By KIP SULLIVAN

Imagine that a drug company released a “study” that claimed to find that if all 75 million Americans with high blood pressure took the drug company’s hypertension drug the nation’s annual medical expenditures would drop by $20 billion. Imagine as well that the “study” failed to take into account the $40 billion cost to patients and insurers of buying all those hypertension drugs. Such a study would be roundly criticized for failing to take into account an essential component of cost – the cost of the intervention that led to lower medical expenditures.

But studies like the hypothetical drug company study appear constantly in the health policy literature. Almost all peer-reviewed papers that examine managed-care interventions – HMOs, ACOs, “medical homes,” “value-based purchasing,” etc – fail to report the cost of the intervention. Instead, they measure only medical costs or medical utilization rates. If they find that costs or utilization rates fell, the vast majority of studies imply or come right out and claim that “costs“ went down. This unethical practice is so widespread and so chronic I propose we give it a name. I propose we call it the “free lunch syndrome.”

In this post I will present four case studies of the free lunch syndrome. The four studies I will examine were written by experts affiliated with Harvard, the Commonwealth Fund, and other well known institutions, and were published in highly regarded journals. They examined managed care proposals elevated to law-of-the-land status by the Affordable Care Act. I will close with a call for research on the incentives that induce so many health policy experts to commit the same mistake over and over.

Hospital readmissions and the free lunch syndrome

Case study number one is a paper  on Medicare’s Hospital Readmissions Reduction Program (HRRP) published in the December 2016 Annals of Internal Medicine. The authors, Jason Wasfy et al., reported evidence that the program has caused reductions in 30-day readmissions of patients hospitalized for acute myocardial infarction (AMI), congestive heart failure (CHF), and pneumonia. In a blog comment  posted a few days after their paper was published, Wasfy et al. claimed, “Our research suggests that penalties such as those imposed by the Hospital Readmissions Reduction Program can play important roles in improving performance [and] lowering costs….”

But Wasfy et al. made no effort to determine what it cost the hospitals to achieve these reductions and whether patient health was improved. Nevertheless, the authors thought it was ethical to claim the HRRP program is “improving performance [and] lowering costs.”

Commonsense and anecdotal evidence tells us that whatever hospitals did to lower readmissions was not performed by elves working for free. Whatever it was the hospitals did (Wasfy et al. admitted they had no idea what that was) cost money.

I wish I could offer you peer-reviewed evidence on what it is the hospitals did to reduce readmissions after July 9, 2009, which is when CMS added AMI, CHF and pneumonia readmission rates to its Hospital Compare Website, or after October 1, 2012, which is when CMS started punishing hospitals for “excessive” AMI, CHF and pneumonia readmissions. But I can’t. US policy-makers and health policy analysts are relentlessly incurious about the costs that pay-for-performance and other managed care schemes force providers to incur. I can, however, offer anecdotes. Here are two.

On October 5, 2012, four days after CMS begin punishing hospitals for “excessive” readmissions under the HRRP, the Medicare Payment Advisory Commission met to discuss yet another evidence-free, “value-based payment” fad – punishing hospitals for “potentially preventable admissions.” Near the end of the meeting Commissioner Peter Butler, a recently retired executive of Rush University Medical Center in Chicago, offered three examples of interventions hospitals deploy when “value-based payment” schemes are imposed on them. Here are the last two:

Another situation, 25 years ago I was in, where you had a very highly incented, primary care motivated, capitated Medicare product…. [T]he ER doctors called the participating primary care physician to come in … before it registered an ED visit to help avoid the admission. It was almost – it was over the top. And if they didn’t get there within the 20 minutes, it converted to an ED visit. I mean, it literally was that kind of management, suggesting the payer incentives do make a difference, to an extreme in that situation.

