Upcoding is a design flaw in risk adjustment. Medicare pays Medicare Advantage plans more for members who look sicker, and it measures sickness by the diagnoses on file. The more conditions a plan documents, the more it is paid. That holds whether or not anyone treats the condition.
The result shows up in the totals. Coding makes Medicare Advantage risk scores about 4% higher than those of similar people in traditional Medicare, even after CMS’s annual adjustment, according to MedPAC. Together with favorable selection, MedPAC expects Medicare to spend about 14% more on Medicare Advantage enrollees in 2026 than it would in traditional Medicare. That gap comes to a projected $76 billion.
Audits and lawsuits treat the extra codes as a compliance problem. They are better understood as a predictable response to how the payment works. A design that pays for documentation will get more documentation.
Where the extra codes come from
Plans have 2 tools that add diagnoses without adding care. A health risk assessment, often done in the member’s home, can record conditions that no treating clinician has addressed. A chart review lets a plan comb through records after the fact and submit diagnoses the original claim did not include.
OIG measured what those tools produce. Diagnoses that appeared only on health risk assessments and the chart reviews linked to them, with no other service record in 2022, drove an estimated $7.5 billion in Medicare Advantage payments for 2023. That finding covered 1.7 million enrollees, according to OIG. In-home assessments and their linked reviews produced almost two-thirds of that total. Just 20 companies drove 80% of it.
Some of those diagnoses are surely real. A home visit can find a condition that a busy clinic missed. But a diagnosis that was never followed by a visit or a prescription did not lead to care. The payment it triggered bought nothing for the member.
OIG recommended that CMS restrict the use of diagnoses reported only on in-home assessments and linked chart reviews. It also recommended audits of those diagnoses. CMS did not concur with either recommendation in 2024.
Audits arrive years later
The main check on coding works after payment. CMS has said it will audit every Medicare Advantage contract each year, as our piece on the expanded audits described. The Justice Department has pursued False Claims Act cases over risk adjustment, as our review of fiscal 2025 recoveries described.
Both tools look backward. An audit reviews a sample of a contract’s diagnoses for a payment year that closed long before. A lawsuit takes years more. Neither changes the incentive in the year a code is submitted. A plan that weighs the cost of an audit against the revenue from coding still sees a return.
Audits also fight about the wrong thing. The arguments turn on sampling methods and on whether a chart supports a code. Those are real questions. They are also the questions that arise when payment depends on paperwork that only one side controls.
A diagnosis that never changes anyone’s care should not change anyone’s payment.
Pay for diagnoses that show up in care
CMS took a step in this direction in April. Starting with 2027 payments, risk scores will exclude diagnoses from chart reviews that are not linked to a specific encounter, with an exception for members who switch plans. CMS will also exclude diagnoses from audio-only visits, according to its rate announcement fact sheet. CMS estimated the chart review change at a 1.53% reduction in payments.
The change targets the loosest source of codes. It leaves the one OIG flagged as largest. A chart review linked to an in-home assessment is tied to an encounter, so the new exclusion does not appear to reach it. The assessment itself still counts too.
The next step would be a simple rule. A diagnosis should count for payment only when it shows up in care during the year. The evidence could be a visit with a treating clinician. A prescription fill or a fitting test result could count as well. An assessment could still find a condition and prompt a visit. It would not raise the payment until the visit happened. The rule would need a sensible window. A condition found in December could be confirmed by a visit early the next year, and the payment could follow then.
The rule has a cost. Some members have real conditions and no care, and the plan would not be paid for them. But a plan that is paid for an untreated condition has no reason to treat it. A plan that is paid only once care starts has every reason to arrange that care. The incentive would point at the member instead of the form. Because 20 companies drove 80% of the payments OIG found, the rule would fall mostly on the plans that lean hardest on assessments.
The rule would not fix everything. Favorable selection accounts for most of the gap MedPAC measured, and it needs its own answer. Corroboration also needs good data. Pharmacy fills and lab results are harder to manufacture than a checkbox on an assessment, especially when the result comes signed by the lab. Our piece on proving facts without the chart described how signed results could travel.
Redlake builds tools that health plans use in utilization review. Every recommendation our tools make shows its reasons, because a decision without evidence cannot be checked. A diagnosis that drives payment should meet the same standard. It should carry the evidence that a member is being treated for it. An audit years later is a poor substitute for a rule that asks for that evidence at the time.