Medical AI Trials Are Leaving Crucial Details Out

Good morning, HealthTech insiders. Good morning, Imagine losing your central vision, then being able to read a word again. Science Corporation says its retinal implant helped a patient with Stargardt disease do just that. It’s an early result from one patient, but one that gives us plenty to talk about.

We’re also looking at what medical AI trials aren’t telling us. A new BMJ review found missing details that make their results harder to judge. Meanwhile, consumer AI companies want access to medical records, raising a question that affects all of us: once we share our health information, how much control do we still have over it?

In today's Signal:

• Why good MedTech fails: I will be speaking at DTD2026 in Munich about the gap between regulatory approval and adoption.

• A retinal implant helps a patient read a word: Science Corporation reports an early milestone in inherited vision loss.

• What medical AI trials leave out: A BMJ review finds gaps in reporting model versions, methods and potential harms.

• Who gets access to your medical records? Consumer AI’s push for health data raises questions about patient choice, privacy and responsibility.

EVENTS & SPEAKING
1. I’m Speaking in Munich About Why Good MedTech Fails

The product works. The clinical evidence is there. The regulatory requirements have been met.

And yet, adoption still doesn’t follow.

On 27 October, I’ll be speaking at The European Conference on Digital Transformation & Data in Medical Technologies (DTD2026) about “Why Good MedTech Fails: The GTM Gap Between Regulatory Approval and Adoption.”

The details:

• Regulatory and technological success can only get you so far. Clinicians still have to use the product. Buyers still have to justify the investment. Reimbursement, where relevant, still has to make sense. And healthcare systems still have to integrate it.

• By launch, many decisions that shape adoption have already been made. Which clinical workflow are you entering? What evidence will procurement, payers and clinicians actually need? Is there a viable reimbursement pathway where one is needed? Who owns the budget? Does your health-economic case stand up to HTA and procurement scrutiny where these apply?

• The value proposition has to survive the buying process. Does it still hold when you move from the clinical champion to the person actually signing the contract? Understanding what each stakeholder needs to say yes is part of bringing the product to market.

The Signal: By launch, many of the decisions that shape adoption have already been made. Go-to-market planning starts with the workflow you’re entering, the evidence you’ll need, who will pay and what each stakeholder needs to say yes.

I’m looking forward to digging into all of this with people building, regulating and bringing medical technologies to market across Europe. Join us on 26-27 October at Novotel München Messe. Register for DTD2026 with code SPEAKER20 for 20% off.

NEUROTECH & VISION
2. A Retinal Implant Helped a Patient With Inherited Vision Loss Read a Word

Science Corporation says its first Stargardt disease patient using PRIMA recognised a letter during his first rehabilitation session and read a word during his second. Implanted in May at Sydney Eye Hospital, the device is now being tested against inherited vision loss after earlier studies in age-related macular degeneration.

The details:

• The patient received the implant in May as part of the PRIMAlia clinical trial at Sydney Eye Hospital in Australia. The company shared the update on October 13.

• PRIMA combines a tiny implant beneath the retina with camera-equipped glasses. The system converts visual information into electrical signals that stimulate surviving retinal cells.

• The update describes one patient’s early rehabilitation experience. The company has not published standardised visual-acuity results for this patient, and the report does not establish lasting benefit across the trial.

The Signal: Different retinal diseases can leave some of the same visual circuitry intact. Learning to use it could open a path to treating several conditions without fixing each genetic cause. And with PRIMA and GS030, the glasses help deliver the treatment, so better optics, software and training could matter as much as what goes into the eye.

AI & CLINICAL EVIDENCE
3. Medical AI Trials Are Leaving Crucial Details Out

A medical AI trial reports promising results. But which model did researchers test? How was it configured? And how carefully did they track harm?

A BMJ review published October 13 found that those details are often missing. Researchers examined 68 trial reports, 15 protocols and 378 registrations evaluating large language models in healthcare.

The details:

• Trial reports met, on average, 40%-59% of reporting requirements, depending on the guideline used.

• Missing information included model versions, how inputs and outputs were processed, and settings that influence the model’s responses. Planning, monitoring and reporting of harms were also frequently incomplete.

• Most published trials were small, single-site studies, primarily testing patient-facing tools for education and management recommendations. These findings concern how studies were reported, rather than proving that the tools were ineffective or unsafe.

The Signal: With medical AI, “clinically tested” needs a follow-up question: which version, under what conditions? If a hospital cannot match the product it is buying to the system researchers evaluated, promising trial results become harder to use. Clear reporting could become a commercial advantage for developers who can show exactly what their evidence supports.

HEALTH DATA & AI GOVERNANCE
4. Consumer AI’s Push for Medical Records Raises Questions About Access and Accountability

Consumer AI companies want easier access to the information that could make their health products more useful. An October 9 KFF Health News investigation describes industry engagement with federal officials over how those connections should work, and the safeguards that should follow the data.