The third situation, I was in [was] a large capitated system. We actually had chest pain clinics set up in ED … to manage the capitated business to help avoid the admission. [pp 137-138, transcript of October 5, 2012 meeting ]

Butler’s anecdotes confirm what anyone with common sense knows – hospital responses to the HRRP and other P4P schemes create costs for someone. The lunch is not free. [1]

ACOs and the free lunch syndrome

Another 2016 Annals of Internal Medicine paper  is the subject of my second case study. This paper bore the extremely misleading title, “Savings from ACOs – building on early success.” The author was J. Michael McWilliams. I have already criticized McWilliams for modeling a simulated version of Medicare’s MSSP ACO program and claiming he studied the real program (see my last post on THCB). Now I have an additional criticism to make: McWilliams has free lunch disease.

McWilliams begins “Savings from ACOs” by claiming Medicare’s MSSP ACO program cut Medicare’s costs by 0.7 percent in 2014 (after taking into account CMS’s bonus payments to the ACOs). His only authority for this is his ACO simulation study , the one I criticized in my last post. According to the CMS data I quoted in that post, the actual MSSP program did not cut Medicare’s costs by 0.7 percent as McWilliams would have us believe, but rather raised Medicare’s costs by roughly two-tenths of a percent over the 2012-2015 period.

After citing that 0.7-percent simulated savings figure, McWilliams steps deeper into the simulated reality he has constructed by arguing that his 0.7 percent estimate is actually an underestimate, and that the “real” (simulated) savings figure is 1.6 percent. McWilliams presents three arguments for this claim:

(1) “ACO contracts probably also affect care of patients that are served by ACOs but not attributed to [them]”;

(2) “ACO spending reductions … reduce ACO benchmarks [for succeeding years]”; and

(3) “Spending reductions by ACOs similarly lower Medicare Advantage spending because payments to Medicare Advantage plans are tied directly to local fee-for-service spending.”

Arguments 1 and 2 are debatable (we have no evidence that ACOs apply ACO magic to all their attributed patients; we have only mixed and debatable evidence for the argument that non-ACO providers adopt ACO magic; and we have no reason to believe that ACOs will continue to participate in ACO programs if insurers keep driving benchmarks down year after year).  But for now let’s assume all three arguments are valid and examine two more fundamental defects in McWilliams’ claim.

First, as I noted above, real-world MSSP ACOs drove Medicare’s costs up, not down. If we apply McWilliams’ three creative arguments to the higher costs generated by the real-world MSSP ACOs, we may conclude the ACOs raised Medicare’s costs by more than double the two-tenths of a percent reported by CMS, or roughly half a percent. Second, McWilliams labors under the free lunch illusion: He thinks that whatever it is ACOs do to lower medical spending, they do with doctors and nurses who work for free, computers that cost nothing to buy and run, etc.

I wish I could tell you we have excellent research on what it costs ACOs to start up and maintain operations, but, as is the case with the intervention costs of hospitals responding to the HRRP program, I must report that ACO advocates and analysts do not give a fig about ACO start-up and maintenance costs. They have spilled oceans of ink about the tiny effect ACOs have had on medical costs (and a thimbleful of quality measures), but for some reason they just can’t find the time to report on the start-up and overhead costs ACOs incur to achieve those tiny effects. [2]

I can, however, report that MedPAC’s staff tells MedPAC that ACO intervention costs equal 1 to 2 percent of ACO medical spending. [3] If we add that 1-to-2 percent to the half-percent increase in the real-world Medicare costs we derived from McWilliams’ clever arguments, the total damage to the health care system (not just Medicare) is 1.5 to 2.5 percent – half a percent increase in the real-world costs incurred by Medicare plus 1 to 2 percent in costs incurred by the ACOs. [4]

ACOs, “medical homes” and the free lunch syndrome

My third case study is a paper published in 2015 in the New England Journal of Medicine  by David Blumenthal and two of his colleagues at the Commonwealth Fund which reviewed the impact of the Affordable Care Act five years after its enactment. In that paper Blumenthal et al. manifested an unusually severe case of the free lunch syndrome. Blumenthal et al. claimed the Pioneer and MSSP ACO programs, and CMS’s Comprehensive Primary Care Initiative (CPCI,“medical home”) demonstration, had all saved money, and they quoted specific dollar figures for each of the three programs. But in all three cases, the dollar figures represented gross savings only. If Blumenthal et al. had bothered to take into account CMS’s payments to the ACOs and “homes,” they would have had to report much lower savings for the Pioneer ACOs and that the MSSP and “home” programs actually raised Medicare’s costs. And when costs incurred by ACOs and “homes” are taken into account, even the Pioneer ACOs are probably raising total costs.

