Opinion: 4 ways to improve ACA risk-adjustment model

The Affordable Care Act (ACA) set up the three “R”s to help stabilize the individual and small group markets.  Two of these programs – reinsurance and risk corridors – are temporary, while the third program – risk adjustment – is permanent.

ToddThe purpose of risk adjustment is to ensure that health plans that enroll sicker enrollees are appropriately compensated. Under the ACA’s risk adjustment program, within each state, plans with healthier enrollees make transfer payments to plans with sicker enrollees. If payments are in line with expected healthcare costs, then plans are encouraged to compete based on cost, quality, and health management, and not to compete for healthier enrollees.

Recent media reports have stated that health plans operating in the individual and exchange markets are experiencing significant financial losses, as a result of lower-than-expected enrollment among younger enrollees, and higher healthcare costs for enrollees. As a result of these challenges, some health plans have exited the market. Such dynamics have renewed the focus on policy solutions to help stabilize these markets, including improving the accuracy of the risk-adjustment model.

KornfieldThe ACA risk-adjustment model is called the Department of Health and Human Services-Hierarchical Condition Categories (HHS-HCC) model and applies to plans sold in the individual and small group commercial market, including plans sold through exchange marketplaces.

In response to concerns about the accuracy of the risk-adjustment model, the federal government held a meeting in March in conjunction with the release of a whitepaper to discuss strategies to improve the model.

Here, we identify four main limitations in the HHS-HCC risk-adjustment model, as well as opportunities for improvement and whether CMS is considering such modifications.

Why risk adjustment is critical

One of the primary goals of the ACA is to increase access to affordable health insurance coverage though several new requirements:

• Insurers are required to offer health insurance to all qualified applicants and cannot set premiums based on health status. As a result, premiums are set based on the expected average costs of the membership.
• The ACA includes a set of provisions designed to make insurance affordable in the exchanges by providing subsidies for lower-income individuals.
• The ACA assesses a penalty to individuals for not enrolling in a plan.

Together, these mechanisms, are intended to encourage enrollment and broaden the risk pool to lower average costs for all enrollees. In this type of market where premiums are not based on health status, risk adjustment plays a critical role in ensuring that health plans are adequately compensated for sicker enrollees and have an incentive to participate in the market and compete based on value.  

The HHS-HCC model is based on the design of the model used for Medicare Advantage. The model groups all diagnosis codes into a set of broad conditions or HCCs (e.g. asthma or congestive heart failure). The model calculates a risk score based on the diagnoses and demographics for each enrollee in a plan and averages those scores for each plan.

These average “risk scores” are then used to determine the transfer payments that are made between plans within states. These transfer payments can have a significant impact on health plan financials but can be hard to predict, particularly if the risk-adjustment model does not accurately predict costs associated with the population. 

Next: Limitations of the commercial HHS-HCC Model

Limitations of the commercial HHS-HCC Model

The HHS-HCC model has several limitations that reduce its predictive accuracy. Improving the accuracy of the risk adjustment program can improve the predictability for plans and reduce financial disincentives to enroll higher cost populations.  

1. The commercial population is different than the large employer group population used to estimate the model

The HHS-HCC risk adjustment model is developed from a database of large employer self-funded commercial claims (MarketScan®).  This database does not reflect the population characteristics of the individual and small group commercial market.

Differences in composition of actual enrollees compared to those used to estimate the model can result in systematic distortions of expected healthcare costs. As shown in the table below, this new commercial market-particularly exchange enrollees that comprised roughly one-third of the individual market in 2014-has a different demographic profile than the large employer population used to create the model.

Within the adult population, both exchange enrollees and the individual market population are significantly older than the employer population. Differences in the age distribution for adults could impact the accuracy of the risk adjustment model, since younger enrollees are healthier and even those with medical conditions are less likely to have comorbidities or experience complications when compared to an older adult population.

Additionally, individual market enrollees have lower incomes than the employer population, which may result in the model under predicting their costs, since disease profiles and utilization patterns are different for lower versus higher-income individuals. 

Exploring ways to estimate the model based on the actual individual and small group commercial population could improve the accuracy of predictions. In its whitepaper, CMS notes this shortcoming of the model.  The agency has suggested that it may consider collecting detailed patient level data from health plans participating in this market to estimate the model in the future.

Notes: MarketScan® data are from 2010 database. Individual and exchange data based on the 2015 Current Population Survey. Exchange population represents enrollees during the 11/15/14 to 2/15/15 open enrollment period, and the individual market data is for 2014 coverage.

