Canada’s Drug Agency sets direction on AI in health technology assessment with new position statement

Canada’s Drug Agency has published a new position statement outlining potential uses, risks, and responsibilities for AI-informed evidence submissions in health technology assessment (HTA).

The guidance is intended to support both internal and external stakeholders in the appraisal of AI-informed evidence, with a strong focus on transparency, appropriateness, and compliance with Canadian regulatory and ethical standards.

“Our organization recognizes that the evidence we assess to help shape our recommendations and advice will increasingly be informed by AI,” the agency stated. “Because of this, we have taken proactive steps to better understand the use of AI methods within the context of HTA and developed a position statement, which details how we will consider evidence that has been generated or reported using AI methods.”

The position statement is structured in two parts. The first outlines how AI may be applied across HTA processes, while the second addresses the responsibilities of those submitting AI-supported evidence.

Potential applications of AI in HTA

AI may support various components of systematic review and evidence synthesis. The agency notes that, “AI methods have the potential to automate various steps in these processes,” including generating search strategies, classifying study designs, screening records, and extracting data. Large language models (LLMs) could also be used to synthesize data into meta-analyses or produce summaries, though these applications are less established.

In clinical evidence generation, AI can assist in trial design, optimizing inclusion criteria, sample size, and retention strategies. Natural language processing may be used to mine electronic health records and generate lay or executive summaries. Additionally, AI can help adjust for data limitations and model complex relationships for improved causal inference.

AI also has growing relevance in real-world data (RWD) analysis and the generation of real-world evidence (RWE). According to the agency, “AI approaches have several potential roles for supporting RWE across numerous stages of evidence generation,” including data cleaning, linkage, and integration of multimodal data sources. These methods can enhance the precision and scalability of analyses, such as estimating treatment effects using advanced machine learning techniques.

For cost-effectiveness evidence, AI may support model conceptualization, calibration, and simulation. LLMs could automate the construction of economic models and facilitate real-time updates. Machine learning can also help optimize simulations, reducing computation time and improving the feasibility of more complex models.

While these applications highlight AI’s potential, the agency stresses that this list, “is neither exhaustive nor [an] endorsement or acceptance of those methods.”

Responsibilities and risk management

The second section of the guidance sets expectations for evidence submitters. “Any potential benefits of using AI methods must be balanced against anticipated and/or known risks,” the agency states, citing concerns such as algorithmic bias, cybersecurity threats, and reduced human oversight.

The agency emphasizes the importance of explainability, accountability, and transparency. “The use of AI methods may introduce added complexity. It is important that the user... clearly justify the use of these methods and outline assumptions... and consider the plausibility of their results.”

Submitters should follow relevant Canadian frameworks, including the Voluntary Code of Conduct on the Responsible Development and Management of Advanced Generative AI Systems and forthcoming legislation such as the Artificial Intelligence and Data Act (AIDA). They are also expected to assess intellectual property rights, licensing agreements, and data protection obligations.

“A capable and informed human in the loop” should remain central to the process, and users must declare AI use, explain the methods, and ensure compliance with all applicable standards. Security risks, such as “data poisoning” or “prompt injection attacks,” must be actively mitigated.

Looking ahead

To inform the development of its statement, the agency referenced a similar document from the UK’s National Institute for Health and Care Excellence (NICE). “While we aligned our statement with all of the positions outlined by NICE, minor additions and modifications were made to contextualize the content to our own organization’s work and the broader HTA environment in Canada.”

The position statement is intended to support both internal deliberative committees and external assessment groups in evaluating AI-supported submissions. It will be reviewed and updated as new evidence and technologies emerge. Updates may also be integrated into the agency’s Methods Guide, which outlines the processes used in HTA activities.

The release forms part of a broader focus by Canada’s Drug Agency on understanding the evolving role of AI in health care. “Canada's Drug Agency has been actively exploring the use of AI and its potential impact on the broader healthcare ecosystem,” the agency noted. Earlier this month, it published its 2025 Watch List, highlighting technologies such as AI for notetaking, disease diagnosis, and remote monitoring, along with emerging implementation challenges.