Regulatory Precedents For Drug Delivery: Uncovering Clues To Successful Submissions

By Tom von Gunden, Chief Editor, Drug Delivery Leader

While I wouldn’t call myself an AI “holdout,” I also wouldn’t (yet) call myself an AI “champion.” My hesitation, perhaps even skepticism, stems primarily from attempted applications of AI tools pointed at a very specific editorial purpose. That is, taking the recordings from my videocast interviews and from online panel discussions in our Drug Delivery Leader Live virtual event series and carving them up into relevant short clips to facilitate viewer convenience in consuming the content. Even with careful instructions about the desired format and focus of the clips, the various AI tools I’ve applied to this task have thus far failed to satisfy. So, I create my own segments.

That being said, I remain open to the power of AI tools for unearthing and organizing for presentation data gleaned from massive information inventories and widespread sources. Thankfully, others more patient and diligent than I are demonstrating impressive results when pointing such tools in ways and in contexts that have even skeptical old me taking notice.

Using GenAI For Precedent Research

One such advocate is combination products consultant Doug Mead. Doug recently reached out to me to make me aware of valuable insights he had been deriving from leveraging GenAI search tools to bolster regulatory intelligence. The bolstering comes by way of search results that surface previous decisions and comments made by regulatory authorities along the way of product approvals, challenges, or denials. As he demonstrates in his consulting practice, as well as encourages biopharma organizations to do themselves, Doug conducts this “precedent research” by framing queries aimed at regulatory agency databases, including, most prominently, those housed at FDA. Among the benefits of applying this approach during product development is avoiding pitfalls experienced by prior organizations navigating similar regulatory pathways with similar or related products.

After seeing some of the GenAI-generated results from Doug’s queries on drug development and delivery topics, I agreed that a helpful service to the industry would be some guidance and modeling on precedent research. To that end, I offered Drug Delivery Leader as a forum for that illumination. For his part, Doug is crafting articles about the approach, the first of which we have published under the title, “GenAI: The Muscle Behind Strong Regulatory Intelligence For Combination Products.”

For my part, I’m using my From The Editor platform to reiterate the importance of strengthening regulatory intelligence. For this installment, I caught up with Doug to further explore the significance of the concepts outlined in his aforementioned introductory article on precedent research.

Below are the video and transcript from my convo with Doug:

Conversation Transcript: Precedent Research With Doug Mead

Tom von Gunden, Chief Editor, Drug Delivery Leader:

Hi, my name is Tom von Gunden, Chief Editor at Drug Delivery Leader. And for today's conversation under the Regulatory Reconnaissance bucket, I am joined by Doug Mead, who is President and Principal Consultant at CP Pathways LLC, a combination products consultancy.

Welcome, Doug.

Doug Mead, President and Principal Consultant, CP Pathways LLC:

Thanks, Tom, and thanks for inviting me.

Yes, well, it's my pleasure to have you here.

 So, Doug, you penned an article for Drug Delivery Leader, which we recently published, and it's called, “GenAI: The Muscle Behind Strong Regulatory Intelligence For Combination Products.”

And for today's conversation, I'd like to talk a little bit more about some of the concepts in that article. It's focused around what you call precedent research, which means diving into regulatory databases to mine information about previous regulatory pathways that might help current organizations see what happened to their fellow submitters and previous experiences with regulatory agencies to determine things that they might learn from.

So, it might seem obvious about why you would want to do that, but what are your thoughts about the critical need to apply AI tools for doing that kind of precedent research?

Sure. Let me try to explain a little bit more about the motivation. What we find when FDA issues draft guidances or [final] guidances — or even ISO standards related to drug delivery devices and combination products — is they provide pretty explicit information about general expectations. But what's difficult is to apply them to specific products.

And what I'm finding in my consulting practice is that a client will be developing a drug delivery device that has unique attributes, or is novel in some way, and they can't really find specific expectations in the guidance documents or the standards.

So, what I try to do to help them is, I go into these FDA and EMA databases to see what other companies have done for similar products. And then, I'm looking for FDA policies that don't come out in guidances. And to look at how I would advise the client to develop either a regulatory strategy or to plan their submission content.

