Introduction
On December 28, 2021, the U.S. Department of Health and Human Services (HHS) issued a proposed 2023 Notice of Benefit and Payment Parameters (i.e., a “proposed rule”) for issuers offering qualified health plans (QHPs) through Federally-facilitated Exchanges (FFEs) and State-based Exchanges (SBEs) on the Federal platform. The proposed rule touches on a range of topics related to risk adjustment, including proposed user fees, recalibration, model updates, new data collection requirements (including enrollee data on race and ethnicity), changes to the risk adjustment data validation (RADV) process, and the recoupment of high-cost risk pool funds. We provide a summary of these proposed changes below, after a brief overview of the ACA risk adjustment program.
Background
The Patient Protection and Affordable Care Act (ACA) included three premium stabilization programs— (1) risk corridors, (2) reinsurance, and (3) risk adjustment—for plans offered through FFEs and SBEs. The risk corridor and reinsurance programs lasted from 2014 to 2016, while the risk adjustment (RA) program continues.
The RA program transfers funds from individual and small group health insurance issuers that enroll lower-than-average risk enrollees to those that enroll higher-than-average risk enrollees. This program is necessary because the ACA requires plans in the Exchanges to comply with rules such as guaranteed issue and community rating that can result in increased financial risk for plans that enroll sicker individuals. The goal of the RA program is to discourage cherry picking—i.e., avoiding enrollment of individuals with high-cost health conditions.
The RA model predicts plan liability (i.e., the costs to the issuer of covering services related to the enrollees’ medical conditions) and works similar to the RA model in Medicare Advantage (MA). That is, the RA model estimates risk scores for each plan enrollee based on their age, sex, and diagnoses grouped into hierarchical condition categories (HCCs). In ACA RA, there are separate models used for adults, children, and infants to account for each age group’s inherent cost differences. There are also some additional differences between these models. For example, the adult model includes additional factors for prescription drug utilization and the number of enrolled months in a benefit year, i.e., enrollment duration, which controls for the lower prevalence of diagnostic reporting associated with enrollees with shorter enrollments. Risk scores are also multiplied by a cost-sharing reduction adjustment, recognizing that enrollees with lower cost-sharing use more services.
Once risk scores for each member are calculated, they are used to determine the relative risk of plans by issuer within specific markets. Risk scores for all the plans within a market associated with a specific insurance issuer are averaged together based on enrollment size. The enrollment-weighted risk scores are used to determine the relative risk levels of each issuer. The level of relative risk for each issuer is then used in the transfer adjustment formula, which determines the payment (or fee) an issuer will receive (or must pay) under the RA program. More details on the risk adjustment processes are provided below.
Risk Adjustment User Fee
In states where the federal government operates the RA program, an issuer of an RA-covered plan must remit a user fee to HHS equal to its monthly billable member enrollment count times the per member per month (PMPM) RA user fee specified by HHS in the annual published final rule. The proposed risk adjustment user fee for 2023 is $0.22 PMPM, down slightly from the user fee of $0.25 per member per month from 2022 (the decrease helps offset the additional costs due to increased enrollment). HHS estimates that it will cost about $60 million to operate the RA program in 2023. The federal government will be operating the RA program in every state and the District of Columbia in 2023.
Recalibration
Consistent with the 2022 payment notice, when calculating the coefficients in the RA models (i.e., the relative value of each of the risk factors in the models), HHS proposes to use the three most recent consecutive years of enrollee-level External Data Gathering Environment (EDGE) data that are available to be included in the calibration. Furthermore, HHS proposes not to update the coefficients for additional years of data between the proposed and final rules even if an additional year of EDGE data becomes available. For 2023, the available enrollee-level EDGE data is from 2017, 2018, and 2019. This means that the coefficients for 2023 will be based on a blend of separately calculated coefficients from these three years. HHS is also soliciting comments on whether 2020 enrollee-level EDGE data should be included in recalibration for future years. Some health plans have expressed concerns about the 2020 data, given that the COVID-19 pandemic changed the way health care was utilized, particularly early on during the pandemic. The 2020 data, if used, would not be incorporated until the 2024 benefit year at the earliest.
Risk Adjustment Model Updates
HHS proposes several changes to the RA models with the aim of increasing the predictive power of the models (i.e., how well they predict the costs of plan enrollees). Many of these changes were considered for the 2022 final rule, but delayed in part to allow stakeholders more time to assess the changes.
