Last week I did a deep dive into the new report from the federal Governmental Accountability Office (GAO) on the Arkansas private option, the state’s unique alternative to Medicaid expansion which uses Medicaid funds to purchase private health insurance for low-income Arkansans (here’s a shorter version of my take, in 23 tweets). GAO argued that in approving the private option, the feds failed to ensure that the policy would be “budget neutral” — budget neutrality means that this new experimental approach is not supposed to be more costly for the federal government than regular old fee-for-service Medicaid expansion would have been. The headline in the D-G the next morning was unfortunately misleading: “U.S. forecasts higher cost for private option.” In fact, GAO did not forecast higher costs for the private option; that’s not the question the agency was examining at all.  The question was the baseline, the “would have been” — trying to project what would have happened if Arkansas had done something the state did not actually do: expand Medicaid in the traditional way. 

The federal Department of Health and Human Services (HHS), in approving the waiver of Medicaid rules (known as an 1115 waiver) for the private option, accepted the controversial assumption from Arkansas that the provider reimbursement rates would be the same under a hypothetical fee-for-service Medicaid expansion as they would under the private option. GAO questions this assumption as a basis for federal budget neutrality, but its report didn’t offer anything new; this is the same debate about the private option and budget neutrality that we’ve had from the beginning.  
This isn’t the debate that’s likely to be most important to the legislature, which will probably focus more on whether the private option is a good deal for the state of Arkansas, a separate question. But it’s an important debate, particularly as other states consider the Arkansas approach. The private option has various potential advantages, but will it cost more than traditional Medicaid expansion? If so, how much? How should the feds decide whether to allow other state experiments? These are big questions for the federal budget going forward if additional states want to try out their own private options — or if Arkansas aims to continue down this path after the three-year life of the waiver ends. 
My previous post on the GAO report included reactions from former Arkansas Medicaid Director Andy Allison, the state official most involved in developing the policy (Allison also had a previous stint as a Medicaid budget analyst at the federal Office of Management and Budget — OMB). I had a long chat with him last week about these issues — and fair warning, as you can probably imagine, things went deep in the wonky weeds. After the jump, a lightly edited transcript of our conversation, with Allison’s take on the GAO report, the role of 1115 waivers in state experimentation, why he stands by the state’s theory of budget neutrality, and the potential for Arkansas to serve as a national experiment (as well as factors that might make Arkansas in particular unusually suited to the possibility of the private option as a cost-effective approach). 


ARKANSAS TIMES: On some level, it seems like GAO is just re-hashing the argument about whether expanding traditional Medicaid would have required raising Medicaid reimbursement rates such that rates would have been the same either way between Medicaid expansion and the private option.

ANDY ALLISON: They’re rehashing one aspect of the debate and the discussion that we had in 2013. But I would amend your statement somewhat. I don’t think that GAO engages in the debate over whether provider rates would actually have to go up in Medicaid if Medicaid was expanded. GAO is making a procedural argument against such hypothetical assumptions.


They’re not actually debating whether this particular hypothetical assumption is appropriate or not. To some extent, I think that they miss, therefore, the whole point – which is to try to project what really would happen if Medicaid were expanded. And allow Arkansas, from CMS’s point of view, to try something different and see if they can do better.

Using GAO’s logic would stymie experimentation because you really can’t take reasonable assumptions of hypothetical circumstances into account. It turns out that an expansion of the traditional Medicaid is itself a set of hypothetical assumptions.


And GAO says this, and CMS [the Center for Medicare and Medicaid Services, the agency within HHS that reviewed the Arkansas private option 1115 waiver] in their back and forth — a back and forth, by the way, that dates all the way back to my involvement in the federal review process of 1115 waivers when I was at OMB.

If you look back at the reference trail for GAO’s critique of CMS’s budget neutrality calculations, you’ll find that it goes back to 1994 when the debate over hypothetical and without-waiver assumptions began. I’d make one more point, which is that GAO very clearly indicates they would like to see CMS apply historical patterns of cost to projections. Now, in the case of Arkansas’s premium assistance waiver, that would tend to lower the without-waiver baseline. But it doesn’t always work that way. In fact, the fundamental problem when I was at OMB and reviewing a skyrocketing number and scale of 1115 waiver submissions, the problem then was that state Medicaid programs were just on the tail end of a historic explosion in Medicaid costs. And GAO’s fundamental critique was the building-in of those high costs and high growth rates to without-waiver projections. For example, the use of a 12 or more percent annual growth rate in Tennessee’s 1115 waiver budget, when one could reasonably question if that was in fact sustainable given that Tennessee was requesting a waiver precisely because it didn’t have the money to fund its Medicaid program. So this logic can go both ways.