The details:

• KFF reports that technology and health industry representatives have used a government-run chat room to engage officials on health-data policy. Requests for simpler or broader access are part of that debate; they should not be treated as an enacted right to unrestricted, permanent record access.

• TEFCA supports individual access services, with applicable identity, privacy and security requirements. A patient choosing an app as the destination for records is making a data-sharing decision; that connection does not establish the app’s clinical reliability.

• HIPAA responsibilities depend on the relationship. HHS distinguishes an independent app chosen by the patient from a service acting on a provider’s behalf. Its guidance on later use or disclosure is specific to HIPAA and does not settle every question of clinical or legal liability.

The Signal: More useful access depends on more than a successful data transfer. Patients need to understand who receives the information and how it may be used. Providers and app developers need to distinguish access rights, privacy obligations and the evidence behind any clinical advice.

👀 WHAT WE'RE WATCHING NEXT

1. The adoption gap: How MedTech teams make the case to clinical champions, procurement teams and budget holders before launch.

2. SURPASS and FDA feedback: Which proposals and comments offer workable ways to accelerate development while maintaining credible evaluation.

3. Patient-directed AI access: Whether policy changes make the recipient’s role, permissions and downstream data handling clearer to patients.

💰 FOLLOW THE MONEY

The deals:

• Confido Health ($12.5M Series A extension): Led by Z47, bringing total Series A funding to $22.5M, to expand CareOS and its AI agents for healthcare operations.

• Readily ($11M Series A): Led by Define Ventures, with Y Combinator and the California Health Care Foundation Innovation Fund participating, to expand healthcare compliance intelligence and enterprise oversight.

• Treso ($3.75M funding): Led by Redesign Health, with Locke Health and KCRise Fund participating, to develop financial software for physician practices ahead of its planned December launch.

The pattern: These rounds back the work around care delivery: coordinating practice operations, keeping up with regulatory obligations and managing finances. The commercial test is whether connected tools reduce administrative work and improve oversight enough to justify adoption.

📆 UPCOMING HEALTHTECH EVENTS

• DTD2026 (October 26-27, Novotel München Messe, Munich): Digital transformation and data in medical technologies. I’ll be speaking on October 27 about the GTM gap between regulatory approval and adoption. Use SPEAKER20 for 20% off registration.

• SURPASS Proposers Day (November 6, Washington, DC & virtual): ARPA-H’s trial-design and research-operations opportunity. Registration deadline: October 28.

• HLTH USA 2026 (November 15-18, The Venetian Expo, Las Vegas): Healthcare innovation, investment and payer/provider collaboration.

🛠️ Trending HealthTech AI Tools

• Cadence HypertensionOS: Clinician-supervised antihypertensive titration with protocol limits and deterministic safety checks; its specified intended use is under evaluation in FDA TEMPO.

• Limbic Unpacked: An AI voice application for structured CBT-based treatment within outpatient talking therapy; selected for TEMPO, with effectiveness for that use not yet evaluated by FDA.

• Dexcom Glucose Health Program: Metabolic monitoring and tailored guidance using integrated health metrics and AI insights; its TEMPO intended use remains under evaluation.

🛠️ Trending HealthTech AI Startups

• Confido Health: Healthcare operations AI company behind CareOS, with a new $12.5M extension bringing its total Series A to $22.5M.

• Readily: Healthcare compliance AI company with an $11M Series A, connecting regulatory requirements with policies, actions and supporting evidence.

• Treso: Financial technology startup for physician practices with $3.75M in funding; its initial launch with design-partner practices is planned for December.

DECISION INTELLIGENCE

Highlights: Deep Dives & Industry Reports

1. Direct AI Record Access: What TEFCA Changes, and What It Does Not.

2. Inside SURPASS: ARPA-H’s Bid to Redesign Clinical Trials.

3. How Should Generative AI Devices Be Evaluated? Inside the FDA Discussion Paper.

That's it for today!

A retinal implant helping someone read a word gives us a concrete glimpse of what medical technology can do. The BMJ review shows how missing details make medical AI harder to assess, while the medical-records debate raises questions about what happens to health information once patients share it.

For healthcare teams, the practical questions are straightforward. What difference will this make for a patient? Can we tell how the technology was tested? Do people understand where their data goes and who can use it? Clear answers give clinicians and patients a better basis for deciding what to trust.

See you tomorrow,

Dr. Dereck Mush, MD, MBA

The HealthTech Signal is independently funded. Sponsorships help us invest in better reporting, research, and a stronger newsletter for our readers. Want to reach HealthTech founders & executives while supporting the Signal? Get in touch.

Was this email forwarded to you? Sign up here.