Musings on healthcare IT, FHIR, EHR systems, and digital health innovation

🔥 FHIR in the hole!
Back to posts

Anyone Can Add Structure... If Plans Publish the Rules of Play

Original on LinkedIn

There is a lot of pressure right now to improve prior authorization. Some of that work is overdue and useful: fewer portals, less faxing, cleaner transactions, better ways to move documentation from the chart into the request. CMS’s interoperability and prior authorization rule, finalized in January 2024, pushes the system in that direction by requiring impacted payers to stand up a Prior Authorization API by 2027. ASTP/ONC’s HTI-4 rule, finalized in July 2025, adds health IT certification criteria tied to the Da Vinci prior authorization workflow stack. All of that may make prior authorization more electronic. None of it, by itself, solves a more basic problem: a coverage rule cannot be encoded if it is not published in enough detail.

I ran into that problem while building a structured coverage model for obstructive sleep apnea from 27 public CMS documents. I picked OSA because it sits at a useful intersection. Some treatment paths rest on national Medicare policy. Others depend almost entirely on regional contractor policy. The billing layer adds its own logic on top. I wanted to know whether the published Medicare rulebook was rich enough to reconstruct the actual decision logic in a form that was usable, comparable, and computable. The resulting prototype is here: https://joshuamandel.com/cms-ncd-browser-demo

Put briefly: The Medicare rulebook and narrative-based APIs were detailed enough for frontier LLM agents to read, process, and structure.

Anyone can add structure, if plans publish the rules of play. That changed how I think about the bottleneck. A lot of the discussion around computable coverage assumes the hard part is format: the policy exists, but it is trapped in prose, so what we need is a schema. After building the OSA model, I think that gets the emphasis wrong. When the source material is detailed and public, structure is the easier part. The harder problem is disclosure.

Medicare has two very different transparency regimes

To see why that matters, it helps to separate two Medicare worlds that often get blurred together.

In traditional fee-for-service Medicare, the coverage rules are at least put on the record. At the top are National Coverage Determinations, which are Medicare-wide coverage policies for particular items or services. Below them are Local Coverage Determinations, written by regional Medicare Administrative Contractors when no national rule exists or when local implementation detail is needed. Then there are companion Articles, which often carry coding, billing, and documentation instructions, and Response to Comments records, which preserve the history of how a local policy was revised.

Medicare Advantage is different. MA plans are still bounded by traditional Medicare for basic benefits. The governing regulation requires them to cover Medicare Part A and Part B services and to comply with national coverage determinations, general Traditional Medicare coverage and benefit conditions, and applicable written local contractor decisions. Plans may create internal coverage criteria only when the relevant Medicare criteria are not fully established, and if they do, those criteria are supposed to be publicly accessible along with the evidence considered, the sources of that evidence, and the rationale for using the criteria. Where a plan is adding criteria to interpret or supplement general Medicare provisions, it must also explain why those added criteria are highly likely to improve benefits more than harms.

On paper, that is a fairly strong disclosure rule. It tells plans when they may fill gaps, and it tells them that if they do, they have to show their work. But implementation is inconsistent at best.

Disclosure rule exists but the public record is patchy.

In May 2025, GAO published a review of nine Medicare Advantage organizations that together covered about 45 percent of MA beneficiaries in 2024. The review focused on selected behavioral health services, and it was not designed to be statistically generalizable, but the findings were still striking. GAO found that posted criteria were often incomplete. Among the eight publicly accessible internal coverage criteria it reviewed, none contained all required components. Some included a summary of evidence. Some included sources of evidence. None included the required rationale. None included the explanation of how the clinical benefits of using those criteria likely outweighed the harms.

Worse, GAO could not locate the public internal coverage criteria used by three of the nine organizations for all selected services. Two of those organizations cited third-party developer issues; one said its contract with a third-party developer did not allow the criteria to be made public. Even when criteria were available, they were often hard to locate, posted in inconsistent places, or not clearly identified as the criteria applicable to MA members. CMS officials told GAO that plans had been confused about what counted as internal coverage criteria in the first place, especially where vendor-developed material had been used for years.

