BRENNEN

Industries / Health / Healthcare / Medical billing

Approval before treatment

Prior authorization, where a payer must agree that a treatment, test or drug is covered before it is given, and the rules for how fast and on what grounds it decides.

7

Milestones achieved, from 1989 to 2026.

2

To watch for: announced, not yet confirmed outside.

2

Open deliverables. Nobody has shipped them.

The road

AchievedWatch forOpen

Achieved

  1. 1989

    Review cuts hospital admissions

    Utilization review is a payer's check on whether hospital care is necessary. It spread among US private insurers in the 1980s with little evidence on its effects. A study of claims from 223 insured groups for 1984 through 1986 found that one commercial insurer's review program reduced hospital admissions by 13%, days in hospital by 11% and total medical spending by 6%.

    ebi.ac.uk

    Credit

    T. M. Wickizer, J. R. Wheeler and P. J. Feldstein (Medical Care, 1989)

  2. 2019

    England lists 17 procedures

    In April 2019 NHS England's Evidence-Based Interventions programme published guidance on 17 procedures, among them tonsillectomy. Some were not to be routinely funded and the rest were to be funded only when a patient met set criteria. A study of hospital records from 2010 to 2020 found that most of the procedures were already becoming less common before the guidance, that the trend changed for none of the 17 afterwards, and that differences between areas stayed the same.

    Lessonpublishing criteria for approval did not by itself change how often the procedures were done.

    pmc.ncbi.nlm.nih.gov

    Credit

    NHS England; Joel Glynn, Timothy Jones and colleagues (PLoS One, 2023)

  3. 2020

    Court narrows deemed approval

    German law requires a sickness fund to decide a member's application for a benefit within three weeks, or five weeks when it asks the Medical Service for an opinion. If it misses the deadline without giving a sufficient reason, the benefit counts as approved. On 26 May 2020 the Federal Social Court, giving up its earlier position, ruled that this deemed approval gives no claim to the treatment itself. It gives a provisional right to obtain the treatment oneself and be repaid, if the member acted in good faith, and the fund may still decide the application.

    Lessona deadline is only as strong as what follows when it is missed.

    gesetze-im-internet.debsg.bund.de

    Credit

    Bundessozialgericht (Federal Social Court), 1st Senate, case B 1 KR 9/18 R

  4. 2021

    Texas issues gold cards

    Texas House Bill 3459 of 2021 requires state-regulated health plans to exempt a doctor or other provider from prior authorization for a service when the plan has approved at least 90% of that provider's requests for it over six months. The exemption is called a gold card. A Texas Department of Insurance survey in January 2023, after the first round of exemptions, found that on average only 4% of providers had enough requests to be evaluated for any service and only 3% received an exemption. A 2025 law, House Bill 3812, lengthened the evaluation period to one year, and plans must report exemption data to the department from 2027.

    Lessonan exemption judged one service at a time can be earned only by providers with enough requests for that service, and few had them.

    content.naic.org

    Credit

    Texas Legislature; Texas Department of Insurance

  5. 2022

    Auditors recheck denied requests

    The Office of Inspector General (OIG) of the US health department took a random sample of 250 prior authorization denials and 250 payment denials issued by 15 of the largest Medicare Advantage insurers in the first week of June 2019. Coding experts and doctors reviewed the files. They estimated that 13% of the denied prior authorization requests met Medicare's coverage rules, and 18% of the denied payment requests met Medicare's and the insurer's own rules. Insurers had applied criteria that Medicare does not, such as requiring an X-ray before a more advanced scan; most wrong payment denials came from human error and system errors.

    oig.hhs.gov

    Credit

    OIG, report OEI-09-18-00260, April 2022

  6. 2024

    California bars algorithm-only denials

    California Senate Bill 1120, signed into law on 28 September 2024, covers health plans and insurers that use artificial intelligence, an algorithm or other software in reviewing requests for care. The tool must base its determination on the individual patient's clinical circumstances, and it "shall not deny, delay, or modify health care services based, in whole or in part, on medical necessity". Only a licensed physician or other licensed health professional competent in the clinical issue may make that decision.

    calmatters.digitaldemocracy.org

    Credit

    California Legislature; bill by Senator Becker

  7. 2026

    Insurers post approval figures

    A CMS rule of January 2024, known as CMS-0057-F, requires Medicare Advantage, Medicaid and HealthCare.gov plans to post their prior authorization figures on their websites each year, the first set by 31 March 2026. AuthDenied, an independent project that collects the postings, counted 92% of the required plans as having published their 2025 figures by 21 July 2026. By its count the median plan denied 7.2% of requests in Medicare Advantage and about 12% to 14% in Medicaid managed care, the Children's Health Insurance Program and marketplace plans. At the median plan only 7.3% of denials were appealed, and 52.6% of those appeals succeeded. These figures are AuthDenied's, as reported by the accounting firm Forvis Mazars.

    cms.govforvismazars.us

    Credit

    CMS; AuthDenied

Watch for

  1. Approvals in real time

    On 23 June 2025 US health insurers, through their trade group AHIP, announced voluntary commitments said to cover 257 million people. Plans would cut the number of services needing prior authorization by 1 January 2026, honor a previous insurer's approvals for 90 days when a patient changes plans, and in 2027 answer at least 80% of complete electronic requests in real time, using FHIR, the common format for exchanging health records. In April 2026 a survey by AHIP and the Blue Cross Blue Shield Association reported that participating plans had removed 11% of prior authorizations, or 6.5 million requests.

