In two Forefront articles on September 29 and September 30, 2021, “Medicare Advantage, Direct Contracting, And the Medicare ‘Money Machine’,” Drs. Don Berwick and Rick Gilfillan argue that Medicare Advantage (MA) plans receive higher premiums relative to what the Centers for Medicare and Medicaid Services (CMS) would pay in traditional Medicare (TM). They attribute this to the incentive in MA to capture diagnoses in order to increase risk scores, and in turn premium payments, for MA plans. As a result, one of their recommendations is to eliminate the disease-based risk-adjustment hierarchical condition category (HCC) risk scoring system.
The debate that Berwick and Gilfillan triggered continued with responses by George Halvorson and Don Crane, and a further response by Berwick and Gilfillan. While the original blog posts by Berwick and Gilfillan addressed the role of risk adjustment in the MA system (the subject of this response) the subsequent debate has been more concerned with the MA business model and its effect on member quality and financial outcomes.
The subsequent debate on the flaws or virtues of MA and capitation shifts the focus from the underlying problem, relatively higher payments in MA versus TM and their cause, and thus fails to provide a meaningful solution. If the relatively higher payments were to go away then MA plans would compete with TM on a level playing field, the original Congressional intent for the two programs. We will therefore respond to the coding and risk adjustment issues raised in the original Berwick and Gilfillan posts, in particular focusing on the root cause of the higher payment, the rationale for disease-based risk adjustment, and a potential solution.
We agree that there is evidence for higher payment to MA plans relative to TM. What is unclear is whether this is due to an overcoding problem in MA or an undercoding problem in TM. The incentive in MA is for physicians to code comprehensively for diagnosis (using International Classification of Diseases (ICD) codes, specifically ICD-10 codes); the incentive in TM is for physicians to code procedures (using Current Procedural Terminology (CPT) codes).
The important purpose of disease-based risk-adjustment is to appropriately pay MA plans for members’ “burden of disease.” Instead of abandoning disease-based risk adjustment, policy makers should deal with the underlying incentives related to coding. The correct short-term solution is to adjust for relatively higher coding intensity in MA. In the medium term, instead of abandoning disease-based risk-adjustment in MA, CMS should introduce disease-based risk adjustment in TM. This would (a) not only address the coding intensity differences between MA and TM but (b) also more appropriately direct resources to providers who are caring for sicker patients within TM.
Then, in the long run, once the two programs are on even footing (i.e., with similar incentives for capturing diagnosis codes), it would be appropriate to look at improving risk adjustment in both programs to account for patient differences, for example in both functional status and social determinants of health (SDOH).
Making The Right Diagnosis: Is There An Overcoding Problem In MA Or An Undercoding Problem In TM?
Drs. Berwick and Gilfillan believe that the diagnostic code creep in MA is a bad thing. Their implicit hypothesis is that correct diagnostic coding occurs in TM. The problem is that in TM, fee-for-service (FFS) physicians are not reimbursed for diagnostic accuracy, but rather are reimbursed for procedures on the basis of resource-based value units (RBRVS), a labor-value theory concept developed by Bill Hsiao in the late 1980s, for each CPT code that they submit. Although a diagnosis is required for a claim to be reimbursed, for physician services only a single diagnosis is required.
Furthermore, reimbursement is based on units of work as embodied in the CPT code rather than the patient complexity (comorbid conditions). A physician can see a complex polychronic patient 100 times per year, providing the diagnosis of diabetes for every visit, and be paid the same amount (for the same CPT code) even though the patient has six other diseases. The patient’s comorbid heart conditions, hypertension, and other conditions are not required for reimbursement and may not be recorded.
FFS reimbursement, therefore, does not provide an incentive for recording patient diagnoses and complexity. The FFS system rewards the “inputs” of care regardless of patient complexity or clinical outcome. A physician treating the same patient in an MA plan is incented to record the full range of the patient’s diagnoses. The problem is more likely under-coding in TM than an over-coding in MA.
For example, let’s take a patient who has three chronic diseases: CHF, diabetes with nephropathy, and peripheral vascular disease. Let’s say the patient is seen three times in a year. In MA, this patient would have all three diagnoses coded over three visits, but the same CPT code for two intermediate and one brief visit could be used for the three visits. Conversely, in TM the same patient would have one comprehensive and two intermediate visits coded but only the diagnosis of diabetes submitted. In MA the physician has an incentive to capture all three diagnoses including complications (diabetes with nephropathy attracts a higher HCC score than diabetes by itself) accurately. In TM the physician is paid based on CPT codes regardless of what diagnosis is recorded. Thus, it is more likely that the same patient’s diagnosis codes are under-coded in TM versus over-coded in MA.
