When the check-engine light comes on, some drivers brace for the worst. Others keep driving — the car sounds fine, and the light has come on before. A mechanic does neither. She plugs in the code reader and asks what tripped it: a loose gas cap, a failing sensor, or a real engine problem. The light is information. It is not a diagnosis.
Prior authorization (PA) metrics went public this spring. Under a federal interoperability rule, Medicare Advantage plans, Medicaid and CHIP programs, and Marketplace insurers had to post their 2025 numbers, including denial rates and how often appealed denials were reversed. The postings cover medical services so far, but pharmacy teams at plans and PBMs have tracked similar numbers for years — and employers have begun requesting the same kinds of rates for their own populations when contracting.
In 2025, the largest insurers denied 12% of standard PA requests in Medicare Advantage — nearly one in eight — along with 14% in Medicaid managed care and 18% in the Health Insurance Marketplace, according to KFF’s analysis of the newly posted data. Across these markets, denials are rarely appealed. When providers and patients did appeal, they usually won in Medicare Advantage — 67% of appealed denials were overturned — and won often elsewhere: 47% in Medicaid managed care and 43% in the Marketplace.
Critics read these numbers one way. Payers read them another.
The critics’ reading
The critics’ reading is straightforward: High denial rates mean PA is blocking needed care. Rare appeals mean the process is too burdensome to fight. And when appeals succeed so often, the initial denials must mostly have been wrong — the few cases that get contested reveal what the rest would have shown. Every gauge points the same direction: a process that denies too much, discourages challenge, and folds when challenged.
A denial rate: harm meter or process signal
To critics, a denial rate measures harm — the share of needed care the payer refused. From inside a plan, the same rate reflects the machinery that produced it: how strict the coverage criteria are, which services require review at all, which providers have been exempted from review, and how well providers know each carrier’s rules. On the pharmacy side, the machinery includes the formulary, tiers, and step therapy rules that a plan sponsor adopts and its PBM helps design and administer — which is why attributing a drug denial rate to “the PBM” is harder than it looks.
Rather than check each carrier’s policy before ordering, some providers may simply submit the request and let the answer come back — in effect, a coverage inquiry. A carrier with stricter criteria can deny more of these requests than a carrier with looser ones, and neither denial rate, by itself, says whether the decisions were right. The public metrics do not report denials by reason, so the same denial rate can reflect different mixes of incomplete clinical information, incorrect information on the request, requests that did not meet coverage criteria, and requests filed too late.
The new data show why denial rates need context. Standard denial rates ranged from 2% to 25% across insurers and markets, with substantial variation even for the same insurer across its own markets. That variation cannot be read as variation in decision accuracy without knowing which services, criteria, and provider submission patterns generated the requests. KFF’s earlier Medicare Advantage work adds a finding that shows why a denial rate cannot be read without knowing how broadly PA is applied: Insurers that required prior authorization for more services denied a smaller share of requests, but the number of denials per enrollee came out similar. A high denial rate on a short list of services can mean the requirement is more selectively targeted, not more punitive.
A low appeal rate: surrender or learning
To critics, a low appeal rate means providers have given up — the process is too slow and too burdensome to fight. In the AMA’s 2025 survey of 1,000 physicians — advocacy data, but directly on the question — 59% of those who do not always appeal said past experience taught them the appeal would not succeed. About half of respondents cited each of two more reasons: too little staff or time for appeals, and patient care that could not wait for the plan’s approval.
But there is a second reading. A denial can teach the provider what this carrier requires before it says yes, and providers learn which denials are worth contesting. On this reading, a low appeal rate can be what provider learning looks like. The appeal rate alone cannot distinguish surrender from learning.
A high reversal rate: confession or correction
A reversal rate is not an error rate, but neither is it meaningless as evidence about error. To critics, a high overturn rate is the confession: If most appealed denials get approved, most denials must have been wrong. But appeals are not a random sample of denials — providers disproportionately appeal the cases they believe are worth the fight. Payers typically cite incomplete clinical information as the most common reason for denials — and most payers attribute some overturns to that information arriving on appeal. Other overturns can reflect an exception, a different clinical judgment, incorrect information on the request, or an error in the initial decision. Some payers categorize those reasons internally; the public overturn rate does not.
Reversal rates may also reflect different review processes. Public reporting does not break out internal and external appeals, and those review processes can vary across lines of business and plans. Medicare Advantage’s automatic independent review is one example of how review requirements can affect the reported rate.
Outcomes or diagnostics
Here is where the two readings part ways. The critics’ reading treats the metrics primarily as outcomes. The payers’ reading treats them as diagnostics: The question is what produced them. A high denial rate paired with a high appeal rate marks a dispute worth investigating; a high reversal rate strengthens the signal, because the initial disposition keeps changing on further review; and the sharpest signals are concentrated ones — a spike in one service, one provider type, one denial reason.
A recent HHS Office of the Inspector General (OIG) analysis of skilled nursing facility admissions shows what that kind of drill-down can reveal. In June 2026, the agency reported that 19 Medicare Advantage organizations denied 12% of requests for skilled nursing facility admission, that enrollees and providers appealed only 18% of those denials — and that 95% of the appealed denials were overturned. OIG concluded that some enrollees had initially been denied medically necessary care and recommended that CMS regularly collect request-level data by service to identify patterns like this one. A 95% reversal rate is not proof that most denials were wrong. Concentrated in a single service, however, it is a much stronger signal.
The diagnostic reading does not guarantee an answer that flatters the payer. Sometimes the drill-down finds missing documentation or an exception pathway doing its job; sometimes it shows that a criterion needs rewriting or a PA requirement is no longer earning its keep; and sometimes it finds denials that should have been approvals.
Massachusetts shows what the diagnostic reading can do with granular data. Some payers apply the same logic internally: a requirement that approves nearly every request invites the question of whether it is worth its cost. Before eliminating PA for a set of routine services this June, the state examined service-level request and approval data and targeted the services with the most requests and the highest approval rates. The state’s own examination identified a limit: A requirement can look useless precisely because it deters — an approval rate cannot count the requests never submitted. A 95% approval rate is not a verdict that review is unnecessary — for the same reason a 20% denial rate is not a verdict that review is abusive.
Signals, not verdicts
The critics’ reading captures real problems: Some PA requirements create unnecessary burden, and the OIG has documented inappropriate denials. Payers read these metrics diagnostically because they have to — they run operations on them. Regulators, employers, and critics can read them the same way. The problem is not using these rates as evidence. It is treating them as self-explanatory.