Every January, primary care practices with Medicare Advantage patients start receiving lists from the plans. The lists are polite. They say things like "suspected conditions for review" or "documentation opportunities," and they name patients who had a diagnosis on a claim in 2024 or 2025 that has not appeared on any claim yet in 2026. A physician in a practice we work with looked at his list last week and said, "They want me to code diabetes on a patient I have not seen since October. Of course it has not appeared yet." He was right, and the list was also right, and the gap between the two is what this article is about.
HCC coding for primary care practices is one of those topics that arrives with a bad reputation because of how some organizations have abused it. Hierarchical condition categories (HCCs) are the diagnosis groupings CMS uses to adjust what it pays Medicare Advantage plans for each member. A member with documented diabetes with complications, heart failure and chronic kidney disease generates a higher monthly payment to the plan than a member with no chronic conditions, because that member is expected to cost more. The plan's payment depends on the diagnoses that appear on claims and encounter data from clinicians like you, every calendar year, starting from zero each January 1.
That creates a legitimate interest and an illegitimate one. The legitimate interest is that conditions a patient actually has, that you actually assess and manage, should be documented and coded accurately, because the plan is paid to manage them and because the patient's record should be true. The illegitimate interest is adding diagnoses the visit did not address, or that the patient does not have, because they carry weight. The documentation rule below is what separates the two.
Key takeaways
- CMS pays Medicare Advantage plans a risk-adjusted amount per member based on demographics and the HCC diagnoses submitted on claims during the calendar year, which is why plans send you lists every January.
- The CMS-HCC model version 28, fully phased in for 2026 payment, removed or reweighted many conditions, so lists based on old assumptions may be out of date.
- A diagnosis counts only if it was documented at a face-to-face or qualifying telehealth visit and supported by evidence the condition was monitored, evaluated, assessed or treated.
- Coding to the highest specificity the record supports is correct; coding a condition the visit did not address is not, and it is the subject of active federal enforcement.
- The workflow that works is pre-visit review of the chronic problem list, documentation of status and plan for each condition addressed, and a coder check for specificity before the claim goes out.
How risk adjustment actually works
CMS pays each Medicare Advantage plan a monthly capitated amount for each enrollee. That amount is the plan's bid for the county multiplied by the member's risk score. The risk score starts with demographic factors (age, sex, Medicaid status, disability) and adds a coefficient for each HCC the member has, with some interactions for combinations. A member with a risk score of 1.0 is average. A member with a score of 2.4 generates 2.4 times the base payment. The diagnoses that feed the score come from claims and encounter records submitted for face-to-face visits (and qualifying telehealth) with acceptable provider types during the calendar year, and the score for payment year 2026 is built from 2025 dates of service.
The word "hierarchical" matters. Conditions are grouped into categories, and within a family of related categories only the most severe counts. Diabetes with chronic complications and diabetes without complications are different HCCs; if both appear, only the higher one is paid. The categories reset every year, which is why the January lists exist.
None of this changes what your practice is paid for the visit under a fee-for-service arrangement. It changes what the plan is paid, and, if your practice participates in a shared savings or capitated contract with the plan, your own results. In value-based arrangements the risk score is the denominator of your performance.
What version 28 changed
CMS finalized a revised model, CMS-HCC version 28 (V28), in the 2024 Medicare Advantage rate announcement and phased it in over three payment years: one third in 2024, two thirds in 2025 and fully in 2026. V28 was built on more recent data and on ICD-10 codes rather than mapped ICD-9 codes, and it made several changes that matter in primary care. The number of payment HCCs rose from 86 to 115, but more than 2,000 diagnosis codes that previously mapped to a payment HCC no longer do. Several conditions that were commonly coded for their weight, including some forms of diabetes with certain complications, angina, atherosclerosis of native arteries without other complications and protein-calorie malnutrition, were removed, constrained (given the same coefficient regardless of severity) or reweighted. Depression was narrowed so that only major depressive disorder with specific severity counts, not unspecified depression.
The practical effect is that lists built on the old model's assumptions about which diagnoses matter are partly obsolete, and any coding education your practice received before 2024 about "high value" conditions should be discarded. What has not changed is the principle. Document what is there, to the specificity the record supports, at a visit where you addressed it.
