How well does the CMS-HCC risk adjustment model predict future expense?

In my previous blog, “Demystifying Medicare Risk Adjustment,” I introduced the model that the Centers for Medicare and Medicaid Services (CMS) uses to predict future Medicare Advantage health expenditures—the CMS-HCC model. The most recent evaluation of the 2018 version reveals that the model fails to account for up to 90 percent of the factors associated with healthcare expenditures. Let’s explore the model further to determine why it falls short of its goal.

Individuals who are 65 years of age, have been deemed disabled, or have end-stage kidney disease are eligible to receive Medicare benefits. Accordingly, CMS has different versions of the model that align with each of these distinct populations. In addition, beginning with model year 2017, CMS began a separate series of models to account for those individuals enrolled in a Dual-Eligible (Medicare and Medicaid) Special Needs plan (D-SNP).

All of the aforementioned models capture the following predictive variables:

• Demographics – only age, sex, Medicaid-eligibility, and whether or not the individual is disabled are captured by the models
• Chronic medical conditions – only those that are documented in an acceptable medical record, during the data collection period, and are considered a Hierarchical Condition Category (HCC) are applied to the model
• Disease Interactions – some HCCs, when combined, are given an additional bump for their associative interactions
• Frailty (when applicable) – this adjustment applies only to those beneficiaries enrolled in one of two special types of MA plans— a Program of All-Inclusive Care for the Elderly (PACE) or a Fully Integrated Dual Eligible (FIDE) Special Needs Plan

Although the model appears to capture a number of variables that would seem to impact healthcare spending, on closer analysis, we can postulate why these variables have limited predictive capability.

Demographics

With respect to demographics, both increasing age and female status are important predictors of healthcare expenditures. However, Medicaid status as an isolated proxy for income level leaves a lot to be desired.  Although eligibility requirements differ by state, in general, one is not garnered Medicaid eligibility unless he or she falls within 200-300 percent of the Federal Poverty Level. Most individuals over the age of 65 require well above 25-30,000 dollars of annual income to live comfortably.  Those who lack for food, water, shelter, or other pressing “comfort” are much less likely to engage in maintaining their health or complying with affordable preventative medicine. Economic studies show a strong inverse correlation between falling income levels and healthcare expenditures likely related to expensive rescue care.

Medical practitioners are increasingly aware of the importance of the social determinants of health in determining one’s perception of health and well-being. With the exception of those that can be directly correlated with Medicaid status, the CMS-HCC model completely ignores these vital drivers of healthcare consumption.  As an example, social isolation likely has no significant correlation with income level. Nonetheless, social isolation can lead to substance abuse, self-neglect and a downward spiral in health.

Numerous additional demographic factors are important to capture in a more comprehensive model. Among these are racial/ethnic identity and geography. Racial and geographic disparities exist for both access and delivery of medical services. In some cases, services are disproportionately withheld from certain racial and ethnic groups, while in particular geographies high-cost services (e.g. surgical procedures) are performed at much higher rates.  Health literacy is another vital driver not currently captured.

Chronic Medical Conditions

The CMS-HCC model focuses entirely on chronic medical conditions. As such, the model fails to account for any acute medical events that can result in expensive medical interventions. Prime examples are heart attacks, strokes, major bone fractures (e.g. hip), and surgical procedures.  While there are HCCs that link to these diagnoses, timing of remuneration lags by at least one calendar year.

To clarify the timing issue, let’s look at an example. Suppose we have a 67-year-old healthy female who falls and breaks her hip in May of 2017. She undergoes an inpatient hospitalization that includes surgery, is transferred to a skilled nursing facility for a two week stay, and subsequently receives home nursing services for an additional month. None of these expenses were factored into the payment her Medicare Advantage plan received to cover her benefits for 2017. In fact, the MA plan won’t see any additional risk-adjusted payment related to the hip fracture until the second half of 2018, long after all claims have been paid.

Additionally, only those chronic medical conditions documented in an appropriate medical record during the data collection period related to payment are included in adjusting for risk. In other words, every year, a given beneficiary’s medical history is expunged (according to CMS)—including, as ridiculous as it seems, an amputated limb.

Frailty and Summary

Frailty occurs when an elderly individual begins to deteriorate both physically and mentally. Frailty is highly associated with disability, falls and hospitalization. However, the CMS-HCC model only recognizes those who are deemed frail and enrolled in one of two special type of Medicare Advantage plan—a PACE or FIDE SNP. Any other frail MA beneficiary won’t receive any “points” towards risk-adjustment.  Clearly, the model is missing a large population of frail beneficiaries.

To summarize, while the CMS-HCC risk adjustment model captures a few important predictors of future medical expense, the model accounts for less than 15 percent of them. By any reasonable statistician’s standards, the model is lacking. As CMS revises the model every year, perhaps future models will resolve some of the issues discussed.

Samuel Young, MD, MBA, FACS, CPE, CHCQM, CRC, is a Clinical Transformation Physician Consultant for 3M Health Information Systems.