This analysis is part of The Leonard D. Schaeffer Initiative for Innovation in Health Policy, which is a partnership between the Center for Health Policy at Brookings and the USC Schaeffer Center for Health Policy & Economics. The Initiative aims to inform the national health care debate with rigorous, evidence-based analysis leading to practical recommendations using the collaborative strengths of USC and Brookings.
The mantra of ‘Repeal and Replace’ has escalated in recent weeks, though what, specifically, the ‘Replace’ component might look like is still unclear. However, many of the current proposals include, at a minimum, some type of continuous coverage provision that allows people with chronic health conditions who have continuously maintained coverage to buy health insurance at standard rates. For example, Paul Ryan’s A Better Way proposal and Tom Price’s Empowering Patients First Act would each prohibit insurers from charging sicker patients more than standard premiums in the individual market as long as they have maintained continuous coverage since before becoming sick.
Such provisions are important to keep patients from seeing their health insurance premiums sky-rocket after becoming sick, which would defeat the purpose of insurance in the first place. However, these provisions also require that insurers sell policies to these patients at premiums that they know will not cover their expected health care spending, generating losses for the insurance company. On its own, this would create a situation where insurers have a strong financial incentive to avoid enrolling these sicker patients.
Risk adjustment combats disincentives to provide coverage for sicker patients
In order to mitigate these incentives for insurance companies to avoid sicker patients, policymakers will need to include a risk adjustment program in any replacement reforms that require insurers to issue insurance to any applicant (also known as “guaranteed issue”) and set limits on adjusting premiums to fully reflect an enrollee’s health status. Continuous coverage provisions are one example of such limits, but risk adjustment will be necessary to combat against adverse selection across a wide range of potential reforms.
A risk adjustment program would make behind-the-scenes financial transfers to insurers to adequately compensate them for enrolling these sicker patients when they are not allowed to charge the individual higher premiums. Risk adjustment will be necessary to promote a well-functioning market where private insurers compete based on the value they deliver and not simply by avoiding sicker patients.
Fortunately, risk adjustment has been used in many contexts, providing experience from which to draw in crafting such a policy. For example, the Affordable Care Act (ACA) includes a risk adjustment program for policies sold in the reformed individual and small group markets. Risk adjustment has also been used for over a decade to adjust payments to private insurers providing coverage in the Medicare Advantage and Medicare Part D programs, as well as in other countries such as Germany, Switzerland, and the Netherlands, which also rely on competition among private companies to provide health insurance coverage to many of their citizens.
The need for risk adjustment in the individual market is underscored by the fact that the ACA’s risk adjustment program has transferred about 10 percent of the total premium dollars collected across competing insurers in 2014 and 2015. However, the ACA’s risk adjustment program has received some criticism, such as complaints that plans are undercompensated for healthier individuals and that the long time lag of data availability is difficult for insurers to manage, particularly for smaller insurers.
The Centers for Medicare and Medicaid services (CMS), which administers the ACA’s risk adjustment program, has engaged in ongoing research and outreach to evaluate and improve upon the risk adjustment program. This work led to several forthcoming changes to the risk adjustment program that were finalized in the 2018 Benefit and Payment Parameters rule. These changes—including adding prescription drug information to the risk adjustment model and incorporating a reinsurance program for very high cost enrollees into the risk adjustment program—are intended to improve the accuracy of transfers across insurers. While we have not yet seen how these modifications work in practice, future risk adjustment programs should draw on the research and public discussion that has gone into developing these improvements and incorporate them into a risk adjustment program that operates in the context of a reformed individual market.
Drawing from these experiences, any replacement of the ACA or reforms to the individual market should utilize a risk adjustment system that:
- Incorporates information from select prescription drug claims; and
- Includes a reinsurance program to reimburse plans for some of the costs from the most expensive enrollees.
And any replacement or reforms should:
- Standardize health plan offerings and set a minimum benefit level, to facilitate both the functioning of risk adjustment and comparison-shopping between plans by consumers.
What does risk adjustment look like under the ACA?
Prior to the ACA taking effect, policies sold in the individual market were generally subject to state regulation and oversight. Most states imposed very few restrictions on how insurers could set premiums and, as such, insurers generally would set premiums to reflect an individual’s expected health care spending. As a result, individuals with chronic health conditions faced very high premiums if, that is, an insurer would sell them a policy covering those conditions to begin with.
