The Recipe for Value Dossier Success:
Should GenAI be Added to the Mix?

The use of AI in HEOR (and everywhere else!) is a hot topic. Coming in at number 3 in ISPOR’s Top 10 HEOR trends for 2024–2025, and often dominating conference agendas, it is clear that it is here to stay. Whilst the use of AI has been explored in the health economic modelling and literature review space (as covered in our ISPOR EU 2023 report), there has been little research to examine its potential in the development of value dossiers. The evolving Market Access landscape is continuing to put greater pressure on manufacturers; the introduction of the EU Joint Clinical Assessment in January 2025 will include developing and submitting a dossier within only 90 days of receiving the final assessment scope. This is just one example of why there is an increased need for the efficient and timely development of high-quality value dossiers that can support a range of activities beyond the ‘standard’ local reimbursement proceedings.

So, could there be a role for GenAI in value dossier development?

We explored the potential of using a generative AI (GenAI) tool, ChatGPT 4.0, at various different stages of the value dossier development process. Given the capabilities of this generic tool, we knew we weren’t going to get the perfect value dossier at the click of a button. However, we saw this as an exciting opportunity to gain insight on what is currently possible as well as what we would want to see in future tools, particularly as they become more advanced with the potential for tailoring to your individual needs/requirements. Given our commitment to high quality, we took a ‘100% Human-in-the-Loop’ approach, with all AI-generated content being thoroughly checked by members of the project team.

Our key findings

1.

Handling confidential information

We are committed to the responsible use of AI, and so did not feed the tool any confidential information for this project. Instead we focused on piloting the tool with sections of the dossier that were based primarily on published data/information (and ensured content was anonymised in cases where the client, product and/or indication were identifiable). Whilst we were still able to gain useful insights, we were restricted with what we could input and this did ultimately limit the usefulness of the tool. It also required additional time to get the content ‘ChatGPT ready’ which created inefficiencies.

2.

Preparing outlines

The first stage of value dossier development is the preparation of an outline to allow stakeholders to agree on the general content, structure and flow. The tool was able to quickly pull together a logical outline that accurately reflected the disease landscape, for example correctly identifying certain areas of unmet need. It was a useful basis for us to work from, however, some adaptation was still required to align the structure and/or content with the product’s value narrative.

3.

Developing summaries

The tool was able to rapidly summarise large volumes of text, for example from scientific articles or pre-written dossier content, and the summaries were generally factually accurate. Although accurate, adaptation was required to ensure the product’s value story and strategy were sufficiently reflected in the summarised content.

4.

Updating the tone and style of text

When clear prompts were used, the tool was able to rapidly update the style/tone of text as requested whilst also generally retaining the accuracy of the content; for example, adapting existing content to make it more patient-friendly. However, some adaptations were required as the tool was not always able to match the tone of the request whilst retaining factual accuracy.

5.

Developing de novo content

We trialled using the tool to generate larger passages of text for the value dossier. Whilst it was able to rapidly generate content that at a surface level appeared to answer the question, substantial updates were required to ensure it conveyed the required messaging in line with the strategy for the product. Additionally, content lacked references that substantiated the claims, so considerable time was spent identifying appropriate references and then subsequently adapting the content.

So, what are we looking for in the future?

To ensure value dossiers are fit-for-purpose, it is critical they include evidence-based content that is written in a compelling manner to best communicate the value of a product, and this is ultimately where the content generated by current AI tools fell down. We are already seeing, and exploring the potential of, more advanced tools that are trained on certain writing styles as well as being able to substantiate claims with robust, publishable references. Such advancements would certainly enhance how we could work with these tools to support value-based content development in the future.

Furthermore, in order to maximise the functionality of tools and continue seeking out new use cases there is a need for a shift in the current restrictions that are in place related to the processing of confidential data and general data security. Developers need to build users’ trust in their tools’ handling of commercially sensitive data within the business sector, so that users can get the best out of these tools.

Through this project we have understood when and how to use current GenAI tools, such as ChatGPT 4.0, in the development of value dossiers as well as identifying areas where we don’t feel these tools are quite ready yet. The release of GPT5.0, expected later this year, will no doubt show us the general direction of travel regarding how these tools are advancing. Whilst the utilisation of AI here did not necessarily create efficiencies or resource savings, it did contribute to learnings around how we would want to optimise AI tools to feed into future projects. Moving forwards we will be continuing to use GenAI where we can as well as exploring how newer bespoke tools could further support our work in the future.

At Costello Medical we are investing in technical innovation, such as AI, and are keen to continue to explore the latest developments and exciting new opportunities in collaboration with our clients. Our strategy is focused on general everyday adoption coupled with sector-specific use cases, such as insight gathering for HTA strategy development, automating elements of literature reviews, conference coverage and insights, and plain language summary generation. If you would like to discuss how we might be able to support you with exploring new and innovative use cases for AI then please get in touch with Lucy Eddowes, Scientific Director, to discuss further.

If you would like any further information on the themes presented above, please do not hesitate to contact Grace Lambert, Consultant – Market Access (LinkedIn). Grace Lambert is an employee at Costello Medical. The views/opinions expressed are her own and do not necessarily reflect those of Costello Medical’s clients/affiliated partners.

Loading