Editor’s Note: In the post below, Amitabh Chandra responds to criticisms of the Dartmouth Atlas and offers his vision of the lessons of the Dartmouth findings on variations in health care costs and practice styles. Watch the Blog tomorrow for a roundtable discussion on Atul Gawande’s New Yorker article on McAllen Texas and the policy implications of the Dartmouth work. Roundtable participants will include Robert Berenson, Elliott Fisher, Robert Galvin and Gail Wilensky.

Since 1973, when Jack Wennberg and Alan Gittelsohn first documented geographic variation in health care, researchers at Dartmouth and their collaborators have compiled a large literature that helps us to understand the phenomenon of regional variation in health care utilization, and its uncertain link with patient outcomes. Over the past three decades, Jack’s pioneering analysis has been replicated at the level of states, regions, cities, hospitals, and even hospitals within regions. It has been examined in different types of patients: aged, newborn, medical, surgical, and the chronically and acutely ill. More than 100 peer-reviewed publications support the finding that more health care spending does not automatically translate into better outcomes, that improvements in health are often caused by lower-cost interventions, that cost growth in health care originates from the use of technologies that are beneficial in some patients but offer great scope for overuse in others, and that there is tremendous geographic variation in the efficiency of the local delivery system that is not explained by patient health, preferences, or malpractice pressure. The last point is the most important one, for it says that regardless of whether the association between spending and outcomes is positive, flat, or negative, there are plenty of providers who’re able to deliver high-quality care without being high-cost suppliers of care.

Today, the Dartmouth Atlas finds itself yanked from the tranquility of Hanover, New Hampshire, and placed squarely in the crossfire of health care reform debates in Washington. Its insights offer enormous promise for managing cost growth in health care. Because health care is almost 20 percent of U.S. gross domestic product (GDP) and is growing at 6.5 percent annually, it is perhaps unsurprising that the Dartmouth group finds itself under attack; all that spending is someone’s income. Some attacks come from beneficiaries of the status quo: financially entrepreneurial providers, such as those in Redding, California, McAllen, Texas, and Elyria, Ohio, have much to lose. Others, including myself, worry that if the Atlas’s findings are misinterpreted or oversimplified, they will result in sledgehammer policies that can damage an already injured health care system.

Here I discuss criticisms of the Dartmouth work and offer my concerns on what should and shouldn’t be concluded from our work. On the Economix blog, Jonathan Skinner  — another Dartmouth Researcher — offers his own thoughts.

Let’s start with the concerns that are easiest to resolve:

1. The Dartmouth work is contradicted by the fact that more spending generates better outcomes. Last week’s Wall Street Journal editorial cautioned against relying on the Dartmouth research program by citing the tremendous benefits of greater medical spending. The point of the Dartmouth work is not that high-cost medical care is automatically ineffective, but that we are at the point of spending where additional spending does not generate additional improvements in outcomes or quality. We’d be thrilled to see more high-tech medicine being used if it delivered as much as benefit as it costs. The Journal cited the work of David Cutler and Mark McClellan to make its point. But if the editors of the WSJ had done a little more homework, they would have learned that an updated analysis of Cutler and McClellan’s work found that an additional year of life now costs Medicare $300,000 — a number that should make any fiscally conservative newspaper very unhappy. David Cutler’s group now shows that these cost-ineffective spending numbers also apply to spending in the non-Medicare population. McClellan’s research has demonstrated that marginal increases in the scope of intensive management for heart attacks does not improve survival, and that the striking improvement in survival after heart attacks comes from the use of low-cost drugs, many of which are off-patent. And finally, a recent paper in the New England Journal of Medicine attributes most of the decline in coronary deaths since 1980 to the use of inexpensive drugs and the better management of cholesterol and blood pressure. Expensive procedures such as bypass and angioplasty (procedures that the same editorial touted as being therapeutic) accounted for less than 7 percent of the decline.

I’m not sure who wrote the WSJ editorial, but I want my money refunded for that day’s paper.

