Methodology · collection, normalization, limitations, corrections
Every number here came from an insurer, a government agency, or a clearly labeled investigation. This page explains what each number means, where it came from, how we translate different insurer labels, and where the public record stops.
We keep five kinds of records separate. A percentage about a claim after care is not interchangeable with a percentage about permission before care. Each result tells you which record you are reading.
Prior-authorization reports from 2025. The first required reports were posted in 2026. They show what services need permission, how often requests were approved or denied, what happened after an appeal, and how long decisions took. The reports are written by insurers and are not checked by CMS before publication. We keep their original wording and point out missing counts, duplicate tables, and files that are hard to preserve.
Claims denied after care. The CMS Transparency in Coverage public-use file, analyzed annually by KFF, shows the share of in-network claims an issuer denied after care was delivered. The latest comparable file currently available in this search covers 2024 activity and HealthCare.gov individual-market plans. It does not cover employer plans, Medicare, Medicaid, or most state-based exchanges.
Medicare Advantage permission-before-care decisions. CMS data, analyzed annually by KFF, shows request volumes, denial rates, appeal rates, and reversal rates by insurer. It usually arrives about a year after the activity.
Independent external reviews. Some states publish decisions from an independent reviewer after a plan’s internal appeal. These records cover only the small share of denials that reach that stage.
Investigations. Congressional reports, court filings, and journalism. An allegation is always labeled as an allegation; it is never presented as a proven result.
Coverage by state
The 30-state file in this search comes from HealthCare.gov. It is useful because every state in it uses the same file and definitions. It is not a list of every Marketplace plan. CMS publishes a separate plan-and-issuer roster for state-based marketplaces, including Washington; that roster tells us which plans exist but does not contain claims or denial outcomes.
Washington and other states run their own marketplaces, so their plan records live in different files. Washington also publishes prior-authorization data, but the public carrier rows are anonymous. A plan can therefore be real and searchable while still having no named, comparable denial rate here.
That does not mean those states have no data. Some publish denied-claims or appeal records using state definitions. We will add a state to a comparable rate only when we can verify the year, definition, denominator, and original source. Until then, “not comparable or not collected here yet” is the honest answer.
The next layer of the record
A state may publish useful denial or appeal records without publishing the same measures that insurers report to the federal file. We keep those records visible, label the difference, and only add a rate to the index when the definition and denominator match.
Washington's Marketplace plan roster is real and searchable, but its public denial outcomes are not named by carrier. The OIC publishes prior-authorization rows as Carrier A–O, not as Premera, Regence, Molina, Kaiser, or UnitedHealthcare.
For a patient: A plan or network can exist in Washington's roster without having a public insurer-specific denial rate. Do not treat an anonymous Carrier A–O row or a complaint count as a named carrier's claims rate.
Covered California publishes denied-claims files, and DMHC publishes external-review outcomes. We have not yet normalized them to the federal file’s definitions.
For a patient: The state files can show claims or external reviews, but they do not describe every plan or every kind of appeal.
New York publishes issuer claim reports and a searchable external-appeal record. The files are state-specific and must stay separate from HealthCare.gov results.
For a patient: External appeals are the cases that reached the state process, not every appeal filed with an insurer.
Pennsylvania’s 2025 transparency report covers 2024 individual-market plans and reports aggregate claims and appeal outcomes. It is not yet an insurer-by-insurer lookup here.
For a patient: The report is a useful state baseline, but it does not tell you what your specific insurer did.
This is a verified source queue, not a claim that these four states are the only states with records. We are adding states as we can preserve the original files, definitions, and retrieval dates.
The work behind the index
The rule created a reporting duty, not a clean database. Each insurer chooses its own website, file type, labels, and level of detail. A number can be technically “posted” and still be difficult for a patient, lawyer, or reporter to check.
Find it. We search the insurer’s public site, subsidiary sites, help pages, PDFs, and pages that load only in a browser.
Record the trail. We save the original link, publication date, capture date, and search path. The public archive and checksums are future releases, not current capabilities.
Compare like with like. We keep the program, year, request type, decision stage, unit, numerator, and denominator attached to every number. A percentage without the counts behind it is incomplete.
Show what is missing. We do not silently repair an insurer’s report. If counts are missing, a table is duplicated, or a file blocks preservation, that defect stays visible.
The translation layer
One company may call a request “standard,” another “routine,” and a third may split it by contract or service. We translate each source into the same plain questions: What was requested? Had care already happened? Which insurance program was involved? What decision was made? Did an appeal change it? How many requests does the percentage describe? The original wording stays attached, so you can check our translation instead of taking it on faith.
Why this is differentiated
The hard part is not displaying a percentage. It is building a reliable trail from a changing insurer website to a shared definition that still makes sense beside another insurer’s number. Each verified source adds a page, a label translation, a scope decision, and a record of what was missing. That growing crosswalk is what lets patients, lawyers, reporters, employers, and regulators compare the record without starting the research from zero.
The insurer reports are self-reported. No regulator checks every line before publication. First-year quality is uneven, and we show the defects we find.
Some Medicare Advantage numbers are reported for a contract, which can cover many plans and states. A contract-level number cannot tell you what happened in every plan inside it.
Drugs are excluded from the federal prior-authorization reporting rule.
Employer health plans are not covered. Public files cover regulated markets, not the inside of a self-funded employer plan.
Denial rates measure how often, not whether the decision was correct. A high rate may reflect different requests or reporting rules. A reversal rate tells you how often an insurer changed a decision after an appeal; it is not a guarantee about your case.
Each published figure names and links to the insurer’s own posting, a government file, or a clearly labeled secondary analysis. We show the activity year, insurance market, unit being counted, publication date, and any caveat. The public archive and version history are not live yet, so the original source link is the current record.
We search public insurer and government pages, download files when possible, and record the path that led us there. We do not bypass logins or CAPTCHAs. If a public disclosure is blocked, we label it obstructed instead of pretending we verified it.
Before we say a report was not found, we search the insurer’s main site, subsidiary sites, public help pages, PDFs, and known reporting locations. We record the searches and dates. “Not found” means not found through that protocol; it does not prove the insurer never posted it.
If we got something wrong, we fix it publicly. Insurers, researchers, and readers can send a correction with evidence through the contact information on the About page. Our target is to apply verified corrections within 48 hours. Material corrections are identified on the affected page with what changed, when, and why.
The Denial Index takes no money from health insurers, their foundations, or their trade associations. Revenue comes from advisory work, employer audits, and data licenses — buyers who want the numbers accurate, not flattering.