CMS clearly saw the discoverability problem. In the proposed rule for contract year 2026, the agency proposed more concrete website requirements: a prominent page listing all items and services with internal coverage criteria, clearer labeling, vendor attribution, no-login access, and pages that were machine-readable, searchable, and downloadable. But CMS did not finalize those provisions in the April 2025 final rule; it deferred them for future rulemaking.

That is the backdrop for the OSA exercise. Traditional Medicare is the relatively good case: the rules may be fragmented and verbose, but much of the rulebook is public. MA is the harder case: plans may be using detailed criteria that are still too hard to find, too incomplete, or too thinly published to inspect.

Why the current standards push does not fix this

This is the part that gets muddled in a lot of health IT conversations.

CMS-0057-F and the related Da Vinci stack improve exchange and workflow. CRD can help an EHR learn whether prior authorization is required and what documentation is needed. DTR can help assemble that documentation in a structured way. PAS can help submit the request and check its status. HTI-4 gives health IT developers certification targets for pieces of that workflow. All of that is useful.

But none of it, by itself, gives you a canonical, inspectable representation of the underlying coverage rule: the threshold, the exclusion, the step-therapy gate, the provider requirement, or the link back to the Medicare baseline.

A questionnaire that asks, “Does the patient have a history of X” is not the same thing as a public coverage object that says how this information will be evaluated together with the rest of the patient history to make a coverage determination.

That is why so much of the current excitement feels slightly off the critical path. Better transport is good. Better workflow is good. But if the underlying rulebook is not public enough to inspect, then the standards stack is moving around an object that still has not been fully disclosed.

Conversation on LinkedIn

25 comments · 75 reactionsReply on LinkedIn
  • Anu RamakrishnanApr 2, 2026

    Director of Product Management

    The age old problem, applicable across all kinds of payer data/requirements - enrollment, auth, allowables and so on

    2 reactions

  • William LaolagiApr 2, 2026

    Interoperability specialist serving the state of California in Riverside and San Bernardino counties

    This sounds great, but I’ve got two big 'what-ifs' from a payor perspective:
    First, the gaming risk. If a COIN agent can see the exact rules, it’ll eventually just learn how to pass the minimum requirements for a 'Yes' without actually giving us the whole clinical picture.

    Second is the versioning. Getting every system to talk to each other is one thing, but keeping them on the same version of the rules at the same time? That’s a massive lift. Without that, I think we might run into more integration headaches.

    If you and the community can pull this off though, this would be amazing, and I believe it's needed.

    2 reactions

    • Josh Mandel, MDAuthorApr 2, 2026

      William Laolagi Re: versioning this feels pretty purely technical; we have so many good tools in software dependency management that I'm quite sanguine on this front (though maybe we have something analogous to supply chain attacks to worry about ;-), in some prompt-injection space?)

      Re: gaming, that is totally right. It's the central question of incentive design. And a hard one. If you have the right game, then everyone scrambling to win should lead to better health...

      1 reaction

    • William LaolagiApr 3, 2026

      Interoperability specialist serving the state of California in Riverside and San Bernardino counties

      Fair point on the tools.

      On the gaming side, my concern is FWA (Fraud, Waste, and Abuse). Working in the public sector, we're under a massive microscope, and that accountability falls on us if the system is drained by bad actors. We want the patient at the center, which means protecting the resources they rely on. If the rules are a 100% transparent 'cheat sheet' for fraud, it’s the patients who lose out in the long run.
      The question now is: how do we solve for this? How do we balance open clinical logic with the necessary 'guardrails' to protect the program? 🤔

    • Josh Mandel, MDAuthorApr 3, 2026 · edited

      William Laolagi I would start from the other direction: in what world does hiding the rules of "how to provide the best-value care" improve the ability of legitimate providers to do just that? Yes fraud, waste, and abuse are problems, but the fraudsters may well be superior at "inferring the hidden rules and gaming them" when compared with legitimate clinicians (who, after all, have to spend their time and energy actually caring for patients).

