Software as a Medical Device (SaMD) developers working with artificial intelligence and machine learning components face a uniquely compressed regulatory timeline. The FDA’s Pre-Submission (Q-Sub) program remains the single most powerful tool for de-risking an AI/ML SaMD filing, yet most sponsors underuse it, treating it as a courtesy meeting instead of a strategic alignment exercise with their ISO 13485 design control system. This 2025 framework shows how to turn Q-Sub feedback into binding design inputs rather than informal guidance.
Why AI/ML SaMD Q-Subs Have Become Higher Stakes
The FDA’s Total Product Life Cycle (TPLC) approach, combined with the Predetermined Change Control Plan (PCCP) guidance finalized for AI-enabled devices, has shifted the burden of evidence earlier in the development cycle. Reviewers now expect sponsors to demonstrate algorithm transparency, training data governance, and post-deployment monitoring plans long before a 510(k) or De Novo is submitted. A poorly scoped Q-Sub can leave these expectations unaddressed, resulting in Additional Information (AI) requests that add six to twelve months to a clearance timeline.
For AI/ML SaMD specifically, the most common costly delays trace back to three predictable gaps:
- Unclear algorithm change protocols that reviewers interpret as uncontrolled
- Insufficient rationale for the patient stratification that drives the Indications for Use statement
- Mismatched terminology between the Q-Sub package and the sponsor’s Design History File (DHF)
Each of these gaps can be eliminated before the meeting if the Q-Sub is treated as a design control deliverable rather than a regulatory checkbox.
Treating the Q-Sub Package as a Design Control Output
The single biggest mindset shift that prevents delays is treating the Pre-Sub meeting package as a formal design control output under ISO 13485 Clause 7.3. When a sponsor’s Design Controls map clearly shows the Q-Sub feedback as a Design Input, the meeting minutes become a Design Output, and the resulting FDA responses become Design Verification artifacts, the entire submission becomes auditable and reviewer-friendly.
Mapping Q-Sub Elements to ISO 13485 Design Controls
Most sponsors produce a Q-Sub package in isolation from their quality system. A more defensible approach is to create a traceability matrix linking each Pre-Sub question to specific Design Control records. The meeting request should reference the relevant Design Inputs, the meeting itself should validate those inputs, and the FDA’s written feedback should be filed alongside corresponding Design Review records.
For AI/ML SaMD, this traceability is especially valuable for the model’s intended use, performance targets, and training/validation data sources — three areas where reviewers consistently probe for rigor.
Pre-Meeting Preparation: The 90-Day Workback
Q-Sub meetings are scheduled approximately 75 to 90 days after FDA receives the package. The compliance strategist’s view is that the real work happens in the 90 days before submission, not the 75 days of waiting. A disciplined workback looks like this:
- Day -90: Lock the specific questions. Limit to four focused, decision-seeking questions. Reviewers disengage from open-ended or exploratory prompts.
- Day -75: Finalize supporting data summaries, even if preliminary. The FDA expects evidence, not aspirations.
- Day -60: Conduct an internal mock Q-Sub with cross-functional reviewers including data science, clinical, and regulatory affairs.
- Day -45: Run the package through an ISO 13485 design review checklist. Confirm terminology consistency with the DHF.
- Day -30: Final quality system sign-off and submission.
The “Four-Question Rule” for AI/ML Q-Subs
Reviewers allocate roughly 60 minutes of substantive discussion per Pre-Sub meeting, covering perhaps 40 minutes of questions in detail. Sponsors who arrive with eight or ten questions find that only two or three receive meaningful feedback. The remaining questions get deferred to the formal submission, eliminating the strategic value of the meeting entirely.
For AI/ML SaMD, the most productive question set typically includes:
- A question about the regulatory pathway (510(k), De Novo, or PMA) tied to a specific predicate comparison if applicable
- A question about the algorithm change protocol or PCCP scope
- A question about the pivotal study or clinical evaluation approach, including whether real-world performance data is acceptable
- A question about labeling expectations, particularly for adaptive algorithms
During the Meeting: Interpreting Real-Time Feedback
FDA reviewers are trained to give directional, non-binding feedback during Pre-Sub meetings. The compliance strategist’s skill is recognizing when an offhand comment reveals a reviewer’s concern that has not yet been formalized. Three signals to watch for:
- A reviewer asks the same question twice. This indicates an unresolved concern that will likely surface as an AI request.
- A reviewer uses phrases like “we would expect” or “we typically see.” These are soft requirements waiting to harden into formal requests.
- A reviewer pushes back on a specific dataset. This signals where additional pre-clinical or clinical evidence will be demanded.
Immediately after the meeting, file the written feedback as a Design Review record. Too many sponsors leave this documentation until after the next regulatory milestone, by which point the institutional memory has faded and the rationale for design decisions has been lost.
Post-Meeting Integration: Closing the ISO 13485 Loop
The FDA’s written responses arrive within 30 days of the meeting. These responses must trigger a controlled cascade within the quality system:
- Open a Design Change Record for any FDA-required modification to algorithm scope, performance target, or labeling.
- Update the Risk Management File (ISO 14971) to reflect any new hazards introduced by FDA-requested changes.
- Re-validate affected V&V activities if the Indications for Use shifts in response to FDA feedback.
- Amend the Software Development Plan if the feedback alters the lifecycle model or verification approach.
A Common Pitfall: Treating Pre-Sub Feedback as Optional
Sponsors sometimes treat Q-Sub feedback as advisory, especially when the feedback complicates their planned submission timeline. This is the costliest mistake in AI/ML SaMD regulation. When the formal submission lands and ignores previously documented FDA feedback, the AI request that follows often quotes the original Q-Sub response line by line, dramatically extending review cycles. Pre-Sub feedback is not formally binding, but ignoring it guarantees delay.
Aligning the PCCP Discussion with Design Controls
The Predetermined Change Control Plan framework is the most consequential 2025 development for AI/ML SaMD sponsors. A Q-Sub is the ideal venue for stress-testing a PCCP draft before committing to it. During the meeting, sponsors should walk reviewers through the planned modifications, performance evaluation methods for each modification, and the impact assessment methodology. Reviewers will indicate which modifications they consider “minor” and which they expect to see escalated for additional review.
The PCCP must also be cross-referenced into the Design Controls. Each planned modification becomes a Design Input. The protocols for evaluating that modification become Design Outputs. The validation that the modification maintains safety and effectiveness becomes Design Verification. This level of traceability is what differentiates sponsors who clear on the first cycle from those who face repeated AI requests.
Conclusion
FDA Pre-Submission meetings for AI/ML SaMD are most effective when treated as formal design control activities rather than informal regulatory touchpoints. Sponsors who build a 90-day workback, limit meetings to four focused questions, document feedback as Design Review records, and align their PCCP with ISO 13485 traceability consistently outperform peers on first-cycle clearance metrics. The framework is straightforward; the discipline required to execute it is what separates successful AI/ML SaMD programs from those mired in avoidable delays.
