Digital health vendor evidence now decides who gets a pilot. In one September 2026 week, PHTI found virtual kidney care companies as a category produced almost no clinical evidence, ARPA-H paired heart failure AI agents with an external evaluator and an FDA deadline, and HealthEx opened a TEFCA route for records to reach Meta's Muse agent. Vendors should publish outcome measures, methodology and consent answers before asking for pilots.
Below: four dated signals, the questions they equip provider and payer buyers to ask, and six steps for pilots, proof content and pipeline. Figures come from primary sources or reputable health IT press.
The signals, dated
| Date | Signal | Why it matters to vendors |
|---|---|---|
| September 8, 2026 | HealthEx connects health records to Meta's Muse agent | Records retrieved through TEFCA can now reach a consumer AI agent at the patient's request |
| September 8, 2026 | Forus raises a $150 million Series C at a $3 billion valuation | Well-funded AI agent platforms for prescription access compete on network reach |
| September 9, 2026 | PHTI releases its virtual CKD management assessment | An independent evaluator judged eight companies as one category, largely on published evidence |
| September 9, 2026 | ARPA-H names ADVOCATE performer teams | Federal clinical AI funding comes with external evaluation and an FDA deadline |
PHTI: a category judged on its evidence
The Peterson Health Technology Institute released its assessment of virtual solutions for chronic kidney disease management on September 9, 2026. It evaluated eight companies as a single category, because they offer similar features and compete head to head for contracts. PHTI did not test the products itself; companies were invited to submit clinical, commercial and economic data. Its summary of findings says these solutions:
- Do not slow disease progression. PHTI found no consistent evidence of slower progression compared with usual care, either through better use of guideline-recommended medications or a slower decline in kidney function.
- Do not lower total healthcare spending. For a million-member Medicare Advantage plan, these companies are responsible for more than $5 billion in annual spending and reduce it by only 0.1%. The full report puts savings per attributed member with CKD at $401 in commercial plans, $53 in Traditional Medicare and $62 in Medicare Advantage.
- Improve transitions to dialysis for a small number of patients. Planned dialysis starts rise, but PHTI says only about 1 in 1,000 managed patients benefits.
PHTI's literature review examined more than 5,400 pieces of evidence and found 34 unique studies that met inclusion criteria, few of which evaluated the specific interventions of the companies assessed. Only three of the eight companies produced peer-reviewed evidence on clinical effectiveness that met inclusion criteria, and PHTI writes that the companies as a category "produced almost no clinical evidence." It also cites the CMS Kidney Care Choices evaluation: after program costs, the net impact of the CKCC option on Medicare was a statistically significant loss of $275.4 million in 2023.
PHTI's guidance points straight at contract design. It asks purchasers to direct more shared savings payments toward utilization-based cost reductions and to add patient-level milestone payments, such as confirmed eGFR and uACR testing with accurate staging, and verified initiation and continued use of guideline-recommended medications. It asks companies to invest in rigorous, comparative research.
ARPA-H: agents funded with an evaluator and an FDA deadline
On September 9, 2026, ARPA-H announced the teams receiving contract awards under ADVOCATE, its program to build an FDA-authorized clinical agentic AI system for heart failure care. The commitment is up to $33.7 million in the first year of the four-year, $62.7 million program. Its structure is a useful template for the questions buyers can ask of any clinical AI:
- A regulatory deadline. The patient-facing agent teams (Atman Health, Tempus AI and Updoc) must submit an FDA authorization package within 24 months of contract award.
- Independent evaluation. The Johns Hopkins University Applied Physics Laboratory is the external evaluation partner, assessing technical performance and clinical outcomes.
- Safety oversight. A Stanford University team will build a supervisory AI system that detects unsafe recommendations and out-of-distribution behavior.
- Real deployment. Duke University will test agents across five health systems on Epic and Cerner/Oracle EHRs. Kaiser Permanente will embed them in Epic workflows across 21 medical centers and more than 260 clinics, using shadow-mode deployments and pragmatic randomized clinical trials.
- Milestones with consequences. Healthcare Dive reported that total award amounts depend on meeting research milestones, and teams that miss them can be removed.
HealthEx and Muse: patient-directed data meets AI agents
On September 8, 2026, HealthEx announced that people can bring their health records to Muse, Meta's new personal AI agent. HealthEx is an Individual Access Services (IAS) provider under TEFCA. People verify their identity, choose what to share, see what they have shared and can disconnect at any time, and the agent checks in before it acts on their behalf.
The TEFCA Recognized Coordinating Entity's FAQ says organizations participating in TEFCA exchange must respond to IAS queries unless an exception applies, and that this includes both payers and providers. It says TEFCA privacy and security requirements generally apply to all participating entities, whether or not they are HIPAA covered entities or business associates, and that IAS providers must obtain express consent for how information is accessed, exchanged, used or disclosed.