Gotcha, and your article goes into detail about the idea and the concept of applying GenAI tools to do this work. But, prior to the existence of those tools, how would people have conducted this kind of precedent research, if they even would have, and to what effect and with what limitations?

Yes, that's a good question. I've been doing this now for 5 or 6 years. And I think it's an important part of my practice to be able to base my recommendations on FDA precedence.

So, historically, I used a Google search, and I would put, in quotes, other reviews, which is the FDA memos coming from CDRH and DMEPA [Division of Medication Error Prevention and Analysis], the Human Factors reviewer at FDA, and then I would put in specific terms like autoinjector stability lots.

It would give me, in the first few pages of a Google search. a list of PDFs that I would then go into, and I would do a word search in the PDF to find the particular snippet of information that might be relevant or not.

And it took hours, Tom. It took hours. And I was never sure whether I was getting all of the information. And what's as important is, what I'm finding now is some of the type B and C meeting minutes coming from FDA that gives them even more information about what FDA expects for a particular product.

Gotcha. And we certainly encourage folks to read your article, which gives them an introductory look at how to start to apply some of these tools. But in addition to reading that article, what are some first steps, initial best practices you would recommend for folks who may not be thoroughly, or even really doing, this kind of Gen AI-based searching of regulatory databases. How do they get going?

Sure, I think it's pretty easy to get started. So, artificial intelligence regulatory search tools are out there. I mean, ChatGPT, Copilot, many of them. And then there are commercial organizations.

 So, what it takes is, you want to develop a skill set, and this is new to regulatory that might not have existed 2 or 3 years ago. What you have to do is be able to frame your question very specifically to get, with these tools, the information you want.

And they have large language models [LLMs] that are very smart. It's kind of surprising, it's scary as well. But, if you begin to ask some questions, you'll see right away where your questions need to be improved. And a lot of these tools will prompt you to ask a different question or ask you if you want your responses refined in any way.

What I found particularly helpful with these tools is to frame the issue first in a couple sentences and then ask the question that educates the large language model to look more specifically for the kind of information that you're really looking for. What you want to do, though, is to look at the results with a sort of a skeptical eye because there is slop, and there is hallucination out there still. It's getting better, much better.

But what you really want to do is go to the source document that they cited and then review that. That's the only way to be sure that the response you got from your question is actually true. And I do that routinely. Even though I think these tools can be 95% accurate and relevant, they're not perfect.

Sure. And then, maybe we'll finish with thinking even further down the road for folks.

So, those are some great tips for individuals who are going to apply these tools, perhaps for the first time. But do you have any suggestions for how to build this as more of an organizational practice or function? How should folks set up to do this on a consistent, regular basis?

Yeah, I think it comes in terms consistent with stages of development. If you're developing a drug delivery device, you want to know the expectations. So, the device team would want to know what's the end game? What do I need to do to get this product ready for submission and approval?

Then it goes to regulatory risk. Everyone is concerned about regulatory risks, what content to put into a submission, and what level of detail is expected. And I typically do searches to interrogate that, to see what other companies have submitted, the kind of information requests that FDA have asked.

And with that information, you can help a client, or I help my clients, by telling them, this was something FDA asked about, you should provide that information up front in your submission.

So, I think regulatory teams and device development teams need to really develop this art as a way to de-risk their development program and their regulatory submissions. I think what you'll find is that, when you start to do these searches, you're going to find out that you didn't know what you didn't know, and it happens to me all the time. I'll do these searches, and I'll uncover things that surprised me that I didn't know, and I've been doing this for 30 years.

So, in some cases, it makes it harder to decide the more information you have about the pathway or the issue you want to address, because you have conflicting information at FDA, or the information that you might want to get from FDA databases is redacted.

So, there are limitations to doing this, but still, overall, you're trying to reduce regulatory risk.

Gotcha. Well, thank you, Doug, for joining me to share a little bit of flavor around the article that you wrote.

And folks in our audience for today: Again, you'll find Doug's article on precedent research on Drug Delivery Leader. And, again, it's entitled, “GenAI: The Muscle Behind Strong Regulatory Intelligence for Combination Products.” We hope to see you land there.