2. Patients’ illness and socioeconomic status invalidate the Dartmouth results. This week’s New York Times cited research by Robert A. Berenson of the Urban Institute and Jack Hadley of George Mason University, which stated that the Dartmouth findings could be explained by “individual characteristics, especially patients’ underlying health status and a range of socio-economic factors, including income.” The Dartmouth view is not that patient illness and income do not affect care — they absolutely do –and I’d be surprised if the principal determinant of health care isn’t patients’ illness and socioeconomic status. But illness, incomes, and malpractice are not explanations for why different hospitals in the same city perform differently on costs and quality, for why measures of quality that are unaffected by patients’ health and income are often negatively related to costs, for why the performance of hospitals within Los Angeles and Manhattan on quality and spending resembles the outcome of a shotgun blast, and for why Miami has a growth rate of Medicare spending that is twice that of San Francisco. It is these geographic variations, which persist even after patient characteristics are accounted for, that are the focus of the Dartmouth work.

The article in the Times didn’t offer any guidance on what work Hadley’s quote was based on. My best guess is that it’s from an earlier Hadley paper whose problems were discussed by us here. In addition, Berenson’s views appear to be more nuanced than captured in the Times piece.

3. Area poverty rates explain the Dartmouth results: spending is higher in poorer areas. The University of Pennsylvania’s Richard `Buz’ Cooper makes this criticism best: “So let’s stop the Dartmouth doubletalk and start addressing the root cause of variation in spending– p-o-v-e-r-t-y.” I’m less sure than Cooper on the certainty of this point: (a) Lifetime Medicare spending is actually higher for beneficiaries who live in richer neighborhoods. (b) Both El Paso and McAllen TX have poverty rates of 27 percent, so that cannot explain why health care in McAllen looks so wacky. (c) I decided to test Cooper’s assertion directly, with real data. The answer: Only 2.6 percent of the variation in Medicare hospital-level spending is explained by the poverty rate of the hospital’s ZIP code. (To further ensure that I’m adjusting for health risk, I used hospital-level spending on decedents adjusted for the presence of comorbidities.)

I would be remiss in not noting other concerns about our work by Cooper, who had two papers published in Health Affairs last fall. He calls our work that uses regression analysis “phony statistics,” “voodoo statistics,” and “statistical permutations.” We published our response to Cooper here and here; a summary is here. His blog is here. Enjoy.

What Do I Believe The Dartmouth Work Tells Us About Reform?

According to the WSJ editorial, “The President’s main case for reform is rooted in false claims and little evidence.” My own thoughts are as follows:

1. Do not cut reimbursements. Many believe that the Atlas tells us to cut spending. Ironically, despite arguing that 20-30 percent of spending might not generate value. Jon Skinner, Elliott Fisher, and Doug Staiger were the first to argue that cutting spending in Miami will not make it Minneapolis. Miami is not Minneapolis with 30 percent more waste. Rather, at least in the context of heart attack treatments, my research with Doug Staiger demonstrates that the two areas have specialized in using different “production technologies” to deliver health care. We find that areas that are more intensive get better outcomes from intensive treatment but and get worse outcomes from low-tech care (perhaps because the specialists who’re good at intensive medicine are not good at offering less-intensive care). One can cut reimbursements, but that does not automatically make a high-spending area adopt the systems, culture, and experience of the low-spending area.

2. Focus on cost growth. I do not believe that the biggest monster under the bed is variation in costs — it’s variation in cost growth, which is as uneven as the variation in costs, but not always determined by the same factors. Differences in growth rates will swamp any differences in spending levels today: when costs are growing at 6 percent for one group of providers and at 2 percent for another, the level of spending for the first will double in 12 years, and for the second in 35 years. One of the things that we’ve learned from the Dartmouth work is that cost growth is not inevitable — while new technologies surely cost more, there are plenty of places that have adopted these technologies without displaying increases in health care spending or decreases in quality. Costs in San Francisco have grown even more slowly than GDP — and no one claims that the health care there is anything less than stellar. So rather than cutting reimbursements, we should think about reducing the growth rate of reimbursements.

