Why average RAF scores can mislead ACO oversight
A lower risk adjustment factor score does not automatically mean a clinic is undercoding. It may simply mean the panel is healthier, younger, or less burdened by conditions that map to hierarchical condition categories. In Medicare Advantage, accountable care organizations, and other value-based care arrangements, that distinction has real financial consequences because RAF is often treated like a proxy for documentation quality when it is really a blend of patient mix and coded disease burden.
The central mistake is simple: raw clinic-to-clinic comparisons assume that all panels are interchangeable. They are not. A practice caring for a community-dwelling, relatively healthy Medicare population should not be expected to resemble a practice with more older beneficiaries, dual-eligible patients, originally disabled beneficiaries, or patients with institutional status and advanced multimorbidity. The CMS-HCC model was built to account for exactly those differences through demographics and diagnosis-based weights. Averages flatten those clinical realities into a single number and invite the wrong managerial response.
A clinic with a lower mean RAF may be doing an excellent job of documentation. A clinic with a higher mean RAF may still be missing important diagnoses. The average alone cannot tell the difference.
RAF reflects both demographics and diagnosis capture
At the patient level, RAF is not a synonym for how many diagnoses appear in the chart. It combines demographic factors, diagnosis-based HCC factors, and applicable interactions. Age, sex, Medicaid or dual status, original disability entitlement, and community-versus-institutional status all contribute to the score. Diagnosis factors are added when ICD-10-CM codes map to HCCs and carry assigned weights, subject to the hierarchy rules built into the model.
That structure creates two very different ways to land on the same low score. One patient may genuinely have few chronic conditions and little HCC-relevant disease. Another may carry clinically meaningful illness burden that was never assessed, documented, coded, or submitted in a way that affects risk capture. Those are not equivalent cases, and a mean RAF obscures the difference.
In practical terms, this is why diagnosis coding cannot be judged by a single average. RAF is not a direct report card on clinician performance. It is a measure of expected cost and clinical risk for a population, and the score only becomes meaningful when the underlying patient mix is visible.
The shape of the distribution tells a more honest story
A panel-level RAF distribution is often more informative than the mean. Many Medicare panels do not form one smooth curve. They may show a low cluster driven largely by demographics, a broader middle group with common chronic disease burden, and a long right tail of medically complex, frail, disabled, or institutionalized patients.
Those patterns matter because they change the interpretation of the clinic’s coding profile. A large low-RAF cluster can mean a practice serves a comparatively healthier population. It can also signal a capture gap if those patients have active conditions that are being evaluated and managed but are not appearing in the risk model. A broad middle group often reflects the expected burden of chronic illness seen in routine ambulatory care. A right tail suggests severe disease, frailty, or institutional status, where specificity in documentation becomes especially important.
Averages erase all of that.
The distribution can even reverse the apparent performance of two clinics. Imagine one site with a mean RAF of 0.85 and another at 1.20. The intuitive conclusion would be that the lower-scoring clinic is undercoding. Yet the lower-scoring clinic may simply care for a younger, healthier population and be capturing diagnoses faithfully. The higher-scoring clinic may care for a sicker, older panel and still leave substantial risk on the table through incomplete coding of diabetes with complications, chronic kidney disease stage specificity, heart failure type, chronic obstructive pulmonary disease, vascular disease, major depression severity, morbid obesity, or substance use disorders.
Why wrong comparisons create wrong incentives
When leadership uses average RAF as a blunt performance metric, the incentives tilt in the wrong direction. Accurate coders can be pressured to “improve” scores they do not control, while clinics with sicker populations can be rewarded for a number that reflects case mix more than documentation quality. In shared savings arrangements, that error can affect benchmarking, resource allocation, and internal physician profiling.
This is not a theoretical concern. A practice can appear underperforming simply because it serves a healthier population. Another can appear strong because it sees more complex patients, even when documentation is incomplete. In both cases, the mean RAF alone fails as a management tool.
The result is predictable: groups may reward the wrong clinics and pressure the right ones.
A better analytic lens for coding performance
A credible RAF review starts by comparing like with like. Clinics should be stratified by age bands, sex, dual eligibility, original disability entitlement, community versus institutional status, new-to-Medicare versus long-standing Medicare status, and prior-year RAF. Disease-specific cohorts provide even more clarity, especially for diabetes, heart failure, COPD, CKD, dementia, and serious mental illness.
That kind of stratification separates patient mix from documentation quality. It also reveals whether low scores cluster in populations where diagnoses are likely being missed or whether the panel simply carries less HCC-relevant disease. From a compliance standpoint, the distinction is essential because overreading a low mean RAF as poor performance can distort audits, feedback loops, and compensation decisions.
The biologic logic behind the model is straightforward. Patients with more chronic disease generate more encounters, more complications, and greater expected cost. The risk model tries to translate those realities into a financial estimate. When the coding process misses active disease, the score undercounts true clinical burden; when the panel itself is healthier, a lower score may be exactly what should appear.
What clinicians and administrators should look at instead
The most useful question is not whether one clinic has the highest RAF. It is whether the distribution of scores matches the population being served and whether documented disease burden is complete within that population. That requires looking beyond the mean to the full panel, the subgroups, and the tails of the curve.
For physicians, that means documentation has to be active, specific, and clinically anchored to conditions actually assessed and managed. For administrators, it means dashboards should incorporate distribution-based analysis rather than a single average. For ACO leaders, it means performance oversight should distinguish between a healthier patient panel and incomplete diagnosis capture before drawing conclusions about coding quality.
The mean can be misleading, especially when panel mix differs from clinic to clinic. The distribution is where the clinical truth lives, and it is the only way to tell whether a lower RAF reflects healthier patients or missed opportunity in documentation.