- Two-stage weighted model specification. A known weakness with the current RA model is that it underpredicts risk for low-cost enrollees, specifically those who are not associated with any HCCs. This model bias is due in part to the fact that roughly 80 percent of ACA Marketplace enrollees do not have HCCs. To address this issue, HHS proposes to estimate the adult and child models in two stages. The approach proposed would have the effect of weighting healthier enrollees (i.e., those without HCCs) more heavily so that the statistical model predicts their costs more accurately. More information on the model specifications and impacts from the change is provided in a technical paper accompanying the proposed rule.
- Interacted factors for “severity” and “transplant” with HCC counts. While the use of the two-stage model improves model prediction for healthy enrollees, HHS note that this change tends to underweight, and therefore underpredict, very expensive enrollees. In part to help address the underprediction of plan liability for the costliest enrollees, HHS proposes to include new severity and transplant indicators in the RA models that interact with HCC counts (i.e., an indicator flagging the presence of at least one severity or transplant HCC is interacted with counts of the enrollee’s HCCs that affect payment). These new interaction factors would account for the fact that costs of certain HCCs rise significantly when they co-occur with other HCCs. These new interaction factors would replace the current severity of illness indicators in the RA models.
- Change in enrollment duration factors (EDFs). As described in the proposed rule, HHS encounter data analysis of enrollee costs and utilization suggested partial year enrollees with HCCs had higher expenditures on average than full-year enrollees with HCCs. At the same time, expenditures for part-year enrollees without HCCs had similar expenditures to full-year enrollees without HCCs. Meanwhile, HHS also found that the current EDFs underpredict plan liability for partial year adult enrollees with HCCs and overpredict plan liability for partial year adult enrollees without HCCs. Enrollment timing (e.g., the beginning or end of the year) also did not affect expenditures. Considering these findings, HHS proposes to remove the current monthly enrollment duration factors of up to 11 months for all enrollees in the adult models and replace them with new monthly enrollment duration factors of up to 6 months only for enrollees with HCCs. HHS considered other durations but found that the monthly average cost variation by number of months enrolled is meaningfully reduced after 6 months.
HHS seeks comment on whether these changes should be adopted individually or only in combination. The accompanying technical paper to the proposed rule summarizes the estimated combined impact of these changes on risk adjustment transfers. The proposed addition of two-stage weighting, new interacted factors along with the removal of the adult models’ existing severity illness indicators, and revisions to the adult model EDFs may have greater impact on issuers with a higher-than-average proportion of enrollees without HCCs, enrollees with a larger number of severity and transplant HCCs, or partial-year enrollees.
Hepatitis C Pricing Adjustments
As noted in the proposed rule, generic Hepatitis C drugs did not become available on the market until 2019. Due to data lags, HHS does not believe this change in pricing for Hepatitis C drugs is adequately reflected in the data years used to recalibrate the RA models for the 2023 benefit year. Consequently, HHS proposes to continue applying a market pricing adjustment to plan liability associated with Hepatitis C drugs. The adjustment would help account for this data lag and more precisely reflect the average cost of Hepatitis C treatments expected in 2023. In addition to more accurately estimating the financial risk associated with Hepatitis C drug utilization, the adjustment will reduce perverse incentives to influence providers’ prescribing patterns that increase enrollee risk scores and produce a more favorable risk adjustment transfer. HHS intends to reassess this pricing adjustment using more recent data in the future as it becomes available.
RXC Mapping for Recalibration
As mentioned above, HHS uses prescription drug utilization, as well as diagnoses and demographic characteristics, to identify and quantify risk factors in the adult RA models. For drug utilization, HHS does this by identifying drugs that appear on claims data, either through National Drug Codes (NDCs) or Healthcare Common Procedural Coding System (HCPCS), and cross-walling them into Prescription Drug Categories (RXCs), which are similar in concept to the HCCs that categorize diagnoses. RXCs are matched to the NDCs and HCPCS applicable to the particular EDGE data year being used for RA model calibration (i.e., calculation of the RXC coefficients). HHS reviews the mappings of drugs into RXCs on a quarterly basis, using criteria such as: comparability of costs for drugs in the same class, the drug’s predictive ability of a disease, consistency between prescribing patterns and treatment of a particular condition, and stakeholder feedback.
Currently, HHS uses the most recent RXC mappings that are available when the enrollee-level EDGE data is first processed for a benefit year during recalibration of the adult RA models. For example, for the 2022 benefit year, the second quarter (Q2) 2018 RXC mapping document for both 2016 and 2017 enrollee-level EDGE data was used, and the Q2 2019 mapping document was used for 2018 data.
