research

Executive Summary 

Policy changes to Medicare Advantage (MA) coding intensity must reflect the diversity of plans in the D-SNP and MA-PD space. Effective coding intensity policies must balance achieving policy goals for accurate MA spending while also accounting for the heterogeneity in the MA and D-SNP markets. To assist policy makers in achieving this balance, the Association for Community Affiliated Plans (ACAP) has developed policy recommendations and policy options for consideration:  

• Continue Allowing D-SNPs to Submit Diagnostic Codes Obtained Through HRAs and Chart Reviews  

• Assess Model Enhancements to Improve the Accuracy of the MA Risk-Adjustment Model and Reduce Reliance on Annual Diagnostic Capture  

• Evaluate the Feasibility and Impact of Adding SDOH-related Metrics to the MA Risk-Adjustment Model  

• Assess Inclusion of Two Years of Diagnostic Data for MA Payments  

• Evaluate Adding a High-Cost Pool to the MA Risk-Adjustment Model  

• Evaluate Replacing the Current Coding Intensity Adjuster with a Tiered Coding Adjustment System  

Additionally, should policymakers opt to pursue an encounter-based risk adjustment model for MA, there are several policy considerations that they should first carefully consider; they are enumerated below.  

Introduction 

A stable and vibrant Medicare Advantage (MA) program is essential to assuring high-quality care for all Medicare beneficiaries. ACAP’s Safety Net Health Plan Dual Eligible-Special Need Plans (D-SNP) members serve individuals dually eligible for Medicare and Medicaid.1 Many of these individuals have complex medical needs. Often, the medical needs of these individuals are complicated by frailty, behavioral health diagnoses, and challenges with social determinants of health (SDOH) such as unstable housing, inconsistent access to transportation, and difficulty accessing healthy food and proper nutrition. On average, dually eligible beneficiaries are costly to both the Medicare and Medicaid programs.2 ACAP’s D-SNP members strive to provide these individuals with the medical care they need, to effectively align their Medicare and Medicaid benefits through close coordination, and to address their SDOH needs. Compared with non-ACAP D-SNPs, ACAP’s D-SNP members take on sicker patients, and spend more of their Medicare payments on patient care (see Figure 1).  

Figure 1. Medical Loss Ratio and Risk Scores, ACAP vs. non-ACAP D-SNP Plans, 2020. 

Medical Loss Ratio   

 

 Risk Scores  

A higher risk score indicates a higher level of medical need among a plan’s patient panel.  

Source: Analysis of CMS MLR and MA-PD bid data performed by ZAHealth for ACAP. 

Policy stakeholders have raised concerns that the current MA payment system creates incentives that result in some MA plans receiving higher capitated payments than their predicted expenditures. Many of these concerns center around coding intensity—the way that MA plans capture diagnostic codes and how these codes impact total MA payments. ACAP supports the development of policies to improve the accuracy and sustainability of the MA program for D-SNPs and traditional MA plans.  

When making changes to the MA program, it is vital that policymakers recognize that there is diversity in the types of organizations that operate D-SNP and MA-PD plans and that a one-size-fits-all approach could cause unintended consequences for Medicare beneficiaries. 

When making changes to the MA program, it is vital that policymakers recognize that there is diversity in the types of organizations that operate D-SNP and MA-PD plans and that a one-size-fits-all approach could cause unintended consequences for Medicare beneficiaries. MA payment changes can impact D-SNPs differently, particularly those integrated with Medicaid (e.g., Fully Integrated D-SNPs (FIDE SNPs), Highly Integrated D-SNPs (HIDE SNPs), and Applicable Integrated Plans (AIPs)). For example, ACAP previously expressed strong concerns about the disproportionately negative impact of the Tukey outlier deletion methodology and the Part D payment system changes on D-SNPs.   

Background on Coding Intensity 

In Medicare Advantage, plans are paid a fixed monthly amount by CMS to cover all Medicare benefits and additional supplemental benefits. These funds help pay for over-the-counter medications and support services aimed at addressing social determinants of health. To ensure there is not an incentive for plans only to enroll the healthiest Medicare beneficiaries, the monthly payment to plans is adjusted based on the demographics and health conditions of their beneficiaries through risk adjustment. MA plans document and submit to CMS the diagnostic codes for each enrollee and receive higher Medicare capitated payments for beneficiaries with more underlying health conditions. The accuracy of MA plans’ diagnostic data is audited by CMS. Plans can face steep financial penalties for errors in diagnostic and claims data.  

