[Published on 25th March 2025]
The use of generative artificial intelligence (GenAI) in health economics and outcomes research (HEOR) (and everywhere else!) is a hot topic for 2025 and beyond. Coming in at number 3 in ISPOR’s Top 10 HEOR trends for 2024–2025,1 and often dominating conference agendas (including ISPOR Europe 2024, as summarised in our previous article), it is clear that it is here to stay.
However, while the use of GenAI for some HEOR applications (notably systematic literature reviews [SLRs] and health economic modelling, as noted in a recent ISPOR Working Group Report)2 has been investigated in a fairly structured manner, the use of GenAI in value and access dossier writing has been less robustly evaluated. With growing pressures on the value and access processes for manufacturers, particularly due to tightening timelines and the introduction of EU Joint Clinical Assessment earlier this year, further exploration of how GenAI can support the timely and efficient delivery of high-quality dossiers is warranted.
At Costello Medical we are committed to investing in technical innovation, exploring the latest technological developments in specific use cases where they can bring the greatest value. As part of this commitment, we recently explored the potential of GenAI to assist value and access writing. We systematically refined and tested GenAI template prompts for generating dossier outlines, section first drafts and section summaries, which were then used as a starting point for further human refinement. We took a ‘100% Human-in-the-Loop’ approach to ensure high quality; all AI-generated content was thoroughly checked by members of the project team for technical and scientific accuracy and AI outputs were used as a starting point for further refinement rather than a final product. We are also committed to the responsible use of GenAI, and therefore use enterprise-grade security across our GenAI tools.
The time invested in drafting and systematically testing comprehensive, detailed prompts using a range of prompt engineering strategies enabled development of prompt sets that were able to generate reasonably high-quality outputs ready for human refinement and which could be adapted as necessary for different settings.
The outline prompt set we developed enabled GenAI to quickly pull together a logical outline that accurately reflected the disease landscape, for example correctly identifying certain areas of unmet need. Human refinement was then required to include any finer details, as well as amending the outline to incorporate important strategic components, but this was fairly quick and relatively simple. AI-generated outlines may be less effective for more complex disease areas, or rare diseases, where there may be a lack of published literature for GenAI to draw from. Careful evaluation and expert input remains crucial to ensure that the strategic narrative is accurately and persuasively conveyed throughout the content.
Overall, GenAI performed well in generating robust first drafts of individual sections which can be used as a starting point for human refinement. However, when trialling prompt sets for drafting of entire dossier sections we found that the GenAI tool could easily become overwhelmed, leading to inaccuracies and low-quality outputs. Retrieval-augmented generation substantially improved the accuracy of the output, and two further approaches were also tested to improve the quality of the output:
As expected, GenAI was able to rapidly summarise large volumes of text, for example from pre-written dossier content, and the summaries were generally factually accurate. Subsequent human adaptation of the generated content was necessary to ensure that the product’s value story and strategy were sufficiently reflected, but the generated content provided a good starting point from which to do this.
Our key takeaway is that investing in generating robust prompt sets can support efficiency gains in dossier writing, by enabling the use of GenAI for initial content development and releasing time for complex, strategic activities. However, it is clear that GenAI should be used to support, not replace, experienced professionals; while GenAI can assist with drafting and summarising content, expert oversight is imperative for ensuring the scientific rigour of the dossier content, as well as alignment to the overall strategy and value narrative for the product. It is also important to bear in mind that one of the most time-consuming aspects of dossier development is the extensive stakeholder review and revisions process once the initial draft is prepared, and GenAI cannot currently significantly shorten this process. Consideration must also be given to ensuring tools are appropriately secure when GenAI is used to draft content based on highly confidential data. Although the use of enterprise-grade tools provides robust data security measures and avoids any training of AI models on inputted content, confidential data should not be uploaded to any AI tools without permission from relevant stakeholders.
Where AI is used to support dossier development it is critical to carefully consider any requirements regarding the use of AI from HTA and regulatory bodies to ensure that dossier content is suitable for adaptation to HTA submissions, for example including appropriate disclaimers of GenAI use. To date, NICE is the only HTA body to have issued a clear statement on the responsible use of AI in evidence generation.3 This states that the use of AI should be transparently declared and only used where it provides value, for augmentation, not replacement, of human involvement.3 Our stance at Costello Medical is closely aligned with this approach.
The anticipated release of GPT5.0 and development of more advanced tools promise to further advance GenAI’s application in value and access writing. We look forward to exploring their potential while continuing to refine and explore our use of GenAI in the future; intentional efforts to understand AI’s strengths and limitations will be key to the successful future use of GenAI in value and access writing.
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, value and access writing to support dossier development, automating elements of literature reviews, conference coverage and insights, and plain language summary generation.
References
If you would like any further information on the themes presented above, please get in touch, or visit our Value & Access page to find out how our expertise can benefit you. Nicola Ashman (Analyst) and Eleanor Atkinson (Senior Analyst) created this article on behalf of Costello Medical. The views/opinions expressed are their own and do not necessarily reflect those of Costello Medical’s clients/affiliated partners.
