5 AI uses in Regulatory Affairs: Current Reality or Future Possibility

There is a lot of chatter about artificial intelligence (AI) in the news, and the pharmaceutical space is no different. AI is increasingly used throughout drug development, from target discovery to post-licencing safety reporting, supply chain management, and PK/PD modelling. As regulatory professionals, we are occasionally required to understand how some fundamental aspect of AI works. For example, this may be necessary to help an innovative new medical device be classified. However, taking a broader perspective, it is insightful to know where AI is at its best in the day-to-day activities of a regulatory professional and where it provides limited functionality. Using our experience, we have outlined use cases where we believe AI can improve quality or efficiency and where there is still a substantial gap between an AI tool and that of an experienced industry professional.

AI Use Case 1: Drug Screening and Target Identification

Recent analyses have tried to assess the number and success of drugs with targets identified by AI and suggested that their success in Phase I trials is superior to that of drugs with targets identified by more traditional means. This research identified 5 ways companies are using AI in the early phase of drug development projects. These are: AI-discovered targets (I), small molecules that were themselves discovered or optimized via AI techniques (II), biologics that were similarly discovered or optimized (III), vaccines similarly discovered or repurposed (IV), or drugs that have been repurposed through AI techniques (V). A deeper look at the AI-discovered targets reveals that a lot of the targets are already known to be implicated in the disease in question. This raises questions about the strength of AI’s involvement in target selection. In addition, questions of sample size potentially invalidate some of the paper’s key findings.

Large pharma and Big Tech Join Forces

Nonetheless, the number of collaborations between big pharma and AI companies points to this being a genuine shift in the paradigm. This suggests AI is generating insights that could not easily be gleaned with traditional techniques. See Nvidia’s alliances with Amgen and Genentec, the partnership between Alphabet subsidiary Isomorphic Labs and Novartis and Eli Lilly, as well as Boehringer Ingelheim’s pact with IBM for examples.

AI-derived drug targets undoubtedly have the potential to make waves in early clinical trials. One example is In Silico Medicine’s novel Idiopathic Pulmonary Fibrosis targeting small molecule ISM001-055. This new drug met primary and secondary endpoints in a recent Phase II trial. Drug screening and target identification therefore appear to be genuinely profitable use cases for AI. For those of us based in Regulatory Affairs, this may feel like a faraway place from where we spend most of our time, but it will become increasingly necessary to be familiar with AI-based target identification methods, especially during the early stages of interactions between pharma and biotech companies and Health Authorities.

AI Use Case 2: Preparing CSRs and Other Large Clinical Documents

Preparing key clinical documents, most notably Clinical Study Reports (CSRs) and patient narratives is typically done by regulatory professionals and medical writers. Preparing these documents is time-consuming and difficult. CSRs can be 10’s of thousands of pages long. The patient narratives, required by International Council for Harmonisation (ICH) E3 guidance, should “describe each death, each other serious adverse event, and those of the other significant adverse events that are judged to be of special interest because of clinical importance”. Depending on the trial scale, this alone can run to more than a thousand patients.

ML and NLP Can Craft Long Regulatory Documents

Various AI writing tools now exist to assist with writing CSRs. These tools use Tables, Figures and Listings to generate results sections and protocols, amendments and the Statistical Analysis Plan to create the methods.

Machine learning (ML) techniques, which analyse structured data, and natural language processing (NLP), which can extract and analyse data from unstructured data sources, combine to create drafts of these documents. Importantly, this takes significantly less time than would be required by a medical writer and generally produces fewer errors. However, crafting the overall narrative, ensuring that key points are stressed, and checking the document for accuracy and consistency of messaging are still the roles of medical writers and regulatory professionals. Indeed, these tools are designed to be operated by people already familiar with writing this kind of document. A lot still needs to be done to make a document ready to be submitted to Health Authorities. Even still, the time saved creating a first draft is not insignificant.

