Expanding global access to oncology trials: mCODE-aligned AI for inclusive patient matching across literacy and resource settings.

18

Background: Clinical trial enrollment remains a global bottleneck in oncology. Despite therapeutic advances, only a minority of eligible patients enroll, largely due to fragmented electronic health records (EHRs), manual chart abstraction, and health literacy barriers. These limitations disproportionately affect patients in under-resourced settings and globally diverse populations. We evaluated an AI-powered platform designed to automatically extract oncology data from EHRs and structure them according to the Minimal Common Oncology Data Elements (mCODE) framework to enable scalable, interoperable trial matching. Methods: A fine-tuned GPT-4o model was developed to process structured and unstructured oncology EHR data. In a validation cohort of 102 patients (randomly sampled from a 3,800-patient oncology database), the system extracted tumor type, stage, disease extent, relapse status, resectability, and genomic biomarkers. Variables were mapped to mCODE 3.0 profiles (Cancer Disease Status, Tumor Characteristics, Genomics) and validated against expert chart review. Primary endpoints were accuracy (concordance) and completeness. Secondary endpoints evaluated interoperability and readiness for downstream clinical trial matching. Results: AI achieved high accuracy for tumor type (98%), extent of disease (90%), and stage (86%). Genomic extraction demonstrated 78% accuracy, reliably identifying common alterations including BRCA1/2 and TP53. Lower concordance in relapse status (77%) and resectability (69%) reflected documentation variability in free-text notes. All outputs were successfully structured within mCODE schemas. Compared with manual abstraction, AI reduced processing time substantially while maintaining clinical validity. Standardized outputs enabled seamless integration into multi-lingual, plain-language trial matching interfaces. Conclusions: AI-driven oncology data standardization aligned with mCODE enables scalable, real-time clinical trial matching across diverse health systems. By embedding interoperability at the data-extraction stage, this framework directly addresses structural and literacy-based barriers to trial access. These findings support the feasibility of AI-assisted infrastructure to democratize precision oncology research beyond geographic and language constraints.