      1 reaction

    • William LaolagiApr 3, 2026

      Interoperability specialist serving the state of California in Riverside and San Bernardino counties

      fair point. I’m not suggesting we hide the 'how-to' for best-value care, that should definitely be out in the open for legitimate clinicians.
      Thinking through it, the shift really needs to be from hiding the rules to verifying the facts. If we open up the logic like you’re suggesting, then 'Proof of Care' has to be the focus. Maybe the move is to lean into the provenance resource or even digital sigs at the resource level.
      If we can create a solid verification layer where we can cryptographically trust that the data is tied to a real encounter, then the 'open rules' become much less of an FWA risk. It gives us(payors) the security we need for program integrity without making the honest providers jump through more manual hoops.
      Does the SMART or HL7 roadmap have a clear path for standardizing that 'Proof of Care' layer yet?

      1 reaction

    • Anna McCollisterApr 3, 2026

      Catalyst, Entrepreneur, 2X Founder, Board of Directors Member, Consultant, Advocate, Health Data Governance, Health IT Policy, Patient Mobilization, Genomics, Trust Frameworks and AI

      William Laolagi Josh Mandel, MD exactly! Solve for the problem in front of you and develop separate processes to identify fraud, rather than live with painful and consequential failures in an effort to prevent theoretical fraud. Fraud will always emerge. That’s a separate problem that requires a separate solution. But if you’re arguing that very consequential, costly and life impacting inefficiency are necessary to prevent potential fraud, you need to find a better way. Not knowing any of the details, I would bet the opacity makes fraud more difficult to detect than it would be if the rules were transparent to the point that it could be automated.

    • Anna McCollisterApr 3, 2026

      Catalyst, Entrepreneur, 2X Founder, Board of Directors Member, Consultant, Advocate, Health Data Governance, Health IT Policy, Patient Mobilization, Genomics, Trust Frameworks and AI

      Anna McCollister I do, however, appreciate the sincerity and honesty of your questions William Laolagi ! :-)

      1 reaction

    • William LaolagiApr 3, 2026

      Interoperability specialist serving the state of California in Riverside and San Bernardino counties

      Anna McCollister This was a great thought exercise and actually led me to build a Proof of Concept using Agentic AI to solve exactly this.
      In the public sector, fraud and abuse aren't theoretical problems, they are a massive accountability we have to manage. My approach moves the security from hiding the rules to verifying the facts. By applying a cryptographic fingerprint to the clinical data straight from the EHR, we can keep the rules 100% open and transparent. If a bad actor tries to game those open rules with fake data, the seal breaks and the rules never even run.
      It gives the honest clinician a fast approval while putting up a mathematical wall against fraud. We stop protecting the secret and start verifying the truth. That is the better way forward.

      1 reaction

    • William LaolagiApr 3, 2026

      Interoperability specialist serving the state of California in Riverside and San Bernardino counties

      Anna McCollister I appreciate the dialogue with Josh Mandel, MD! This has been a great thought exercise, and I just want to bring up some points that may be helpful to think about as we try to push healthcare forward.

      1 reaction

  • Healthcare Strategy & Product Management| AI Platforms | Health Tech

    For Medicare and Medicaid plans the state specific plans will have the rules already defined. It’s a good place to start.

  • Brendan KeelerApr 2, 2026

    Interoperability Practice Lead at HTD Health

    Scarcity by design, unfortunately.

    https://healthapiguy.substack.com/p/artificial-scarcity-meet-artificial

    1 reaction

  • Lucia SavageApr 2, 2026

    Smart ideas to fix healthcare. Board Member. Advisor. Expert advisor on public policy, AI, privacy, cybersecurity, and the future of healthcare. I answer questions for CNN and Senators. I write my own posts, not AI.

    So true. It’s not like a coverage rule is top secret IP. It’s not really good science unless it’s based on what the relevant medical specialists say in other published (peer reviewed) materials. (Oops did I say that out loud?)

    3 reactions

  • Lucia SavageApr 2, 2026

    Smart ideas to fix healthcare. Board Member. Advisor. Expert advisor on public policy, AI, privacy, cybersecurity, and the future of healthcare. I answer questions for CNN and Senators. I write my own posts, not AI.

    Can I add that although it is an under-enforced right, and likely under-litigated, too, people in ERISA employee welfare benefit plans have a legal right to get copies coverage rules when they are disputing a denied claim for benefits. Theoretically it’s possible to crowd-source this info. But as I said under enforced and under-litigated. Plus version control as the rules do evolve.

