Clinical LLM Inventory and E-CAF Scores Supporting Data for "A Technical Taxonomy and Evidence-Based Framework for Assessing Clinical Large Language Model Systems"

OVERVIEW This dataset accompanies the paper "A Technical Taxonomy and Evidence-Based Framework for Assessing Clinical Large Language Model Systems." It contains the systematic inventory of clinical Large Language Model (LLM) systems and their Evidence-Based Capability Alignment Framework (E-CAF) scores used to support the analysis. FILES • inventory.csv — 42 clinical LLM systems identified through a structured PubMed search. Columns: system, source_paper. • ecaf_scores.csv — Binary E-CAF scores (0/1) per system across six capability dimensions, with textual justifications from the source papers. Columns: system, taxonomy, dimension, score, justification. 252 entries (42 systems × 6 dimensions). • README_dataset.txt — Detailed column definitions, taxonomy abbreviations (H-LLM, V-LLM, H-AS, V-AS), and scoring rubric. E-CAF DIMENSIONS • Clinical Specialty Grounding (C) • Clinical Time-Series Analysis (L) • Systemic Architecture Complexity (A) • Real-World Data Validation (R) • Traceable Governance and Accountability (G) • Adaptive Post-Deployment Alignment (P) METHODOLOGY Records were retrieved via PubMed using a structured search string with the free full-text filter (108 initial results). Screening against inclusion/exclusion criteria — focus on LLM-based clinical decision support; exclusion of genomics, mental health, psychiatry, dentistry, oral health, pediatrics, and cadaver studies — reduced these to 81 candidates and, after deduplication of substantially similar deployments, to a final inventory of 42 representative tools. Each tool was then scored on the six E-CAF dimensions using a binary rubric (0 = no documented evidence, 1 = documented evidence), with justifications drawn from the source paper. REUSE The dataset supports replication of the paper's findings and can be reused to (a) extend the inventory with new clinical LLM systems, (b) re-score existing systems under updated criteria, or (c) benchmark new evidence-reporting frameworks against E-CAF. FUNDING This project has received funding from the European Union's Horizon Europe research and innovation programme under the Marie Skłodowska-Curie grant agreement No. 101168344. Views and opinions expressed are those of the authors only and do not necessarily reflect those of the European Union or the European Research Executive Agency (REA). Neither the European Union nor the granting authority can be held responsible for them. RELATED PUBLICATION The accompanying paper is currently under review at AIDI 2026. A DOI will be added as a related identifier upon publication.