3. Be realistic about the gains from bundled payments. Bundled payments combine reimbursements for inpatient, outpatient, and home health into a single payment. If bundled payments work (they did reasonably well with the introduction of the Medicare inpatient hospital prospective payment system, or PPS), we will realize fairly large one-time savings as hospitals figure out ways to cut out waste in follow-up visits. But I have three concerns about how much cost savings we’d actually get from this innovation: (1) Bundling payments for, say, the first 30 days of care assures us that there will be a spike in utilization on day 31. Nor do these relatively short bundles do anything about a large source of cost growth, which is care provided in days 30-365 after the acute care hospital admission. (2) Cost growth in health care does not come from the greater use of inpatient services — almost all of it is in the use of outpatient services, imaging, and office visits that are not associated with an inpatient visit. For these services there is no inpatient hospitalization to bundle them up with. (3) Never, ever, ever, underestimate the ability of the system to respond to bundled payments by producing more bundles  — there is an unlimited reservoir of patients and body parts for providers to image, diagnose, and treat.

4. Think accountable organizations. The key Dartmouth message is that regardless of whether the association between spending and outcomes is positive, flat, or negative, there are plenty of providers who’re able to deliver high-quality care without being high-cost suppliers. Reform efforts should focus on encouraging these organizations, which are accountable for all of my care, to replicate. But why not simply play with the fee schedule for providers? One of principal lessons of our work is that geographic variation in care is driven by the use of procedures and therapies where medical textbooks and randomized controlled trials are silent. Office visits, specialist consultations, and the use of imaging diagnostic services (as opposed to major surgeries) are major determinants of cost variation and cost growth. We do not know how to figure out the right rate or correct reimbursement for these types of services. On the other hand, because integrated delivery systems have the right incentives to figure out the optimal rates of these therapies, they will do this better than the rest of us. The best part is that they already exist: Geisinger, Intermountain Health Care, and Mayo Clinic. I don’t know the answer to whether we can get such places to replicate or expand  — but I do know that providers respond to incentives.

5. Worry about spillovers. One concern with our work is that we’ve almost exclusively studied the Medicare program, and we do not know how Medicare payment reform will affect the care of patients in the commercially insured population. The Dartmouth group has hypothesized that physicians use a common set of decision rules to treat all patients. This is Jack Wennberg’s “practice style” hypothesis, and it would mean that Medicare reforms would generate additional cost savings because of spillovers onto the commercial plans. But the spillovers could work in the opposite direction if providers ramp up utilization in the non-Medicare programs. I don’t think that we know the all the answers here. One Dartmouth paper that compared Medicare and commercial insurance spending for very sick patients supports the “practice style” view, but another by Buz Cooper does not. There are problems with Cooper’s analysis, but I don’t believe that this automatically invalidates his point. Michael Chernew, Joseph Newhouse, and I are working hard on this problem as I write this, and we hope to have an answer soon. But if offsetting spillovers exist, we still have a policy solution that will reduce the externality: put Medicare beneficiaries into plans that also provide care to the under sixty-five.

The keys to unlocking efficiency in American health care do not all reside in the Dartmouth work. There are inefficiencies that have nothing to do with geographic variations in utilization, for they are present in every health care market. We must not ignore them: tax reform, insurance market reform, malpractice reform, and a greater role for smart demand-side cost management should be key elements of successful reform. Nothing in the Dartmouth work suggests otherwise.

The Dartmouth message is that the savings from these efforts are small relative to what we could get out of focusing on unwarranted variation in utilization. Congress needs to encourage Minneapolis-type care to emerge in Miami, while curtailing the incentives for Minneapolis to morph into McAllen. This involves doing much more than offering piecemeal adjustments to the reimbursement system. It requires rewarding the creation and expansion of already existing delivery systems that have the right incentives to check costs while delivering quality. It requires competition between these delivery systems for patients and their premiums. It requires that Congress and the president have the courage to fight the beneficiaries of the status quo, who have already begun to launch a bitter fear-campaign against attempts to implement accountability and value. These are the lessons that I’ve learned from the Dartmouth research program.