In 2006 stand-alone prescription drug plans (PDPs) became a new source for prescription drug coverage for Medicare beneficiaries. PDPs weren’t just new to Medicare, they were a new type of insurance product in general. No such thing had previously existed in the commercial market. Since level of prescription drug use is relatively easy for an individual to predict there was good reason to worry that PDPs would fail due to severe adverse selection. In a paper by Steve Pizer, me, and Roger Feldman we predicted the likely selection experience of PDPs and found that they could weather the degree of adverse selection they would experience (Pizer, Frakt, Feldman. (2008). Predicting Risk Selection Following Major Changes in Medicare. Health Economics 17(4).)
Some background is warranted to explain the paper’s findings in detail. Adverse selection occurs when an insurance product is purchased by individuals who are on average more risky than expected. The insurer sets the product’s premium based on the expected level of risk (expected claims). If actual risk of those who purchase the product is higher (adverse selection), the product is at risk of failure since it does not have enough revenue to cover claims.
Adverse selection is the reason stand-alone drug plans didn’t exist before. Prescription drug use is not like use of other health services in that it is very predictable. Individuals generally know their pattern of drug use much more accurately than they know if they’ll be hospitalized, for example. Therefore, to insure the expenses of drug use is a dangerous game. It is very likely that only individuals with high expected use will purchase stand-alone drug coverage. The insurer will experience adverse selection, raise its premiums in response, thereby driving out the relatively lower risks, and experiencing selection more adverse. This is the classic insurer death spiral.
For this reason, Medicare’s drug program (Part D) includes a variety of measures that reduce the amount and consequences of adverse selection. One is a late enrollment penalty. Others include risk adjustment, risk sharing, and reinsurance. In addition, PDP premiums are subsidized at a rate of 74.5% (of basic coverage, not including enhancements), making a PDP a good value even for beneficiaries with relatively low drug use.
In the paper, we predicted that PDPs would experience adverse selection. In fact, drug expenditures for PDP enrollees were predicted to be 78% higher than those for Medicare HMO enrollees. However, we also found that the PDP market would likely be able to absorb the degree of adverse selection that we predicted they would experience: on the whole the PDP market will be stable. Death spirals, if they occur, are likely to be seen only for plans that offer extensive enhanced benefits (such as the Humana Complete plan that entered the market in 2006 with full brand name coverage in the gap–it subsequently raised premiums and dropped brand name gap coverage.)
Careful readers may have noticed that I used terms like “predicted” and “likely” in the foregoing paragraph. Why didn’t we use actual enrollment to see what plan selection experience has been? We couldn’t. The data weren’t available at the time of the study. Moreover, they’re still not available. The only way to estimate PDP selection experience is to simulate it based on models of beneficiary decision making among other plan types for which data are available. That’s precisely what we did.
Using Medicare Current Beneficiary Survey data, we developed a model of Medicare beneficiary insurance decisions based on plans that existed in 1998-2001. Those plans include HMOs, Medigap supplements, and traditional fee for service (FFS) Medicare. The characteristics of choices available today are included among the choices that existed in 1998-2001, only mixed up in different ways. That is, there were choices then (HMOs) and now (still HMOs) that restrict provider choice. There were choices then (FFS, Medigap+FFS) and now (FFS, FFS+PDP, FFS+Medigap+PDP) that do not restrict choice. There were choices then (some HMO and Medigap plans) that offered drug coverage and the same is true today (some HMOs and all PDPs).
By estimating beneficiary response to the characteristics of options in 1998-2001 it is possible to predict their response to new options with the same characteristics bundled differently. That’s, more or less, what we did in the paper. Someday, when the data are (finally) available, we’ll be able to re-estimate our model and see the extent to which predictions of beneficiary decision making differ in the pre- versus post-Part D era.