CMMI ACCESS Program: Can Well-Intentioned Goals Conceal Adverse Incentives?

Header:  CMMI produces a JAMA Op Ed on its "ACCESS" technology and chronic care program.  I used an AI-generated essay to explain, then critique the program in some detail.

_______

This week, JAMA publishes an important paper from CMMI, which appeared online May 11.  See:

• Outcome-Aligned Payments for Technology-Enabled Care—A New CMS Approach to Paying for Chronic Disease Care in Medicare.  
• (2026)  Shiff J & Sutton A.  JAMA 335:1932-34.
• Online here.

I've been experimenting with longer-form AI-mediated writing, and have created/supervised a ten-page, 5000-word essay in response:   

The ten-page essay is in the Google cloud as a PDF here.

The Essay Summary is here:

• CMS’s ACCESS model is a serious attempt to modernize Medicare payment for technology-enabled chronic care. Rather than paying for visits, devices, app clicks, or care-management minutes, ACCESS pays organizations for measurable improvement in chronic disease outcomes — an elegant and potentially important idea. 
• But it also exposes a deeper CMMI paradox: under-managed Medicare patients may need more care, not less. Better management can mean more drugs, physical therapy, behavioral health, monitoring, labs, and specialist follow-up. Judged too narrowly on savings, ACCESS may reward low-cost "metric improvement" while discouraging the costly care activation patients actually need. 

I asked Claude Opus and Chat GPT to make a classification system for the 102 PLA codes being considered for this summer's Annual Lab Meeting, which will price the 102 tests by the crosswalk or gapfill methods between now and November.

I noted that one approach is repeated binary decisions (human vs microbial; then microbial as pathogen panels vs other; and so on), but I emphasized that other classification schemes were possible.    

• Here is a cloud zip file with (1) the original CMS excel, (2) the Claude excel, (3) the Chat GPT excel: 3 files 1 zip.
• Note that both AI's produced complex Excels with many rows, columns, and tabs.  Chat GPT even included a bar chart.

###

 

AI Attempts to Classify 100 New PLA Codes — Gets Migraine

The AMA CPT Editorial Panel creates special proprietary lab codes called PLA codes, or “Proprietary Laboratory Analyses.” In many cases, these describe tests unique to one laboratory, although some are FDA-approved, kit-distributed tests. There are now roughly 700 PLA codes, but no stable public classification system for them.

For the June 10, 2026 CMS Annual Laboratory Meeting, CMS posted a list of 102 new PLA codes to be priced by crosswalk or gapfill between now and November. The CMS spreadsheet includes a “Category” column, but it has not yet been populated. Public presentations proceed code by code, but by the July expert advisory meeting, CMS will impose some kind of category structure on the group.

So I asked two AI systems — Claude Opus and ChatGPT — to classify the 102 codes using the long descriptor column. As a starting point, I suggested one possible approach: repeated binary decisions, such as human versus microbial; then microbial as pathogen panel versus other; and so on. But I also emphasized that other classification patterns might work better. The challenge will be that PLA codes are not just “oncology” or “infectious disease.” They mix analyte, technology, clinical intent, specimen type, and algorithmic output in ways that do not fit neatly into one tree.

 

News Summary: WaPo: HHS Wants to Bring Physicians and AI Together

 

On June 4, 2026, Washington Post has a lead aritlce, "Push to bring AI doctors into American medicine," by Elizabeth Dwoskin.

___

Chat GPT summarizes;

The Washington Post article is worth reading because it shows how quickly AI medicine is moving from speculative futurism into federal and state policy. The main story is not just that chatbots may help patients, but that some officials and entrepreneurs are now openly contemplating AI systems that diagnose, triage, refill prescriptions, and eventually practice medicine with limited human oversight.

The article centers on Amy Gleason, now advising HHS on AI, and describes a broader Trump administration push to integrate AI into health care through 

• prescription-refill pilots, 
• cardiovascular triage research, 
• FDA digital-health fast tracks, 
• Medicaid reimbursement for AI wellness tools, and 
• possible future regulation of “independent AI doctors.” 

The article also gives WaPo readers the essential counterweight: physicians, researchers, and licensing boards warning that chatbots still make diagnostic errors, perform unevenly in real-world patient conversations, and may blur the line between advice and unauthorized medical practice.

