A new NBER working paper (gated, with abstract) but available in a similar form here, concludes that the risk adjustment techniques used to determine Medicare Advantage pricing “didn’t work.” Rather than make the sick more attractive to capitated Medicare Advantage plans, they triggered a large reallocation of Medicare funds from the relatively sick to the relatively healthy. They also increased Medicare spending by an estimated $30 billion in 2006.
The authors conclude that CMS is unlikely to be able to improve its risk adjustment performance. Worse, ObamaCare is set to propagate the CMS failure through the private sector. The authors note that ObamaCare encourages the use of risk adjustment techniques using “similar ‘criteria and methods’ to those used in Medicare Advantage” by 2014.
Advocates for the capitated care that is the center of Medicare Advantage have long argued that paying health plans a fixed amount per person would reduce expenditures. It would, the story goes, reduce costs by giving plans an incentive to control costs and focus their spending on keeping people healthy.
Despite evidence that existing preventive care has a limited ability to stave off expensive illnesses, enthusiasm for capitated care was so high that officials rushed to push theU.S.population into it. Belief in the capitated care plan efficiency was so strong that the county benchmarks developed in the 1980 set the first Medicare Advantage reimbursement rates at just 95 percent of per capita county fee-for-service costs. When, against all odds, Medicare Advantage failed to expand rapidly enough to satisfy enthusiasts, the 95 percent benchmarks were raised to 103 percent of average fee-for-service spending.
Eventually even diehard capitated managed care enthusiasts had to admit that paying the same flat fee for everyone had the undesirable side effect of encouraging health plans to discriminate against the sick. To correct for this problem, advocates proposed risk adjustment, the use of statistical models to adjust individual capitation rates based on their prediction of the amount of future medical care a person with given characteristics would likely need. Although the authors don’t discuss it, most private medical plans at the time adjusted for risk using experience rating. They looked at the past medical costs of specific individuals, plus additional variables, to determine the rates they should charge for covering future care.
In the beginning, CMS risk adjustment consisted of not-so-fancy statistical models that don’t work very well. In the 1980s and the 1990s it used such variables as age and gender and disability and Medicaid status. The resulting model managed to explain one percent of the variation in payments in the fee-for-service population. The traditional measures of risk used by the people running private health plans were far better at predicting individual costs, and Medicare Advantage plans apparently were able to target relatively healthy individuals. Their success is indicated by the fact that Medicare Advantage participants had medical costs that were an estimated 20 to 37 percent below those of the people remaining in fee-for-service.
In 2000, CMS added health status to its risk adjustment model. As Brown et al. explain, the new model distills the “roughly 15,000 possible ICD-9 codes into seventy disease categories.” This improvement allows CMS to explain 11 percent of the variation in fee-for-service costs in the following year. Unfortunately, the formula also “systematically under-predicts costs for those with above-average costs, and over-predicts costs for those with below-average costs.” For example, capitation payments generated by the new model cover only 86 percent of costs for the highest-cost quintile of fee-for-service members.
Given that the pricing formula makes the most expensive 20 percent of the Medicare population a sure source of financial loss for capitated Medicare plans, the plans have a strong incentive to find ways to keep expensive people from enrolling and to encourage them to disenroll if they are already enrolled. They also have an incentive to exploit the gray areas in coding by diagnosing patients “more aggressively.” The authors mention a variety of ways that plans can attract the healthy rather than the money-losing sick. InGermany, the heavily regulated health plans respond more quickly to inquiries from people in low-cost areas of the country. Plans can also offer benefits likely to be valued by healthy members — 57 percent of Medicare Advantage plans offered free or discounted gym memberships in 2010, while discouraging high-cost members with much higher cost sharing for serious medical conditions.
As one would expect given the incentives created by capitated care and the risk adjustment the government uses to price it, the authors find that self-reported good health is a better predictor of satisfaction with Medicare Advantage plans than of satisfaction with Medicare fee-for-service. They interpret their results as “suggesting that [Medicare Advantage] plans focus resources on keeping their healthier enrollees relatively happier than their sicker enrollees.” They also find that sicker Medicare Advantage plan members are more likely to switch to fee-for-service.
Although the authors do not discuss the implications of Medicare Advantages’s risk adjustment failure for Accountable Care Organizations, the implications are obvious. They do point out that the risk adjustment requirements in ObamaCare will ultimately require that the government collect cost data from private plans. “[T]here will be no analogue to the [fee-for-service] Medicare population, on which the government has extensive cost and claims data that it can use to estimate a predictive cost model,” they write.
ObamaCare’s structure will likely compromise medical privacy as the bureaucracy demands access to every shred of personal health data in a futile effort to shore up an academically popular (managed competition) model that does a fine job of serving the healthy and a poor job of serving the sick.