Presentation (CTI)

OBJECTIVES: Agentic artificial intelligence (AI), defined as systems capable of autonomous goal formulation, multi-step task execution, and adaptive decision-making, is increasingly described in healthcare analytics and research workflows. While these capabilities enable advanced automation in evidence generation, their application in health economics and outcomes research (HEOR) raises challenges related to transparency, reproducibility, and regulatory governance. This study aimed to develop a risk-stratified framework to support the responsible use of agentic AI in HEOR, aligned with good research practices and health technology assessment (HTA) requirements.
METHODS: A structured narrative synthesis was conducted using peer-reviewed literature and policy frameworks (2019-2025), including sources on agent-based AI systems, large language model-based agents, AI governance models, and HEOR methodological guidance. Evidence was reviewed to characterize: (1) applications of agentic AI in healthcare and HEOR; (2) risks associated with autonomy, opacity, and system evolution; and (3) governance approaches for responsible deployment. Findings were mapped to key HEOR functions, including evidence synthesis, economic modelling, and decision support, and synthesized into a conceptual governance framework.
RESULTS: Agentic AI shows applicability across multiple HEOR use cases, including iterative evidence synthesis, simulation-based modelling, and scenario analysis. Key risk domains include limited transparency due to non-deterministic behaviour, uncontrolled system evolution, data governance risks (e.g., potential leakage), and challenges related to auditability and accountability. The proposed framework comprises four governance pillars: (1) risk-proportionate autonomy aligned to decision impact; (2) human-in-the-loop oversight at defined checkpoints; (3) structured documentation of data, models, and decision processes; and (4) integration with organizational and regulatory governance structures.
CONCLUSIONS: The use of agentic AI in HEOR requires a shift toward governance-centered implementation. Embedding proportional autonomy, human oversight, and transparency within HEOR workflows can support responsible adoption while preserving scientific validity, reproducibility, and trust in healthcare decision-making.