Again, the fundamental problem here is the narrowness of the 1115 waiver budget neutrality calculation and process to support real state experimentation. Which the private option is. More or less, you can only establish waiver budgets using a fact base if a fact base is allowed to accrue through such demonstrations. Fortunately, CMS has as many as 50 states to work with. … Allowing Arkansas to experiment with a very new approach to coverage through the private market place seems a very reasonable investment on the federal government’s part, should it work in a payoff a hundred times larger across the country.

What do you mean 100 times larger?


Arkansas is roughly one one hundredth of the country.

You’re saying the payoff if the private option ends up being more cost-effective than traditional Medicaid expansion.

Yeah. It goes both ways. Number one, is this an experiment? Yes. Do we know exactly how much it’s going to cost? No. That’s why this is a demonstration and that’s why we have to rely on projections. The fundamental question of whether it’s worth it from CMS’s point of view could well be informed by Arkansas’s size and willingness to really invest in genuine reform, not only in the insurance market but in the health care marketplace.

The core of what you’re saying is that real experiments have value, and you will never be able to do an experiment if your entire process of estimating costs is historical data. If you close yourself off within those confines, you’ll miss all kinds of opportunities.

Yes, if past is always prologue you’ll never be able to demonstrate anything new.

It does seem to me that GAO is implying that the hypothesis underlying this experiment had insufficient data or proof. They’re not diving in and saying we think this won’t happen, but they’re saying that Arkansas didn’t make the case, right?

Oh, we made the case. This actually isn’t about the case that Arkansas made. If you look at the methodology of the report, they say very explicitly that they only talked to CMS or HHS officials. Had they actually wanted to review the case that Arkansas made and to review our assumptions and really evaluate the private option, I’m sure they would have talked to Arkansas officials. This really has more to do with a long-standing question and point of view in GAO’s mind [about] hypothetical costs – which by the way aren’t limited to simply growth rates and baseline costs, but that’s the subject here. That in turn goes to the fundamental nature of 1115 waivers, because again, if you cannot include any hypothetical assumptions than of course you’ll never be able to demonstrate anything of significance. I think everybody involved understands and acknowledges that the private option is a significant and meaningful experiment.

But that could also go both ways — we can acknowledge the value of experiments based on hypothetical assumptions but still worry about a scenario where those hypothetical assumptions were groundless and ended up costing the feds more money, right?

That’s not just hypothetical, I’ve watched that happen in Medicaid for many years. At scales, by the way, many times larger than this. Partly because of Arkansas’s size. And partly because we’re really only talking about the difference between one provider rate and another — compounded with the difference, potentially, with how those costs and services are administered. Which is an order of magnitude smaller in scale than, for example, a hypothetical assumption about whether a state could continue to spend a billion, or 3 or 4 billion dollars more, in supplemental payments to public hospitals when CMS or Congress has just invalidated the funding source for those. Which is more or less how Romneycare got started.

There’s an inherent tension with 1115 waivers — state experimentation versus policymaking beyond what Congress intended.

Absolutely. If you go back to CMS’s first policy statement about the private option and about premium assistance in the Marketplace, it said two things: one, we expect to approve only a handful of these. Two, we’re only going to approve them for three years. That wasn’t because they only had three years worth of confidence. That was also because CMS’s waiver budget math will change beginning in 2017 with the Section 1332 Wyden waivers, when the premium impact and other non-Medicaid impacts can be taken into account.

Beginning in 2017, GAO would have to take the full effect of the private option into account. I think this takes us back to a very important distinction between GAO’s review and our own legislature’s criteria when it first approved the private option. You’ll recall the debate over whether it would save the state money didn’t last very long. It was plainly obvious that it would save the state money. Immediately the question became – is it going to work? Is it going to save the federal government money over time. Is this experiment worth it to the federal government? That isn’t the question that was posed to GAO, that isn’t the question that GAO answered, and I think limits the relevance of GAO’s methodological or regulatory critique of CMS to Arkansas’s actual policy. If taxpayers are the point — federal or state — I think we’re going to have look somewhere else for a real assessment of the private option.

GAO says, “HHS approved a spending limit for the demonstration that was based, in part, on hypothetical costs—significantly higher payment amounts the state assumed it would have to make to providers if it expanded coverage under the traditional Medicaid program—without requesting any data from the state to support the state’s assumptions.” 

