Using AI to Aid in Device Development and Regulatory Submissions

By Doug Mead, Co-Founder, Rhizome AI and President & Principal, CP Pathways, LLC

Recently, I had the pleasure of presenting at the Pharma Ed Combination Products Summit 2025 in Philadelphia on a topic that's become increasingly important in our industry: leveraging artificial intelligence to streamline device development and regulatory submissions. As regulatory professionals, we're constantly seeking ways to make the submission process more efficient while reducing risk. In this short synopsis, I'd like to share some insights from that presentation.

The Regulatory Research Challenge

Let's start with a fundamental question: What are we trying to accomplish with regulatory research? At its core, our objectives are to:

·         Understand FDA's and EMA's specific expectations for product testing and submission content

·         Assess how similar products were developed, as documented in FDA reviews

·         Consider regulatory precedents when developing submission strategies

·         Mitigate the regulatory risks in our development plans

·         Keep our product teams and management informed about evolving regulatory expectations

The challenge we face is multifaceted. While FDA guidances and standards provide valuable advice, they rarely address specifics related to unique product designs, indications, or novel testing approaches. The context we need often comes from FDA advice posted in Type B/C/D meeting minutes or FDA reviewers’ information requests that go well beyond published guidances.

The Goldmine in Drugs@FDA

The Drugs@FDA database contains a wealth of information that can provide crucial context beyond what's found in guidances. This database includes:

·         Delivery device performance testing that was advised, performed, submitted, questioned by FDA, and ultimately justified

·         Details on EPRs, ISO preconditioning, functional stability, EMI, software validation, and more Process validation runs and control strategies that passed FDA scrutiny

·         Human Factors Validation Study results, including use errors reported, FDA concerns, and mandated labeling revisions

·         Information Requests during reviews and Complete Response Letters

·         Part 4 cGMP summary deficiencies and CDRH Pre-approval Inspection information

The problem? This database is massive, unstructured, and can be extraordinarily time-consuming to search effectively.

Current Search Options and Their Limitations

When it comes to navigating FDA's database, we have several options, each with significant or some limitations:

·         FDA Drugs@FDA Search: Limited capabilities that only allow searching one database at a time via product name or NDA/BLA number. Extremely time-consuming.

·         Google: Limited search functionality that doesn't work for EMA EPARs. It's difficult to know where to look inside the PDFs, making this approach time-intensive.

·         General AI tools (Copilot, ChatGPT, etc.): These provide well-worded responses but are often factually incorrect and difficult to verify.

·         Specialized startup offerings: These can work extremely well when properly designed for regulatory affairs, but can also miss the mark by being poorly validated, overly complex, or prohibitively expensive.

Designing an AI Tool for Regulatory Intelligence

At Rhizome AI, we approached this challenge by establishing clear "design inputs" for an AI search tool specifically tailored to regulatory professionals:

·         Properly indexed and OCR'd data files organized in a dedicated cloud server with regular updates

·         "Fit for purpose" search algorithms customized and validated for regulatory content

·         Ability to search all databases simultaneously or select specific datasets

·         Responses using Large Language Models customized to address regulatory questions with precision

·         Robust security features including authentication, isolated usage per user, and full encryption

·         No shared machine learning that would benefit other users

·         Citations that link directly to pages in files with relevant text

·         Copy/paste functionality for easy reporting to teams and management

Our development process involved making thousands of queries and checking for relevance and accuracy with regulatory experts. We maintain a validation database with scoring to assess the impacts of search algorithm changes.

Practical Applications

The applications for this technology are extensive. Regulatory professionals can quickly find information on:

·         Post-market commitments

·         Required stability data

·         Essential Performance Requirements

·         Bioequivalence standards

·         Clinical trial study endpoints

·         PAIs that led to 483s

·         Extractables and Leachables expectations

·         Dose bracketing strategies

·         Established Conditions

·         510(k) Indications for Use

·         Complete Response Letter content

·         Endotoxin requirements

During the presentation, I was joined by Rhizome AI Co-founder, Chetan Mishra, to perform an on-line demonstration of the search tool.  We picked a question about FDA’s reliability requirement for PFS needle safety devices. The search results found specific examples where the numeric requirements were stated by FDA in multiple review memos.   

Looking Ahead

As this technology continues to evolve, several questions emerge:

·         Will AI tools make device development and regulatory submissions easier?

·         Will they make submissions better and less risky?

·         Could AI programs eventually write test reports and regulatory submissions?

·         Will these tools make Regulatory teams obsolete?

My perspective is that while AI will transform how we approach regulatory intelligence, it won't replace the critical thinking and strategic judgment that experienced regulatory professionals bring to the table. Rather, it will allow us to focus our expertise on higher-value activities by reducing the time spent on information gathering and increasing the time available for analysis and strategy development.

The future of regulatory affairs isn't about replacing humans with AI, but about creating a powerful partnership where AI handles the labor-intensive research tasks while regulatory professionals leverage their expertise to make more informed decisions more quickly.

Doug Mead is Co-Founder of Rhizome AI and President & Principal of CP Pathways, LLC. For more information about Rhizome AI's solutions for regulatory intelligence, visit https://rhizomeai.com/.