Presentation (CTI)

OBJECTIVES: Healthcare decision-making relies on consistency across clinical evidence, economic evidence in HTA submissions. However, contradictions in assumptions, definitions, or conclusions across documents often go undetected until late stages. This study evaluated an AI-supported approach to systematically identify internal and cross-document inconsistencies within HTA evidence packages.
METHODS: An AI-based document analysis framework was applied to linked sets of clinical study reports, economic model documentation, and HTA submissions for selected case studies. Documents were parsed to extract key claims, assumptions, definitions, and quantitative statements. Logical and semantic consistency checks were performed across documents to flag potential contradictions, including mismatched population definitions, inconsistent time horizons, conflicting clinical assumptions, and divergent interpretations of uncertainty. Identified issues were reviewed by senior HTA experts for relevance and severity.
RESULTS: The AI framework identified multiple categories of contradictions across documents, including inconsistencies in comparator definitions, outcome measurement timepoints, subgroup specifications, and interpretation of uncertainty. Several issues were classified as high-impact, with potential implications for HTA credibility if left unresolved. Expert review confirmed that many flagged contradictions would have been difficult to detect through standard manual review due to document volume and fragmentation. False positives were primarily related to contextual nuance requiring human interpretation.
CONCLUSIONS: AI-supported contradiction detection can strengthen internal coherence of HTA evidence packages by systematically identifying misalignments across clinical, economic, and HTA documents. Used as a quality assurance tool alongside expert review, this approach may reduce downstream HTA risk and improve transparency and consistency in evidence submissions.