Nonresident Fellow - Economic Studies, Center for Health Policy, USC-Brookings Schaeffer Initiative for Health Policy
The ACA significantly changed the regulation of health insurance in the individual and small group markets beginning in 2014. These new federal rules include guaranteed issue and adjusted community rating regulations, meaning that insurers can no longer deny coverage nor vary premiums according to an individual’s health status. Instead, insurers can only adjust premiums according to an enrollee’s age (within limits), geography, family composition, and tobacco use. Insurers can also charge more for more generous plans.
These regulations significantly improved the availability and affordability of insurance policies for sicker individuals. However, they also created a market where the premium revenue that an insurer collects for a given enrollee no longer necessarily reflects that individual’s actual health risk. Two individuals who are the same age and live in the same neighborhood are charged the same premium even if one has several expensive health conditions and the other does not.[i]
Absent any other interventions, an insurer would likely lose money on the former and make a profit on the latter, thereby creating a strong incentive to try to avoid enrolling the sicker individual. Without risk adjusted transfers, these same regulations intended to improve availability of health insurance products for sicker individuals would create a market where an insurance product sold to these very people would not be financially viable because insurers could not charge premiums high enough to cover their expected losses.
In order to mitigate these concerns, the ACA included the creation of a risk adjustment program, whereby payments are collected from insurers that enroll healthier than average individuals and distributed to insurers that enroll sicker than average individuals within each state and each market (i.e., individual vs. small group). The goal of this program is to adequately and fairly compensate insurers for the health risk that they enroll.
To do so, the risk adjustment program uses a predictive algorithm that incorporates information on individuals’ demographics and health conditions to predict variation in expected spending that insurers are prohibited from using to adjust premiums. Payments are collected from plans that enroll healthier than average individuals and transferred to plans that enroll sicker than average individuals. Unlike risk adjustment in the Medicare context, payments are entirely financed by assessments on other insurers within a state’s market and do not involve any direct outside funding.[ii]
What have we learned from risk adjustment under the ACA?
Three key lessons can be gleaned from our experience with risk adjustment in the ACA to date.
Health risk varies considerably and plan selection is non-random in the individual market
(i.e., risk adjustment is really important)
The ACA’s risk adjustment program transferred about 10 percent of the total premium dollars collected in the individual market across plans in 2014 and 2015. This suggests that there is significant variation in health risk in the individual market; this is perhaps not surprising, given that the individual market serves as somewhat of a catch-all, providing coverage to many different types of people, ranging from healthy individuals between jobs to very sick individuals who were previously uninsured or insured through high risk pools.
The magnitude of transfers across insurers also highlights that selection into competing plans is not random. Instead, certain insurers or products seem to attract sicker individuals while others seem to attract healthier individuals. This could be due to factors like broader provider networks and brand recognition attracting sicker individuals or targeted marketing attracting healthier individuals. Taken together, the significant variation in health risk and non-random selection into plans that we have recently seen in the individual market reinforces the importance of incorporating a risk adjustment scheme.
Any replacement option that limits insurers’ ability to adjust premiums according to enrollee health status should be coupled with a risk adjustment scheme to ensure fair compensation for insurers.
There is a lot of churn in the individual market
In addition to the variation in health risk seen in the individual market, experience from the first few years of the exchanges has demonstrated that there is a significant amount of churn in the individual market. For example, the average length of enrollment in Covered California is reported to be around two years.
This fact is relevant for risk adjustment because data on an individual’s health conditions is necessary in order to be able to compute enrollee risk scores. Most risk adjustment schemes used in other contexts (e.g., Medicare) use what is called a prospective model, where this year’s diagnoses are used to predict next year’s spending. However, there is much less enrollee churn in these Medicare markets, meaning that prior-year data is available for nearly all beneficiaries.
Risk adjustment under the ACA, however, has used a concurrent model, where this year’s diagnoses are used to predict this year’s spending. Using a concurrent model was necessary for the initial years of the exchanges because there was no existing data to use from prior years, though some analysts suggested that perhaps the program could be transitioned to a prospective model after several years of built up experience.
Concurrent models generally have higher predictive accuracy, though prospective models are generally thought to better reflect the selection incentives that an insurer faces when enrolling an individual for the following year. Additionally, concurrent models can have the downside of providing plans with a more direct incentive to intensify their coding in order to increase their risk scores.
However, there is currently no centralized collection of data on enrollee health risk across insurers in the exchanges. Thus, the high turnover in the individual market, both across insurers in the market as well as in and out of the market altogether, suggests that it may be necessary to use a concurrent model indefinitely.
