Handling FDA Additional Information Requests (AI Requests) in 510(k) Submissions

## Introduction: Why AI Requests Can Make or Break Your 510(k) Timeline An Additional Information (AI) request is one of the most consequential events in a 510(k) review cycle. The FDA issues an AI request when the reviewing division determines that the submission, as filed, lacks sufficient information to complete its substantial equivalence determination — the core legal standard under 21 CFR 807.87 requiring that a predicate device (a legally marketed device to which your device is compared) be identified and that your device be shown not to raise new questions of safety and effectiveness. When an AI request arrives, the clock effectively resets on your review timeline, and how your team responds determines whether you reach clearance or face an unnecessary second request or outright Not Substantially Equivalent (NSE) decision. FDA's Standard Review Program goal is to issue a final decision within 90 FDA days. The AI request pauses that clock. Per FDA's published performance metrics under MDUFA V (the Medical Device User Fee Amendments reauthorized through FY2027), FDA suspends the review day count from the date the AI request is transmitted to the date FDA receives your complete response. Incomplete or poorly organized responses restart this sequence, and a second AI request — while not prohibited — signals serious substantive gaps that can damage the submission's prospects. In FY2025, FDA data showed that submissions requiring two or more AI cycles had an average total review time exceeding 200 FDA days, roughly double the program goal. Understanding the mechanics, content expectations, and response strategy for AI requests is therefore not a procedural nicety — it is a core competency for any regulatory affairs function supporting 510(k) submissions. This guide covers what triggers AI requests, how to interpret and triage them, how to construct a response that closes the review efficiently, and where companies most commonly lose time and credibility during this phase. --- ## What Triggers an AI Request: FDA's Review Logic FDA reviewers work from a structured checklist framework. For 510(k)s, the primary operative document is FDA's guidance **"Refuse to Accept Policy for 510(k)s"** (updated February 2019), which governs the initial acceptance review. However, AI requests occur after acceptance — they arise during substantive review when a reviewer identifies one or more of the following: * **Missing or ambiguous technological characteristics data.** Under 21 CFR 807.87(f), the submission must describe technological characteristics and compare them to the predicate. If that comparison is incomplete — for instance, if the submission says materials are "equivalent" without providing compositional data — an AI is likely. * **Performance testing gaps.** FDA's guidance **"Recommended Content and Format of Non-Clinical Bench Performance Testing Information in Premarket Submissions"** (September 2023) sets detailed expectations. Missing test protocols, incomplete statistical justifications, or unreported failure modes are common triggers. * **Software deficiencies.** For devices containing software, FDA's guidance **"Considerations for the Design, Development, and Analytical Validation of Software as a Medical Device (SaMD) Used in Artificial Intelligence and Machine Learning"** (2024) and the foundational **"Software as a Medical Device (SaMD): Clinical Evaluation"** guidance have raised reviewer expectations substantially. Missing Software Description documents, cybersecurity documentation consistent with FDA's **"Cybersecurity in Medical Devices: Quality System Considerations and Content of Premarket Submissions"** (September 2023), or inadequate SOUP (Software of Unknown Provenance) documentation are frequent AI drivers. * **Biocompatibility gaps.** If ISO 10993-1 biological evaluation reports are missing endpoint justifications or use outdated testing matrices, FDA reviewers will request the missing elements specifically citing **"Use of International Standard ISO 10993-1 Biological Evaluation of Medical Devices"** (September 2020). * **Labeling inconsistencies.** Indications for use that don't align precisely with the predicate's cleared indications, or instructions for use that reference off-label applications, routinely generate AI requests. FDA typically issues a single consolidated AI request per review cycle. Reviewers are instructed to gather all outstanding questions before transmitting, which means the document you receive often contains questions from multiple scientific disciplines — mechanical engineering, software, biocompatibility, and clinical — simultaneously. --- ## Anatomy of an AI Request: How to Read It Correctly An AI request is transmitted via the FDA's eSTAR system or, for older submissions, via written correspondence. It will contain: 1. **A cover letter** identifying the submission number (K-number), the assigned reviewer(s), and the response deadline. 2. **Numbered deficiencies**, each typically organized by topic area and citing the specific regulatory or guidance basis for the request. 3. **A response deadline**, historically 180 calendar days from the date of the AI request. Failure to respond within this window results in automatic withdrawal of the submission under 21 CFR 807.87(l). ### Parsing Each Deficiency Correctly Read each deficiency for three components: (1) what information FDA says is missing, (2) the regulatory or scientific basis cited, and (3) what format or level of detail the reviewer implies is expected. For example, a deficiency reading *"Please provide the full test report for electrical safety testing per IEC 60601-1, including raw data and pass/fail criteria"* tells you the reviewer has seen only a summary — they want the complete report with all annexes. Do not assume a deficiency is minor because it is brief. A one-sentence AI question about your predicate selection can require a 20-page comparative analysis if the reviewer is questioning whether substantial equivalence exists at all. ### Requesting a Clarification Call If any deficiency is ambiguous, FDA's **"FDA and Industry Procedures for Section 513(g) Requests for Information and Meetings"** framework and the **"Formal Meetings Between the FDA and Sponsors of Medical Devices"** guidance (April 2021) support requesting an informal clarification call. For 510(k)s specifically, you can contact the reviewing branch directly via the contact information on the AI letter and request a brief teleconference before drafting your response. This call is not recorded in the docket but can prevent you from producing an entire response section that misses the reviewer's actual concern. Use it. --- ## Structuring Your AI Response for Maximum Clarity The response document is not a letter — it is a structured technical document that will become part of your official submission record. Organize it as follows: ### Response Format Best Practices | Section | Content | Notes | |---|---|---| | Cover letter | Submission number, response date, summary of attachments | Keep to one page | | Deficiency index | Numbered table matching FDA's original question numbering | Helps reviewer navigate | | Individual responses | One section per deficiency, FDA question quoted verbatim, then your response | Never omit the quoted question | | Attachments | New data, test reports, revised labeling, updated software docs | Label each attachment to its deficiency number | **Quote every FDA question verbatim before your response.