For readers interested in FDA policy, digital health, medical liability, access to care, or reimbursement, this is an early map of a major coming policy battlefield.

 ###

Chat GPT also researched Amy Gleason's background, as follows:

Amy Gleason began in nursing, including emergency-room nursing, then moved into health-care technology, especially electronic medical records, practice-management systems, interoperability, and patient access to data. An archived White House profile from the Obama-era Precision Medicine Initiative says she “began her career in nursing” and then spent years building and implementing EMR and practice-management technologies.

Her personal story is central to her public identity. Her daughter Morgan developed a rare autoimmune disease, often described as juvenile dermatomyositis / juvenile myositis, and Gleason became intensely focused on the failures of fragmented medical records and data access. The WaPo article says Morgan later uploaded years of medical records into ChatGPT, leading to a different diagnostic framing and trial eligibility.

Professionally, she co-founded or helped lead CareSync, a health-care coordination / records company, and later worked in digital-health roles including Russell Street Ventures, Main Street Health, and CareBridge.

In government, she worked with the U.S. Digital Service [bringing tech experts and government together, under Obama), including health-data modernization projects. She is now described as Acting Administrator of the U.S. DOGE Service, and Strategic Advisor to CMS, focused on modernization, interoperability, and AI/data systems in health care.

So, in one sentence: Amy Gleason is a former nurse turned health-tech executive and patient-data advocate whose personal experience with her daughter’s rare disease pushed her toward interoperability, federal digital services, and now AI-driven health-system modernization.

AMA's New Digital Pathology Codes in "Category III": What If They're Not on the CLFS?

Key Lesson:

In 2026, AMA CPT began assigning whole-slide-imaging digital pathology codes to Category III, rather than the PLA pathway used for earlier WSI codes. That matters because PLA lab codes can move onto the Clinical Laboratory Fee Schedule and be priced by crosswalk or gapfill, while Category III codes are usually not nationally priced by CMS. 

If software-intensive WSI services instead fall into the Medicare RVU/practice-expense system, they face a known trap: CMS may count only technician time and tiny amortized equipment costs, while rejecting per-use software fees. CPT 92229, autonomous retinal imaging, is the warning case.  In the past, CMS has elected to leave such codes "carrier priced" rather than underprice them by 80% - but this does nothing to fix the underlying problems for software-intensive services.

####

####

How You Get CMS to Liberalize a Burdensome Rule (Case Study: Physical Therapy Orders)

Why You May Care...

 Many stakeholders approach CMS each year with a request to liberalize one or another rule - be it an administrative policy, an NCD, or some other problem.   While the example below is not related to molecular laboratories (my main readers), the example DOES show how multiple stakeholders coordinated - and over several years - to get  a meaningful policy change at CMS.

What's It About? Physical Therapy?

Here's the status quo before 2025.  A physician sees a patient who needs physical therapy.   The physician writes a REFERRAL or ORDER for a physical therapy evaluation.  The physical therapist spends 30 to 60 minues with the patient, doing careful tests and examples, and develops a plan of care.  The PT transmits the plan of care to the physician to countersign.   (This countersignature must be re-affirmed every 90 days).

Here's what people got.   Beginning in January 2025, the physician issues a ORDER or REFERRAL for physical therapy.  The physical therapist does the exam and prepared the report and plan of care.   It's transmitted back to the referring MD.  NEW RULE:  As long as the physician doesn't reach out and tell the P.T. there is a problem, the P.T. can ASSUME the plan of care is OK with the physician.

While that may sound like  a small change, I suspect PT stakeholders were very happy to get it.

Below I give a detailed Chat GPT summary of the rulemaking, requests, and the rationales discussed.  

Find the rulemaking at 89 Fed Reg 97710, 12/9/2024, specifically 97912ff (about eight pages).

 Take Home Lessons:  How It Worked [Chat GPT]

CMS Releases Data for "87798" Billing in 2024

On May 21, 2026, CMS released data by lab and by CPT code (also by physician and CPT code) for Part B payments in CY2024.

There's been a lot of discussion of changing payment policies for code 87798 in the past six months; 87798 can be use as a "probe" of the data set.