Alternatively, policymakers could implement a central collection of medical diagnoses that would move with enrollees across insurers. This could enable the use of a prospective model based on prior year health claims without having to exclude the many enrollees that changed insurers or plans in the given year. However, such an effort would likely need to collect data beyond just the individual market due to the significant enrollee churn across the individual, employer-sponsored, and Medicaid markets. Due to patient privacy and other concerns, we suspect that such an undertaking is unlikely.
The high turnover in the individual market will likely necessitate the use of a concurrent risk adjustment model indefinitely.
Product standardization facilitates risk adjustment
In addition to imposing guaranteed issue and adjusted community rating, the ACA also standardized insurance products sold in the individual and small group markets. Plans are all required to cover a set of “essential health benefits” and to do so at pre-specified levels of generosity, known as metallic tiers including bronze, silver, gold, and platinum. Each of these metallic tiers represents an increasingly generous plan—or higher actuarial value—meaning that the plan covers a higher percent of expected spending ranging from 60 percent for bronze plans to 90 percent for platinum plans.
Plan-covered (rather than total) spending is what is relevant in terms of adequately compensating plans. More generous insurance plans tend to attract sicker individuals—the risk adjustment program directly accounts for this non-random selection. Specifically, the ACA’s risk adjustment program redistributes funds across plans in different metallic tiers, in order to have premiums reflect differences in benefit generosity rather than differences in the health risk of who chooses more vs. less generous plans. [iii]
However, even after accounting for this adverse selection of sicker individuals into more generous plans, these plans are expected to have higher plan-covered spending both because they have a higher actuarial value (and thus by definition cover a higher percent of expected spending) and also because people with more generous coverage tend to use more healthcare services due to lower out-of-pocket payments for care.
Fairly compensating more generous plans for this higher expected plan-covered spending is complex. This is because computing predicted plan liability is not as straightforward as simply multiplying total spending by the actuarial value of the plan, because most health insurance plans include deductibles. This creates a non-linear cost-sharing structure where enrollees with lower spending pay a much higher share of expenditures than enrollees with higher spending. This means that to predict total plan-covered spending, you can’t simply predict total spending and then multiply that by the plan’s actuarial value, because where an individual falls in the spending distribution will affect the predicted plan liability.
Accurately adjusting expected plan liability to account for the actuarial value of the plan in a reformed individual market with no pre-specified requirements on the level of benefit generosity would get very complicated very quickly. For example, despite the fact that the ACA imposes a very high level of product standardization, the ACA’s risk adjustment scheme actually requires fifteen different calibrations of the risk adjustment model to accurately predict plan liability for each combination of actuarial value (bronze, silver, gold, platinum, and catastrophic) and age group (adults, children, and infants).
Similarly, the ACA’s risk adjustment program is facilitated by the fact that all plans are required to cover the same essential health benefits. If a reformed individual market instead allows insurers to choose what benefits to cover, adjustments would be necessary to ensure that risk adjusted payments were being made only for benefits that the plan actually covered, which would be very complicated to implement. Standardizing insurance products also facilitates competition by allowing consumers to compare policies more easily, and may also reduce insurers’ ability to segment the market by attracting beneficiaries with a certain risk profile. For example, an insurer could attempt to deter sicker enrollees by not covering certain benefits, like prescription drugs or maternity care.
Reform proposals should standardize policies sold in the individual market, such that health insurance plans must cover a minimum set of health care services and be offered at specified levels of benefit generosity.
How can the ACA’s risk adjustment model be improved?
Experience with the ACA to date also highlights two ways risk adjustment can be improved, of which reformers should take note.
Incorporating Prescription Drug Claims
Some analysts have suggested that the ACA’s risk adjustment program tends to undercompensate plans for healthier enrollees without chronic conditions and overcompensate plans for sicker enrollees who do have chronic conditions.
This is particularly important due to the fact that the vast majority of enrollees in the individual market today do not actually have any chronic conditions that affect their risk adjustment score; these enrollees are adjusted on age and gender alone.
To date, the ACA’s risk adjustment program has relied only on diagnoses recorded in an individual’s medical claims in the same calendar year. That is, an individual’s insurer will not receive a risk-adjusted payment for that individual for a given chronic health condition unless he or she has seen a provider in that year who has correctly coded that condition in their medical claims.
In the 2018 Benefit and Payment Parameters rule, CMS finalized regulations following lengthy discussions about improvements to the risk adjustment program. One of the major forthcoming changes is to incorporate health status information from prescription drug claims starting in 2018. That is, if an enrollee takes a prescription drug that strongly indicates that they have a chronic health condition (e.g., insulin for diabetes, antiretrovirals for HIV/AIDS), then that enrollee would receive a condition-related adjustment even if they do not have a diagnosis recorded in their medical claims in that year.
