In the previous articles of our Beyond FDA series, we explored ten realities that many Healthcare AI companies discover too late: FDA clearance does not guarantee market adoption. Great products fail without a reimbursement strategy. AI accuracy alone is no longer enough. Clinical validation is what earns trust. Trust itself comes from Key Opinion Leaders, not marketing. Clearance, validation, and adoption are three separate pillars that must be earned independently. The best algorithm still loses to the best-integrated one. CIOs buy risk reduction before they buy innovation. Even a validated, cleared tool can stall waiting for PACS and EHR governance approval. ACR alignment is becoming a competitive advantage in radiology AI. And most vendors still confuse detection accuracy with genuine clinical workflow fit. This brings us to a regulatory shift that will reshape how most of the above even gets evaluated going forward: The FDA is rewriting how it clears AI that keeps changing after clearance.
(Beyond FDA Series, Part 11)
The Problem the Old System Was Never Built For
Traditional FDA clearance was designed around a simple assumption: the device you validate is the device you ship, permanently. Any meaningful change requires a new submission.
AI doesn't work that way. Models get retrained. Performance improves as more data accumulates. Vendors want to update algorithms the same way they'd push a software update: but under the traditional 510(k) framework, even a modest retraining could trigger a brand-new clearance process.
That mismatch has been one of the quiet, unresolved tensions in AI regulation for years. The Predetermined Change Control Plan (PCCP) is the FDA's answer to it.
What a PCCP Actually Does
A PCCP lets a manufacturer pre-specify, at the time of original clearance, exactly what future changes to the algorithm are anticipated, and exactly how those changes will be validated, so that certain updates can happen without a brand-new submission each time.
✔ Defines the change boundaries upfront: what the model is allowed to learn or update, and what it isn't
✔ Pre-commits the validation method: how each anticipated change will be tested before it's deployed
✔ Builds in transparency: a documented, auditable trail of what changed, when, and why
✔ Enables continuous improvement without repeated full submissions: as long as changes stay within the pre-approved plan
This is a fundamentally different regulatory posture: instead of regulating a fixed product, the FDA is regulating a bounded process of change.
Why This Is a Bigger Deal Than It Sounds
For AI vendors, a well-constructed PCCP can mean the difference between a model that's frozen in time at the moment of clearance, and one that keeps improving as real-world data accumulates, without months of resubmission every time.
For hospitals and CIOs, it raises a new diligence question: does this vendor have a credible plan for how their AI is allowed to change, or is "continuous learning" just a marketing phrase with no governance behind it?
For the industry more broadly, this shifts competitive advantage toward companies that treat their change-control plan as seriously as their original clinical validation: because increasingly, that's what regulators, hospitals, and IT review boards will be evaluating.
What Founders Get Wrong About PCCPs
-
Treating the PCCP as a formality to satisfy at submission, rather than a real operational commitment the company has to live inside afterward
-
Writing change boundaries so vague they don't survive FDA review, or so narrow they don't provide meaningful flexibility
-
Underestimating the monitoring infrastructure required to prove, after each change, that performance stayed within the pre-specified bounds
A PCCP is only as good as the governance and monitoring systems built to actually execute it.
Final Thought
510(k) clearance answered whether a static device was safe and effective at one point in time.
PCCP answers whether a company can be trusted to keep changing its AI safely, indefinitely.
510(k) regulated a product.
PCCP regulates a process.
As more AI vendors adopt this pathway, the differentiator won't be who has the best model at launch: it'll be who has the most credible, well-governed plan for what that model becomes next.
Next in the Beyond FDA Series (Finale):
The Four Pillars Of Healthcare AI Adoption
#HealthcareAI #PCCP #FDARegulation #SaMD #RegulatoryStrategy #MedicalImaging #DigitalHealth #HealthcareInnovation #ArtificialIntelligence #MedTech #RadiologyAI #QscriptionTechnologies