Summary
In Part II of this series, Leon shifts the focus from initial regulatory positioning to one of the most underestimated and often misused steps in the FDA journey: the pre-submission (Pre-Sub). While many teams treat it as a procedural checkbox, this article makes a compelling case for why it should instead be approached as a strategic inflexion point in AI/ML device development.
The blog opens with a grounded observation: by the time teams reach the Pre-Sub stage, many of the most critical decisions: intended use, product code, and regulatory pathway, have already been made. Yet, it is precisely here that uncertainties around validation strategy begin to surface. For AI/ML-enabled devices, validation is not a one-size-fits-all exercise. It sits at the intersection of technical performance, clinical relevance, and regulatory expectations, and getting it wrong can significantly impact timelines, costs, and approval outcomes.
Leon introduces a key distinction in how AI/ML systems are evaluated: stand-alone performance evaluation (SAPE) versus validation approaches that require clinical investigations. Some device categories, particularly in areas like radiology, come with more defined expectations, including structured reader studies. Others rely on evolving or draft FDA guidance, making interpretation and justification more complex. The takeaway is clear: validation is not just a technical exercise; it is a regulatory strategy decision that must be aligned early and communicated effectively.
A particularly candid moment in the article stands out: Leon reflects on a past submission where this aspect was underestimated. That experience becomes the anchor for one of the blog’s central messages: if there is one place where teams lose time, it is in misjudging validation expectations. This is where the Pre-Sub becomes critical; not as a formality, but as a mechanism to de-risk assumptions before they become costly mistakes.
The article also highlights the expanding landscape of AI/ML-enabled medical devices, spanning use cases from detection and diagnosis to prediction and treatment support. With most falling under Class II classifications, the variability lies not in the category itself, but in how performance is demonstrated and justified. This variability reinforces the need for structured dialogue with the FDA, especially when guidance is incomplete or open to interpretation.
What emerges throughout the piece is a consistent theme: clarity beats completeness. A well-prepared Pre-Sub is not about presenting a finished story; it is about asking the right questions, framing the problem correctly, and aligning early with FDA expectations on validation design. Done right, it reduces rework, shortens timelines, and strengthens the overall submission strategy.
This summary captures the essence, but not the depth, of Leon’s experience-driven insights. To fully understand how to structure your Pre-Sub, avoid common pitfalls, and define a validation strategy that stands up to FDA scrutiny, read the full article here: AI/ML Systems, preparing an FDA submission Part II: ‘Pre-sub strategy’