    Achieved whenthe figures plans must now post each year, or a regulator's audit, show the cut and the 80% real-time share for 2027.

    ahip.org

    Claimed by

    AHIP and the Blue Cross Blue Shield Association, June 2025 and April 2026

  2. AI review in traditional Medicare

    CMS's WISeR model runs from 1 January 2026 to 31 December 2031 in Arizona, New Jersey, Ohio, Oklahoma, Texas and Washington. One technology company per state, such as Cohere Health in Texas, reviews requests for selected services, among them skin substitutes, nerve stimulator implants and knee arthroscopy for osteoarthritis, using AI with a clinician checking the result. CMS says this will speed decisions and ease paperwork for providers. Outside reports so far conflict: in Texas about 62% of requests were approved on first submission and 84% after a doctor's review, by Washington Post reporting relayed by PYMNTS in June 2026, and the Medicare Rights Center said in September 2026 that released records show widespread delays and thousands of denials.

    Achieved whenCMS's evaluation or a government auditor publishes approval rates, decision times, the share of denials overturned on appeal and spending for the six states, set against comparison states.

    pymnts.comcms.gov

    Claimed by

    CMS, for a model that began in January 2026

Open

  1. 01

    A coverage decision test set

    A public set of invented prior authorization requests for services whose coverage rules are public, starting with the services in Medicare's WISeR model. Each case has an invented chart summary and the correct answer under the published rule: approve, deny, or ask for a named missing document. Two clinicians agree each answer. Then score AI models on the set and publish their errors by type, wrong denials first.

    The brief
    Credit · unclaimed
  2. 02

    Scoreboard for approval interfaces

    A public page showing which insurers' prior authorization interfaces exist and pass the open conformance tests. From 1 January 2027, CMS-0057-F requires the affected insurers to run a Prior Authorization API, a software interface that tells a provider's system whether approval is needed, takes the request and returns the decision with a reason. Run the US government's open test kit against each insurer's test endpoint, using invented patients only and never a live system, and record which tests pass.

    The brief
    Credit · unclaimed

Open 01 · the brief

A coverage decision test set

Credit · unclaimed

The first to publish is credited on this line, and the deliverable moves from Open to Achieved.

What to build

A public set of invented prior authorization requests for services whose coverage rules are public, starting with the services in Medicare's WISeR model. Each case has an invented chart summary and the correct answer under the published rule: approve, deny, or ask for a named missing document. Two clinicians agree each answer. Then score AI models on the set and publish their errors by type, wrong denials first.

Why it is still open

Published work stops short of this. An April 2026 preprint built synthetic requests from 186 passages of Medicare coverage text for 10 procedures, to train a system that looks up policy, not as a clinician-labeled test. A 2024 preprint reported 86.2% accuracy for GPT-4 on checklist items in the authors' own setup. The Duke Institute for Health Innovation checked its own drug approval assistant against 83 past cases for three drugs and reports that it got the approval status right in all 83, but those cases are real patient files and are not public. The companies reviewing requests under WISeR have published no accuracy figures. Searches in November 2026 found no public set of cases labeled by clinicians.

dihi.orgarxiv.org

Built with what exists

The Medicare Coverage Database, which holds the national and local coverage rules; invented patient summaries written by clinicians, with no real records; general AI models reached through their public interfaces, and open-weight models run locally; a spreadsheet and a short Python scoring script.

cms.gov

  1. First step

    Pick one WISeR service, read its Medicare coverage rule, write 20 invented requests and have two clinicians label them separately.

  2. Done looks like

    300 cases across at least five services, with agreement between the two clinicians reported, at least four models scored, and a wrong-denial and a wrong-approval rate for each service, all published with a DOI.

Open 02 · the brief

Scoreboard for approval interfaces

Credit · unclaimed

The first to publish is credited on this line, and the deliverable moves from Open to Achieved.

What to build

A public page showing which insurers' prior authorization interfaces exist and pass the open conformance tests. From 1 January 2027, CMS-0057-F requires the affected insurers to run a Prior Authorization API, a software interface that tells a provider's system whether approval is needed, takes the request and returns the decision with a reason. Run the US government's open test kit against each insurer's test endpoint, using invented patients only and never a live system, and record which tests pass.

Why it is still open

The test kit for the Da Vinci Prior Authorization Support guide, the standard these interfaces follow, is public but rated "Low" maturity at version 0.16.0, and it publishes no results by insurer. A rule CMS proposed in April 2026, CMS-0062-P, would have insurers report their interface endpoints to CMS, which would give a list of addresses but not whether each one works. A search in November 2026 found no list of insurers that pass.

cms.govinferno.healthit.gov

Built with what exists

The Inferno Da Vinci PAS Test Kit, open source under the Apache 2.0 license; the Da Vinci PAS guide, version 2.2.1; insurers' developer portals and test endpoints; a web agent such as Browser Use to find and register on the portals (see the Note "Deleting 80% to start over").

github.com

  1. First step

    Run the test kit against the Da Vinci project's public reference server and publish the result in the format the scoreboard will use.

  2. Done looks like

    For the 50 largest affected insurers, a table with the endpoint, the test kit version, the date and the count of tests passed, the insurers with no test endpoint listed as not testable, and a rerun each month through 2027.