Find more about the data set here:

https://www.discoveriesinhealthpolicy.com/2026/06/cms-releases-rich-cloud-database-for.html

87798

First, we searched the dataset for all providers billing 87798.  There were about 1400 total providers of 87798 services.  Payments per provider ranged from $24M down to $109.  Total payments for 87798 in all 50 states were $436,743,071.  $436M.  Of that $436M, $139M or 32% went to TX and $36M to PA, $34M to FL, $23M to NJ.  This shows strong representation of Novitas and FCSO states for 87798 payments in CY2024.

Among all codes, 87798 was $436M, 81528 Exact Cologard was $306M, and 81479 was $572M (prior blog).

Services per bene - about 20 labs billed >30 services per year per bene, including 3 that billed >60 services per year per bene (87798).

The top providers of 87798 received $24M, $21M, $12M and $8M.  

Note that this is only about 16% of all 87798 payments.  Services per patient per year ran from 9 to 20.  The four labs were in TX (2 labs) and PA and FL.  

Top Four Labs Billing 87798: Their Overall Billings

We used these four labs to go back to the database and get all the CMS payments to these labs.  

The four labs are the "new universe" of claims.  The four labs were paid $131M in total, and got half of their Part B revenue from 87798 ($65M).

After getting 50% of revenue from 87798, they got 12% for candida amp probe (87481), 4% for 3-5 Resp Virus panel (87631), and about 4% each for Strep B Amp Probe (87653) and MultiOrg Amp Probe (87801).  These five tests brought in about 74% of all the annual revenue of the 4 labs.

Special Mention to Moldx Labs

Although it's not a big lab, I feel like honorable mention should go to MolDx Labs, in North Hollywood CA, which was paid $2.7M in 2024 by Noridian and MolDx, 58% for large panels of code 87798.   

Look up the NPI - from March 2024 - it's really "MolDx Labs." 

Predicting the Scientific Future: From UCSF's Robert Wachter

Robert Wachter of UCSF is both a medical leader and a futurist. In his newest book, A GIANT LEAP, he discusses AI in medical care.  From his early AI experiences in 2022, he extracted the following.

The story of whether AI would finally transform healthcare would mostly be about whether the healthcare system could implement these tools in ways that would produce better outcomes for patients, lower costs, and some relief for beleaguered doctors and nurses. 

And that, in turn, would be determined -- as much by history, politics, economics, pride, regulations, leadership, lawsuits, guilds, culture, workflows, inertia, greed, hubris, vibes, and zeitgeist as by graphics processing units, diffusion models, and neural networks.

Wachter, A Giant Leap, 2/2026, Preface.

CMS Releases Rich Cloud Database for CY2024 Claims, Medicare Part B

 If you love CMS Part B data, Christmas comes every May or June, when CMS releases extensive cloud data for all labs and all  physician providers of every CPT code.   

CMS classes this as;

Data.cms.gov

>> Provider Summary by Type of Service

>> Medicare Physician and Other Practitioners [incl labs]

Find it here:

https://data.cms.gov/provider-summary-by-type-of-service/medicare-physician-other-practitioners/medicare-physician-other-practitioners-by-provider-and-service

The 2024 data set was released on May 21, 2026.  The same source has year-by-year back files to 2013.  the 2024 data has 9,781,673 rows.   You use it by filtering - for example, every lab that got paid $1 or more for code 81479 (filter on HCPCS = 81479).

For my blog on microbiology, 87798, here.

##

I filtered for codes ending in M or U, plus 811, 812, 813, 814, 815.   This misses the molecular microbiology codes, which are in different ranges up in the 87000-87999 range.  

The codes I did filter - without 877 microbiology - were paid $2,813,342,286, meaning, amost 3 billion dollars.

This produces lines as "lab x code" so, for example, 81479 is split over many lines, many labs.

The top code was 81528, Exact Sciences, $306M, or 11% of all molecular payments.  Next ccame Natera, CareDx, and Caris, all for code 81479, respectively for $104M, $102M, $100M.

The top 10 codes were paid $1.2B or 41% of all molecular payments. About 30% of the top 20 lines were 81479 payments. 81479 providers were paid $572M, 95% to the top 12 billers of 81479. 81479 used only in MolDx states. Below: Click to enlarge.

click to enlarge

##

I've heard there was a recent boom in uncontrolled (no LCD) payments for 81419, epilepsy panel.  I used Ctl-Alt-L to turn the highlighted "CODES" column into a drop-down-box selection column and checked only 81419.   Let's remember, just because codes like 81419 epilepsy, 81443 expanded carrier panel (e.g. cystic fibrosis), or 81440 mitochondrial genes, rapidly rose greatly in utilization and primarily in TX and FL, where Novitas did not have controlling edits, does NOT mean that anything inappropriate was occurring.   

