Iterative AI needs deliberate change governance
Medical AI developers may want to refine a model, adapt an implementation, or improve performance after initial release. These changes can affect device behavior and the evidence supporting it. A predetermined change control plan, or PCCP, is not a general exemption from oversight. FDA describes it as a plan included in a marketing submission that addresses certain planned modifications, the methods used to develop and validate them, and an assessment of their impact.
Describe the boundary of the planned changes
A useful lifecycle plan distinguishes what may change from what remains fixed. Inputs, supported use, processing, output behavior, and dependencies can all matter. The FDA, Health Canada, and MHRA guiding principles emphasize focused, risk-based, evidence-based, transparent, lifecycle-oriented planning. The details are product-specific. Organizations should seek qualified regulatory advice rather than assuming that one approved plan can be copied across unrelated models, indications, or deployment configurations with different risks.
Connect development to controlled implementation
Verification, validation, release approval, and monitoring should form a coherent process. Record the data and configuration used to assess a modification, along with acceptance criteria and decisions. Define how an unsuccessful change is stopped or reverted where applicable. Hospitals and other buyers should ask how they will learn which version is operating and what has changed. Communication about device updates should be understandable without implying that users must reconstruct the manufacturer’s entire development process.
Evaluate network and software changes together
A model update is not the only modification that can matter. Runtime changes, input processing, connected interfaces, and hosting arrangements may affect the supported system. Review the applicable plan and obligations for the specific change rather than judging it by its marketing label. Future network readiness belongs in that controlled assessment. A credible lifecycle strategy supports useful improvements while preserving traceability and evidence, instead of treating continual software delivery as permission for unrestricted clinical experimentation.