Insight

Considerations for Incorporating Prescription Drug Data in Exchange Risk Model

The Centers for Medicare and Medicaid Services (CMS) recently released a white paper on the risk-adjustment methodology used by CMS in the Exchanges, and is seeking public comments through April 22. Unlike the other two pieces of the “3Rs”—the risk corridors program and the reinsurance program—the risk adjustment program will be a permanent fixture for plans offered through the Exchanges. CMS is seeking to improve the risk adjustment model in order to—in the administration’s view—more accurately and appropriately spread the risk and financial burden across all the insurance plans in an effort to keep the marketplace from collapsing.

Background

As AAF has previously detailed, the risk mitigation programs included in the Affordable Care Act (ACA) were vital to countering the insurance reforms also included in the ACA in order to prevent a “death spiral.” Proper functioning of all these provisions is critical to the success of the market. Primarily, because insurers are no longer able to adjust premiums based on an individual’s health status, i.e. the financial risk they pose—completely undermining the premise on which insurance markets stand—the risk mitigation programs were created to fill the hole that was created. So after the risk corridors program was only able to pay out 12.6 percent of claims made for 2014 and the reinsurance program has failed the last two years to raise as much revenue as was required by law, CMS is anxious to get this last remaining piece right.

The Current Risk Adjustment Model

The risk adjustment program is supposed to transfer funds among insurers in order to compensate issuers enrolling individuals with above-average health risks with payments from issuers enrolling individuals with below-average health risks. In order to do so, a risk score is calculated for each individual enrolled in an Exchange plan, similar to the practice used in Medicare Advantage and the Part D Prescription Drug Program.

The current model consists of two components: one formula to calculate an individual’s risk score and another to calculate an insurer’s transfer amount. A person’s risk score is calculated using a concurrent model incorporating an individual’s demographic information and medical diagnoses.[1] Concurrent models use current year data to estimate what an individual’s expenses will be or should have been in that same year. Additionally, the model is designed to reflect plan liability, as opposed to total expenditures, in order to account for differences in actuarial values across metal tiers and the fact that plans don’t pay the full cost of a patient’s care. Because of vastly different expected costs, separate models were developed for the adult, child, and infant populations. Disease severity was taken into account by creating categories of diagnostic groups which tend to denote severe illness based on historical claims data. Higher costs resulting from particular disease interactions are also accounted for using regression models. A hierarchy was assigned to the various categories such that the highest cost diagnosis would take precedence in determining an individual’s risk score.

An insurer’s transfer amount—how much they either owe or are entitled to receive—is based on the difference between the estimate of what the plan’s premiums would be had the insurer been allowed to adjust premiums based on the insured population’s health status and what the plan’s premiums actually are, as allowed by statute.

Under this program, each state operates as its own risk pool in that the calculations are based on the population insured through the Exchange in each state separately and transfers are only made between insurers in each state. Risk scores are therefore relative to those of the other individuals in the state and payments are budget-neutral within a state; no more is paid to insurers than is paid by insurers. However, that is in no way to say that every insurer is ultimately made whole of any financial losses incurred as a result of enrolling more expensive individuals.

Problems with the Current Risk Model

The risk adjustment model is not perfect; it cannot be—risk, by definition, involves uncertainty. However, given the extent to which insurers must rely on this program, efforts should be made to make the model as accurate as possible. To do so, the model must be based on the most relevant and appropriate pieces of information which will most closely predict an individual’s health care costs.

While the current model does incorporate patients’ diagnoses, it does not include all diagnoses. Those selected for inclusion are diagnoses that are related to conditions believed to be “associated with systematic selection risk of enrollees or providers” in order to discourage insurers from actively avoiding potential enrollees with certain high-cost conditions. However, some diagnoses not included could still result in high costs and there are no adjustments for such occurrences. Further, diagnostic codes for some patients may be incomplete or inaccurate. Partial-year enrollees may be most likely to have incomplete diagnostic data, and data has shown that individuals who enrolled during special enrollment periods had significantly higher costs than those who enrolled before the start of the plan year.

Considerations for Updating the Risk Adjustment Model

One principle of risk adjustment specified by CMS is that diagnostic categories should predict medical expenditures, including drug expenditures. In 2011, CMS began using prescription drug event data in the Medicare Part D risk adjustment model, and they are now considering using such data to enhance the Exchange risk adjustment model.

Using prescription utilization data has several benefits. Prescription data can fill in the blanks for patients whose diagnoses have not been adequately reported. Prescription utilization data can indicate the severity of an individual’s disease state (presumably, someone taking prescription medicines for a disease is faring worse than someone who has been diagnosed but is not taking medications; alternatively, one medicine may be more likely to be prescribed than another based on the severity of a disease, allowing for relative comparisons between patients). Conversely, even if two patients are suffering equally from a disease, their medical costs could be vastly different if one is taking medications and the other is not. Prescription data is more readily available than diagnostic data and is standardized, whereas diagnostic data often is not, making it more difficult to build a model around. Finally, incorporating prescription utilization into the risk adjustment model will lessen the disincentive to cover high-cost medications that currently exists.

The incorporation of prescription data will not solve all the problems, though, and may create problems of its own. As occurred with the incorporation of diagnoses in the risk adjustment model, there will be a new incentive to prescribe more and costlier prescriptions. Plans would have less incentive to aggressively manage drug utilization, and increased demand for medicines could lead to increases in price.  However, providers do have a responsibility to act in the best interest of their patients, and inappropriately prescribing medications would certainly not meet this standard. Further, because plans are typically responsible for a portion (sometimes a majority) of a drug’s cost, the perverse incentives created will be limited.

Consideration should also be given to factors that influence drug utilization and the impact that will have on risk adjustment payments if such data is incorporated into the model. Plans with higher cost sharing and/or lower-income enrollees may have lower utilization not because their patients are healthier but because of affordability issues. While lower-income individuals are eligible for cost-sharing reductions intended to mitigate this issue, such individuals are not automatically enrolled in the plan that allows them to receive this additional subsidy. Perhaps a baseline should be determined before a new model is implemented to account for possible utilization variations among metal tiers.

Another complication arises from the fact that many drugs are indicated for multiple conditions, each of which may have wide variations in its associated expected costs. Further, drugs are often prescribed off-label which could also lead to inaccurate expected cost calculations.

Conclusion

While the incorporation of prescription utilization data will likely increase the predictability of patients’ expected costs, the potential difficulties in accurately and appropriately setting the parameters without creating more problems than benefits makes it likely this will not a be one-time adjustment. Plans should be prepared for multiple adjustments, which may lead to continued instability in the Exchange market.

[1] This is in contrast to a prospective model, as is used in Medicare Advantage and Part D, in which diagnoses from the past year are used to predict expected costs for the upcoming year. Such a model would be difficult, particularly in the first year of Exchange operation, since prior year data would be difficult if not impossible to collect, primarily because the Exchange population is not likely to be stable from one year to the next.

Download PDF

Disclaimer