| Condition family | V24 treatment | V28 treatment | Documentation that matters |
|---|---|---|---|
| Diabetes | Separate HCCs for with and without complications, different weights | Constrained: all diabetes HCCs carry the same coefficient | Type, control status, and each specific complication with its linkage |
| Heart failure | One HCC | Split by type and acuity into several HCCs | Systolic, diastolic or combined; acute, chronic or acute on chronic |
| Chronic kidney disease | Stages 3 and above counted | Stage 3 split into 3a and 3b; stage 3a weight reduced | Stage from the most recent eGFR, restated at the visit |
| Depression | Major depressive, bipolar and paranoid disorders in one HCC | Only specified major depressive disorder with severity; unspecified depression removed | Severity, episode and remission status |
| Angina and atherosclerosis | Counted | Removed as payment HCCs | Code them if present; they no longer carry weight |
| Morbid obesity | Counted with BMI 40 or above | Still counted; BMI documentation required | Diagnosis plus BMI in the note, addressed at the visit |
The documentation rule: MEAT
A diagnosis is supportable for risk adjustment when the note shows that the condition was monitored, evaluated, assessed or treated at the visit. Coders call it MEAT. Monitored: signs, symptoms, disease progression noted. Evaluated: test results reviewed, response to treatment considered. Assessed: the condition's status stated (stable, worsening, at goal). Treated: medication continued, adjusted, started, referral made, counseling given. Any one of those, documented for that condition at that visit, supports coding it. A condition that appears only on the copied-forward problem list, with no mention in the assessment, does not meet the standard, and coding it is the error that drives most audit findings.
This is also why the physician's objection to the January list was correct. The patient has diabetes. It was documented in October. It has not been addressed at a 2026 visit because there has not been a 2026 visit. When the patient comes in for the spring follow-up, the diabetes will be assessed and coded, and it will count. The list is not asking you to code it now without a visit; it is telling you the plan's data does not yet show it this year. Read the lists as scheduling prompts for patients with chronic conditions who are overdue, not as coding instructions.
The enforcement environment is real. The Department of Justice has pursued False Claims Act cases against Medicare Advantage organizations and provider groups for unsupported diagnoses, and OIG audits of plan risk adjustment data have repeatedly found diagnoses on claims that the medical record did not support. Chart reviews that add diagnoses without a visit, one-way audits that only add and never delete, and coding from problem lists have all featured in those cases. A practice's protection is the note: if the assessment says what was done about the condition, the code stands.
A capture workflow for a small practice
Start the day before the visit. A pre-visit review, by a nurse or medical assistant, pulls the chronic problem list and the last year's diagnoses, checks which have not been addressed at a visit this calendar year, and puts a short prompt in the pre-visit note: "Chronic conditions not yet addressed in 2026: type 2 diabetes with CKD stage 3b, chronic systolic heart failure, major depressive disorder recurrent moderate." The physician sees the list before walking in. That is the entire intervention on the clinical side, and it is also good medicine, because those are the conditions that should be reviewed anyway.
During the visit, the physician addresses the conditions that are relevant and documents each with a status word and a plan, as described in our note on E/M leveling: "CKD stage 3b, stable, eGFR 42 in December, continue ACE inhibitor, recheck in six months." That single sentence meets MEAT, supports the specific ICD-10-CM code (N18.32) and contributes to the E/M level. Conditions not relevant to the visit are not addressed and not coded, and that is fine.
After the visit, a coder or a coding-aware biller checks the claim for specificity before it goes out. Unspecified diabetes when the note says type 2 with CKD; unspecified heart failure when the echo says systolic; depression unspecified when the note says recurrent moderate. Those are not upcoding, they are coding what the record already says, and they are the most common gap we find. A quarterly internal audit of 20 charts per physician, checking that every coded HCC is supported by MEAT in the note and that every documented chronic condition was coded to specificity, keeps both sides honest. Practices that want an outside review of that sample, or a team that runs it, can fold it into the medical coding work we do, and our RCM training courses include a risk adjustment module for coders new to it.
Questions we hear
The plan is offering to pay us for completing their "health assessment" forms. Should we?
Read the form and the terms carefully, and ask counsel if the payment is tied to the number of diagnoses. Completing a structured annual assessment during a real visit, with the patient present and the conditions actually evaluated, is legitimate and can be useful. Signing a pre-filled list of suspected conditions without a visit is the pattern that has drawn enforcement. The difference is whether the assessment happened.
Does any of this matter if we do not have value-based contracts?
Less, but not nothing. Accurate, specific diagnosis coding is correct regardless of who is paid for it, and it affects the medical necessity support for tests and referrals, the quality measure denominators the plan reports on your patients, and your standing when the plan evaluates practices for future contracts. It just does not change your fee-for-service payment directly.
Can we code a condition from the specialist's consult note?
For your own claim, the diagnosis has to be supported by your own encounter documentation. You can and should incorporate the specialist's findings into your assessment ("cardiology confirmed chronic systolic heart failure, EF 35 percent; stable on current regimen") and then code it, because you evaluated and managed it. Copying the consult's diagnosis list into your claim without addressing the conditions does not meet the standard.
What to do this week
- Take the January lists from your Medicare Advantage plans and turn them into a scheduling list of patients with chronic conditions not yet seen in 2026.
- Add a pre-visit step where a nurse or medical assistant lists chronic conditions not yet addressed this year in the pre-visit note.
- Give physicians the one-sentence pattern for each chronic condition: status word, evidence, plan.
- Have the coder review the last 20 Medicare Advantage claims per physician for specificity gaps between the note and the code.
- Discard any coding guidance in the practice that predates the V28 model and predates 2024.