HHS proposes to still conduct quarterly reviews, but would recalibrate its adult RA models using the fourth quarter RXC mapping document for each benefit year of data included in the applicable benefit year’s model recalibration. This new approach would begin with recalibration of the 2023 models, in which case the Q4 RXC mapping documents from 2018 and 2019 would be used for those years of enrollee EDGE data. However, the Q2 2018 mapping document will be used for the 2017 EDGE data since that is the most recent RXC mapping document that was available at the time of first processing that data. With this approach, HHS aims to limit the volatility of some RXC factors from year-to-year and to ensure better representation of the utilization and costs observed for the underlying drugs in use in that year for the condition.
HHS also notes in the proposed rule that regardless of the version of the RXC mapping document used during the annual adult RA model recalibration, there may be a small number of drugs that still require additional analysis and consideration given the changes that can occur in the market between the data year and the applicable benefit year of risk adjustment. As such, HHS also proposes to continue to make targeted changes to drug mapping as needed. This might occur in extenuating circumstances where HHS believes there is significant off-label prescribing, substantial changes in clinical indications or practice patterns, or instances where a drug or biosimilar cost is much higher or lower than those in the same prescription drug category.
Payment Transfer Formula
HHS proposes to continue using the State payment transfer formula that was finalized in the 2021 payment rule. HHS will not republish the formula in future payment notices unless changes are proposed.
As discussed in the Background section above, once risk scores are calculated for each plan’s enrollees, HHS feeds these into its payment transfer formula to determine, for each State’s market rating areas, the relative risk for issuers in that area. In this way, a per member per month amount to be transferred among issuers is calculated based on each plan’s total member months for the plan year. The amount of the transfers is reduced by 14 percent to account for administrative costs that do not vary with claims and are normalized to the average premium in the state. Also, the costs of enrollees whose claims exceed $1 million are excluded when calculating enrollee-level plan liability risk scores. Plans would be compensated directly for 60 percent of costs in excess of $1 million (more on this later).
Issuers offering coverage in multiple rating areas within a given State have multiple transfer amounts that are grouped into a single invoice. Transfers within a given State net to zero; therefore, they are budget neutral to the federal government.
Requests for State-Specific Adjustments
In the 2019 payment rule, HHS gave States the flexibility to request a reduction to their RA transfers by up to 50 percent of the premium used in the applicable plan year. The goal of state-specific adjustments is to allow for RA transfers that more precisely account for differences in the risk in a state’s market. Currently, States must submit requests with data demonstrating that state-specific factors warrant an adjustment and that an adjustment would have a de minimis effect on premium increases. HHS can approve or deny requests and can approve a reduction that is lower than what a State requested, if warranted.
To date, Alabama has been the only state to make such requests. In 2020, 2021, and 2022, Alabama requested to significantly reduce transfers, with the rational being that the RA program is not working as precisely as it should because Alabama has a dominant carrier in its individual and small group markets.
HHS is now proposing to repeal the option for States to request a reduction in RA transfers beginning in 2024, given the low interest from states. The proposed rule includes an exception for any state that previously asked for such a reduction—which would be limited to Alabama. However, such requests are not guaranteed, and HHS could also update the criteria used for consideration. They are currently reviewing Alabama’s request to reduce transfers by 50 percent in both the individual and small group markets.
New Data Collection Requirements for Risk Adjustment
HHS proposes that plans collect and disclose five new data elements through the EDGE server: including (1) zip code, (2) race, (3) ethnicity, (4) enrollment in an individual coverage health reimbursement arrangement (ICHRA), and a (5) subsidy indicator. HHS also intends to extract three new data elements issuers already provide through the EDGE servers as part of the RA data submissions, including (1) plan ID, (2) rating area, and (3) subscriber indicator.
HHS recognizes that there may be additional burden on plans to collect the additional data elements; however, the belief is that these data illuminate important governmental interests, such as helping to monitor market trends and to inform future policy aimed at better addressing disparities in health care. Some of these new data elements—race, ethnicity, ICHRA indicator, subsidy indicator, and subscriber indicator—would be included in the limited data set that HHS makes available to researchers upon request (Plan ID, zip code, and rating area would be excluded from the limited data set to avoid providing potentially identifiable information). HHS also proposes to expand permitted uses of the RA data, potentially allowing EDGE data to be used not only for the purposes outlined in the 2020 payment notice but also to inform policies and program integrity for other HHS federal health programs (including those outside of the commercial individual and small group markets).
