GLOBAL TRENDS AND DISPARITIES IN THE USE OF CLINICAL OUTCOME ASSESSMENTS IN CLINICALTRIALS.GOV, 2015-2025: AN AI-ENABLED LANDSCAPE ANALYSIS
Author(s)
Pauline Flajolet, PharmD, PhD1, Francesca De Giorgio, PhD1, Tilly Stott, MPH1, Sylwia Chorylo, MA1, Karolina Bogusz, PhD1, Caprice Sassano, MPH2, Nadine Gabriele Kraft, MSc1, Hannah Lewis, PhD3, Gilyana Borlikova, PhD4.
1Mapi Research Trust, Lyon, France, 2ICON plc, Blue Bell, PA, USA, 3ICON plc, London, United Kingdom, 4ICON plc, Dublin, Ireland.
1Mapi Research Trust, Lyon, France, 2ICON plc, Blue Bell, PA, USA, 3ICON plc, London, United Kingdom, 4ICON plc, Dublin, Ireland.
OBJECTIVES: Clinical Outcome Assessments (COAs) are essential to patient-centered research, yet their use at scale remains poorly understood. Previous analyses have been limited to specific therapeutic areas or predefined instrument lists. We aimed to provide the first large-scale analysis of COA use across ten years of ClinicalTrials.gov and to evaluate how different COA definitions impact prevalence estimates.
METHODS: We sourced 374,024 studies from ClinicalTrials.gov (all trials with first posted date between 1-Jan-2015 to 6-Nov-2025). A specialized AI pipeline extracted COAs from outcome measures and reconciled the named instruments to PROQOLID, the reference-standard library of COAs. Analyses examined COA use by study characteristics, endpoint positioning and therapeutic areas from MeSH. Sensitivity analyses compared broad AI-extracted and stricter reconciled COA definitions. Manual review was conducted on a randomized sample of 1000 trials to validate performance.
RESULTS: The proportion of trials including at least one COA extracted by AI increased from 45.6% in 2015 to 56.8% in 2025. COA use was higher in interventional versus observational studies (56.9% vs 35.8%). Marked therapeutic differences were observed, with highest prevalence in mental disorders (83.5%) and nervous system diseases (70.5%), and lower use in neoplasms (42.8%). Among all trials using COAs, 61.4% included a COA as a primary endpoint and 71.9% as a secondary endpoint. Stricter PROQOLID-reconciled definitions reduced absolute COA prevalence in trials from 29.1% (2015) to 40.3% (2025), while preserving temporal and structural trends across study characteristics.
CONCLUSIONS: COAs are increasingly embedded in clinical trials but remain unevenly distributed across therapeutic areas and study designs. AI enables a comprehensive and scalable view of COA use across clinical research. Although absolute prevalence estimates varied substantially across COA definitions, temporal and structural patterns remained consistent. These findings support the robustness of the landscape analysis and highlight the need for clearer COA classification and taxonomy frameworks for large-scale automated analyses.
METHODS: We sourced 374,024 studies from ClinicalTrials.gov (all trials with first posted date between 1-Jan-2015 to 6-Nov-2025). A specialized AI pipeline extracted COAs from outcome measures and reconciled the named instruments to PROQOLID, the reference-standard library of COAs. Analyses examined COA use by study characteristics, endpoint positioning and therapeutic areas from MeSH. Sensitivity analyses compared broad AI-extracted and stricter reconciled COA definitions. Manual review was conducted on a randomized sample of 1000 trials to validate performance.
RESULTS: The proportion of trials including at least one COA extracted by AI increased from 45.6% in 2015 to 56.8% in 2025. COA use was higher in interventional versus observational studies (56.9% vs 35.8%). Marked therapeutic differences were observed, with highest prevalence in mental disorders (83.5%) and nervous system diseases (70.5%), and lower use in neoplasms (42.8%). Among all trials using COAs, 61.4% included a COA as a primary endpoint and 71.9% as a secondary endpoint. Stricter PROQOLID-reconciled definitions reduced absolute COA prevalence in trials from 29.1% (2015) to 40.3% (2025), while preserving temporal and structural trends across study characteristics.
CONCLUSIONS: COAs are increasingly embedded in clinical trials but remain unevenly distributed across therapeutic areas and study designs. AI enables a comprehensive and scalable view of COA use across clinical research. Although absolute prevalence estimates varied substantially across COA definitions, temporal and structural patterns remained consistent. These findings support the robustness of the landscape analysis and highlight the need for clearer COA classification and taxonomy frameworks for large-scale automated analyses.
Conference/Value in Health Info
2026-11, ISPOR Europe 2026, Vienna, Austria
Value in Health, Volume 29, Issue 12S
Code
MSR118
Topic
Clinical Outcomes, Methodological & Statistical Research, Study Approaches
Topic Subcategory
Artificial Intelligence, Machine Learning, Predictive Analytics
Disease
No Additional Disease & Conditions/Specialized Treatment Areas