The benefits of using prescription drug claims in the risk adjustment program include reducing the reliance on physicians and other providers to capture a patient’s diagnoses in their medical claims in every year, helping to differentiate severity across patients within a condition, and providing more timely access to data.
The potential drawbacks include concerns regarding prescription drugs not specifically being associated with a single condition (due to multiple on- or off-label indications) and the potential incentive to over-prescribe.
Prescription drug information has been used for the purposes of risk adjustment in the international context with success for several years. However, international differences (such as off-label usage patterns) suggest that ongoing research will be needed to evaluate the impact of including prescription drug information on improving the accuracy of risk adjustment in the US. However, the necessary reliance on concurrent, within-year data suggests that incorporating limited prescription drug information will help improve risk adjustment and payment accuracy.
Prescription drug claims for a select group of drugs that strongly predict the presence of a single health condition should be included in risk adjustment.
Incorporating high-cost cases into the risk adjustment formula
Some individuals have extremely high medical expenditures, which are generally not adequately compensated for even with risk adjustment. This liability can be particularly problematic for small insurers, which may not have a large enough pool of enrollees over which to spread such risk. Even with the best data and risk adjustment algorithms, it is very hard to predict which enrollees will have extremely high costs.
One way to address this issue is to include a reinsurance scheme within the context of the risk adjustment program, whereby plans are compensated for a portion of claims that any enrollee incurs over a given—typically very high—threshold. This limits plan liability and generally improves the performance of a risk adjustment model.
The ACA included a transitional reinsurance program for the first three years of the exchanges (2014 through 2016). This program provided plans with generous reinsurance coverage (with a threshold of $45,000 in 2014 and 2015) and was financed by assessments on all insurers, including self-insured products and other employer-based plans. Thus, the ACA’s reinsurance program not only protected plans from high cost claims, but also subsidized premiums in the individual market.
The termination of this transitional reinsurance program increases potential plan liability for high cost enrollees. With limited political feasibility of an ongoing externally-financed reinsurance program, the 2018 Benefit and Payment Parameters rule called for the creation of a reinsurance scheme to be included within the risk adjustment program. This reinsurance program would cover 60 percent of the costs for an enrollee whose claims exceed $1 million in a given year and is financed by drawing funds from payments into the risk adjustment program, across all states. This very high threshold is likely to be relevant only for a handful of enrollees; however, it will provide assistance for insurers who happen to enroll these few individuals.
The potential benefits include improving the functioning of the risk adjustment algorithm and reducing significant plan liability. The potential drawbacks include limiting plan incentives to manage costs for enrollees with extremely high costs and perverse incentives for some providers to employ very high charges, though maintaining some percent of plan liability above the reinsurance threshold can help alleviate these concerns.
Reinsurance for high cost claimants should be included in risk adjustment. An externally-financed reinsurance program would likely result in more generous insurer protection and reduce premiums in the individual market overall. However, if external financing in infeasible, reinsurance can be financed within the risk adjustment program. If a very high reinsurance threshold is used, pooling across states is likely to be necessary given how few individuals would likely reach such a threshold.
Any health reform proposals that include continuous coverage provisions or other limits on insurers’ ability to adjust premiums according to enrollee health status will require risk adjustment so that insurers are fairly compensated for the health risk that they enroll. Recent experience from the ACA further suggests that risk adjustment is particularly important for the individual market due to the significant variation in health risk and non-random selection into plans.
In order to facilitate a well-functioning individual market for health insurance, therefore, a robust risk adjustment system will be necessary and policymakers should standardize the products sold in this market. Additionally, risk adjustment models should incorporate information from select prescription drug claims and include a high-cost reinsurance program.
The author did not receive financial support from any firm or person for this article or from any firm or person with a financial or political interest in this article. She is currently not an officer, director, or board member of any organization with an interest in this article.
[i] Assuming they purchase the same plan at the same level of benefit generosity. They could potentially pay different premiums out-of-pocket if they have different income levels or family structures and receive different levels of federal subsidies. However, the total premium collected by the insurer would still be the same if the benefit generosity is the same.
[ii] The Medicare Advantage and Medicare Part D programs increase or decrease payments from the federal government to health insurance plans to reflect the health risk that they enroll. This is distinct from the risk adjustment program included in the Affordable Care Act, where money is transferred across insurers directly in a “zero-sum” fashion.
[iii] Catastrophic plans are excluded from this redistribution.