Separately, HHS asks for comment on how to encourage consistent use of z codes to support risk assessment analyses. Z codes are a subset of ICD-10 encounter reason codes that document social determinants of health. HHS has begun conducting analyses with z codes reported on the EDGE data but notes limitations due to inconsistent use by providers.
RADV Error Rate Calculation
To ensure that RA transfers are accurate, RA data collected from issuers must be validated, first by an independent validation auditor retained by the issuers and then by HHS. The issuer provides the auditor with demographic, enrollment, and medical record documentation for a sample of enrollees selected by HHS. The review by HHS verifies the accuracy of the independent auditor’s findings.
HHS then uses data from both reviews to conduct error estimation and calculate an error rate for each issuer. These error rates are used to calculate adjustments to an issuer’s RA transfer payments. The error rates are used to prospectively adjust the issuer’s risk scores. For example, the 2017 RADV results will impact 2018 risk scores, therefore impacting 2018 risk transfer amounts (collected and paid in 2019).
In the proposed rule, HHS proposes to change its RADV error rate calculation methodology beginning with the 2021 benefit year. A couple of the changes involve the so-called super HCCs, which are essentially HCCs that have similar effects on an enrollee’s risk score—i.e., for the more technically savvy, this means that they have similar coefficients in the RA models. Super HCCs also form what is known as a coefficient estimation group. The error rates for RADV are calculated by these coefficient estimation groups.
One of the changes is to extend the use of super HCCs to apply to coefficient estimation groups throughout the RADV error rate calculation processes, in order to make the calculations more consistent with the enrollee RA calculations. This can help avoid a scenario where an enrollee is recorded as having multiple conditions in a coefficient estimation group during the RADV process, which could require the issuer to validate both conditions to avoid receiving a RADV adjustment to the enrollee’s risk score, even though the enrollee only received the coefficient for one of those conditions in the enrollee’s RA risk score calculation (this is because the RA risk score calculation de-duplicates conditions in coefficient estimation groups in the same way that multiple HCCs that share a hierarchical relationship are deduplicated).
The other change related to super HCCs is to specify that the super HCC will be defined separately according to an enrollee’s age group model. Currently, the super HHCs are defined based on the adult model. This was done because of concerns that using the child and infant models separately would result in some infant model super HCCs with very small sample sizes, leading to less stable failure rate group assignments year over-year (HCCs are grouped into low, medium, and high failure rate groups depending on how often HCCs pass their validation in the audits). However, a recent analysis shows that, contrary to expectations, using each model’s factor definitions separately to define super HCCs could in many cases provide more stability than using only the adult models.
In addition to these changes involving the super HCCs, HHS proposes to constrain to zero any outlier negative failure rate (i.e., this happens when there are more instances of an identified HCC found during medical record audits than reported in the EDGE data) in a failure rate group, regardless of whether the outlier issuer has a negative or positive error rate. This refinement is consistent with HHS’s intent to reduce potential incentives for issuers to use RADV to identify more HCCs than were reported to their EDGE servers for an applicable benefit year.
A final note on the proposed RADV changes is that they would not alter the underlying RADV data, including the amount of data HHS collects or the second review conducted.
Recoupment of High-Cost Risk Pool Funds
Starting in 2018, the federal government created a national-level, high-cost risk pool to help further limit the risk associated with extremely high-cost enrollees in risk adjustment transfers, as well as to reduce any potential incentives for plans to avoid such enrollees. In particular, the high-cost risk pool can help control for enrollees that are not high-risk, but who may still have extremely expensive claims. Such enrollees are less likely to be controlled for by the RA model used to calculate transfer payments. For high-cost enrollees, whether they are high-risk or not, a plan will receive an adjustment to their transfer amounts equal to 60% of the costs above a defined threshold (the threshold is proposed to be maintained at $1 million for benefit year 2023). To maintain budget neutrality, the total amount of paid claims costs above the threshold is calculated to determine the amount to be transferred. Then, an adjustment is calculated as a percentage of an issuer’s total premiums in each market. Once determined, this amount is added to or subtracted from an issuer’s transfer amount calculated by excluding costs above the threshold the high-cost enrollees.
In the proposed rule, HHS proposes to address an issue regarding the disbursement of high-cost risk pool (HCRP) payments or charges recovered during an audit. If HCRP payments were already calculated for the current benefit year, HHS would use the recouped funds to reduce the HCRP charges for all insurers during the next benefit year. If HCRP payments were not already calculated, the recouped funds would be disbursed in the next benefit year in the form of reduced charges for all insurers. The proposal would not change the amount of HCRP payments or charges made.