    4 reactions

  • Full Stack OB/GYN. Healthcare Disruptor. AI x Health Ambassador. Human in the Loop. Wartime Consigliere.

    Triple whammy of unpublished and deliberately opaque coverage rules, mid-season changes (there’s a TON that went into effect 4/01) , and perpetual moving goalposts 🎯
    … and when all-else-fails just claw it all back

    1 reaction

  • Epic, AI & cognitive sustainability

    If health plans publish their approval logic, then healthcare systems could tailor documentation accordingly to get approved.

    That would make the approval rate too high. Health plans would have to change the logic again. It would be a constantly moving target.

    2 reactions

    • Josh Mandel, MDAuthorApr 2, 2026

      > could tailor documentation

      That would be playing correctly, if it's truthful and accurate. We might still need to have hard conversations about what's worth paying for (and how much)... but it's better than allocating based on who *talks about their care* the best.

      2 reactions

  • Imran QureshiApr 3, 2026

    CTO & Chief AI Officer @ b.well Connected Health

    Agree. We need transparency not format standardization.

    2 reactions

  • Gunjan V.Apr 3, 2026

    Developer @ GMG ArcData | Digital Platforms, NLP

    Josh Mandel, MD - aren't these just clinical guidelines that've been adopted by the plans? Evernorth and others publish these

  • Nidhi PengoriaApr 3, 2026 · edited

    Product Leader | HealthTech | Interoperability

    For Medicare and Medicaid plans, payers are already publishing PA criteria in their provider portals, adding transparency but shifting the burden to providers navigating complex, payer-specific rules. While some discretion exists in documenting clinical necessity, payers are increasingly applying structured, evidence-based criteria that limit subjectivity. With advances in AI, it will be interesting to see how health plans are motivated to expose AI agent-ready APIs or MCP servers to streamline PA — reducing administrative burden on both sides. That said, HIPAA compliance, algorithmic bias, and fragmented AI governance across federal and state regulators remain real barriers to meaningful adoption.

    2 reactions

  • stephen wardApr 3, 2026

    Innovative Healthcare IT Executive | CTO & Cloud Architect (AWS) | AI & RAG Specialist | Telehealth & FHIR Interoperability Leader

    prior authorization is the "problem" unless your dealing with a fake patient and fraudulent doctor. Insurance shouldn't be deciding care plans. Formularies and other blocking techniques.

  • Chief Medical Officer at Foci Labs

    can't some of this be solved with better contracting? Health plans act the way they do in part because they write contracts very much in their favor to act with virtual impunity for extractive fees - if these were challenged or rewritten to force transparency on items such as prior authorizations, would this be addressed? Granted, a consequence might be the plan "pulling out" of a market because it is no longer as profitable, but given the current state, perhaps to be expected

  • Healthcare AI Subject Matter Expert & Benchmark Evaluation Specialist at Mercor

    In my state, NJ we are on a WISER AI model and if you don't know what's acceptable to the AI Wiser model, appeals are even futile. I created a model with codes to ensure even before submitting, it's a higher acceptance outcome especially now that CMS0057 is a factor on the front end. The codes that are acceptable and their combinations are built into WISER, which means there can be an algorithm (codes), that will be accepted knowing how the AI is set up. Great read.

    2 reactions

  • Reuven SchnaiderApr 3, 2026

    Senior Project Manager | Healthcare Interoperability | FHIR | AI | Digital Health Innovation

    It sounds very similar to the challenges we face in Israel. While we have a centralized national health basket, each HMO operates with its own internal 'black box' of clinical validation logic, especially for complex oncology protocols.
    ​The real breakthrough won't just be a digital pipe for requests, but making the HMOs' coverage rules computable. By leveraging FHIR’s Clinical Reasoning resources (like PlanDefinition and Library) combined with CQL, we could shift from a reactive 'wait and see' process to real-time transparency.
    ​Imagine an oncologist receiving a CDS Hook notification during the ordering process, confirming that the chemotherapy protocol meets the HMO’s specific criteria based on the patient’s latest lab results. That’s the level of transparency we need to reduce the administrative burden and, more importantly, the time-to-treatment for our patients.

    1 reaction