They didn’t have to request it, we’d already published it all. Long before the waiver was actually submitted. As you know, because you were the one asking, among others. Those assumptions were all crystal clear and public, as they should have been, because our legislature demanded them.

Okay, but here’s where we always get a little bit stuck. You know that there are lots of people – informed experts – who think that assumption was not warranted. You’re aware of that.

Absolutely. And that would have been a very interesting thing for GAO to have focused on. They didn’t choose to. That’s a legitimate debate, which, by the way, likely only an experiment and time and variation across states will demonstrate. That’s what it takes to establish the fact base that GAO would like CMS to always rely on.

I take your point, but my read on GAO is that they’re implying that yes, you explained where your assumptions are coming from but that you didn’t make a sufficient argument for those assumptions, or the feds didn’t demand a sufficient argument for it. 

But that’s GAO’s opinion, that’s not something they audited. If they had  audited it, the report would have had all the assumptions, a reasoned consideration, opposing viewpoints. That’s just not what they looked at. They really focused on process and definition.  It’s hard to know [what GAO looked at]. It’s not written into their report. This [debate] was all very public — and is still accessible on the web. That sort of demonstrates the point that GAO’s review was more of a process audit, not a substantive review. A substantive review would have taken into account all available information. A process review would intentionally narrow itself to the 1115 process itself.

The core issue here is that y’all proposed a theory of how rates would operate in this counterfactual world of traditional Medicaid expansion that didn’t happen, and that HHS found that basic line of argument plausible.

Yes, and by the way, that has now been written into a second waiver. Pennsylvania’s waiver is based on exactly the same logic, except in that case they did away with the numbers entirely and just accepted the logic, thereby saying it’s by definition budget neutral.

GAO is at least implying that it projects the private option to cost 24 percent more than Medicaid expansion. That’s what it all boils down to — GAO says we don’t have enough to run with your assumption and the feds say they did have enough to run with that assumption.

You know, I think it’s slightly different than that. I think GAO is saying 1115 waiver demonstrations should not be used to experiment with new programs and policies. Therefore only historic patterns of cost and growth should be used to project their cost. CMS believes that demonstrations should allow for experimentation, therefore requiring hypothetical projections be used essentially on both sides of the equation, with and without the waiver. I do think having read through it, it would be a mistake to infer…it may be the case that that’s what the authors of the report believe. I don’t know. The report isn’t about our assumptions. It’s really not about the theory of the private option. It really doesn’t get to the heart of the private option. It’s about GAO and their view of 1115 waivers versus CMS’s, and that debate runs across party lines through presidents dating back at least to George H.W. Bush.

But they are explicitly saying that the feds should not have deemed this budget neutral, and the way they got to budget neutrality was via an assumption that in GAO’s view they didn’t request proper data or documentation for.

The issue is that they don’t really identify which documents and assumptions, etc., CMS should have requested. I hate to over-read GAO’s findings here. What is clear to me is that they have a different definition of budget neutrality. And plainly a different assumption about the nature of 1115 waivers than CMS (in every administration).

The assumption that is a point of debate here is a theoretical hypothesis. GAO might say it would be perfectly reasonable for a state to say, ‘we assume this would be the cost under traditional Medicaid expansion and here’s X, Y, and Z reason that goes beyond historical rates. Here’s the experience of State X, and here’s data on it.” Which is different in kind than what Arkansas did, which is to say, we have this big-picture theory for how we think things are going to work in practice in Arkansas if the state expands Medicaid. 

The construct of your comparison, I follow it. The content though, belies the argument. You and I have had this conversation. GAO only alludes to it in a footnote. Remember that the federal standard — mandate and entitlement — within Medicaid…Under federal requirements for those who are funded by Medicaid, individuals need to have access at a level which is equivalent to others in their geographic region. That is a critical flaw in the commercial rate critique, which we faced in Arkansas from day one. I don’t even remember how many people I have had that conversation with, at the national level, press, CBO, administration, etc. Not the GAO yet, but maybe they’ll call. 

I think it’s a fairly powerful argument — what level of payment does it take to get you that level of access? States have struggled with that in court for years, including in Arkansas, where we’re under consent decrees as a result. It really is an open question, when you’re expanding coverage at the proportion that Arkansas did – now more than 200,000 individuals, a fifteenth of the state, it is a very reasonable question whether or not you can secure access from the state’s providers at the same level of payment. Providers had a certain balance between private and Medicare and other forms of payment such as Medicaid, but when you increase coverage by that amount – to a different population, by the way, with a different pattern of use by providers – it is not at all reasonable to assume that those new providers or that increased payer mix of Medicaid within their business model, fits. That’s actually a pretty aggressive assumption.

