HTA Coordination Group issues guidance on AI use in Joint Clinical Assessment dossier preparation

The Member State Coordination Group on Health Technology Assessment has adopted its first guidance on AI use in preparing Joint Clinical Assessment dossiers. The document sets out high-level principles for the responsible use of AI, emphasizing that developers remain accountable for all submitted content and that appropriate human oversight is essential throughout the dossier preparation process.

The Baseline

• The HTA Coordination Group has issued its first guidance on the use of AI in preparing Joint Clinical Assessment dossiers under the EU HTA Regulation.
• The guidance sets out principles of accountability, transparency, and reporting, emphasizing that health technology developers remain responsible for dossier content and that appropriate human oversight is required throughout preparation.
• The document aims to support the responsible use of AI while maintaining the scientific quality, methodological rigor, and legal compliance of JCA submissions.

As implementation of the EU Health Technology Assessment Regulation (EU HTAR) progresses, the Member State Coordination Group on Health Technology Assessment (HTACG) has issued its first guidance on the use of AI during the preparation of Joint Clinical Assessment (JCA) dossiers. The guidance is intended to support the responsible use of AI while making clear that health technology developers (HTDs) remain responsible for the scientific quality, accuracy, and completeness of their submissions, regardless of whether AI is used.

The guidance applies specifically to HTDs, whose dossiers form the basis of JCAs, rather than to the assessment process itself. It reiterates that dossiers must remain complete with respect to the available studies and data relevant to the assessment and that evidence must continue to be analyzed using appropriate methods to answer all assessment scope questions. The document also makes clear that the use of AI does not change developers' legal obligations or the applicability of existing HTACG methodological guidance.

For the purposes of the guidance, AI is defined as all machine learning models, including generative AI models, regardless of model architecture. While the HTACG recognizes that AI "has the potential to increase the efficiency of processes which are part of JCA," it also warns that it "carries the risk of compromising the completeness, methodological quality and rigor of a JCA." The guidance therefore states that AI "should be used in an accountable way and transparently to ensure its outputs are accurate and reliable, thus upholding the scientific quality of all the steps that lead to a JCA."

The guidance is built around three overarching principles:

1. Accountability
2. Transparency
3. Reporting

Under the principle of accountability, HTDs remain responsible for the content, methods, and findings of the evidence synthesis presented within the dossier, including decisions about whether and how AI is used, the impact of AI on the evidence synthesis, and "the validity of any output resulting from AI use." Developers are also expected to ensure that AI complies with applicable legal requirements, including the EU AI Act, copyright, and data protection legislation, recognizing that JCA dossiers will ultimately be published.

The HTACG places particular emphasis on human oversight, stating that "human oversight should be maintained throughout any steps which are assisted by AI in the preparation of the JCA dossier" and that "none of the steps in the dossier preparation or analysis process should be fully automated without a human being ultimately responsible for the quality and accuracy." It also notes that when AI is used, it should remain possible to verify compliance with existing methodological guidance.

Under the principle of transparency, developers are expected to clearly document where AI has been used during dossier preparation. Examples include information retrieval and study screening, data extraction, risk of bias assessment, analysis, and reporting. The guidance further states that "any outputs or judgments that are either AI-generated or AI-informed... should be transparently identified in the dossier," while clarifying that AI used solely to improve spelling and grammar does not need to be disclosed.

Under the reporting principle, dossiers must identify, at a minimum, the AI tool used, including its name, version, date, developer, and intended purpose. The guidance also identifies documenting adaptations made to commercially available AI tools as good practice and states that prompts used during dossier preparation should be recorded and made available if requested during the JCA process.

The guidance comes as AI is being adopted across a growing range of evidence generation activities supporting HTA. Pharmaceutical companies and evidence developers are increasingly exploring AI to support systematic literature reviews, evidence synthesis, medical writing, data extraction, regulatory documentation, and activities supporting JCA readiness, including early PICO identification and prioritization, automated literature screening, comparative evidence synthesis, and the analysis of large real-world datasets that may ultimately contribute to HTA submissions.

Although the guidance does not address these individual applications directly, it sets expectations for how AI should be used whenever it contributes to the preparation of a JCA dossier. The publication also reflects a broader movement towards establishing governance for AI across HTA. Recent work has explored how AI can support literature reviews, evidence synthesis, medical writing, and early JCA preparation while emphasizing the importance of appropriate safeguards. Speaking to The Evidence Base earlier this year, Dalia Dawoud, Research Principal, HTA Policy and Strategy, Cytel and a member of ISPOR’s GenAI for HEOR SLRs Task Force said that reporting standards for AI were intended "primarily to be guardrails due to the critical concerns around transparency, reproducibility, and trustworthiness of AI."

Rather than restricting the use of AI, the HTACG guidance establishes principles for its responsible application during JCA dossier preparation. As AI becomes more widely integrated into evidence generation, the emphasis is on ensuring that AI-assisted work remains transparent, scientifically robust, and subject to appropriate human oversight, with HTDs retaining responsibility for the quality and integrity of every submission.