If there is one thing we have learned as health economists it is that whatever incentive scheme is placed in front of medical providers, some of them will find a way to game the system, making money without delivering value. We should note that this is actually not unique to health care firms, for example financial firms respond similarly to changes in financial regulations. It is only that many people cling to the belief that doctors and hospitals won’t respond to profit incentives.
The most recent, and largest, of all incentive schemes is the Medicare Accountable Care Organization (ACO) Shared Savings program. Sure enough, there is way to game this system that will result in large profits for some, without necessarily decreasing overall health care spending. It is a bit risky, and requires some careful calculation, but the payoff can be huge. So, at the risk of sounding like a late night infomercial, if you want to learn how to make a cool million bucks (or a lot more, depending on your skill and risk tolerance) and have a loose moral compass, then read on.
To explain our plan, we must first refresh your memory about the key details of the Shared Savings program. The Center for Medicare and Medicaid Services (CMS) has already approved several hundred ACOs, which come in all shapes and sizes. Some are organized by hospitals, others by multispecialty group practices, and still others simply represent loosely affiliated physicians. Importantly for our scheme, there is no regulatory hurdle that we are aware of that prevents a lay person from organizing a number of independent physicians into an ACO.
Patients are currently “assigned” to ACOs retrospectively, based on the delivery of primary care. In a nutshell, if providers in the ACO delivered at least 50 percent of the primary care services to a Medicare enrollee in the previous year, then that enrollee is assigned to the ACO for the current year. For each assigned enrollee, CMS predicts their annual health spending, using a risk prediction algorithm that accounts for the patient’s age, sex, and diagnostic conditions (based on the past year’s diagnostic codes that appear in Medicare’s billing records). If, during the current year, actual spending is less than predicted spending, CMS shares the savings with the ACO. Under current regulations, ACOs receive up to 60 percent of the savings and they are also responsible for up to 60 percent of losses if their beneficiaries end up having above average spending. The absolute magnitude of these gains and losses are capped, but the stakes are still large.
In a seemingly perpetual search for slower growth in healthcare costs, this is not the first time we have tried to introduce market incentives and reward providers who hold costs down. These shared savings are simply a variant of capitated payment systems in HMOs with two important differences. First, traditional capitated system only covered a portion of medical fees, i.e. all professional fees but not the more expensive facility fees. Second, under traditional capitation providers captured 100 percent of their savings (of an admittedly much smaller pie). Under this system it was well known that providers with relatively healthy patients stood to make a lot of money. Therefore, HMOs used proprietary risk prediction models in combination with various rules of thumb, combined with tough negotiation strategies, to set their capitated rates. Even so, many providers made a lot of money; a favorable patient mix was the provider’s best friend.
In creating the shared savings program, CMS faces the same problem previously faced by HMOs – rewarding efficiency without encouraging risk selection – as well as many new ones. Unlike private firms, CMS cannot rely on rules of thumb, because government agencies can never be seen to make arbitrary decisions. So CMS relies on a state of the art risk prediction algorithm (albeit one whose parameters are well known). However, it is hard to predict medical spending with purely objective data in billing records and as a result even the state of the art is mediocre at best. The R-squared for a regression of actual spending on predicted spending is no higher than about 0.4. That means that the risk prediction algorithm explains no more than 40 percent of the variation in health spending. What about the remaining 60 percent? Some of it may be pure random noise – which would be impossible for CMS or any provider to predict. For example, even the healthiest woman with the best prenatal care will sometimes give birth to a child who requires costly neonatal care.
More troubling is that some of the remaining variation is likely very predictable to individuals who have more information about the patient than the statisticians performing the risk adjustment. A risk prediction model may know that a certain percentage of women of child bearing age will, in fact, have a child in the next year. Each woman likely knows this probability with far greater certainty. For most other health conditions, patients may be less well informed than the statisticians. But their doctors, well that is a horse of a different color. Physicians are likely both more informed than patients about their broad health risks and more informed about their patients than the statisticians at CMS.
Historical precedent shows that this is likely the case. For example, consider the long standing controversy about third party utilization review. Independent panels establish norms for the appropriate treatment protocols based on available statistical research. This type of research is quite similar to the process used by CMS for risk prediction. Many physicians are reluctant to strictly adhere to these norms, arguing that their specific treatment plans account for idiosyncratic differences among their patients that are not captured by the statistical models.
Want another example? Consider the objections that providers raise to the creation of report cards for physician or hospital quality. Again, these report cards are risk adjusted, but the adjustments are necessarily imperfect. Some providers treat patients who are sicker than the data indicate, and this adversely affects their rankings. This is not just a statement of physician preferences. There is considerable evidence that providers game this system by avoiding patients who are more severely ill than is suggested by the risk adjusters. (For example, see the relevant section in this lengthy review article.)
These examples strongly suggest that physicians are able to out predict statisticians. We should be clear, this is not an indictment of the quality of the statisticians but simply reflect the fact that physicians have better information about the underlying health of their patients then is available through billing records. In most settings we find this greater familiarity with our health by our physicians comforting.
Of course, physicians were also able to out predict statisticians under traditional capitation plans. But the stakes in earlier capitation programs were small in comparison to the stakes in shared savings – professional fees versus the whole medical care kit and caboodle. Under shared savings, the right patient can be worth thousands of dollars. In addition, savvy providers in the capitated world were matched in a fairer fight against profit maximizing HMOs, whereas under shared savings government regulations often force CMS to fight with at least one hand tied behind his back.
And these conditions create the million dollar opportunity.
Anyone can access and use the CMS risk adjustment model or something very close to it. A clever statistician with decent access to extensive clinical data – the kind of data that a physician can help provide — can determine whether a given physician’s patients tend to cost more or less, on average, than is predicted by the model. Or physicians might just know this information at a gut level and may be willing to bet on their own instincts. The savings could be due to the physician’s efficient decision making, but is perhaps more likely due to the patients’ good health. It really doesn’t matter. Either way, there are opportunities to profit at CMS’ expense.
If we are correct, then the following is a formula for success. First, gather together millions of dollars. Second, approach a series of physicians who likely have healthy patients. Third, gather clinical data from candidates and identify those physicians whose patients consistently cost less than predicted by the CMS algorithm. Alternatively, invite physicians to identify themselves as efficient, with the proviso that they must take an ownership stake in the ACO. Fourth, organize these physicians into an ACO, using the millions of dollars as seed money to employ the physicians (or as a cushion in case it turns out that we are full of hot air.) Finally, wait for the checks from CMS to roll in and get ready for a life on a beach filled with those drinks with the tiny little umbrellas.
You may be wondering why we don’t undertake this plan ourselves. Perhaps it is because our moral compasses are firmly pointing north. (Yes, there are Republicans with moral compasses pointing north.) The adage “it takes money to make money” also has something to do with it. Maybe it is that our tolerance for risk doesn’t extend to exposing ourselves to the capricious whims of regulatory agencies. But if any of you want to give our idea a try, you have our permission if not our blessing. And remember that Dranove loves expensive dark chocolates (no nuts please) and Garthwaite has a taste for premier cru Burgundies (red or white). Our addresses are not hard to find.