Risk Adjustment Validation Under Medicare Advantage: Part I

Introduction

Medicare Advantage (MA or Medicare Part C) is quickly becoming the program of choice for people eligible for Medicare.  The number of people enrolled in MA has doubled in the last decade, amounting to over 22 million today (34% of all Medicare enrollees).  Unlike the traditional Medicare program, which is administered by the federal government through the Centers for Medicare & Medicaid Services (CMS), the MA program is administered by private health insurance plans.  These plans are paid a per member per month (PMPM) fee—commonly referred to as a “capitated” fee— by CMS to manage the health care needs of their enrollees.

A vast majority of Medicare enrollees have one or more chronic conditions (e.g., diabetes, hypertension, kidney disease, etc.).  In order to pay for the additional health care services and care coordination needed to manage chronic conditions effectively, plans receive a higher capitated fee for these enrollees. The process of adjusting the capitated rate based upon an MA Enrollee’s health status is called risk adjustment.

Risk Adjustment Data Validation (RADV)

Risk adjustment data validation (RADV) is the process that CMS uses to ensure the accuracy of the data submitted by MA plans on the health status of its members.  Diagnosis codes (represent as ICD-10 codes) are the main pieces of information submitted by health insurance plans administering Medicare Advantage (MA Organization or MAO).  It is these codes, along with certain demographic information (age, gender, Medicaid enrollment status, and disability status) that are used to risk adjust the capitated monthly payments.

Validation of the risk adjustment data not only ensures that correct payments are made to MA plans, but also that appropriate health care resources are available for MA plan members.  The RADV process relies heavily on medical record documentation prepared by healthcare providers during face-to-face encounters with MA Enrollees.  During a RADV audit, CMS uses Hierarchical Condition Categories (HCCs – To be explained further in the next section) to represent one or more chronic conditions supported by medical record documentation.

Diagnosis codes submitted by MAOs are mapped to HCCs when the risk adjustment is calculated for the capitated monthly payments.   When an MA Enrollee’s medical record documentation is reviewed during a RADV audit, the HCCs resulting from data submitted by the MAO during a reporting period are compared to those HCCs supported by the related medical record documentation to determine if a discrepancy exists.

CMS currently relies on a manual process to conduct medical record documentation reviews, which is extremely labor-intensive.  Thus, CMS requests medical record documentation for a relatively small sample of MA members as part of the current RADV process (about 6,000 of the 22 million MA enrollees).

Hierarchical Condition Categories (HCCs)

CMS categorizes the submitted diagnoses into groups of clinically related conditions called Hierarchical Condition Categories (HCCs).  There are currently 79 HCCs.  Some of the more common HCCs include chronic obstructive pulmonary disease (COPD) and congestive heart failure.

Not only does each HCC represent a group of conditions that are clinically related, but they also have similar cost patterns.  Hence, some conditions are represented by more than one HCC, as different symptoms of the condition can require varying intensities of treatments.  For example, there are three HCCs for diabetes:  diabetes with acute complications, diabetes with chronic complications, and diabetes without complications.  Hierarchies ensure an MA enrollee is coded for only the most severe manifestation among related diseases, resulting in the name “Hierarchical” Condition Categories. 

HCCs focus on conditions that are chronic since they are used to be predictive of future costs.  In contrast, acute illnesses and injuries, such as influenza or a broken arm, are not considered for HCC categories since they are not predictive of future healthcare costs. 

It is the HCCs that are used to calculate a risk score for each MA member.  Each HCC has an associated value called the relative risk factor, which contributes to the member’s risk score.  HCCs that represent more severe conditions, such as cancers, quadriplegia, and pressure ulcers, have higher relative risk factors.  Many MA members will have multiple HCCs assigned to them, which further increases their risk scores and associated payments, as the relative risk factors are additive. (See below for Illustration of How HCCs Impact Payments to MA Organizations) 

Potential Future of RADV Audits

Healthcare information technology is at an inflection point where it is feasible to eliminate the burden of manual RADV audits. The state of artificial intelligence (AI) and electronic health record (EHR) interoperability provides the capability to implement an automated RADV audit with a staggering amount of productivity improvement (See below for more information onArtificial Intelligence in Healthcare and Interoperability with Electronic Health Records). Please read more about how RAPIDS has become a leader in applying AI and EHR to relieve the burdens of manual RADV audits.

Additional Information

Illustration of How HCCs Impact Payments to MA Organizations

Mr. Smith is a 71-year-old man that lives in Bronx, NY.  The base payment rate for the MA plan that he is enrolled in is $916 per member per month (PMPM).  During various encounters with clinicians during the year, he was diagnosed with prostate cancer, major depression, rheumatoid bursitis, and Crohn’s disease.  These diagnoses are all documented in the medal records for Mr. Smith.  The table below depicts how these diagnoses map to an HCC, as well as the relative risk factor associated with the HCC.

The relative risk factors are cumulative.  Adding up the relative risk factors for Mr. Smith’s HCCs, along with the relative risk factor for being a 70-75-year-old male (0.395) results in a risk score of 1.608.  The base payment rate is multiplied by the risk score to estimate the PMPM payment that the MA plan will receive for Mr. Smith.  In this case, the PMPM would be $1,473 (1.608 x $916).

Now let’s assume that none of the clinicians that treated Mr. Smith discussed his mental well-being.  In this scenario, the major depression diagnosis is not documented.  This means that his PMPM payment would be $314 less (0.343 x $916), which amounts to $3,770 annually.  If there were 300 members with similar missing documentation, the total cost to the MA plan would be over $1.1 million.  Hence, insufficient medical record documentation can lead to substantial financial impacts on plans.

Artificial Intelligence in Healthcare

AI can be applied in the form of deep learning for computers to acquire knowledge by processing millions of complex medical records with a moderate amount of computing power. With even less computing power, the machine-learned expertise achieves medical record review outcomes that are more accurate than manual reviews. We have already seen examples of this in the field of radiology to interpret images and produce diagnostic reports automatically. Healthcare institutions are actively using FDA-cleared image-analysis applications powered by AI to care for patients.

Interoperability with Electronic Health Records

According to 2017 data from the Office of the National Coordinator of HIT (ONC), 96% of hospitals and 80% of office-based physicians have adopted a fully certified electronic health record (EHR). Also, the 21st Center Cures Act of 2016 requires regulations to be put in place to reduce the burden and accelerate interoperability, and the rules proposed by ONC, HHS, and CMS were finalized on March 9, 2020 and will go into effect May 2020 . Compliance is required in six months, along with compliance reporting by EHR vendors every six months, and EHR vendors have 24 months to develop and deploy technology to support the new standards. The top 10 EHR vendors, which have close to 70% market share, already support many of the proposed rules.