Next: High rates of partial-year enrollment skews estimation of health plan risk

2. High rates of partial-year enrollment skews estimation of health plan risk

Another factor that may result in inappropriate risk adjustment is the high rate of churn within the individual market.

Enrollees in the individual market may shift in and out of coverage within the year as a result of changes in employment, income, and availability of other sources of coverage (e.g., Medicaid or employer coverage). 

Specifically, the risk model predicts costs assuming full year enrollment-with spending spread evenly throughout the year. For individuals enrolled for only part of the year, their costs are often concentrated during the period of their enrollment, which means the risk-adjustment will undercompensate the plan for that period of participation.

Initial Avalere analysis of Inovalon’s Medical Outcomes Research for Effectiveness and Economics Registry® (MORE²) for individual market enrollees from 2014 and 2015 finds that over 30% of enrollees in the individual market are enrolled for less than one year.  On average, part year enrollees have costs that are 18% higher than what would be predicted based on the risk-adjustment model for full year enrollees. 

These initial findings suggest that the model may not be fully capturing or predicting cost differences for these partial-year enrollees.  CMS has suggested making changes to adjust for the partial year enrollment costs, although use of the MarketScan data could mean that these adjustments may not truly capture the churn in the commercial market.

3. Diseases included in the model are not accurately capturing the health needs of the population

Similar to the Medicare Advantage model, to develop the HHS-HCC model, the federal government worked with a clinical panel of experts to group diagnosis codes together and then selected a subset of disease groups for inclusion in the model.

Through this process, however, the model ended up having a very small proportion of individuals (only 19.2%) who had at least one HCC.

Although the HHS-HCC model includes over 3,000 diagnoses, the grouping methodology and approach based on the Medicare model could mean that diseases more often found in the commercial population could be excluded. 

CMS selected 127 of 264 HCCs for inclusion. By including only 127 HCCs, the model may underpay plans without HCCs because their risk scores are artificially low due to a high concentration of expenditures in such a small number of enrollees.

Moreover, if the risk adjustment methodology fails to account for the actual conditions of the population, the model may not appropriately compensate plans for differences in costs of individuals without HCCs.

Providing more transparency on how HCCs are selected for inclusion in the model could help stakeholders assess options to improve the model.  CMS did not address disease grouping in its whitepaper.

Next: By excluding prescription drug information to identify an enrollee’s health conditions, the model does not take a holistic view of healthcare costs

4. By excluding prescription drug information to identify an enrollee’s health conditions, the model does not take a holistic view of healthcare costs

Premiums charged by plans in the individual and small group markets are designed to cover all expenditures, including medical and prescription drug costs.

The model takes this into account by using medical and prescription drug costs to predict total relative healthcare costs associated with health conditions. The HHS-HCC model does not, however, use prescription drug information to identify health conditions.

Prescription drug information can be a useful and reliable source of information to identify health conditions that may not be identified with diagnosis codes. Lack of diagnosis codes may be particularly acute for this population, given high rates of churn in the market.

Additionally, prescription drug use may help classify the severity of disease. For example, individuals could be stratified based on their use of medications that correspond to different levels of disease progression. Under the current model, plans may be systematically undercompensated for enrollees who use high-cost medications because they are not used to inform disease identification and severity.

Adding prescription drug information to the model could improve the accuracy of the model and result in broader prescription drug coverage for enrollees. CMS’ whitepaper considers four options for adding prescription drug utilization as a factor to the risk adjustment model to improve prediction of health conditions, severity, and costs.

Future outlook

Stability in the individual, small group, and exchange market relies heavily on improving the accuracy of the risk adjustment model.

CMS noted in its whitepaper that it is considering modifying its risk adjustment model and sought comments on modifications, such as making the risk adjustment model more accurate for partial year enrollees as well as considering prescription drug utilization as a predictor.

Improvements like these will allow plans to receive payments that better reflect the needs of their enrollees, help patients get access to appropriate care, and foster a more stable and competitive individual and small group market focused on value vs risk aversion.   

Any formal changes to the model for 2018 would be released by CMS in the proposed Notice of Benefit and Payment Parameters, which should be published by the fall of 2016.

April Todd, senior vice president, leads Data Analytics and Product Development for Avalere. Her team provides creative data driven insights and product solutions to inform strategic decisions for complex business challenges and policy issues through the use of various data assets.

Tom Kornfield, vice president, Avalere Health, provides strategic advice to clients on the impacts of health reform, Medicare Advantage and Part D policy.  Prior to joining Avalere, Tom was a Senior Advisor in the Center for Medicare and Medicaid Innovation at CMS where he assisted in developing alternative payment models in Medicare Advantage.