AI Use Case 3: Comment Resolution Meetings (CRM)

Moving on to a broader topic that many in the regulatory space will be familiar with – the comment resolution meeting. CRMs can be in regard to a Pre-Submission briefing book, a large clinical document (CSR or DSUR), or even just a cover letter. Regardless, meetings between various stakeholders intended to agree on the contents of a given document are currently not under the scope of any AI tool. These meetings require the direct interaction of many people from various functions, including regulatory, medical affairs, and data.

Managing People

Each of these people has a range of experience in their given role and the aims and needs of their own organisation (including budget restrictions). This means that perspectives are frequently not aligned. Making progress in these meetings requires active listening, considered communication and a deep technical understanding of the topics in question. AI-based functions such as the auto-transcript feature in Microsoft Teams can help make this kind of meeting easier. Even still, there is no substitute for the open exchange of ideas and knowledge that this kind of meeting facilitates.

AI Use Case 4: Divestments

Divestment is the process of selling established products in order to streamline or refocus the overall product portfolio of a pharmaceutical company. When the product in question is marketed globally, divesting it successfully to another company is a complicated process that needs careful coordination between regulatory, supply, and safety teams. Local affiliates/partners and the Health Authorities in the affected regions also need to be factored into planning. The first step involves clustering Market Authorisation Transfers (MATs) and deregistrations for each country under the scope of the divestment. These are grouped into time windows during which the transfer or deregistration must occur. This is determined by factors including MAT requirements, ongoing regulatory activity and supply chain/bridging stock availability. While this feels like the kind of activity AI could help with, there is currently no tool available to perform this task.

Lots of Moving Parts

One reason is that MAT requirements are constantly changing, and understanding these requirements often requires clarification from local affiliates. When the diverse range of ongoing activities on a global scale is considered, an even more complicated picture emerges. In practice, divesting a global product is essentially a huge exercise in communication: communicating the very specific local MAT requirements and fulfilling them, communicating with local Health Authorities (HAs) where there is ongoing regulatory activity, and communicating with supply to organise the transfer of the product with minimal disruption to stock. Put simply, there are too many factors for any AI tool to take into account, even before things like country-specific communication styles or language barriers are considered.

AI Use Case 5: Response to Questions

Responding to HA questions following a Marketing Authorisation Application (MAA) or New Drug Application (NDA) is required at key time points following submission to answer important points raised by the assessing HA. The answers to these questions are often dispersed through many different source documents or need to be gleaned anew from existing data. Answering the questions well in the allotted timeframe is challenging and requires coordination of multiple departments. These include quality, regulatory, and clinical functions. Negotiation skills are required to know when to push back on a request and when to concede.

Cultural Differences

It is also very often the case that the client and the HA are not from the same part of the world, which means language and cultural considerations must be taken into account. It is conceivable that an in-house AI-chatbot style tool trained on a company’s own source documents could help with answering the questions themselves. In practice however, this is beyond the reach of most small/medium and even some large organisations.

Summary

To conclude, AI already plays an important role in the early stages of the drug development process. This role will only increase in the future. Generating the first draft of large regulatory documents is also comfortably within the scope of AI tools, though not without the guidance of someone with professional know-how. Working in these areas, we are likely to see an evolution of roles as AI-based technologies become increasingly integrated. The hope is that they can remove the monotonous and least enjoyable tasks. This would leave brain power free for strategic, big-picture activities, though how this pans out in reality is unpredictable. What is clear is that where there is a need for high-level communication, negotiation, the interaction and integration of multiple stakeholders and perspectives, as is often the case in the world of regulatory affairs, AI tools in their current form are much less useful.

If you’re looking to stay ahead of the curve in the ever-evolving regulatory landscape, we’re here to help. Our commitment to innovation ensures that we are always exploring the latest developments and integrating them into our regulatory strategies. By staying informed and adaptable, we maximise our approach to regulation, helping you navigate challenges and ensure compliance at every stage. Contact us today to learn how our expertise can help you stay ahead. Email DLRC at hello@dlrcgroup.com or use the link below.