The number of diagnostic codes captured for MA enrollees is referred to as coding intensity. On average, more diagnostic codes are documented for MA enrollees compared with beneficiaries in traditional Medicare (i.e., Medicare fee-for-service (FFS)).3 The higher degree of coding intensity contributes to higher total spending on the MA program, relative to total Medicare FFS spending. By law, CMS must lower payments to MA plans to account for the difference in average diagnostic coding practices between MA and Medicare FFS. This is currently done through the Coding Intensity Adjustment (CIA) factor, which is set to a statutory minimum of 5.91 percent.  

The Medicare Payment Advisory Commission (MedPAC) recently found that, on average, MA coding intensity is likely higher than the current CIA factor of 5.91 percent.4 However, MedPAC also found that many plans are likely coding less intensely than the current CIA factor. Moreover, coding intensity can vary significantly by geography.5 It is outside of this policy brief’s scope to provide an analysis of the differences in coding intensity by plan type. While MedPAC estimates that MA risk scores had approximately a 6-percentage point increase from 2017 to 2021 relative to FFS,6 ACAP’s D-SNP risk scores declined by 3 percent over that same period. This suggests that ACAP’s D-SNP members have lower levels of coding intensity than other MA plans.  

 

Coding Intensity Policy Options 

The following coding intensity policy recommendations and considerations are intended to serve as a menu of options to achieve the policy goal of accurate payments to MA plans, while acknowledging the diversity of plans participating in D-SNP and MA programs, and the variation in coding practices across the industry. 

(1) Continue Allowing D-SNPs to Submit Diagnostic Codes Obtained Through HRAs and Chart Reviews  

Previously, policymakers considered not allowing MA plans to use diagnostic data obtained from Health Risk Assessments (HRA) or retrospective chart reviews for payment purposes. ACAP believes this policy would have a disproportionately negative impact on D-SNPs, particularly those operated by Safety Net Health Plans. Thus, we strongly recommend that MA plans continue to be permitted to submit diagnoses from HRAs and retrospective chart reviews for payment purposes.  

Dually-eligible individuals tend to receive care from safety net providers such as clinics and federally qualified health care centers. ACAP’s D-SNP members that work with these providers describe them as optimized for direct patient care, but not optimized for documentation and coding (as compared to providers practicing in other settings). ACAP’s D-SNP members report utilizing chart reviews to assess the accuracy of the diagnostic codes on the claims data by comparing the patient’s medical record with the diagnostic information on the providers’ claim.  

Through diagnostic information obtained through HRAs and retrospective chart reviews, ACAP’s D-SNP members obtain a more accurate and complete set of diagnostic data on their dually eligible members than is available from provider claims alone. This data is used by CMS to adjust the plans’ Medicare payments and is used by the plan to connect the member with needed care and for care management. Prohibiting D-SNPs from using diagnostic data from HRAs and retrospective chart reviews would inappropriately reduce Medicare payments to D-SNPs. The reduced Medicare payments would not be accurate, and would not reflect the true cost of the dually eligible individuals’ needed care.   

(2) Assess Model Enhancements to Improve the Accuracy of the MA Risk-Adjustment Model and Reduce Reliance on Annual Diagnostic Capture  

An accurate risk-adjustment model is necessary for the sustainability of the MA program. CMS significantly improved the accuracy and predictive power of the MA risk-adjustment model for dually-eligible individuals by introducing the six-segment model in 2017. However, there are additional changes that can be made to the MA risk-adjustment model that will both improve the model’s accuracy and reduce the model’s dependence on annual diagnosis capture.   

The bullets below present options to potentially improve the predictive power of the MA risk-adjustment model and incrementally reduce the importance of clinical diagnostic capture to the risk-adjustment model. ACAP recommends that policymakers assess and evaluate each of these policy options on both their impacts to the predictive power of the MA risk-adjustment model and on addressing concerns around coding intensity. Importantly, policymakers must assess the impact of these policy options on all MA plans, and separately on all D-SNPs and D-SNPs integrated with Medicaid (i.e., FIDE SNPs, HIDE SNPs, and AIPs).  