Top payments for 81419 epilepsy panel were $73M with with half of national payments going to the top 6 epilepsy panel labs.

Note the states; TX, FL, NJ, PA, TX, FL, FX, FL.  

Nearly all labs getting paid for Epilepsy Gene Panel 81419 in Medicare, were under the Novitas and FCSO MACs, which had in recent years paid around a billion dollars for 81408 (Tier 2 code) and adjacent codes.  That's a code not covered by any other MACs, and one whose payments were apparently finally stopped by OIG.

While payments stopped for 81408 around 2023, by 2024 labs provided market access for genetics to patients under a different uncontrolled code [no LCD], 81419, literally just ten digits away in the code book.   

click to enlarge

I took the top 4 labs for 81419 (as above - FL, FL, PA, TX) and pulled all of the billing for these four NPI's for 2024.   The four labs were paid $27M for 81419 and $118M for all genetic codes.  Other leading codes from these four labs were 81440 (mitochondrial genes), 81443 (inherited conditions aka expanded carrier panel, CF etc), 81162 (BRCA).

At a glance, these genes seem medically unrelated.  One lab billed the same number of patients for 81443 and 81448, and the same number for 81181, 81183, 81343.  

###
AI CORNER

###

I gave the data to both Claude Opus and Chat GPT.  Claude noted that many of the labs - about a dozen out of 48 nominally independent labs  - billed the exact same charge, to the penny for 81419.   That's a clever observation and one I never have looked for.   Of course, that could happen for many reasons and I do not view it as suspicious.  Claude also wrote, "A single elderly Medicare beneficiary rwho medically requires an epilepsy panel + a 100-gene mitochondrial-disease panel + a prenatal-style carrier screen is not a workup that exists in nature."   Claude also asserted that many of the single-gene codes were already inside 81443, something that CMS edits should have known.   (I have not manually checked that.)

Of the four labs we checked based on high 81419 billing, one billed only 3 CPT codes, one billed 18, and the other two (both in the same city) billed about 65 codes each.  

Chat GPT studied the Excel's and opined, "The data do not look like an organic epilepsy-testing market. They look like a rapid, concentrated exploitation pattern around a high-paying, apparently weakly edited genetic CPT code, with the strongest signals in Florida/Texas, and especially around a small number of labs whose broader code portfolios show high-volume use of many unrelated genetic codes, not a coherent epilepsy-testing service line."  It tallied 7 of 48 labs billing 81419, as all being in Deerfield Beach FL.  Labs billing the same charge were as little as 1000 feet apart.

A New Commercial Source for Rapid Complete Medicare Claims Research LUMA CLAIMS

CMS has some free databases for Part B claims - like this one for annual Excel spreadsheets and this one for a cloud database x CPT code x Provider Name.  And CMS sells anonymous data files of claims data, too.

Now there's a new vender in town.  

I'll copy below an email I got today about LUMA CLAIMS dot COM.   It includes nearly real-time CMS claims processing data (through February 28, about four months ago).  Wow.    Pricing info was not readily available.

###

Continuing our partnership with medicareclaimreports.com (now https://lumaclaims.com/), Parrish Law Office is pleased to announce that initial claim approval/denial information for ALL Medicare Part B CPT codes is now available from January 2020 through February 28, 2026. Using a new interactive interface, custom reports examining approval differences by Medicare Administrative Contractor, provider, and State, and time based analyses can be created. Charts, graphs, and raw data of the reports can be downloaded. The data is particularly useful in identifying approval/denial trends and issues, and where Medicare has not issued a local or national coverage determination. 

We use these reports to show that Original Medicare has been covering something when a denial asserts that the item is experimental/investigational.

If this information would be useful in your business, please contact Parrish Law Office for more information or click on this link: www.lumaclaims.com

Debra M. Parrish, Bridget Noonan

Parrish Law Offices     www.dparrishlaw.com

###

https://lumaclaims.com/