Presentation (CTI)

OBJECTIVES: The introduction of Joint Clinical Assessment (JCA) in Europe poses significant challenges for pharmaceutical developers, particularly in demonstrating the value of treatments against all existing management strategies across European member states. A key preparatory step for a successful JCA review is the provision of comparative data, either directly collected or synthesised. Given the strict timelines, developers must ensure early mapping of PICO (Population, Intervention, Comparator, Outcomes) questions to facilitate evidence synthesis. An AI-powered algorithm has been developed to identify PICO questions, requiring validation against traditional targeted searches.
METHODS: Three orphan diseases, cystinosis, urea cycle disorders, and tyrosinemia, were selected to test the algorithm. These diseases were chosen due to their similarity as inborn errors of metabolism. A targeted review for all EU member states was conducted for each disease to determine PICO questions, which were then compared to those generated by the AI algorithm. The time spent developing PICO questions was recorded, with a maximum of 4 hours allocated for targeted searching across all 27 EU member states. For validation, the "intervention" in PICO was regarded as a new intervention.
RESULTS: The largest differences in PICO formulation were observed in the comparator (C) category. Both methods identified common elements for the patient (P) and outcome (O) categories. However, targeted searches failed to identify 25-33% of comparator strategies highlighted by the algorithm. Each AI-driven search was completed in under two minutes, with an additional 60 minutes per disease required for consistency checks. These findings suggest the algorithm achieves higher accuracy than traditional methods while significantly reducing time requirements.
CONCLUSIONS: The AI-powered algorithm demonstrates potential as a more accurate and resource-efficient alternative to traditional methods. However, further validation is necessary, it should be used alongside human-led searches to ensure robustness, with subsequent testing of PICO outputs by clinicians in relevant markets to confirm accuracy.