Here is an example of Blumenthal et al.’s free-lunch logic. They stated the CPCI “has reduced monthly Medicare expenditures per beneficiary by $14, or 2 percent,” but conveniently failed to mention that CMS paid the “homes” $20 per beneficiary and the net effect, therefore, was a loss for CMS (see p. xvi of Mathematica’s first-year evaluation , the very document Blumenthal et al. cited). Free lunch disease does not get much worse than this.

When Ted Marmor and I submitted a letter  to NEJM pointing out that Blumenthal et al. failed to mention CMS’s payments to ACOs and “homes,” Blumenthal et al. replied with more disingenuous arguments. They claimed that they did mention the $20 per beneficiary payment in an online appendix to their article and, in defense of their free-lunch ACO “savings” estimates, they misrepresented a CMS document Ted and I cited. [5] They just could not bring themselves to discuss the issue Ted and I raised, namely, it is extremely misleading to conceal or ignore the intervention costs of ACOs and “homes” and to claim that the gross savings represent net savings or losses.

The Alternative Quality Contract and the free lunch syndrome

My last case study examines a 2012 Health Affairs paper  by Zirui Song and seven colleagues. In this paper, Song et al. claimed repeatedly that the Alternative Quality Contract (AQC), an ACO set up by Blue Cross Blue Shield (BCBS) of Massachusetts in 2009, “lowered medical spending.”

Like Blumenthal et al., Song et al. failed to measure (1) the shared-savings and other payments BCBS made to the participating physician groups and (2) the costs those groups incurred to provide whatever services it is ACOs provide. However, unlike Blumenthal et al., Song et al. had the integrity not only to mention the bonuses and other payments BCBS made to providers, but to say as well that those payments “probably” exceeded the reductions in medical costs. “[T]otal payments to groups from Blue Cross Blue Shield of Massachusetts, including surplus sharing, quality bonuses, and infrastructure support, probably exceeded the savings achieved by most groups that year,” wrote the authors.

But despite having the backbone to warn readers BCBS’s payments to providers “probably” exceeded the reductions in medical costs, Song et al. couldn’t refrain from stating repeatedly – first in the title, then in the abstract, and then in the text – that the AQC cut “spending.” And when Song et al. were challenged by a letter to Health Affairs from Rachel Nardin et al., Song and co-author Michael Chernew responded with bafflegab. In their letter, Nardin et al. noted what I have just stated – that Song et al. claimed BCBS’s “spending” went down when in fact, as Nardin et al. put it, “[BCBS’s] total costs under the AQC went up by some undisclosed amount, not down.”

But like Blumenthal et al., Song and Chernew just refused to concede that conflating gross with net spending is misleading and that a few edits would have avoided this problem. Instead, Song and Chernew argued that conflating medical with total spending is ok because (forgive me, this will make no sense, but I’m merely the messenger here) readers want to know how physician “behavior” changes in response to ACO incentives, and readers can’t know that unless Song et al. celebrate the reduction in medical spending and ignore or downplay the intervention costs.

This is free lunch disease in its worst form. How is the paper by Song et al., and Song and Chernew’s response, any different from the laughable hypothetical drug “study” I described at the outset? It isn’t. Like the hypothetical drug company, Song et al. went out of their way to induce readers to think the intervention in question created net savings when in fact the reverse happened – total spending went up when the intervention costs were included – and when they were challenged, they dodged the issue and spouted nonsense.

Etiology of free lunch disease

The relentless spread of free lunch syndrome among “value-based payment” advocates and allegedly objective analysts cries out for analysis. What causes this form of groupthink? What are the incentives that cause so many intelligent men and women to pretend that the extra services provided by ACOs, “homes” and other creatures from the managed care menagerie are free or are so inexpensive they can be ignored?