It’s not aggressive in a state with a proportionally lower rate of poverty and a higher historical base of Medicaid coverage. But our expansion was very large in proportion to the size of the state, the provider community, and overall health care spending. And I think raises a real question about where rates should be in the end, and about where they would have had to end up were Medicaid simply expanded.

In the scenario where Medicaid was expanded with traditional Medicaid expansion—given all the things that you’ve said—if the rates were too low and access was too low, what would the trigger be? A lawsuit?

Yes, it could be a lawsuit or simply a mutual acknowledgment by providers, legislators and the state that a provider tax should be increased to fund an increase in Medicaid rates. That general taxes should be increased in order to accomplish that. These are things that have actually happened in this state and in other states repeatedly over time. The only question is what rate increase of change might a very proportionally large expansion cause.

This is the part where people get skeptical. There are lots of scenarios where Medicaid really does under-reimburse because of state budget realities and political realities. Isn’t it possible that—at least unless and until someone sued—that the state could just short the reimbursements?

In economics one of the things empirically they would teach us as we did the calculus of change is to be very careful not to assume the calculus applies to large-scale step function changes. At the margin is not at all the same thing as an increase in coverage of 200,000 people.

In other words, there would be a kind of crisis that the legislature would almost have to react to?

It’s happened before and will happen again in publicly funded programs like this.

You’re arguing that, at least in Arkansas, it would be a shock to the existing supply of providers.  

No question about that. Let’s just be very pragmatic about this. There is a social contract implicit with providers in almost any state when it comes to the Medicaid program. It’s an alternative to free. It’s an alternative to uncompensated care. It isn’t necessary to assume that it takes commercial rates to pay for all individuals in the Medicaid program. But the second issue is what’s your starting point. We had an amount of access and we had relative stability. I wasn’t getting sued as Medicaid director with the scale of program we had pre-expansion. It’s just as difficult to argue that we could add 200,000 without upward pressure on price as it is for us to argue that it would have to go all the way up to commercial. And we didn’t. We did not ever contend that rates would have to rise all the way to commercial. It’s not clear whether GAO understands that or not.

You assumed that it would have to be the same—that the two numbers would meet somewhere in the middle?

Right. Let’s just stick with what we thought we’d spend on the private option. Remember what our assumption was about those underlying provider rates. We never assumed they would have to rise all the way to commercial. We assumed that they would have to rise to something in between Medicaid and commercial. That’s a pretty critical point, which underlies the assumption that we may in fact come to see in Arkansas that the rest of the Marketplace benefits from this blended market. Blended market leads to blended rates. It’s not clear whether GAO gets that because their language suggests – the language just says “commercial rates.” It never refers to any blend whatsoever.

You didn’t purport to estimate where in between it would fall, simply that it would be the same?

We had to pick the number. That’s why we have a waiver cap. Underlying that is, I believe it was a 5 percent weighted average provider rate decrement versus commercial.

If rates are lower than expected in 2015, would that generally signal the sort of thing you’re talking about?

Yes. It’s not just about the medically frail. It’s not just about risk segmentation. The risk segmentation is useful mainly in calming the fears of private actuaries who’d never before seen the private option. Long-run, health risk will drive cost no matter what. The medically frailty diverts those who need extra care to the source of that extra care and also provides certainty to traditionally risk-averse people like private actuaries. But in the end the premiums will go up or down based on actual risk, and of course Medicaid costs for those medically frail will go up or down. That’s not what we argued was the source of the premium savings. What we argued was if you have a blended insurance market, you will end up with a blended provider rate as well. And that’s where the premium savings actually come from.

Our argument was, there might be a little more spend on the Medicaid side, but the feds get it back on the premium side. GAO didn’t focus on this, maybe they will in a future report. What they might have focused on is the inadequacy of 1115s to handle the real federal budget impact of an experiment as bold as the private option, which it doesn’t. Because you can’t cross that boundary at 138 percent of poverty. The Whyden waivers and the 1332 budget math are big enough for it. We all get to re-do this in 2017 and beyond.

