PROSPECTIVE AI-ENABLED PICO PREDICTION FOR EU JOINT CLINICAL ASSESSMENT: BLINDED VALIDATION USING THE FIRST PUBLISHED JCA REPORT ON TOVORAFENIB
Author(s)
Turgay Ayer, PhD1, Sumeyye Samur, PhD1, Ismail Fatih Yildirim, MSc1, Mine Tekman, PhD1, Jag Chhatwal, PhD2.
1Value Analytics Labs, Boston, MA, USA, 2Associate Professor and Director of the Institute for Technology Assessment, Harvard Medical School / Massachusetts General Hospital, Boston, MA, USA.
1Value Analytics Labs, Boston, MA, USA, 2Associate Professor and Director of the Institute for Technology Assessment, Harvard Medical School / Massachusetts General Hospital, Boston, MA, USA.
OBJECTIVES: To evaluate whether an AI-enabled, multi-agent platform can prospectively predict PICO frameworks for EU Joint Clinical Assessments (JCAs) under Regulation (EU) 2021/2282, using the first published JCA report on tovorafenib (Ojemda) for pediatric low-grade glioma as an external validation benchmark, under conditions designed to prevent data leakage.
METHODS: ValueGen.AI, an agentic AI architecture was applied to tovorafenib for BRAF-altered relapsed/refractory pediatric low-grade glioma (pLGG). To prevent data leakage, the AI platform was restricted to pre-JCA sources only: national HTA guidance, clinical practice guidelines, pivotal trial evidence, and regulatory documents from European HTA bodies including NCPE (Ireland), IQWiG (Germany), HAS (France), and AIFA (Italy). The published JCA report -- endorsed by the HTA Coordination Group on April 30, 2026 and published June 9, 2026, defining 3 populations and 8 PICOs -- was withheld from the AI and used solely as the validation benchmark. Concordance was assessed across populations, comparators, and outcomes.
RESULTS: The AI platform correctly predicted all 3 molecularly stratified population definitions (BRAF fusion/rearrangement, BRAF V600E mutation [age >1 year], and BRAF V600 non-E/fusion [age >=6 months]). Key comparators were anticipated, including individualized chemotherapy regimens (vinblastine; carboplatin plus vincristine; TPCV) and dabrafenib plus trametinib, though some less common comparators (e.g., everolimus, bevacizumab plus chemotherapy) were not predicted. Outcomes aligned with the JCA, including overall survival, progression-free survival, objective response, health-related quality of life, disease-specific symptoms, and safety.
CONCLUSIONS: Under conditions designed to prevent data leakage from the published JCA, AI-enabled PICO scoping demonstrated high concordance with the first EU JCA report. This approach may help health technology developers prospectively anticipate JCA scope and streamline evidence preparation under the EU HTA Regulation.
METHODS: ValueGen.AI, an agentic AI architecture was applied to tovorafenib for BRAF-altered relapsed/refractory pediatric low-grade glioma (pLGG). To prevent data leakage, the AI platform was restricted to pre-JCA sources only: national HTA guidance, clinical practice guidelines, pivotal trial evidence, and regulatory documents from European HTA bodies including NCPE (Ireland), IQWiG (Germany), HAS (France), and AIFA (Italy). The published JCA report -- endorsed by the HTA Coordination Group on April 30, 2026 and published June 9, 2026, defining 3 populations and 8 PICOs -- was withheld from the AI and used solely as the validation benchmark. Concordance was assessed across populations, comparators, and outcomes.
RESULTS: The AI platform correctly predicted all 3 molecularly stratified population definitions (BRAF fusion/rearrangement, BRAF V600E mutation [age >1 year], and BRAF V600 non-E/fusion [age >=6 months]). Key comparators were anticipated, including individualized chemotherapy regimens (vinblastine; carboplatin plus vincristine; TPCV) and dabrafenib plus trametinib, though some less common comparators (e.g., everolimus, bevacizumab plus chemotherapy) were not predicted. Outcomes aligned with the JCA, including overall survival, progression-free survival, objective response, health-related quality of life, disease-specific symptoms, and safety.
CONCLUSIONS: Under conditions designed to prevent data leakage from the published JCA, AI-enabled PICO scoping demonstrated high concordance with the first EU JCA report. This approach may help health technology developers prospectively anticipate JCA scope and streamline evidence preparation under the EU HTA Regulation.
Conference/Value in Health Info
2026-11, ISPOR Europe 2026, Vienna, Austria
Value in Health, Volume 29, Issue 12S
Code
MSR31
Topic
Methodological & Statistical Research
Topic Subcategory
Artificial Intelligence, Machine Learning, Predictive Analytics
Disease
Oncology, Pediatrics