Is There More to AI in HEOR Than Literature Reviews?

While ISPOR Europe 2025 showcased artificial intelligence (AI) moving from an experimental add-on to a core capability in certain HEOR workstreams, the buzz around AI’s impact on dossier and model development appears to have died down for now. Compared with the 60 studies reporting on AI in evidence synthesis at ISPOR this year, only 18 focused on the application of AI in economic modelling and 19 on AI in value dossier development.¹ Here, we pull together the latest insights from the conference, taking a front-line view of what works, what doesn’t, and what needs governance as the use of AI in HEOR is set to increase.

Beyond Literature Reviews – Where is the Promised AI Progress in Dossier Development?

In an issue panel discussing the integration of AI into market access workflows, an opening poll found that only 15% of the audience thought that genAI would bring most benefit to global value dossiers, compared with 45% for literature reviews.² Interestingly, 41% felt confident that AI would improve and accelerate patient access to new health technologies across the next 5 years.² Surely these gains cannot be realised through SLR efficiencies alone? We presented research exploring the quantitative and qualitative impact of AI in value dossier drafting to determine where the greatest value currently lies, and what continued research is required to refine this approach.

There were a handful of issue panel discussions exploring the potential for AI to streamline dossier reviews both internally during development and externally during HTA processes. Although benefits in terms of the reduction in development times and HTA backlogs were highlighted, there was a lack of clarity on how AI would be utilised and the quantitative metrics which would measure its impact. In particular, what governance would be required to ensure that quality remains unaffected, and how as an industry do we ensure sufficient human involvement and oversight? Ultimately, speed is valuable only if it is accompanied by sustained quality and traceability.

AI’s Fight Versus the Modern Health Economist: Still in the Early Rounds

Throughout the conference, various issue panels, posters and training courses also focused on the use of AI within health economic (HE) workstreams. Despite not dominating the AI conversation at Glasgow 2025 to the same degree as the use of AI in literature reviews, AI in HE shows promise to improve the efficiency of some current workstreams but remains in the shadow of continued concern regarding the reliability of results and how much ‘value add’ AI provides versus an experienced health economist. Here, we provide commentary on two specific HE workflows.

How Competent is AI Conducting Modelling Tasks?

During Sunday morning’s short course, BMS and Estima Scientific presented an introduction to applied generative AI for HEOR, including an example of an AI tool to complete basic model adaptations.³ An Excel model could be fed into the AI tool alongside detailed instructions regarding which inputs to update and with what values. The AI tool then successfully updated the model with new inputs. Posters presented throughout the conference showed how AI can develop code to build economic models in R or run model stress tests to identify potential errors.

There is clear potential for AI-driven time savings versus a purely human approach for simple tasks such as adapting model inputs for multiple geographies or indications, or building simple proof of concept models. However, the examples of currently developed proof of concept AI tools have largely been constructed using idealised scenarios which are not transferable to everyday HE practice. Our Health Economists have worked directly with these current AI tools, and our experience suggests they cannot handle dynamic Excel formulas well – a mainstay of modern health economic modelling. Technical QC’s are still 100% necessary to ensure the reliability of AI-based updates, at which point how much time is actually being saved? As any health economist will tell you, there is no one size fits all approach to modelling – will AI tools become flexible enough to deal with more nuanced modelling tasks? This question is reflected industry wide, evidenced in the recent research conducted by NICE as part of their HTA Innovation Labs initiative.⁴

AI’s Place Constructing Nuanced Reports

On Tuesday afternoon, Issue Panel 118 discussed the use of an AI tool to streamline the writing of a survival analysis report.⁵ The premise: can AI take fitted survival curves, write a first draft survival analysis report to be reviewed by an expert health economist, and subsequently write a final draft. The AI tool was tasked with selecting the correctly fitted curve according to NICE TSD14 guidance capturing both the logical and nuanced flow of a human author.⁶ Again, the promise of using an AI tool to significantly streamline report writing, often a time-consuming task, is clear. Outcomes of the issue panel were in part positive – the AI-generated report followed a very well-structured layout capturing the key detail needed. Unfortunately, when the report was reviewed by an expert, the accuracy was questionable. Multiple detailed reviews (which were fed back into the AI tool) were required to produce a finished product. On the one hand, this demonstrates a very successful proof of concept to use AI to help begin to write detailed reports used within HE workstreams. On the other, if an AI drafted report requires multiple detailed reviews from an expert to achieve something final, are there actual tangible efficiencies? By developing AI tools to streamline workflows typically conducted by junior team members, are we at risk of deskilling our workforce over time or will our skills change alongside AI’s development – it seems there is still progress to be made and increased adoption across the industry to find an optimal solution.

The Road Ahead: Balancing Speed, Rigour, and Regulation

Whilst there is clear promise of potential to streamline workflows, actionable tools that can be used across the industry are still lacking. We are yet to see AI tools able to construct nuanced arguments or consider detail trade-offs, something crucial for evidence-based decision making. As with any emerging technology, an important question remains around whether AI will deskill the workforce. To ensure sustainable development of the technology, it is important we continue to trial new and innovative tools and AI use cases within existing and new workflows whilst reskilling our workforce to adapt alongside AI’s development whilst retaining key technical skills.

At Costello Medical, we are embedding AI where it meaningfully reduces time and preserves or enhances quality, while maintaining transparent documentation and a robust human-in-the-loop. This means living governance that evolves with new tools, ongoing validation across domains, and a culture of careful scrutiny, especially when our work is subject to review by regulatory or HTA bodies.

We’re cautiously optimistic, but we are still a long way away from AI making meaningful impacts in the complex and detail-heavy world of HEOR, with its strict quality requirements for regulatory or reimbursement purposes.