A large-scale dataset of structured clinical case narratives and question-answer pairs from case reports

This dataset is a large-scale linked resource derived from full-text PMC case reports. It contains structured article-level records, case-level clinical narratives, and linked question-answer items for downstream biomedical NLP and clinical reasoning tasks.

The current release contains 102,489 retained PMC case report articles, 135,149 structured case records, and 828,985 linked QA items.

The release includes five core record types: article-level records preserving titles, abstracts, and hierarchical body sections; case-number records defining article-to-case structure; case-details records storing structured clinical narratives; task-plan records documenting QA planning; and QA-level records containing diagnosis SBA, management SBA, NLI judgement, and open-ended clinical QA items.

Records are linked through stable article_id, case_id, and qa_id identifiers. The dataset is intended to support case-grounded question answering, structured clinical narrative analysis, information extraction, retrieval, summarization, and biomedical reasoning research.

The release also includes documentation, schemas, and summary files for reuse and validation.