In a Health Affairs Forefront article in 2020, Rick Kronick (a critic of MA risk adjustment and former CMS official) gives a telling example of Medicare FFS under-coding: “In TM, approximately 40 percent of beneficiaries coded with quadriplegia in a 12-month period do not have a diagnosis of quadriplegia appear on any claim in the subsequent 12 months.” It is difficult to argue that the corresponding patient in an MA plan with an annual quadriplegia code is over-coded.
The Rationale For Disease-Based Risk Adjustment In MA Plans
Congress intended that the MA Risk Adjustment (RA) calculation produce actuarial equivalence between payments under the FFS system and MA, resulting in a statutory requirement that “health insurance companies participating in the Medicare Advantage program be paid in a manner that ‘ensures actuarial equivalence’ to traditional fee-for-service Medicare,” (Taylor, 2017) as described in 42 U.S.C. §§1370a-7k(d)(1)). Because plans cannot adjust benefits to ensure equivalence, an adjustment has to be applied to the plan’s revenue stream (amounts paid to the plan by CMS). The key reference is 42 USC §1395 w-23 at page 2904:
(C ) Demographic adjustment including adjustment for health status
(i) In general
Subject to subparagraph (I), the Secretary shall adjust the payment amount under subparagraph A(i) and the amount specified under subparagraph B(i), B(ii), and B(iii) for such risk factors as age, disability status, gender, institutional status and such other factors as the Secretary determines to be appropriate, including adjustment for health status under paragraph (3) so as to ensure actuarial equivalence. [emphasis added] The secretary may add to, modify or substitute for such adjustment factors if such changes will improve the determination of actuarial equivalence.
At the heart of MA revenue is the plan’s risk score. Risk scores are based on CMS’s risk adjustment model which has been in use (with updates) for decades. Earlier risk adjustment models for MA in the 1980’s and 1990’s were based only on demographic data (age and gender), “dual” eligibility for both Medicare and Medicaid, institutionalization, and the original reason for Medicare entitlement (e.g., age, disability or end-stage renal disease).
Early on, the beneficiary’s disease profile was not included in risk adjustment. However, the health care system exists to take care of patients’ needs—their diagnoses—and resource utilization is correlated to the burden of disease. So, in the late 1990s CMS (then the Health Care Financing Administration) developed the HCC risk adjustment system. HCC is a grouper model that maps selected diagnoses from more than 80,000 ICD-10 codes into 86 condition categories. Using a multi-variate logistical regression model, each category is then assigned a weight based on its relationship to TM resource utilization.
CMS added HCCs to its health plan risk-adjustment system in 2004 to accomplish two overarching policy goals: first, to ensure that health plans that had a sicker population (“adverse selection”) would get higher reimbursement to reflect the disease burden of its enrollees; second, to remove the incentive for plans to focus their enrollment efforts on skimming the healthy by giving plans with healthy beneficiaries lower payments. This skimming certainly adds no value to the system and created immediate profits for MA organizations.
Since the advent of disease-based risk adjustment, the MA program has been taking on older and sicker patients. Appropriate risk-adjustment allows prospective, value-based clinical models to flourish, especially for beneficiaries that need it the most: frail, complex, elderly patients. For the providers of care, it represents a significant enhancement of personalized care for the senior population, providing better access, better care, and better benefits to the sick beneficiaries who need it the most. And, in turn, for MA plans, risk-adjusted payment allows them to implement population health programs such as patient monitoring, palliative care, meals, transport, and the like, that individual physician practices cannot offer.
The Short-Term Fix: Adjusting For Diagnostic Coding Intensity Differences Between MA And TM
While the public policy reason for introducing disease-based risk adjustment is clear, compelling and on-going, CMS has long recognized that the application of the disease-based risk adjustment model will encourage differences in coding intensity at the program level between MA and TM. MA plans are likely to record a more complete range of diagnosis codes than FFS Medicare (see for example Kronick & Welch, 2014; Geruso & Layton, 2018; GAO 2012 & 2013; Frogner et al., 2011; MedPAC 2012; MedPAC 2017 & 2018, Jacobs & Kronick, 2018, Kronick 2017.) Applying to the model a broader range of recorded diagnoses for an individual with otherwise equivalent risk factors (including disease burden) is likely to result in a higher risk score for the individual with the wider range of diagnoses. (The effect of a broader range of recorded diagnoses will depend on the mapping of diagnoses to HCCs, however. Diagnoses that map to the same HCC would not affect the member’s risk score; diagnoses that map to a more severe HCC within the same diagnostic category, or that map to a different HCC would increase the risk score.)