• Evaluate the feasibility and impact of adding SDOH-related metrics to the MA risk-adjustment model. Since social determinants of health significantly impact overall health, better utilization of SDOH information in the MA risk-adjustment model could reduce the importance of diagnostic codes to the risk-adjustment model, while still maintaining the accuracy of the risk-adjustment model for MA plans.   

While individual SDOH metrics7 are currently not widely collected by MA plans or Medicare providers with enough consistency for inclusion in the MA risk-adjustment model, we encourage CMS to investigate the predictive power of SDOH proxies in the risk-adjustment model. Examples of these proxies may include Healthy Places Index, county designation (large metro, rural etc.), months of Medicaid enrollment in the last year, beneficiary household size, zip code, median income, or Gini coefficient. SDOH proxies should be evaluated both individually and collectively.   

• Assess Inclusion of Two Years of Diagnostic Data for MA Payments. ACAP recommends that policymakers consider using two years of diagnostic data in the risk-adjustment model. This policy option has been modeled by MedPAC and recommended multiple times since 2021.8 In considering this policy option, policymakers should ensure that this policy change does not have the unintended negative effect of incorrectly reducing payments for D-SNPs and MA plans with higher proportions of new enrollees or enrollees with multiple complex chronic conditions.9   

• Evaluate Adding a High-Cost Pool to the MA Risk-Adjustment Model. The value (i.e., weights) of diagnosis codes in the risk-adjustment model can be influenced by very high-cost beneficiaries. These outliers may skew the weights of those diagnoses, and thus, result in average Medicare payments for those codes that are higher than the median costs for those diagnoses. In addition to resulting in higher Medicare payments for some diagnoses, outliers also decrease the predictive power of the risk-adjustment model.   

Allowing average costs to be closer to median costs (by carving out the outliers of the highest-cost cases) would better align the MA risk-adjustment model’s predictions to plans’ actual costs.10 For example: CMS could set aside the costs of the highest X percent (e.g., 1, 2, or 3 percent) of beneficiaries into a high-cost pool; the coefficients for each diagnosis in the risk-adjustment model would then be lower, on average; when a high-cost beneficiary reached a certain expenditure amount, funds from the high-cost pool would be used to reimburse that beneficiary’s MA plan directly. This concept of a high-cost pool is similar to that of reinsurance, but would be built directly into the risk-adjustment model.  

Policymakers should consider the impact of this approach on the accuracy of the risk-adjustment model prediction for all MA plans and separately for integrated D-SNPs (FIDE SNPs, HIDE SNPs, and AIPs). Because high-cost pools can be tailored, some of the MA risk-adjustment model segments could include or exclude a high-cost pool, or the model segments could have different levels of outlier exclusion (e.g., 1 percent, 2 percent or 3 percent set aside for the high-cost pool).  

(3) Evaluate Replacing the Current Coding Intensity Adjuster with a Tiered Coding Adjustment System  

As we noted earlier, there is large variation among plans in coding intensity. Conceptually, policymakers concerned about coding intensity could tailor the coding intensity adjuster based on individual plan’s level of coding intensity. However, without the ability for a plan-specific coding intensity adjuster, some stakeholders have floated the idea of a tiered coding intensity adjuster. Under this concept, MA plans would be placed in “tiers” – such as high, average, or low-level diagnostic coders – and the coding intensity adjuster would be different for each tier. A benefit of this policy could be that a lower coding intensity adjuster than the current 5.91 percent adjuster, would be applied to plans that are capturing diagnoses at a similar level to FFS.  

However, many technical challenges remain to developing a tiered coding system. Most notably, it is unknown how policymakers would measure each MA plan’s level of coding intensity, or which levels of coding intensity would be considered high, average, or low. Another outstanding question is whether the value of the coding intensity adjuster in each tier would vary geographically to reflect geographic coding pattern differences. With respect to D-SNPs, a tiered coding intensity adjuster would have to be refined, such that it does not penalize D-SNPs for having sicker populations, and thus having higher risk scores than simple demographic factors would predict. Finally, the accuracy of benchmarking coding levels for integrated D-SNPs (FIDE SNP, HIDE SNP, and AIP) may not be possible in many geographic areas, where, by CMS policy design, most dually-eligible beneficiaries in Medicaid managed care plans will likely be enrolled in the integrated D-SNP rather than in Medicare FFS. Thus, integrated D-SNPs may not have a sufficient FFS cohort for an accurate comparison.  

