BRENNEN

Industries / Health / Health insurers and PBMs

Doctor and hospital networks

Choosing which doctors and hospitals a plan pays for and on what terms, steering members toward them, and keeping the published list of them accurate.

6

Milestones achieved, from 1973 to 2025.

2

To watch for: announced, not yet confirmed outside.

2

Open deliverables. Nobody has shipped them.

The road

AchievedWatch forOpen

Achieved

  1. 1973

    Employers must offer HMOs

    The Health Maintenance Organization Act of 1973 (Public Law 93-222, 29 December 1973) funded new HMOs, plans that give care through their own salaried doctors, medical groups or associations of independent practices in return for a fixed advance payment, with grants, loans and loan guarantees. Its section 1310 required employers with 25 or more employees that offered a health plan to include the option of a qualified HMO in the area, and, where both kinds existed, one built on staff or group doctors and one on independent practices.

    Lessona mandate to offer a tight network did not make most workers choose one; in 2025, 12% of US workers with employer cover were in an HMO and 46% in a PPO, a plan that also pays part of the cost outside its network, by KFF's survey.

    govinfo.govkff.org

    Credit

    US Congress

  2. 2008

    Family doctor first

    On 1 July 2008 a contract for family-doctor-centred care (hausarztzentrierte Versorgung, HzV) took effect in Baden-Württemberg under section 73b of Germany's social code, between the AOK sickness fund there and family doctors' organizations. It is voluntary for doctors and patients, and members who join see their family doctor first; 1.25 million were enrolled by 2014. Evaluators led from Heidelberg compared 580,924 adults who joined with 862,237 who did not: by 2010 participants made 38% more visits to their family doctor and fewer visits to specialists, and their drug costs were lower, but their hospital admissions did not differ significantly. A later report, presented in September 2014, said about 9,000 admissions had been avoided in 2011 and 2012.

    Lessona gatekeeper changed visits and prescribing within two years, while effects on hospital use showed up only in later data.

    ebi.ac.ukaerzteblatt.de

    Credit

    AOK Baden-Württemberg, MEDI Baden-Württemberg and the state branch of the German family doctors' association; evaluation by Joachim Szecsenyi's department at Heidelberg University Hospital and Ferdinand Gerlach's institute at the University of Frankfurt

  3. 2012

    Doctors share Medicare savings

    In the Medicare Shared Savings Program, launched by CMS in 2012, accountable care organizations (ACOs), groups of doctors and hospitals held to a spending target for their traditional Medicare patients, keep part of any savings below the target. Using claims from 2009 to 2013, McWilliams, Hatfield, Chernew, Landon and Schwartz found that ACOs entering in 2012 cut spending by $144 per beneficiary in 2013, about 1.4%, while those entering in 2013 saved almost nothing.

    Lessonsavings were larger in independent primary care groups than in groups tied to hospitals.

    ebi.ac.ukhealthcarefinancenews.com

    Credit

    CMS; evaluation by J. Michael McWilliams and colleagues (New England Journal of Medicine, 2016)

  4. 2016

    Half the listings wrong

    CMS checked Medicare Advantage plans' online provider directories in three rounds from 2016 to 2018, reviewing listings for cardiologists, oncologists, eye specialists (ophthalmologists) and primary care doctors. The share of listed locations with at least one error was 45.10% in the first round, 55.07% in the second and 48.74% in the third, which covered 10,504 locations at 52 organizations from November 2017 to July 2018. Errors meant the doctor was not at that location, the phone number was wrong, or the doctor was not taking new patients when the directory said so; plans' error rates in the third round ranged from 4.63% to 93.02%.

    Lessonthe error rate did not fall across three rounds of reviews and compliance actions, though CMS notes that the rounds' methods differed slightly.

    cms.gov

    Credit

    CMS

  5. 2024

    Most Swiss accept limits

    Swiss basic insurance lets members trade free choice of doctor for a lower premium by joining an alternative model, such as a family-doctor model or an HMO. The share of insured people in such models rose from 1.7% at the end of 1996 to 80.5% at the end of 2024, 7.28 million of 9.05 million people, by the Federal Office of Public Health's statistics, which count people who combine such a model with a higher deductible in the same group.

    bag.admin.chbag.admin.ch

    Credit

    Swiss insurers; Federal Office of Public Health

  6. 2025

    Free-choice policies disappear

    Dutch standard policies come as restitution policies, which pay for any provider the member chooses, in-kind policies built on the insurer's contracted providers, and combinations of the two. After years of decline, no insurer offered a restitution policy for 2025; insurers blamed the cost of bills from non-contracted community nursing and mental health care, by the NZa's report of May 2025. The share of insured people on budget policies, which offer fewer contracted providers or pay less than 75% of the average contracted rate for others, rose from 18.4% in 2021 to 28.5% in 2025.

    Lessona policy that pays any provider in full carries the bills of every provider the insurer has not contracted with.

    zoek.officielebekendmakingen.nl

    Credit

    Dutch insurers; NZa

Watch for

  1. Record year for Medicare ACOs

    CMS said on 17 November 2026 that 472 accountable care organizations, 80% of those in the Medicare Shared Savings Program, earned $4.9 billion in performance payments for 2025, covering 84% of 10.9 million assigned beneficiaries, and called it the highest rate of savings since the program began in 2012. For 2024 it had reported $2.4 billion saved for Medicare. CMS measures savings against benchmarks set by its own formula.