To compensate for the resultant higher payment to health plans, CMS applies a Coding Intensity Factor (CIF) to a health plan’s revenue. The Medicare Advantage CIF is a mandated reduction made to all Medicare Advantage plan risk scores to account for differences in coding patterns between MA and FFS Medicare coding practices on which the HCC weights are based. CMS implemented the CIF in 2010 at 3.41 percent. The CIF has changed slowly over subsequent years and is now 5.91 percent – substantially less than projections by the Medicare Payment Advisory Commission (MedPAC) and others of the growing difference between TM and MA.
To date CMS has pursued two approaches to the Berwick-Gilfillan objection to Medicare Advantage plan reimbursement. One is to find a way to increase the CIF adjustment. But this approach continues to reward plans that have invested the most in risk adjustment factor coding and not better care. The other approach is to change risk adjustment inputs to move away from claims to physician encounter data that are assumed to be more reliable and independent than health plan reporting (claims). Both approaches focus on MA only and neither approach addresses the root cause of the problem: under-reporting of diagnoses in TM.
Considering Disease-Based Risk-Adjustment In TM: The Purpose Of Risk Adjustment Is To Direct Resources To Patients According To Their Disease Burden
Most health policy experts would agree that risk adjustment is needed, and they would agree that disease-based risk adjustment, in addition to other factors, should be a crucial part of risk adjustment. The challenge is that beneficiaries do not make diagnoses; physicians make diagnoses and record them in claims (that are subject to scrutiny and audit). So, an accurate disease-based risk adjustment system cannot rely on patient self-reported diagnoses as Berwick and Gilfillan suggest but rather requires clinicians to make and record all diagnoses. The consequence is that financial incentives in MA for complete diagnostic coding versus lack of these incentives in TM result in differences in diagnostic coding capture between the two programs.
The policy goal is to pay MA plans the actuarial equivalent for similar patients in TM. One way to deal with this is through coding intensity adjustments noted above. The other way to deal with the differences in coding intensity due to the differences in payment incentives is to use incentives in the FFS Medicare system similar to those in MA to ensure complete diagnostic coding.
Using physician-documented diagnoses is and should be a central pillar to any risk adjustment system in the future and should be used not only in MA but also TM. In MA, the purpose of disease-based risk adjustment is to reallocate existing resources to plans that are taking care of sicker patients. The same concept is true for TM. Disease-based risk adjustment should be used within TM to re-allocate resources to providers who are taking care of sicker patients. In TM, the same HCC system could be used at the premium level (in CMS’s direct contracting initiative) or at the episode level in their bundled payment initiatives, or at the service level for physician reimbursement. Irrespective of the initiative, complete diagnostic coding should be required.
Furthermore, from a value-based outcome perspective (capturing and documenting the correct diagnoses) in a value-based world is a more valuable and clinically coherent activity than capturing CPT codes. Capturing and coding the correct diagnoses would create more complete problem lists which would drive better comprehensive care for all of the patient’s diseases. It would also allow rational comparisons between quality of care in the two systems: Without complete diagnostic coding in TM, it is difficult to compare treatments and quality outcomes between patients in different systems.
With equal incentives to code and report diagnoses in TM, both the undercoding issue in Medicare FFS and the overpayment issue in MA would be eliminated. Disease-based risk adjustment in both MA and TM would encourage physicians to simply “make the right diagnosis” and accurately capture and code that diagnosis regardless of what system they are practicing in.
The Future: Improving Risk Adjustment For Both Medicare Advantage And TM
The growing importance of risk adjustment in both MA and TM for CMS reform priorities raises a further complication: disease-based risk adjustment alone is likely biased against populations from diverse racial and ethnic backgrounds, lower socioeconomic status, and rural areas. In particular, HCC risk adjustment appears to do a relatively poor job of capturing beneficiary complexity and resource needs related to race, ethnicity, socioeconomic status, and functional status.