** This practice, while not formally required, is standard in high-functioning regulatory operations because it eliminates any ambiguity about which question you are answering and prevents reviewers from having to cross-reference the original AI letter constantly. **Be direct about what changed.** If you are providing a revised labeling document, state explicitly which sections changed, what they previously said, and what they say now. Track-changes versions of labeling are appropriate and appreciated. **Do not volunteer new indications or expand your device description** in an AI response. Doing so may require FDA to issue a new acceptance review or, in some cases, treat the response as a new submission. If new clinical data or a design change is genuinely necessary, discuss this with your reviewer before submitting. --- ## Timelines, Extensions, and Partial Responses ### Managing the 180-Day Response Window Under 21 CFR 807.87(l), you have 180 calendar days to respond to an AI request. FDA does not grant formal extensions of this deadline, though in practice, if you contact the reviewing division before the deadline and explain that a specific test is ongoing, reviewers may agree informally to await your complete response. Get any such agreement in writing (email from the reviewer is sufficient) before relying on it. ### Submitting a Partial Response Some companies submit partial responses to close out completed deficiencies while additional testing continues. FDA's **"The 510(k) Program: Evaluating Substantial Equivalence in Premarket Notifications [510(k)]"** guidance (July 2014) — still the primary operative guidance for 510(k) review standards — does not prohibit partial responses, but practice varies by division. Contact your reviewer before submitting partially; some divisions will not resume the review clock until all deficiencies are addressed. ### When a Second AI Request Arrives A second AI request almost always reflects one of two situations: your response introduced new ambiguities, or the reviewer's original question was broader than their first wording conveyed. Treat a second AI with heightened urgency. Request a clarification call immediately. Consider whether the submission strategy needs adjustment — for instance, whether the predicate selection requires revisiting or whether clinical data now needs to be added under FDA's **"Factors to Consider When Making Benefit-Risk Determinations in Medical Device Premarket Approval and De Novo Classifications"** framework. --- ## Common Pitfalls Medical Device Companies Encounter * **Responding to the literal question rather than the underlying concern.** An AI question about statistical sample size in a bench test often signals that the reviewer does not believe the test results are generalizable. Answering only with a statistical justification without addressing the test's clinical relevance leaves the underlying concern unresolved. * **Providing new data without updated summaries.** Submitting a 200-page test report in response to a biocompatibility question while leaving the original biocompatibility summary unchanged creates an internal inconsistency that triggers follow-up. * **Changing the predicate device in the AI response without explanation.** If your AI response switches to a different predicate, provide a full comparative analysis. Do not assume the reviewer will infer why the change was necessary. * **Missing the 180-day deadline.** This results in automatic withdrawal. If your submission is approaching the deadline and testing is incomplete, consider voluntarily withdrawing before the deadline expires — you can immediately refile without prejudice, preserving your ability to request an expedited review under FDA's Breakthrough Device Program if applicable. * **Submitting through the wrong eSTAR pathway.** AI responses must be submitted as supplements to the original submission, not as new 510(k)s. Errors in submission routing cause administrative delays of one to three weeks while FDA reroutes the document. --- ## Practical Recommendations for Your Regulatory Team 1. **Build an AI response SOP before you file your first 510(k).** Define internal roles: who triages the AI request, who owns each deficiency type (software engineer for software questions, testing lab coordinator for bench test gaps), and who signs the cover letter. 2. **Maintain a living data package throughout your review period.** Tests that are still running when you file should be tracked so you can provide results rapidly if an AI request arrives. 3. **Use pre-submission meetings proactively.** FDA's **"Requests for Feedback and Meetings for Medical Device Submissions: The Q-Submission Program"** guidance (February 2020) provides a formal mechanism to get written feedback on your testing plan, software approach, or predicate strategy before filing. A well-executed Q-Sub reduces AI probability significantly. 4. **Log every reviewer interaction.** Keep contemporaneous records of all phone calls and emails with FDA during the AI response phase. These records are valuable if a dispute arises about what was agreed. 5. **Hire a technical writer for complex AI responses.** A response involving 10 or more deficiencies across multiple disciplines benefits from a structured document review before submission. Clarity in presentation directly affects reviewer confidence. --- ## Key Takeaways * **An AI request pauses your review clock and resets your timeline** — treating it as a routine administrative step rather than a critical technical event is the single most common cause of preventable review delays. * **Read each deficiency for the underlying regulatory or scientific concern**, not just its literal wording; responding to the surface question without resolving the reviewer's actual concern almost always produces a second AI request. * **Quote every FDA question verbatim in your response**, organize attachments by deficiency number, and explicitly identify what changed in any revised documents — these structural practices materially reduce review friction. * **The 180-calendar-day response deadline is hard** — if you cannot respond completely in time, voluntary withdrawal and immediate refiling is preferable to an automatic withdrawal, which carries more administrative complexity. * **Pre-submission engagement via the Q-Submission Program is the most effective long-term strategy** for reducing AI request frequency; the investment of four to six weeks in a pre-sub pays dividends across every submission your program files thereafter.