Agentic induction of common data elements and information extraction from sarcoma case report abstracts
arXiv cs.CL · 14 Nov 2026, 04:23 AEST
Large language model agents successfully derived a common data element schema from sarcoma case report abstracts. The study demonstrated high accuracy in information extraction, with a 94.9% agreement rate in sampled abstracts.
Key facts
- An agent proposed 72,909 candidate CDEs from 24,588 PubMed abstracts; agentic consolidation yielded 1,570 CDEs.
- We retained 53 supported by at least 50 abstracts and attempted evidence-linked extraction across the corpus.
- Processing succeeded for all retained CDEs in 79.2% of abstracts, including responses indicating non-mention.
- Two human reviewers approved 83.0% of retained CDEs for relevance, 67.9% for clarity, and 56.6% for correctness.
- Errors concentrated in anatomical and temporal fields.
Read aloud
Large language model agents successfully derived a common data element schema from sarcoma case report abstracts. The study demonstrated high accuracy in information extraction, with a ninety four point nine percent agreement rate in sampled abstracts. An agent proposed seventy two thousand nine hundred nine candidate CDEs from twenty four thousand five hundred eighty eight PubMed abstracts; agentic consolidation yielded one thousand five hundred seventy CDEs. We retained fifty three supported by at least fifty abstracts and attempted evidence-linked extraction across the corpus. Processing succeeded for all retained CDEs in seventy nine point two percent of abstracts, including responses indicating non-mention. Two human reviewers approved eighty three point zero percent of retained CDEs for relevance, sixty seven point nine percent for clarity, and fifty six point six percent for correctness. Errors concentrated in anatomical and temporal fields.
Sources in this cluster
- Agentic induction of common data elements and information extraction from sarcoma case report abstracts
arXiv cs.CL · 14 Nov 2026, 04:23 AEST