In some ways we’re in exactly the same position we were [before the GAO report]: There’s an experiment happening. Y’all have a theoretical framework for how you believe this will function as compared to traditional Medicaid expansion. The obvious way to compare that would be a comparison with other states. And we’ll see what happens.


GAO just says, here’s what the cost would look like if you take out the theoretical assumption that the state made. 

It’s a little bit like saying, let’s assume that the Arkansas private option was to fail. Does Arkansas’s 1115 waiver meet the budget neutrality test? Well, no. Not if it fails.

What do you mean?

They assume historic spending in Medicaid. That means the private option fails. They assumed the result in their math. That’s why I would call it definitional, not substantive.

Well, you could have a scenario where you stayed under the budget neutrality cap, but you did a comparison in three years time to other states that did a traditional Medicaid expansion, and you might conclude that the theoretical idea of what would happen was wrong.

Here’s the issue. We will never, looking solely at Arkansas data and experience, we’ll never know whether the private option worked. Other than we obviously know it has engendered dramatically more competition and access to insurance than has existed. But there’s no way — because our experiment affects the entire state — we’ll never have a counterfactual, we’ll never know what an expansion in Arkansas would have looked like with just a traditional Medicaid and if we had tried to do that without raising provider rates. We’ll never know that. The only way to demonstrate this — and the real test, empirically, that will be done of the Arkansas private option — will be taking advantage of differences of how states expanded and when they expanded.

Right. But it is possible — granting that you will never know the true counterfactual looking solely at Arkansas data — if we look at comparisons to other states, it is possible that we could conclude that the private option was more expensive than Medicaid expansion would have been even if it stayed under the budget neutrality cap.

Sure, that’s possible.

That seems to me to be what GAO is saying. The test itself was set a level that was too high.

Yeah, but here’s the difficulty in accepting the legitimacy of that interpretation of GAO’s argument. I’ll restate it to you: “Because an experiment might fail, it shouldn’t be done.”

But I think what they’re saying is that the way we’re going to test whether the experiment was a failure or not –they’re saying that the test itself is insufficiently rigorous.

No, they’re saying that the test of the rigor of the projection was insufficient. Then they say that the projection should be based on the fact base — that is demonstrated trends, not hypothetical trends. Which, in other words, means, if you have not already proven that the experiment succeeds, then you cannot do it. Therefore they have a definitional problem with the scope and nature of 1115s. That’s really what’s going on here. Because you cannot allow large-scale demonstration and state-level experimentation without experiments. That’s definitional.

I think it circles back to this tension that we talked about at the beginning about experimentation versus policymaking. In general, with 1115 waivers, if we’re saying that the aim is for them to be budget neutral, we want to make sure that you are testing that budget neutrality in a reasonable way.

Absolutely, that’s about evaluation. The test is the evaluation. We do the best we can using Arkansas data. But what I have said repeatedly to CMS – how should CMS evaluate the effectiveness and the impact of the expansion? I said, don’t ask states to do anything. Just collect the data. Highly detailed, time-stamped data from all 50 states and use the variation in what they do to demonstrate the results of its effectiveness in each state. Because the individual state-level evaluation is fundamentally flawed because of the lack of a counterfactual. There’s no control group within Arkansas for Arkansas’s statewide decision – and by the way, the same applies to payment reform. You cannot demonstrate or test or experiment with payment reform. For exactly the same reason. You can’t ask a doctor to do something for 5 or 10 or 15 percent of their business for three years and then fundamentally change it. They’re going to unbuild the rooms they just built? They’re going to fire the case managers and nurses they just hired?

We do have an unusual scenario for looking at expansion — not that it’s perfect, but you will have the experience of Mississippi not expanding at all, the experience of Kentucky expanding in a traditional way, Arkansas expanding in this unique way.

Those governors [in non-expansion states] better hope this doesn’t work. Because we will eventually know the impact of their decisions, as well as ours.

These policy debates we are having now are not theoretic. They are real. Real money. Real health care. Real people. Real differences in impact on the poor, on government programs, and on private insurance markets. The data will come, and so too will accountability for me and for policymakers — in states that expanded, in states that didn’t, and in states that expanded in a very different way like Arkansas

We will all be held accountable. For me, that accountability will be to cost, to market impact. For governors in non-expansion states that accountability will be to those who would have benefited from expansion, foremost low-income adults. 

We have the potential for a big national experiment.

I know it was self-serving, but at the time, I made the argument in 2013, of all the experiments that I’ve seen, [the private option] is not only conceptually promising but imminently testable — in a market one one hundredth of the country.