Industries / Health / Healthcare / Pharmacy / Specialty pharmacy
Approval from the drug plan
Prior authorization for a specialty drug is the work of getting a drug plan or a national payer to agree, before a costly drug is dispensed, that it will pay for it for this patient, done by the prescriber's office and the pharmacy through forms, phone calls and electronic requests.
7
Milestones achieved, from 2006 to 2025.
3
To watch for: announced, not yet confirmed outside.
2
Open deliverables. Nobody has shipped them.
The road
AchievedWatch forOpen
Achieved
2006
Pilots reject the procedure standard
In 2006 drug prior authorization ran on paper forms, fax and phone. A task group at NCPDP, the body that writes US pharmacy data standards, collected 350 approval forms holding 1,750 questions from 53 payers, and found that criteria and wording differed from plan to plan. The Medicare e-prescribing pilots of 2006 then tested electronic requests, and the investigators agreed that the only prior authorization standard named under HIPAA, X12 278 version 4010, "is not adequate to support drug PA because it was designed for procedure or DME PA." NCPDP set out to build a drug standard of its own.
LessonThe hard part was the questions, not the wiring, because every plan asked its own in its own words.
Credit
NCPDP Workflow-to-Prior Authorization Task Group, led by Tony Schueth of Point-of-Care Partners; the pilot teams led by RAND, Brigham and Women's Hospital, Achieve Healthcare, Ohio KePro and SureScripts
2012
Italy writes registries into law
Italy's medicines agency, AIFA, runs online monitoring registries for many new high-cost medicines: doctors and pharmacists at authorized centres enter each patient, and the entry is checked against the conditions for payment. A law of 7 August 2012 (no. 135) made the registries part of the national health service's information system. They also carry deals such as payment by result, under which the maker refunds the cost of treatments that fail. By AIFA's count the platform links about 3,500 health facilities, 32,857 doctors and 2,318 pharmacists. A 2025 review of eight countries found registry coverage high because entry is required by law and tied to payment; England's Blueteq approval forms and Belgium's Chapter IV requests, by contrast, are submitted once at the start and do not show whether the drug was then used.
LessonAn approval record tied to payment gets filled in, but a record filled in once says nothing about what the patient received.
Credit
AIFA; Italian Parliament; Steven Simoens and colleagues (Frontiers in Pharmacology, 2025)
2013
Drug approvals get a standard
In July 2013 NCPDP approved electronic prior authorization (ePA) transactions inside SCRIPT, its standard for electronic prescriptions. A prescriber's software could now ask the plan whether approval is needed, receive the plan's questions, answer them and get a decision without a fax. Networks grew up to carry these messages. One of them, CoverMyMeds of Columbus, Ohio, was bought on 3 April 2017 by McKesson, the drug distributor, for a net $1.3 billion, with up to $160 million more tied to performance; McKesson describes it as selling ePA to "pharmacies, providers, payers, and pharmaceutical manufacturers."
Credit
NCPDP, under president Lee Ann Stember; CoverMyMeds; McKesson
2017
An auditor times drug approvals
Ontario's Exceptional Access Program pays, case by case, for drugs that are not on the province's public formulary, its list of covered drugs. In 2016/17 about 65,850 people received about 580 such drugs at a cost of about C$810 million, by the Auditor General's count. The auditor found that requests for the two most-requested biologic drugs, adalimumab and infliximab, took on average 34 to 41 business days, about seven to eight weeks, from receipt to decision. In 2015/16 the ministry met its own response targets 48% of the time, against a target of 85%, and it reported time per step rather than the patient's whole wait. An online request channel, SADIE, was due in October 2018; the ministry now recommends it for faster answers.
LessonA payer that counts only the steps it controls can report progress while patients wait two months.
Credit
Office of the Auditor General of Ontario; Ontario Ministry of Health
2018
Congress orders electronic drug approvals
Section 6062 of the SUPPORT Act, signed on 24 October 2018, required Medicare Part D, the drug benefit that private plans run for Medicare, to provide by 1 January 2021 for secure electronic prior authorization requests and replies. It said a fax, a payer's own web portal that does not meet the adopted standards, or an electronic form does not count. CMS's rule of 31 December 2020 named NCPDP SCRIPT version 2017071, allowed its use from 2021 and required Part D plans to support it from 1 January 2022. The rule binds plans, not prescribers: "Prescribers remain free to use non-electronic means," and plans must still accept faxed requests.