Drs. Berwick and Gilfillan acknowledge this and consequently urge consideration of reforms in risk adjustment for MA plans that incorporate these adjusters (functional health status, socio-economic factors, etc.) more accurately and completely. But assuming we can identify less-biased ways of measuring the association of these adjusters with resource needs than is currently possible in HCC adjustment, incorporating the improved methods only in MA payments would create a new kind of bias. In short, if we only adjusted in MA, the result would be a different distribution of spending across these expanded “HCC Plus” subpopulations than in the corresponding FFS population.
Should MA be the primary or only policy vehicle to shift additional resources to such patients? Without a more effective mechanism for capturing these expanded factors affecting health needs in TM, and for implementing comparable resource shifts in FFS, it will be hard to achieve Berwick and Gilfillan’s goal of better care models in TM—where even modest shifts to innovative models have proved challenging.
Finally, it is hard to imagine a risk adjustment system that uses only functional status or only SDOH. A poor person who cannot walk up a flight of stairs would not have the same health care utilization needs compared with a poor person who cannot walk up a flight of stairs due to arthritis but also has CHF, diabetes and COPD. Disease, functional status, and SDOH act independently, but they are correlated to health status and needs. Each by itself does not fully predict care and resource needs and all three together are more likely to be better correlated with future spending.
Recommended Path Forward
In the short run, rather than abandoning disease-based risk adjustment in MA, which would return us to a world where MA plans are incentivized to “skim the healthy” and plans that experience adverse selection would be underpaid, we believe that CMS could use the tools it has in place to deal with coding intensity issues and continue to move to physician-reported encounter data. Indeed, Berwick and Gilfillan acknowledge that this is a viable short-term solution: “In the absence of eliminating the HCC system, CMS should move immediately toward a more realistic Coding Intensity Factor (CIF) …” (page 10/18 part 2). Increasing the CIF would be consistent with the statutory requirement of actuarially equivalent payments between MA and FFS Medicare, but it is not a solution to under-diagnosis in TM.
In the medium term, rather than throwing out the MA disease-based risk adjustment approach, CMS should continue down the path marked by the CMS Innovation Center of using disease-based risk adjustment approaches in Medicare FFS. We can imagine a world where it will not matter whether physicians are practicing in FFS or managed care or both; all physicians will have incentives to code diagnoses accurately and completely, and then care for all the conditions of the complex patient, with reimbursement appropriate to the patient’s disease burden.
Berwick and Gilfillan overlook the important incentive provided by the capitated reimbursement of the MA program: Capitation provides strong incentives to a plan to manage patient conditions while encouraging preventive care to avoid later complications. Applying the same incentives to FFS physicians would eliminate the differential between MA and TM and make the programs actuarially equivalent for patients with the same diagnoses. Providing incentives to FFS physicians would encourage these physicians to make a correct diagnosis as soon as possible and treat all the patient’s problems.
In the long run, the best risk adjustment methodology for both programs would include disease status and both functional status and SDOH. There is compelling evidence to support the addition of SDOH (see Irvin et al., 2020; Gruenberg et al., 1996; Chen et al., 2015; Johnston et al., 2020; Center for Innovation in Medicare Advantage 2021, Johnston et al., 2018) and functional status to improve the accuracy of risk adjustment. SDOH and functional status could be obtained through patient self-reported instruments, as Berwick and Gilfillan suggest. We would applaud CMS’s efforts to add these and other risk-adjusters to physician-reported, disease-based risk adjustment.
Acknowledgements: The authors would like to thank both Ms. Nancy-Ann DeParle and Dr. Mark McClellan for their review and contributions. Ms. De Parle was HCFA (Health Care Financing Administration, now CMS) Administrator when the original HCC methodology was developed. She is currently a Managing Partner & Co-Founder at Consonance Capital Partners. Dr. McClellan was CMS Administrator when disease-based (HCC) risk adjustment was implemented. He is currently the founding Director of the Duke-Margolis Center for Health Care Policy.
Conflicts of interest: Dr. Kang is CEO of WellBe Senior Medical which provides longitudinal geriatric care to underserved, frail, complex, homebound Medicare Advantage beneficiaries. His company’s primary care model would not be possible without adequate risk adjusted payments.
Santa Barbara Actuaries, Dr. Duncan is president, uses risk adjustment for its analytical and actuarial services to healthcare management organizations.