today.
The other aspect of this that shouldn’t be overlooked is that the new primary care models may actually increase the overall cost of care in year 1. This is because they are stuck performing ‘catch-up’ care. Meaning that the cost model from the population in a previous year is artificially low because people aren’t getting all of the preventative care that they need. They’re skipping the colonoscopy or other screenings in previous years. This has the effect of artificially reducing medical spend in previous years, which is used for the baseline. When those individuals enter these new primary care models, they have to catch-up on all the care that they did not receive over the past few years. The result is that these models actually spend more than anticipated in year 1. Note this is ultimately a good thing, because it should translate to reduced long term spend. However, the financial models fail to capture that, which presents a challenge.
Competing Initiatives
It is also worth noting that any insurer or employer that a primary care company partners with is going to have a whole host of initiatives already in place aimed at reducing downstream costs. Those initiatives are also attempting to improve outcomes and reduce costs. Insurers have entire population health and clinical teams internally attempting to implement initiatives to reduce costs in the health system.
This is challenging both from a data perspective and a practical perspective. From a data perspective, it is hard (more for political reasons than intellectual reasons) to determine how to allocate savings. Which initiative actually resulted in the cost savings? From a practical perspective, the insurance initiatives and primary care efforts can often conflict. For example, when trying to convince an individual to get a colonoscopy at an imaging center instead of a hospital, a nurse from an insurance company and their primary care doctor would likely take two different approaches. Coordinating those teams and approaches is a challenge.
It is doubly challenging when there are multiple initiatives because it is impossible to prove a counterfactual in these scenarios. For instance if somebody didn’t show up in the ER, how do you know for sure that it was your intervention that caused them to not show up? How do you know that if your intervention didn’t occur that they would have ended up in the ER? It is impossible to prove, which is why data is typically used in aggregate. Interestingly, at Harken we were able to use some longitudinal data on members to make a good argument (i.e. if someone shows up in the ER once a week for 3+ years and then completely stops all ER visits and they start showing up in a primary care clinic once a week, seems like something happened there). When there are competing interventions, this becomes even more complicated to address. Who knows what intervention caused the changed behavior?
The Primary Care Cost Model Assumption
It’s worth noting again that the entire cost model built in Part I of this post — i.e. that these models increasing costs by $25.00 PMPM — relies on the presumption that all other existing primary care spend drops to $0.00 PMPM. In other words — any person who uses this primary care model does not spend anything on primary care outside of the model. If existing primary care spend does not go to $0, the costs of the model continue to rise. In practice, that spend probably isn’t going to drop to $0 immediately, unfortunately. People have primary care doctors before signing up for these models, or don’t realize everything they have access to as part of this model, and still end up spending money on primary care elsewhere. It’s probably not a significant amount of money, but it is likely that traditional primary care spend doesn’t go from $25 to $0 PMPM immediately, it probably goes from $25 to $5 PMPM. This needs to be accounted for in the cost equation. This means that instead of increasing costs from $25.00 to $62.69, the costs would go up to $67.69 PMPM, which makes it all the more challenging to prove the model is cost neutral.
The Net Result
All of those impediments create a scenario where it is difficult to demonstrate cost savings versus the increase in cost caused by these primary care models. The theoretical $31.40 PMPM of savings dwindles away very quickly. These challenges are difficult to overcome, and make it unlikely it can be shown that the model either reduces costs or is cost neutral in year 1. I have a hunch this issue led, at least in part, to the demise of some of the venture backed approaches to building new primary care companies.
The Takeaways
It is a pretty challenging landscape that these primary care models face in terms of trying to demonstrate the cost savings they’d need to be implemented within the existing system. What to do about these issues? Here are a few thoughts on steps forward:
- Stop trying to demonstrate cost savings or cost neutrality from the get go. These models are more expensive than traditional primary care when initially implementing them. They should be. Primary care is underfunded in this country, and comparing these new models to the existing primary care infrastructure just undermines the effectiveness of the new model. The challenges of proving out a cost reduction of $25 — $50 PMPM immediately are huge. Focus on implementing a few core interventions that reduce cost, collect and demonstrate data on those interventions and go from there. Acknowledge that the first few years may increase costs on the whole because you’re investing in fixing the upstream that will eventually result in downstream savings.
- Target high cost populations. There’s a reason why these models are taking off in Medicare Advantage. Those members cost more than a commercial population, which makes the higher primary care spend more appealing for insurers. This is also the population where there are the most opportunities for interventions to prove these models reduce spend. Targeting the Apple / Silicon Valley crowd sounds great from an innovation perspective, but they are young and healthy and don’t spend anything on healthcare. If you can convince them to spend out of pocket on this stuff, great, but don’t expect insurers or others to pay for them.
- Start small and invest time up front figuring out the details. This all sounds great in theory — but don’t forget that the devil is in the details. Data sharing is important, hard to get right, and will drive clinicians crazy if not done correctly. Downstream network utilization management is complicated and challenging because of existing agreements and use patterns already in place. Understanding who the population is and the cost / use expectations is no small challenge either. Take the time to work through these challenges up front so that you’re not blindsided by them later. Don’t underestimate the difficulties to be had trying to integrate different data sets.
- Don’t throw the baby out with the bathwater. These models are expensive, and when push comes to shove it’s easiest to ‘give’ on the things that make these models different. It’s really hard to prove that integrating behavioral health into a primary care practice is beneficial, and it is obviously more cost effective to not have a behavioral health professional in the primary care office. But anecdotally we know it changes the game to integrate behavioral health into primary care. Giving in on those things that make primary care models different than the existing system because of short term financial pressures is a long-term losing proposition.
- Find the right partners that have a long term view and believe in this vision of the future. Commercial insurance generally has a one year time horizon on their financial models — that is the length of a policy. This is particularly true on the Obamacare Exchanges; member churn is very high and insurers generally can’t plan on retaining a member for more than a year. In that scenario, there is no incentive to reduce costs in years 3–5, as the insurer doesn’t benefit from any savings they’re generating, they’re just eating the short term costs. Instead, focus on finding partners that have a long term view. For example, this model is growing in the Medicare Advantage space, at least in part because member retention is very high, making insurers more likely to benefit from the downstream cost savings.
- Don’t try to grow too quickly. These models are new and haven’t crossed the chasm yet into mainstream popularity. The general population is not thinking about how they are desperately seeking a better relationship-based approach to their primary care doctor — it’s not a problem that the world knows they need solved. Yes, once people experience these models, almost everyone loves them, but before they’ve experienced it, they don’t understand what it is. The majority of people who purchased Harken didn’t even know we had primary care centers that they could visit. And they probably already had another doctor that they know who they’d go see if they had an issue. So, slow down those growth expectations, even when you have large employers super excited to sign up. And, as a corollary, be wary of VC funding to fund short term losses associated with building a ‘platform’ that is going to require rapid expansion. Focus on profitable growth(or at least with an eye on how to achieve profitability) from the start. See, for example, how Oak Street has grown over the last five years.
These economic challenges are going to inhibit the growth of new primary care models from crossing the chasm into mainstream healthcare delivery in America in the immediate future. Until there are either regulatory changes (i.e. allowing people to use HSA dollars for direct primary care) that encourage individuals to sign up directly or more payment model reform that makes it more appealing for insurers, the economics of these primary care models will make it difficult for them to take hold with insurers or employers. They’ll continue to grow in areas where the economics are particularly favorable (i.e. Medicare Advantage) or on the periphery in bets that people are willing to place.