LessonThe mandate made electronic requests possible across Part D but left the fax machine legal.
Credit
US Congress; CMS
2018
Step therapy reaches Part B
Step therapy is a plan rule that a patient must try a preferred, usually cheaper, drug first and move to the prescribed one only if the first does not work. In 2012 CMS had barred Medicare Advantage plans, the private plans that replace traditional Medicare, from requiring it for Part B drugs, the drugs given in a clinic or infusion centre. On 7 August 2018 CMS rescinded that guidance and let these plans use step therapy for Part B drugs from 1 January 2019, including requiring a Part D drug before a Part B drug, provided the plan offered patients drug-management care coordination and rewards for taking part.
Credit
CMS, under Administrator Seema Verma
2025
Certified e-prescribing must include ePA
ASTP/ONC, the US health IT office, certifies the software clinicians use. Its HTI-4 rule, published on 4 August 2025 inside a CMS hospital payment rule and in force from 1 October 2025, requires certified e-prescribing software to move to SCRIPT version 2023011 by 1 January 2028, and software certified on that version must support the electronic prior authorization transactions, which had been optional for certification since 2020. The rule also adds a criterion for real-time prescription benefit tools and three criteria for approval interfaces built on FHIR, the common format for exchanging health records.
Credit
ASTP/ONC; CMS
Watch for
Drugs join the approval rule
CMS's January 2024 rule on electronic approvals, CMS-0057-F, left drugs out: "this provision does not apply to prior authorization decisions for drugs." A rule CMS proposed on 14 April 2026, CMS-0062-P, would require Medicare Advantage plans, Medicaid and CHIP programs and their managed care plans, and plans on the federal HealthCare.gov exchanges to support NCPDP's SCRIPT transactions for electronic approval of drugs in the pharmacy benefit, and to add drugs in the medical benefit to their FHIR approval interfaces, from 1 October 2027. Medicaid and CHIP would answer drug requests not already under a deadline within 24 hours, and exchange plans within 72 hours, or 24 hours when urgent. Every drug denial would carry a specific reason, and plans would post their drug approval and denial figures.
Achieved whenThe rule is final and the first posted figures show drug requests answered electronically within those times.
Claimed by
CMS, April 2026
Approvals in 18 seconds
Surescripts, an e-prescribing network, says its Prior Authorization Automation returns approvals in seconds when all of a plan's criteria are met. Its 2025 impact report, dated March 2026, gives 68,000 prescribers with access in 42 health systems, 104 medications covered (up from 40 in 2024), a median of 18 seconds to approval when all criteria are met, and automatic approval for 34% of requests for the medications in scope. It does not say how these figures were measured.
Achieved whenAn outside study of health-system or plan records compares time to approval, denials and abandoned prescriptions for automated requests against faxed and portal requests for the same drugs.
Claimed by
Surescripts, March 2026, repeated in trade press in May 2026
A federal step therapy exception
The Safe Step Act, introduced in the US Senate on 18 September 2025 by Senator Lisa Murkowski with 19 cosponsors, would write into ERISA, the federal law for employer health plans, a duty to grant an exception to step therapy when the required drug has already failed, is likely to cause harm, would cause harmful delay, or when the patient is stable on the prescribed drug. Plans would answer within 72 hours, or 24 hours when delay could seriously harm the patient, and an exception would last at least one year.
Achieved whenIt is enacted and the Labor Department issues the rules it calls for.
Claimed by
Senator Murkowski and cosponsors, September 2025; the bill was referred to the Senate health committee
Open
01
Same drug, many plan rules
A public, machine-readable table of the prior authorization and step therapy criteria that every Medicare Part D plan sponsor posts for 50 high-spending specialty drugs, for plan year 2027. Each criterion is coded into fixed fields (diagnosis, prescriber specialty, drugs to try first, tests required, length of approval, renewal rules), linked to the page it came from, and marked where it goes beyond the FDA label. Repeat each October for the next plan year.
