Presentation (CTI)

OBJECTIVES: With the introduction of the Joint Clinical Assessment (JCA) as part of the EU Health Technology Assessment Regulation (Regulation (EU) 2021/2282), early preparation has become crucial for assessing the potential Population, Intervention, Comparator, and Outcome (PICO) criteria that will be requested by member states. Efficient methods of navigating the data to estimate likely PICOs for a given product are required. This study aimed to assess if an AI tool can predict PICOs in oncology.
METHODS: A process leveraging natural language processing techniques, was designed to support early health technology assessment (HTA) planning by identifying PICOs based on precedent HTA assessments and clinical guidelines from select EU member states and non-EU members in Canada and the UK. The process mimics a manual review in which analysts identify patterns, such as commonly accepted comparators and population definitions, to inform expected PICO criteria. These elements are then ranked using a rules-based algorithm to identify likely PICOs. To assess the accuracy of this approach, outputs were assessed against the PICO exercises published by the Member State Coordination Group on Health Technology Assessment (HTA CG) in February 2025.
RESULTS: For advanced or unresectable hepatocellular carcinoma in the first line, the AI-driven process instantly identified all comparators, 93% of outcomes and 77% of populations and comparator combinations that were identified in the HTA CG exercise. For KRAS G12C mutated non-small cell lung cancer after at least one prior systemic therapy, the AI-driven process identified all comparators, 93% of outcomes and 92% of populations and comparator combinations that were identified in the HTA CG exercise.
CONCLUSIONS: With this methodology, an accurate view of the PICO landscape for treatment can be developed instantly enabling a more rapid and efficient JCA preparation process.