Moving on from the maybes: Real-life uses of generative AI in Medical Affairs

Generative AI (genAI), the subset of machine learning led by ChatGPT, Claude, Gemini, etc., needs little introduction. These tools still dominate headlines and were a hot topic at the Medical Affairs Professional Society (MAPS) 2024 EMEA meeting.

Rightly so, as Medical Affairs Professionals produce a lot of written and visual content for communicating with HCPs and patients – how is this essential work improved, or even threatened, by these new genAI tools?

MAPS has been educating and sharing expertise on genAI since ChatGPT first exploded onto the scene in early 2023, first with a webinar, followed by panel discussions at both the US and EMEA meetings in 2023. This year, as the dust settles, we focused on sharing the learnings from a year of pilots and projects to help others begin or accelerate their journey.

Authors:

Natasha Hansjee (Roche)
Emma Vitalini (Amgen)
Jessica Ingram (Camino)
James Turnbull (Camino)

Download GenAI Cheat Sheet

MAPS EMEA Session Slides

Starting Small: Case Study and Initial Steps

Despite the excitement and widespread use of genAI, the applicability of these general-purpose tools to specific tasks within Medical Affairs remains unknown in many cases. A 2023 report from Harvard Business School¹ called this the “jagged technological frontier”: where difficult knowledge tasks are sometimes perfect for giving to genAI, while with other, seemingly similar tasks, it completely fails.

This means that the best way to apply these tools is through pilots and experimentation: start small but it’s important to start.

One use case, which is an ongoing project, is using genAI to assist the process of determining competitor intelligence through publications.

The project began with a one-time search through all the literature on a specific competitor, a common task for pharma to outsource to a medcomms agency. However, as it was looking at purely public data, and was a process which was severely limited by the human capacity to read a finite number of papers in any given day, we augmented the project with genAI.

We divided the task into 18 steps, from data collection, to categorization, summarization, finding themes, and developing a slide deck summary. Within the initial pilot, just two of these steps were supported by genAI, but those were key filter points in being able to consider a much larger set of initial papers (1600 on first pass) from which to extract the key data.

The ongoing project from this pilot now considers many more sources, across multiple brands and competitors. The initial tool has continued to develop, and as genAI tools have got better at summarizing larger documents, we have augmented four steps in the process of 18. This iterative approach, making the most of new features as they develop, is an essential way to consider genAI in your projects.

When considering an AI project, we suggest three main criteria, which must all be considered equal:

Efficiency: Maintain quality whilst reducing timelines and costs
Value: Enrich outputs or accomplish things that were not possible or practical without genAI
Innovation: Demonstrate the potential of genAI to inform future adoption, creating a competitive advantage

Balancing these factors helps scale a project from pilot to ongoing development, while demonstrating the necessary outcomes.

Many pharma companies are providing teams with internal chatbots, either as pilots or company-wide. These could be Microsoft CoPilot, which is integrated into their Office 365 suite, or the enterprise version of ChatGPT Pro.

Some companies, such as Merck with their ‘MyGPT’, have created custom chatbots built atop the developer interface for GPT-4 and other models, which can connect to internal documentation to increase the chatbot’s relevant knowledge².

RocheGPT, meanwhile, has been put in place to support teams with extracting structured data from scientific articles and medical reports³.

Moderna gave staff the ability to create their own custom “GPTs”: chatbots pre-loaded with access to a specific set of content and a system prompt to direct the model to act or reply in a certain way. This resulted in 750 custom bots being built and used across the organization⁴.

In all these examples, there is a need for clarity and training on how to get the most out of these models for employees. Each individual within an organization knows their own job and subtasks better than any other, and like at Moderna, should be empowered to experiment with AI augmentation of these tasks. But basic training on prompting the model with context, tone, audience and length etc, as well as understanding its limitations and unavoidable biases, is essential.

Use Cases: Ideas and Examples

The three routes to implementing genAI within medical affairs are:

Public tools

ChatGPT, Claude, Gemini, etc. can be very useful in your day-to-day life without using confidential work-related content. Helping you finish a sentence, clarify a social media post, or coaching you through your 1-to-1 with a manager. You should add these tools to your daily ways of working

Custom tools

Building custom tools on top of the APIs provided by Microsoft, AWS, or similar allows you to solve specific challenges with genAI. Here you can use confidential data and can build in safeguards to minimize hallucinations (although you should still always be responsible for the output and regular monitoring to prevent or minimize the introduction of bias). This creates efficiency, value and innovation within your team

Enterprise solutions

As your enterprise adopts genAI, it will provide you with custom chatbots, or tools that integrate with your existing workflow (such as Microsoft CoPilot). While these tools will not be specifically for Medical Affairs teams, it’s important to be part of the rollout and share your learnings. Consider how within Medical Affairs you can build your own backlog of use cases, how you can improve data and information quality, and how you will share capability harmonization and build.