The briefCredit · unclaimed02
Answer the plan's questions
A public set of made-up patient charts for specialty drugs, each paired with a question set built from a Part D sponsor's posted criteria (diagnosis, drugs tried and why they were stopped, test results, prescriber specialty), with the correct answer to every question and the correct outcome: meets the criteria, does not, or information missing. Score AI systems on filling in the questions from the chart, the step that automatic approval depends on. No real patients or real requests are used.
The briefCredit · unclaimed
Open 01 · the brief
Same drug, many plan rules
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A public, machine-readable table of the prior authorization and step therapy criteria that every Medicare Part D plan sponsor posts for 50 high-spending specialty drugs, for plan year 2027. Each criterion is coded into fixed fields (diagnosis, prescriber specialty, drugs to try first, tests required, length of approval, renewal rules), linked to the page it came from, and marked where it goes beyond the FDA label. Repeat each October for the next plan year.
Why it is still open
CMS's free formulary files show only whether a drug needs prior authorization or step therapy, not the criteria. The nearest study, by Huseyin Naci, Aaron Kesselheim and colleagues (JAMA Health Forum, 2023), read 638 policies from the eight largest Part D insurers for drugs approved from 2013 to 2017, using 2020 formularies. It found that 40% of the drugs were covered by at least one large plan with criteria stricter than the label, and that total requirements ranged from 8 at Kaiser Permanente to 187 at Humana. It reports counts for one year; no public, machine-readable copy of the criteria was found, and a GitHub search in November 2026 found no such dataset.
Built with what exists
CMS's Part D formulary files; each sponsor's posted criteria documents; a web agent such as Browser Use to find and download each sponsor's criteria file (see the Note "Deleting 80% to start over"); a low-cost AI model for repeated extraction, such as Meta's Muse Ember or Claude Haiku 5.5 (see the Note "The jobs agents repeat"), drafting coded rows for a pharmacist to check; FDA labels through openFDA; a CSV file in a public GitHub repository.
First step
Download the current formulary file, list the sponsors that require prior authorization for five specialty drugs, and code those sponsors' criteria by hand to settle the codebook.
Done looks like
50 drugs across every sponsor that posts criteria for 2027; every coded field linked to its source page; a second pharmacist re-codes a random 10% with agreement (Cohen's kappa) reported; the label comparison done by two pharmacists; the 2028 refresh published by 31 December 2027.
Open 02 · the brief
Answer the plan's questions
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A public set of made-up patient charts for specialty drugs, each paired with a question set built from a Part D sponsor's posted criteria (diagnosis, drugs tried and why they were stopped, test results, prescriber specialty), with the correct answer to every question and the correct outcome: meets the criteria, does not, or information missing. Score AI systems on filling in the questions from the chart, the step that automatic approval depends on. No real patients or real requests are used.
Why it is still open
Surescripts reports automatic approval for 34% of in-scope requests but publishes no test of how its answers are drawn from charts. Duke's AI tool for drugs such as Ocrevus, Botox for migraine and Pluvicto got the approval status right in all 83 past cases checked, but those cases are Duke patients' records, the tool needs Duke Health authorization, and the drugs are medical-benefit drugs, not pharmacy-benefit ones. The 2006 pilots already found coded questions "too clinical for MAs to complete accurately, and too general to be autopopulated". NCPDP's own question-set format can be read only by its members.
Built with what exists
Synthea for made-up patients in FHIR format; chart notes written by clinicians; question sets written as FHIR Questionnaires in the HL7 Da Vinci DTR format, which is open; general AI models through their public interfaces, and a small open model fine-tuned on one graphics chip with a separate set of made-up training charts (see the Note "One GPU to fine-tune"); a spreadsheet and a short Python scoring script.
First step
Take one drug's posted criteria from three Part D sponsors, write 20 charts, and have two pharmacists answer them separately.
Done looks like
300 charts for 30 drugs; every question answered independently by two pharmacists, with agreement reported; at least four systems scored per question and per outcome, with the rates of wrong "meets criteria" and wrong "does not meet" reported separately; cases, answers and code public under CC BY.