Post-Clearance Reality: What AI Medical Devices Owe Investors

Getting an AI medical device from concept to a tool doctors actually use is a slog, and that’s true even for the ones that clear the FDA’s hurdles. If you’re on a regulatory strategy team or you review evidence for a payer, you have to get this. You have to see that a demanding clearance pathway is just the ticket to the game. It doesn’t mean the product will stay on the field or ever be commercially viable.

The Rigorous Pathway Completed by Pear Therapeutics

Look at Pear Therapeutics. They were a big name in digital therapeutics and took their Software as a Medical Device (SaMD) products through the wringer to get them cleared. Their platforms reSET and reSET-O, for substance use and opioid use disorder, got the FDA’s nod, which meant they’d passed muster on clinical validation and safety after extensive trials and paperwork. This set a high bar for other AI tools in mental health. A key part of the conversation then, as now, was the FDA’s Predetermined Change Control Plan (PCCP), which is the agency’s way of dealing with adaptive AI/ML devices that learn and change. FDA guidance on AI/ML medical device change control The whole point of the PCCP is to let these algorithms get better over time without forcing the company to go back to square one with a new premarket submission for every single model tweak, all while keeping the FDA in the loop.

What the PCCP Framework Commits a Cleared Product To

A PCCP commits a company to ongoing evidence generation and a controlled evolution of its product. It’s not a one-and-done checkbox. As Michael Pencina has argued, building durable evidence is a continuous job. So for a cleared AI/ML SaMD, the PCCP spells out exactly what kinds of changes you can make to the algorithm, what data you can use to make them, and what performance level you have to keep. This locks the product into a specific lifecycle: you monitor, you evaluate, you adapt within the agreed-upon guardrails. A big part of this is constantly watching for algorithmic drift, making sure the model doesn’t start failing just because the real-world patients it’s seeing now look different from the original training data. Michael Pencina’s work on evidence and evaluation in AI/ML For any regulatory team, the message is clear: FDA clearance is day one of a long-term job gathering evidence and managing quality. This is where your Quality Management System (QMS) and standards like ISO 13485 become your bible, because they provide the structure for documenting all your post-market surveillance and change control.

The Commercial Reality: When Clearance Isn’t Enough

Even after all that work, Pear Therapeutics went bankrupt. Their bankruptcy is a brutal lesson: getting regulatory clearance for a SaMD, even with a solid PCCP, doesn’t guarantee anyone will pay for it or that doctors will keep using it. The clearance pathway validates safety and efficacy at one moment in time, but that validation has a short shelf life in the real world. This forces a distinction that regulatory teams and payer reviewers need to internalize, the difference between getting a product cleared and building the durable evidence that convinces the market to adopt and pay for it. And Pear isn’t alone. Other high-profile AI health companies that couldn’t answer the evidence question in the market also folded. Forward, which wanted to upend primary care with its AI model, shut down all its locations in November 2024. Olive AI, which was automating administrative tasks in hospitals, went under in October 2023, its good parts sold for scrap. What these failures show is that getting from clearance to being a normal part of clinical practice requires proving value in messy real-world settings, getting payers to reimburse, and actually fitting into a doctor’s chaotic workflow, all of which go far beyond the controlled environment of a clinical trial.

Clearance and Durable Evidence: Two Distinct Projects

The takeaway from these flameouts isn’t that the regulatory process is broken. It’s that getting clearance and building a layer of durable evidence are two completely different jobs. Regulatory clearance, a 510(k), De Novo, or Breakthrough designation, proves a device is safe and effective at the moment of submission. It’s the starting gun, an official signal that the product isn’t snake oil. But durable evidence, especially for an AI that’s constantly changing, demands continuous generation of real-world evidence (RWE). This is the messy data from EHRs, patient registries, and insurance claims that shows if the tool actually helps patients, saves money, or improves outcomes across different hospitals and populations, long after the initial trial ends. Payer guidelines on real-world evidence for digital health So if you’re on a regulatory or payer review team, this split is everything. Clearance gets you in the door. Durable evidence is what keeps you in the building, convinces doctors to use your product, and gets payers to write the checks. It has to answer the tough questions about long-term performance and economic value. When you’re looking at an AI health solution, you have to evaluate both the rigor of its initial clearance and the seriousness of its plan for generating post-market evidence. That second project, tracking the ongoing proof of value and safety long after the FDA’s letter arrives, is exactly what we follow here at AI Health Risk Monitor.