    Achieved whenan independent evaluation with a comparison group, of the kind published in 2016, finds net savings to Medicare for 2024 and 2025 after the payments to ACOs.

    healthcarefinancenews.com

    Claimed by

    CMS, November 2026, as reported by Healthcare Finance News

  2. Directory accuracy scored publicly

    Section 6220 of H.R. 7148, the appropriations law signed on 3 February 2026, called the REAL Health Providers Act, requires Medicare Advantage plans from plan year 2028 to verify each listing at least every 90 days, mark listings they cannot verify, remove a departed provider within 5 business days, protect members from out-of-network cost sharing when they relied on a wrong listing, and report a yearly accuracy score from a random sample. CMS must post the scores in a machine-readable file for plan years from 2029.

    Achieved whenCMS publishes the first scores and an outside check of sampled listings agrees with them.

    govinfo.govchiefhealthcareexecutive.com

    Claimed by

    US Congress, February 2026

Open

  1. 01

    Network breadth from public files

    A yearly measure, for every plan sold on HealthCare.gov, of how many local doctors and hospitals are in its network, by specialty and county, built only from public files: the provider files each insurer must publish, listed in CMS's machine-readable URL file; the national provider register (NPPES); and CMS's public data on clinicians who bill Medicare, as a count of doctors actually in practice. Publish the results each plan year with the code.

    The brief
    Credit · unclaimed
  2. 02

    Rosters into directories, scored

    A public test of software that turns provider rosters into directory updates. Medical groups send insurers rosters, spreadsheets listing each clinician's locations, specialties, phone numbers and whether they take new patients, along with joiners, moves and departures, and the insurer must turn each into correct directory changes. Build 500 invented rosters from 100 invented groups in the messy forms they arrive in (different column names, merged cells, PDFs, scans and emails), each with a hand-made answer key of the right changes in the HL7 Plan-Net format, and score AI pipelines and rule-based tools on getting every change right. Report departures separately, since the 2026 law gives plans 5 business days to remove them.

    The brief
    Credit · unclaimed

Open 01 · the brief

Network breadth from public files

Credit · unclaimed

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

What to build

A yearly measure, for every plan sold on HealthCare.gov, of how many local doctors and hospitals are in its network, by specialty and county, built only from public files: the provider files each insurer must publish, listed in CMS's machine-readable URL file; the national provider register (NPPES); and CMS's public data on clinicians who bill Medicare, as a count of doctors actually in practice. Publish the results each plan year with the code.

Why it is still open

The most recent national measure, KFF's analysis of August 2024, covers plans from 2021 only and rests on a CMS research file on doctors' practices (MD-PPAS) and commercial directory files from Ideon, so it cannot be rerun from public files alone. It found that enrollees' networks included 40% of local doctors on average, and that 23% of enrollees were in plans with a quarter or fewer. An open toolkit released in October 2026, plannet-directory-check, reads the same insurer files, but to measure what records leave out, not network breadth.

kff.orggithub.com

Built with what exists

CMS's Marketplace public use files, including the machine-readable URL file; NPPES; CMS's public "Medicare Physician & Other Practitioners" data; the plannet-directory-check code; Python; a web agent such as Browser Use to check a sample of listings against each insurer's own search page (see the Note "Deleting 80% to start over").

cms.gov

  1. First step

    Download the provider files of every insurer in one state from the machine-readable URL file and compute primary care breadth for each network against NPPES.

  2. Done looks like

    Breadth for every HealthCare.gov plan network in plan year 2027, for primary care, five specialties and hospitals, with the national averages compared to KFF's 2021 figures as a check of method, a hand check of 300 listings against insurers' websites, and code and results published.

Open 02 · the brief

Rosters into directories, 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 test of software that turns provider rosters into directory updates. Medical groups send insurers rosters, spreadsheets listing each clinician's locations, specialties, phone numbers and whether they take new patients, along with joiners, moves and departures, and the insurer must turn each into correct directory changes. Build 500 invented rosters from 100 invented groups in the messy forms they arrive in (different column names, merged cells, PDFs, scans and emails), each with a hand-made answer key of the right changes in the HL7 Plan-Net format, and score AI pipelines and rule-based tools on getting every change right. Report departures separately, since the 2026 law gives plans 5 business days to remove them.

Why it is still open

From plan year 2028 the REAL Health Providers Act makes Medicare Advantage plans verify every listing every 90 days and report an accuracy score that CMS will publish. A vendor executive writing in July 2026 said most plans still rely on manual roster ingestion and updates. Vendors sell roster automation, but a search in November 2026 found no public test set and no independent accuracy figures for it.

govinfo.govchiefhealthcareexecutive.com

Built with what exists

The HL7 Da Vinci PDex Plan-Net format that insurers' public directory interfaces use, as read by plannet-directory-check; invented clinicians and groups only, with identifiers checked not to exist in NPPES; AI models that read spreadsheets, PDFs and scans, including small low-cost ones for the repeated work (see the Note "The jobs agents repeat"); Python for scoring.

github.com

  1. First step

    Write 20 invented rosters for four invented groups in four layouts, holding 200 changes, and score one AI model on them.

  2. Done looks like

    500 rosters holding about 5,000 changes, each answer checked by two people; at least four pipelines scored on precision and recall for joiners, changes and departures, with missed departures reported on their own; data and scoring code public.