Many Medical Affairs teams are starting their AI journey with training on using the public tools or internal chatbots so that the whole team is rapidly building knowledge and confidence in using AI. From here, they quickly move to workshops around what challenges they can augment with custom genAI tools.

Pharma projects using custom tools that we’ve seen launched recently include:

Developing core narrative and modular content plans

Scrutinizing market trends and stakeholder insights to craft tailored core narratives and content plans that resonate with your target audience. GenAI can also optimize content distribution strategies by predicting engagement patterns and suggesting timely updates.

“Build your own” learner journey

Analyzing individual performance data to address specific knowledge gaps and preferences by serving users with tailored content, in a format supporting the individual’s preferred learning style or availability (e.g. audio vs visual). Feedback can be real-time and adaptive to further enhance skill development and knowledge retention.

Literature review and alert emails

Augmented mapping of the competitor data landscape to drive strategic decision-making, train teams, and keep them up-to-date as new data are released.

Insight analysis

Parsing and summarizing quantitative and qualitative data from a variety of sources to highlight where insights may be available to Medical Affairs teams.

Global to local omnichannel adaptation

A web-based tool that is loaded with template global materials, as well as local strategies, local compliance guidelines, and approved indications. The tool provides first-draft adaptations of materials and can take feedback on local learnings to improve future outputs.

ProductBot

A (crucially) internal-only chatbot loaded with the knowledge around a launch product. Used by medical teams and their vendor agencies for quick reference.

Integration, Implementation, and Impact

Developing KPIs for your AI-augmented projects is essential, but can also be tricky. Unlike most other technologies, measuring (and showing) the metrics of an AI project is difficult, as it is highly subjective to the user.

Often AI reduces the burden of a regularly repeated task (such as drafts of meeting reports) and in these examples, user-reported efficiency gains can be helpful, especially if captured after a period of use (e.g. monthly).

Showing the value of projects augmented with genAI is often done through comparing with the ‘traditional’ methods – e.g. using AI to parse larger datasets, find more potential insights, or even tackle unexplored avenues and options that, without the support of genAI, would normally be ignored. It also adds an unprecedented turn of speed to the parsing of large and potentially currently inaccessible data volumes, enabling rapid insight generation and increasing agility of decision-making.

Similarly, on a personal level, you can add value through augmenting your work with extra content and flair. For example, there are few internal projects for which you would normally commission a photoshoot, but instead of wordy slides, AI image generators can support your presentation with photorealistic visuals. Of course, some projects still require the expertise of a real photographer, but there’s added value through the easy avenues which AI offers all users throughout the organization.

Remember, just like a human, it can make mistakes or fall shy of the mark, so always check the outputs of genAI and don’t cut yourself out of the loop completely.

Challenges and Ethical Considerations

With the fast-paced release schedule of AI technologies, it’s often worrying that the moment you commission a project, the underlying technology will be superseded by the next frontier model. While this is a risk with any technology, you can mitigate some of these. Firstly, you should assume that the models are getting better, and cheaper, and therefore build your AI project in as modular a style as possible. With the technical know-how, there are also plenty of ways to allow change of the underlying model, so that a shift from version 4 to 5, or indeed sideways to a competitor model (or even a self-hosted model) are all possible.

There are also many ethical considerations with using AI in your organization, and guidance for employees on these is essential. The EU have enacted the AI Act as a first regulation on AI broadly⁵. For healthcare, the WHO has published a broad 40-point guide on ethical use of genAI⁶ and companies such as Roche have also published their employee guidance publicly⁷.

Future Trends

The recent release of “GPt-4o” (the ‘o’ stands for ‘omni’) has shown that we are already past the world of Large Language Models, and multi-modal is the future. GPT-4o can take text, image, live video, and spoken input all at the same time, and output using these modalities too. The ability to rapidly parse spoken and visual data will have many applications in healthcare, but also in Medical Affairs: supporting the unique and complex needs of communicating science to increasingly diverse audiences, and analyzing the impact of those in real-time.

A further avenue of genAI, which is only now being explored, is AI coaching. Helping you learn and practice skills, set goals, or work through challenges with colleagues, a well-built and monitored coach can have endless patience and can be personalized with its approach.

What you should do next

We already know that genAI is transforming Medical Affairs. From assisting with competitor intelligence to developing tailored content plans and learning journeys, the potential—and executed—applications are vast. Yet the transformation is not complete; as the technology continues to evolve, it is crucial for Medical Affairs professionals to embrace and adapt alongside it to make the most of the efficiencies, value creation, and innovations genAI can offer. By staying informed, aware of the ethics, experimenting, and sharing experiences of both successes and failures, we can collectively navigate this frontier and harness the power of genAI to revolutionize the way we work.

If you’ve not yet started with genAI, we have created a one-page cheatsheet to help you get started. Download it here.