BRENNEN

Industries / Health / Healthcare / Pharmacy / Specialty pharmacy

Taking in referrals

Taking in referrals is the work of turning a new specialty prescription, usually sent with a drug maker's enrollment form and pages of records, into a patient whose insurance has been checked and who is ready to receive a first shipment.

5

Milestones achieved, from 2018 to 2024.

2

To watch for: announced, not yet confirmed outside.

2

Open deliverables. Nobody has shipped them.

The road

AchievedWatch forOpen

Achieved

  1. 2018

    Cost decides who starts

    Jalpa Doshi and colleagues at the University of Pennsylvania studied insurance claims for 38,111 people with a new, processed prescription for one of 38 cancer pills in 2014 and 2015. Overall 18% never picked up the drug within 90 days. After adjusting for other factors, the rate was 10.0% when the patient owed $10 or less and 49.4% when the patient owed more than $2,000.

    LessonA prescription can pass the insurance check and still never become a first dose, because what the patient will owe decides whether the drug is collected.

    doi.org

    Credit

    Jalpa A. Doshi, Pengxiang Li, Hairong Huo, Amy R. Pettit and Katrina A. Armstrong

  2. 2019

    Coverage shown while prescribing

    A Medicare rule of May 2019 required every Part D drug plan to offer, by 1 January 2021, a real-time benefit tool: software inside at least one prescribing system that shows the doctor, for one patient, whether a drug is covered, what it costs and what approvals it needs. This moves part of the benefits investigation (finding what a plan covers and what the patient owes) to the moment of prescribing. A June 2024 rule requires these tools to use one standard, the NCPDP Real-Time Prescription Benefit standard version 13, from 1 January 2027. US only.

    federalregister.govfederalregister.gov

    Credit

    Centers for Medicare and Medicaid Services (CMS); National Council for Prescription Drug Programs (NCPDP)

  3. 2021

    A standard for enrollment data

    In 2015 consultants writing in Pharmacy Times said "the vast majority of prescriptions for specialty medications are faxed to the pharmacy," often on a drug-specific intake form, because the e-prescription standard had no place for lab results, coverage details or financial-assistance information. On 21 June 2021 HL7 and NCPDP published the first trial-use version of the Specialty Medication Enrollment guide, a shared electronic format for the clinical, insurance and consent data that pharmacies and drug makers' support programs ask for after the prescription itself. A second version followed on 4 May 2023. Surescripts had launched a commercial service, Specialty Patient Enrollment, in January 2020, with Accredo, Kroger, AssistRx, CareMetx and PharmaCord taking part.

    hl7.orgpharmacytimes.comhealthdatamanagement.com

    Credit

    HL7 Pharmacy Work Group; NCPDP Work Group 18 (Specialty Requirements for ePrescribing); Surescripts

  4. 2021

    One clock for turnaround

    In 2021 the members of the Pharmacy Quality Alliance (PQA), a US quality-measure body, endorsed a single definition of specialty pharmacy turnaround time: weekdays from the first date a new prescription appears in the pharmacy's system to the date it is ready or scheduled for delivery. In a pilot published in 2026 and funded by Pfizer, PQA staff rechecked the raw data of five specialty pharmacy organizations for 2021 to 2023. Mean turnaround was 2.42 days, and at one organization the recheck changed its own figure by close to a full day.

    LessonThe clock starts when the pharmacy enters the prescription, so days lost before that, to missing pages or unanswered faxes, are not counted.

    pmc.ncbi.nlm.nih.gov

    Credit

    Pharmacy Quality Alliance; Heather Gibson, Lynn Pezzullo and colleagues

  5. 2024

    Nine to fifteen days

    A Columbia University Irving Medical Center team followed 1,024 new prescriptions for cancer pills at Columbia and NewYork-Presbyterian in 2018 and 2019. 86% of them reached the patient within 90 days. The median wait from prescription to receipt was 9 days when no copay assistance was sought and 15 days when it was, 8 days for Medicaid and 10 days for Medicare or commercial insurance. Waiting for prior authorization, a payer's approval before it will pay, was not linked to longer waits in this study.

    pmc.ncbi.nlm.nih.gov

    Credit

    Morgan Lichtenstein, Dawn Hershman and colleagues, Columbia University Irving Medical Center

Watch for

  1. AI phones the insurer

    Infinitus said its AI agents, which handle calls to insurers to check a patient's drug benefits and follow up on approvals, completed more than 60,000 tasks for one specialty pharmacy in the first half of 2025, and that each benefits call saved staff an average of 29 minutes. The pharmacy is not named and no accuracy figures were given.