Most analysts who publish in health policy journals, especially ACO proponents, are obsessed with incentives, especially financial incentives. They routinely accuse doctors and hospitals of caving in to financial incentives at the expense of payers and patients. But managed care proponents show utterly no interest in looking in the mirror and asking what incentives influence their profession and whether those incentives might be warping their judgment. The prevalence of the free lunch syndrome is circumstantial evidence that health policy entrepreneurs and analysts are influenced by a common incentive or set of incentives. I believe money, tenure, and status are among those incentives. Research on my hypothesis is at least as important as research on the incentives that influence physicians and hospitals. I urge the health policy community, including the foundations and other institutions that finance health policy research, to get on with it.

[1] Here is another comment by a hospital executive about hospital responses to “value-based performance” schemes. The executive in this case was Dr. Lara Gotein, medical director at Christus St Vincent Regional Medical Center in Santa Fe, New Mexico. In a 2014 article  in JAMA Internal Medicine in which she commented on a paper that reported on the perceptions of CMS’s “quality” measures held by hospital executives, Dr. Gotein observed that hospital responses to the HRRP and other P4P schemes include short-changing patients whose care is not being measured, and changing coding and documentation policies. “Lindenauer et al surveyed hospital leaders (chief executive officers and executives responsible for quality) about publicly reported quality measures required by the CMS,” wrote Gotein. “Although most respondents said that they used the measures extensively, more than half were concerned that the measures encouraged teaching to the test, and almost half reported trying to maximize performance primarily through changes in documentation and coding.”

[2] We do have some gray-literature reports on absolute sums of money ACOs incur to start up and maintain ACOs, but to my knowledge no one has translated those estimates into percent-of-medical-spending estimates.

[3] Here is an example of a statement by MedPAC staff that ACO overhead is 1 to 2 percent of medical spending by the ACO. At the September 11, 2014 MedPAC meeting, commissioner David Nerenz asked MedPAC staffer Jeff Stensland if “we know anything about” ACO “overhead.” Stensland replied, “[P]eople we talk to and the data we have seen, it looks like maybe 1 to 2 percent of your spend, that that’s what they’re spending on their ACO to operate it….” (p. 133 of the transcript ).Stensland also reported, “[I]f you averaged everybody [that is, all ACOs] so far, at least in the first year of the program, the share of savings, on average, that they get is going to be less than their administrative costs of being in it….” (p. 144)

[4] ] Of course, it’s possible the ACOs finance some of their 1-to-2-percent by short-changing patients, for example holding ACO “attributees” in hospitals for “observation” instead of admitting them, or driving away sicker “attributees.” If hospitals finance all of their 1-to-2 percent overhead costs that way, then total spending in the short-term would not rise. It might rise over the longer term due to the damage done to patient health.

[5] In their New England Journal of Medicine paper about the ACA, Blumenthal et al. claimed CMS’s MSSP program had saved $700 million and the Pioneer program $385 million. Both of those numbers were free-lunch  numbers – they were gross savings, not net savings to Medicare, and did not take into account ACO start-up and maintenance costs. 

In our letter to NEJM, Ted Marmor and I noted that fact and stated that the MSSP program had raised Medicare’s costs by 0.2 percent.  We cited a 2015 report by CMS’s Office of the Actuary (OACT). Rather than simply concede the fact that the MSSP program had raised costs and that they had misled readers by citing gross rather than net figures, Blumenthal et al. claimed that the OACT report concluded “the MSSP have been shown to produce savings…” That was false. I urge readers to go to page 4 of the report and read this statement: “[T]he MSSP beneficiaries in the program’s first performance period (covering April 2012 through calendar year 2013) exhibited total spending that was only 0.5 percent below the combined benchmark, or slightly less than the offsetting cost of resulting shared savings payments (net of shared losses) that represented about 0.7 percent of the combined benchmark.” If we subtract savings of 0.5 from payments to ACOs of 0.7, we get a loss to Medicare of 0.2 percent.

The OACT report also examined a simulated version of the MSSP which did not simulate shared savings, and reported modest gross savings for the MSSP ACOs. Blumenthal et al. may have been referring to this simulation of gross savings. If they were, they were being doubly disingenuous. Citing a SIMULATION that examined GROSS savings to demonstrate that the REAL-world MSSP achieved NET savings is disingenuous twice over.