Could GenAI Dramatically Reduce The Cost Of Regulatory Medical Writing?

By Punya Abbhi, COO and Co-Founder of Celegence

Today, Generative AI (GenAI) technology is disrupting the way all kinds of organizations filter, find, and process knowledge and content, to deliver summaries, analysis, and insights in a highly automated way.

The output — generated in an uncannily human way — can be striking. But the life sciences industry’s ability to trust that output, and use it for content which has a bearing on the safety and efficacy of drug products, depends directly on how the technology is applied and controlled. And this is where some biopharma companies are now running into difficulties, having started to build their own internal GenAI expertise in a bid to transform their multiplying regulatory and medical writing workloads.

The life sciences industry is ripe for GenAI-based process transformation, with its extensive, rigid, and repeat activities including marketing authorization applications and life cycle maintenance, and post-market safety report generation. If purpose-built applications armored with large language models (LLMs) could even just collate an initial draft of these documents (with expert oversight), it would alleviate at least some of the burden on time-poor regulatory professionals.

The Catch: Knowing What To Aim For

Preparation is everything, though, and this is where drug companies are running into some early problems. This is because GenAI needs a lot of ‘shaping’ to be of reliable use, even with the benefit of LLMs (vast data banks) to draw on.

Irrespective of how intuitive GenAI’s natural language processing and deep learning capabilities might be, algorithms must still be shown what to look for and how to distill and generate information and data in a way that is correct, fit for purpose, and acceptable to health authorities. This demands not only relevant GenAI skills but also deep first-hand knowledge of pharma’s specialist vocabularies, required templates, and subtle variances in expectation of each target market; along with a strong feel for specific medical writing best practices linked to each medical writing use case.

At the same time, vast reams of successful example content will be required to coach a GenAI model in what constitutes optimal output for the given use case. There needs to be a robust understanding of formal terminology (and its abbreviations and variants); of how to decipher tables, listings and figures; and of meaningful correlations between data and a drug or treatment. Only in this honed, specialist context can AI tools be trusted to interpret complex data and point to a logical outcome.

Research Reveals A Skills Gap In Life Sciences

All of the above equates to a lot of multi-disciplinary specialist knowledge for a pharma company to amass. Yet the need is real, and growing. Medical writing for regulatory purposes remains a critical area requiring support, due to the increasing pressures on professionals’ time.

This is confirmed in a 2024 survey conducted with the Regulatory Affairs Professionals Society (RAPS), in which the appetite to intelligently automate some facet of the medical writing process is revealed to be significant and increasing. Over half (57%) of surveyed companies cited planned investment in technology to improve medical writing over the year ahead. The primary medical writing needs were linked to clinical study protocol/report writing and drafting of regulatory documents.

A similar proportion of regulatory professionals indicated a desire to harness AI in data extraction (56%) and information summarization (53%), where just 9% to 10% are using AI for those purposes today — specifically within the context of medical writing. Twelve percent of respondents said they were actively in the process of incorporating AI into automated report generation from multiple sources, which is the ultimate opportunity on offer. Despite these ambitions, more than half (53%) said their organizations did not have sufficient knowledge to deploy AI technologies.

Compliance Considerations

Further barriers to AI uptake revolve around confidentiality and the scope for inadvertent breaches, linked in part to how the technology is ‘prompted’ to call up information. Because the LLM models are effectively a black box, there are concerns that applying public cloud-based GenAI models could compromise internal data security. Such concerns are readily addressable in a ‘closed’ processing environment tailored to life sciences use cases, which ensures that all data interactions remain within a secure and controlled ‘space’ for data confidentiality and integrity purposes. IT departments need to be careful about how exactly they provision for GenAI applications.

Additional considerations include traceability and auditability: the ability to see where extracted data has come from or from which source content summarization has been achieved (for instance, clear links in the final report to original documents). This is crucial to build confidence in solutions, so that they are seen to add significant value and genuinely save time (if they require continuous, attentive supervision, this may not bear out).

Seizing The Opportunity To Act

The assurance that there are solutions that can be validated for priority pharma medical writing use cases will be a significant facilitator for companies, as medical writing workloads continue to soar.

In looking for maximum value from an effective GenAI medical writing automation capability, pharma companies should consider not just efficiency gains, but also the scope for improvements to consistency, and the opportunity to deliver tighter output once systems have deduced what ‘good’ looks like, from extensive exposure to both approved and rejected documents.

Investing in R&D or partnering with external technology solution or service providers to leverage GenAI across a variety of use cases should ultimately enable better use of scientific experts’ time, as more time is freed up for strategic thinking. Closer collaboration with clinical development, for example, is one area that could benefit from reduced time constraints.

In fact, GenAI promises transformation in early clinical evaluation. Here, there is an opportunity for the technology to scour and summarize the vast wealth of existing information online to help focus planned research. AI-assisted search across multiple sources offers both to summarize headline findings such as alternative therapies to a specific drug or device, while signposting the best path to follow for clinical research.

Feeling The Way

Trying out GenAI capabilities offers companies a low-risk way to explore what’s possible, and to estimate the difference it could make in everyday workflows. Given the progress of industry trailblazers like Moderna (which already has more than 750 active use cases of AI to improve process efficiency), it is important to at least understand what is possible to avoid the risk of falling behind.

Even if pharma companies lack sufficient R&D resources to trial the various possibilities themselves, co-designing a pilot solution with a technology partner is one practical way forward, to determine how powerful and accurate the technology can be and how quickly it is capable of learning and adapting. Often, it’s only from seeing what rapidly-evolving GenAI tools can do that makes the opportunity fully apparent.

About The Author:

Punya Abbhi is chief operating officer & cofounder at Celegence. She was previously a life sciences consultant at Capgemini Consulting, and before that spent time at Kates Kesler Organization Consulting, and Gartner, and worked for a specialist life sciences service provider. Punya holds an MBA from INSEAD and a BA in Politics, Philosophy, Economics (PPE), with a minor in cognitive science from the University of Pennsylvania.