    Achieved whenA pharmacy or outside reviewer publishes how often the agents' benefit answers matched a human re-check, and the change in days from prescription to first fill.

    hitconsultant.net

    Claimed by

    Infinitus, with IBM Consulting, September 2025

  2. In-house pharmacies start sooner

    Vanderbilt University Medical Center reported that of 1,466 new specialty prescriptions written in its cancer, inflammation and multiple sclerosis clinics in 2021 and 2022, 6.5% were never filled when sent to its own health-system specialty pharmacy, against 9.9% when sent to outside specialty pharmacies. Median time to first fill was 3 days against 4. The lead and senior authors work for Vanderbilt Specialty Pharmacy, the pharmacy being compared.

    Achieved whenAn independent study across several health systems finds a similar gap.

    doi.org

    Claimed by

    Katie Cruchelow, Autumn Zuckerman and colleagues, Vanderbilt University Medical Center, March 2026

Open

  1. 01

    Made-up referral packets, scored

    A public set of 300 invented specialty pharmacy referral packets, each a fax-like PDF of 5 to 15 pages: prescription, drug maker's enrollment form, insurance card, clinic notes, lab results and consent page, with smudges, skew and handwriting. Each comes with an answer key of every field in the HL7 Specialty Medication Enrollment format, and a scoreboard of how well extraction tools fill it.

    The brief
    Credit · unclaimed
  2. 02

    Enrollment forms in one format

    A public catalog of drug makers' enrollment forms (the start forms a doctor's office fills in to send a patient to the maker's support program and specialty pharmacy) for 100 limited distribution drugs. Each field on each form is mapped to the matching element of the HL7 Specialty Medication Enrollment guide, or marked as having none, with counts of pages, fields, wet-signature demands and fax-only return.

    The brief
    Credit · unclaimed

Open 01 · the brief

Made-up referral packets, scored

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 300 invented specialty pharmacy referral packets, each a fax-like PDF of 5 to 15 pages: prescription, drug maker's enrollment form, insurance card, clinic notes, lab results and consent page, with smudges, skew and handwriting. Each comes with an answer key of every field in the HL7 Specialty Medication Enrollment format, and a scoreboard of how well extraction tools fill it.

Why it is still open

Consultants wrote in 2015 that most specialty prescriptions arrive by fax. The closest public work is a proof of concept by GitHub user debamit (November 2026) that reads one legacy referral form template, with no scored test set; faxtract (September 2026), which scores extraction on invented radiology referrals, not pharmacy packets; and a Hugging Face set of 223 page images of English cancer referral forms with no dataset card or licence. None is a specialty pharmacy packet with an answer key that several tools are scored on.

github.comgithub.comhuggingface.co

Built with what exists

Synthea, an open-source generator of made-up patients; Python with ReportLab and Faker to lay out pages; image noise scripts to imitate fax; the HL7 Specialty Medication Enrollment profiles as the field list; OCR such as Tesseract or docTR; vision-language models, including small open models fine-tuned on one graphics chip (see the Note "One GPU to fine-tune").

  1. First step

    Write the field list from the guide's second-version profiles and generate 20 packets on four layouts.

  2. Done looks like

    300 packets built on at least 20 different form layouts, every answer key checked by two reviewers with their agreement reported, and at least three extraction tools scored field by field, with the packets, keys and scoring code released under an open licence.

Open 02 · the brief

Enrollment forms in one format

Credit · unclaimed

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

What to build

A public catalog of drug makers' enrollment forms (the start forms a doctor's office fills in to send a patient to the maker's support program and specialty pharmacy) for 100 limited distribution drugs. Each field on each form is mapped to the matching element of the HL7 Specialty Medication Enrollment guide, or marked as having none, with counts of pages, fields, wet-signature demands and fax-only return.

Why it is still open

The HL7 guide sets out the data to exchange but does not map any real enrollment form, and its pages say nothing about which forms or how many were studied. Surescripts' enrollment service fills forms inside its own network and publishes no catalog. A GitHub search found no public catalog of these forms.

hl7.orghealthdatamanagement.com

Built with what exists

A web agent that drives Chrome to find and download each maker's public form, such as Browser Use (see the Note "Deleting 80% to start over"); pdfplumber to read form fields; a CSV file on GitHub; the HL7 guide's published profiles.

  1. First step

    Take the drug names from the TRICARE limited distribution list and download and map the first 10 forms by hand.

  2. Done looks like

    100 forms, each with a source link and download date, every field mapped or marked unmapped, two reviewers independently mapping 20 forms with agreement reported, and a published share of form fields that have no element in the standard.