Lessons learned from multiple AI/ ML 510(k) submissions, and feedback from the FDA. How a small team can build a framework to create focused lean submissions with data risk driven decision making.

PCCP - We have tasked the engineers to be buidling with a long term view, with that have drawn up the PCCP. The strategy we have used is to have a 510(k) for the initial submission that is based on the MVP, but to then have a fast follow PCCP to cover all the proposed changes that we think will come about in the coming years. This can included intergrations, web portals, additional inputs, we have tried to strike a balance between freedome to innovate and regulatory constraints. With this in mind we can update on a clear planned schedule much faster and withiout the need for new 510(k) sumissions. 

Risk analysis - we score risk analysis in the usual manner with P1 x P2 x severity, however we have added scoring for detectability and fixability. This then give the engineers a clear motivation to create internal alerts and monitoring to ensure that any issues are highligted to appropriate teams imminently and possible patient exposure is limited. We have additionally used updated P2 scoring to be more specific on possible severity in line with Bijan Elhai's book Safety risk management for medical devices. This approach has enabled use to use risk to drive future developments to lighten the regulatory load and free engineers time.

Automations - we have internally created risk management tools to automatically scan all SOUP in our devices assigned it a software risk score and then assess the possible implications of SOUP failure and map these to exisiting risk scoring. This in combination with our updated risk approach has driven areas of focus and improvement, with clear visibility across teams. Whilst this is not AI specific it has provided an additional layer of visibility and security within our product.