Presentation (CTI)

OBJECTIVES: To evaluate the feasibility and performance of a fully automated, AI-driven pipeline for the identification, summarization, and dissemination of new literature relevant to Health Economics and Outcomes Research (HEOR), and to propose a scalable framework for future AI-assisted evidence communications.
METHODS: A proof-of-concept system was developed, integrating advanced machine learning (ML) classifiers for literature screening and a generative AI engine for content synthesis. The system was configured to identify studies relevant to predefined HEOR topics using a custom classifier trained on PICOTS criteria. Newly indexed publications were retrieved from major bibliographic databases over a 3-month period. The classifier’s performance was assessed against a gold standard set of manually screened articles. Selected articles were processed by a generative AI module, which produced structured summaries and narrative syntheses. Outputs were evaluated for accuracy, consistency, and readability by a panel of HEOR experts.
RESULTS: The AI classifier achieved a mean specificity of 99% and sensitivity of 87% in validation against the reference set. The generative AI module produced newsletter-style summaries judged to be accurate and consistent with manual extractions in 93% of cases. Automated outputs demonstrated high readability and required minimal post-editing. The modular framework supported rapid adaptation to new HEOR topics and evolving inclusion criteria.
CONCLUSIONS: This pilot study demonstrates the feasibility of an end-to-end AI-driven literature review and synthesis pipeline for HEOR communications. The approach shows strong potential to reduce manual workload and accelerate knowledge dissemination, while maintaining methodological rigor. Future work will focus on expanding use cases, refining content evaluation metrics, and exploring integration with existing HTA and HEOR workflows.