Regulation — pharma-specific
FDA PCCP Draft Guidance — the envelope of allowed AI modification
The Predetermined Change Control Plan is FDA's answer to "how does an AI model change over its life without a new submission each time?" For sponsors, the PCCP is where "we monitor drift" becomes a specific + auditable commitment — and where silent drift outside the envelope becomes a live regulatory question.
What the guidance actually says
"A PCCP describes the planned modifications to the device that the manufacturer intends to implement … the associated methodology to develop, validate, and implement those modifications … and an assessment of the impact of those modifications."
FDA Guidance: Marketing Submission Recommendations for a PCCP for AI/ML-Enabled Device Software Functions (2023 draft, referenced in ongoing rulemaking)
"Modifications that are outside of the PCCP … would require a new marketing submission before implementation."
FDA PCCP Guidance — the envelope-breach trigger
What this means in plain English
Two obligations for any sponsor whose trial AI has a PCCP on file (or whose vendor claims one):
- Real-world model behavior must stay inside the pre-approved PCCP envelope. Distributional drift of the magnitude in this sample may fall outside.
- The sponsor must be able to detect + document envelope breaches. Silent breaches become a live compliance question the moment anyone asks.
Sponsors often assume "the vendor holds the PCCP." The sponsor's own reliance on the AI in a regulated trial is a separate matter — and BIMO inspects sponsors, not just device manufacturers.
What triggers the exposure in the sample
A 53 percentage-point exclude-lane rate shift on P3 is not standard PCCP-envelope drift. It is behavior the pre-market submission almost certainly did not contemplate at that magnitude for the elderly-multimorbid subgroup. The moment the sponsor knows — or should have known — a PCCP breach is a live question, the CAPA + notification timelines begin.
What the $499 Snapshot shows against this rule
- Timestamped drift-detection evidence — the "when did the sponsor know" record
- Per-patient-group detail sufficient to characterize likely envelope-breach magnitude
- Model-version pinning — the exact classifier build the drift attaches to
- Independent-verifier signature — supports the good-faith stance that the sponsor investigated
See the drift-detection chart →
$499 Snapshot. 3 business days.
Independent evidence of when the sponsor learned + what the pattern looked like. The timeline record that documents the good-faith investigation stance.
Buy $499
Snapshot credit applies to Baseline ($2,500) or Enterprise Attestation ($35-55K) upgrade within 30 days.