 

 

 

As additional work is conducted to better refine a tiered coding-intensity adjustment, ACAP has several guiding principles:  

• Integrated D-SNPs (FIDE SNPs, HIDE SNPs, and AIPs) should, by policy, be placed in the lowest coding intensity adjuster tier. CMS has already acknowledged that the MA risk-adjustment model does not adequately account for the complexity of this population (e.g., a frailty adjuster is applied to FIDE SNPs based on the level of frailty in their population), and that – as discussed above – there may not be an accurate FFS cohort of dually eligible individuals in many geographic areas. Therefore, any comparison in coding intensity to other plan types will not be appropriate.  

• A tiered coding intensity adjuster should account for changes in the population served by an MA plan, such as plans with new members or those expanding into new areas. Changes in the underlying plan population could affect the risk profile of a population in a way that could be independent of any change in coding intensity.  

• The tiered coding system should account for sub-populations of Medicare beneficiaries that are truly sicker than their demographics would indicate. This is a concern because many models that attempt to determine coding intensity compare diagnoses of demographically similar cohorts. However, those demographic cohorts may not fully capture the factors that contribute to the higher illness burdens seen among some sub-populations of dual-eligible beneficiaries.  

• The tiered coding system should include flexibility for smaller MA plans that serve populations different from traditional MA-PDs. This would be necessary because in smaller populations, a few individuals could increase average risk scores in a way that may appear to be higher coding intensity. There will always be some fluctuations in diagnoses from year-to-year that need to be allowed for when determining any coding intensity tiers.  

(4) Policy Considerations for an Encounter-Based MA Risk-Adjustment Model  

At this time, ACAP does not recommend policymakers change the MA risk-adjustment model to be based on MA encounter data rather than on Medicare FFS (currently, the relative weight of factors in the risk-adjustment model is developed based on Medicare FFS spending data). However, should policymakers pursue such a policy, there are several considerations we recommend be taken into account in the design of an encounter-based MA risk-adjustment model.   

An encounter-based risk-adjustment model would utilize Medicare Advantage encounter data to calibrate the risk-adjustment model. Today, while MA encounter data is used for calculating the risk scores, the model is calibrated using FFS data. The appeal of an encounter-based MA risk-adjustment model is that, in theory, it would negate the problem of coding differences between MA and FFS, since MA payments would no longer be influenced by FFS coding levels.   

However, if not designed correctly, an encounter-based MA risk-adjustment model could result in a system where MA payments are shifted away from plans that do not capture all their enrollees’ diagnosis codes, and toward plans that are aggressive coders. This redistribution of payments would, in essence, reward plans for coding aggressively.  This raises real concerns about the impact of this distribution on plans with less coding intensity – as such a redistribution could threaten the sustainability of those plans.   

To address these concerns, ACAP proposes that any encounter-based risk-adjustment model include the following features:  

• The encounter-based risk adjustment model must not advantage organizations that currently have the most aggressive diagnosis capture practices; it should not, in effect, transfer money from industry-average coders to those that invest the most in capturing diagnoses.  

• An encounter-based model should not destabilize the MA market in certain geographies because of geographic differences in coding practices.  

• An encounter-based model should reflect differences between the FFS and MA populations. Because statutory payment benchmarks are based on a percentage of FFS, any encounter-based model must pay based on the average FFS beneficiary. For example, if MA plans are more efficiently treating patients compared with FFS, that value of MA should not be discounted from any payments to MA plans.  

• Such a model should at least maintain the existing model segments used in risk-adjustment today.  

Conclusion 

ACAP is committed to working with policymakers and stakeholders to ensure the sustainability of the Medicare Advantage program, especially for plans serving the most vulnerable dually-eligible beneficiaries. ACAP understands the concern that coding practices increase total MA payments relative to FFS. However, it is critical to remember that the degree of coding intensity varies across types of MA plans, geography, and populations served. For this reason, ACAP proposes policy solutions and considerations on coding intensity that continue to encourage the identification and management of chronic conditions, but that do not widen the gap in coding intensity between MA and FFS. For coding intensity policies to be effective over the long term, those policies must account for the different types of MA plans and the impact of coding intensity policies on D-SNPs.