Forus: agent platforms selling on reach
Also on September 8, Forus raised a $150 million Series C at a $3 billion valuation, led by Bain Capital Ventures. Its AI agents handle prior authorization, financial assistance and specialty pharmacy routing for prescriptions. The announcement leads with network reach: providers in all 50 states, patients in 85% of US residential ZIP codes, more than one-third of US rheumatology providers and 9 of the top 15 global biopharma companies. Smaller vendors competing for the same workflows will need proof that is just as specific and verifiable.
What provider and payer buyers will ask for
None of these sources is a procurement rule. Together they give a CFO, CMIO, payer medical director or privacy officer public precedent for sharper questions.
Evidence that matches the claim
PHTI judged a whole category largely on published, comparative evidence, and it notes that without company-specific evidence it is hard to tell whether one offering does better than the broader evidence base. Expect buyers to ask what the comparison group was, whether results are peer-reviewed or independently evaluated, how long patients were followed, and whether savings are absolute or measured against a growing benchmark.
Independent evaluation and safety oversight
For AI that touches clinical decisions, ADVOCATE gives buyers a federal model to point to: who evaluated the system besides the vendor, how unsafe outputs are caught, when human clinicians step in and what the regulatory pathway is. ARPA-H's description of Updoc's approach, a clinician-built rules system that validates every proposed action against approved protocols before execution, shows the level of detail buyers can compare against.
Milestone-based pilots and contracts
PHTI recommends patient-level milestone payments tied to actions that predict better outcomes, and ARPA-H ties funding to milestones. A pilot judged on whether the team liked the product will look weak next to one with defined milestones, a verification source and a decision date.
Data use, consent and action approval
Patient-directed routes such as HealthEx and Muse mean records can reach AI tools at the individual's request. Expect buyers to ask how consent is captured and revoked, how identity is verified, what data is used for, whether an agent ever acts without approval and how every action is logged.
Six steps for AI and digital health vendors
- Define outcomes firstSet outcome measures, comparison group and time frame before the pilot starts.
- Publish your methodsPut methodology and results pages online, limits included.
- Offer milestone pilotsTie pilot success to verified milestones and a decision date.
- Prepare data answersDocument consent, identity, data use, TEFCA and HIPAA positions.
- Map the committeeReach clinical, finance, IT, privacy and payer roles by name.
- Be answer-readyStructure evidence so AI search tools quote it accurately.
1. Define outcome measures before the pilot. Agree in writing on the primary outcome, the comparison group (usual care, matched controls or a pre-period), the data source and the follow-up period. Choose measures a buyer can verify in its own claims or EHR data. PHTI notes that interpreting eGFR slope requires two to three years of follow-up, which is why early, verifiable milestones matter for pilots that last months.
2. Publish methodology and results pages. Give each study or evaluation its own dated page: design, population, comparison, time period, who ran the analysis, peer-review or independent evaluation status, results and what did not improve. Link every outcome claim to these pages, give each a clear path to request the full report or a methodology call, and apply conversion rate optimization as you would to a demo page.
3. Offer milestone-based pilots. Write a pilot charter with a few milestones, how each will be verified, a decision date and what happens if a milestone is missed. Borrow PHTI's logic: reward early, verifiable actions that apply to many patients, not only late outcomes that affect a few. For payers, show how each pilot measure connects to utilization and total cost of care.
4. Prepare data governance, TEFCA and HIPAA answers. Write down, in plain language, whether you sign business associate agreements, whether you participate in TEFCA or rely on a partner that does, how you verify identity and capture and revoke consent, whether patient data is used for model training or other secondary purposes, whether any agent acts without clinician or patient approval and how actions are logged. This is not legal advice. Have counsel confirm how HIPAA, the TEFCA Common Agreement and information blocking rules apply to your product.
5. Target the right committee roles. Map the clinical sponsor (CMO, CMIO or service line chief), finance and value-based care leads who will test your savings method, quality and patient safety, IT and EHR integration owners for Epic or Oracle environments, and the CISO and privacy officer. On the payer side, add medical directors and network or contracting leads. Give each role the proof it needs, then run the program as account-based marketing against a named list. In our ITT Digital case study, a focused healthcare IT program produced 54 leads in a year at $250K to $300K ticket sizes. For HLTH USA (November 15 to 18, Las Vegas), book evidence briefings rather than booth demos; our Q4 2026 B2B event calendar has outreach timing.
6. Build answer-ready content for AI search. Committee members may ask AI assistants whether a vendor has peer-reviewed evidence, who evaluated it and how it handles consent. Put a short, dated, cited answer at the top of each evidence page, use question headings that match those queries and keep numbers identical everywhere they appear. Our guide to AEO for healthcare IT services covers trust pages and case study structure.
This is not legal advice
This post summarizes public reports, agency announcements, company releases and TEFCA guidance as of September 14, 2026. It is not legal or regulatory advice. Confirm FDA, HIPAA and TEFCA questions with qualified counsel.
Where Lemniscate fits
Lemniscate Growth builds revenue pipeline for health IT services firms, digital health companies and AI vendors selling to providers and payers. See our healthcare IT practice. Book a free growth audit and we will review your evidence pages, pilot offer and buying committee map, and show which accounts and roles to reach first.
