Abstract
Clinical case reports (CCRs) provide an important means of sharing clinical experiences about atypical disease phenotypes and new therapies. However, published case reports contain largely unstructured and heterogeneous clinical data, posing a challenge to mining relevant information. Current indexing approaches generally concern document-level features and have not been specifically designed for CCRs. To address this disparity, we developed a standardized metadata template and identified text corresponding to medical concepts within 3,100 curated CCRs spanning 15 disease groups and more than 750 reports of rare diseases. We also prepared a subset of metadata on reports on selected mitochondrial diseases and assigned ICD-10 diagnostic codes to each. The resulting resource, Metadata Acquired from Clinical Case Reports (MACCRs), contains text associated with high-level clinical concepts, including demographics, disease presentation, treatments, and outcomes for each report. Our template and MACCR set render CCRs more findable, accessible, interoperable, and reusable (FAIR) while serving as valuable resources for key user groups, including researchers, physician investigators, clinicians, data scientists, and those shaping government policies for clinical trials.
Introduction
Clinical Case Reports (CCRs) are fundamental to medical literature, often acting as the primary evidence for rare diseases, unexpected drug reactions, or innovative treatments. Despite their value, CCRs are notoriously difficult to utilize for large-scale research. Most exist as unstructured free-text, and standard indexing systems—such as PubMed’s MeSH terms—typically provide only document-level summaries.
Currently, a researcher seeking specific cases—for instance, a 40-year-old male with a rare mitochondrial disease treated with a specific pharmaceutical—must manually review hundreds of individual documents. This creates a significant "knowledge gap" between the clinical data generated and our ability to transform it into structured, searchable knowledge.
Methodology
To bridge this gap, the researchers developed a standardized metadata template designed to make CCR data FAIR (Findable, Accessible, Interoperable, and Reusable). The methodology integrated human clinical expertise with computational validation across several stages.
The process began with the identification of 15 key clinical concepts, ranging from demographics and signs/symptoms to interventional therapies and patient outcomes. A team of 12 medical experts, including clinical fellows and researchers, then manually reviewed 3,100 reports. Rather than simply tagging documents, they extracted specific text segments corresponding to each clinical concept. Finally, the extracted data was standardized by mapping it to formal ontologies such as ICD-10-CM and MeSH.
The MACCRs Dataset
The primary output of this work is the Metadata Acquired from Clinical Case Reports (MACCRs) dataset. This resource provides a "gold standard" for the medical community, containing over 32,000 expert-annotated clinical concepts.
The dataset includes:
- 3,100 curated CCRs covering 15 major disease groups.
- 750+ reports on rare diseases, offering critical insights into "small n" conditions.
- The RMD Subset: A specialized collection of 246 mitochondrial disease reports, enriched with specific ICD-10 diagnostic codes.
- Validation Corpus: The authors validated the MACCRs set against 1.89 million CCRs indexed in MEDLINE to ensure the sample was representative of the broader medical literature.
Key Insights and Results
The structuring of metadata across thousands of reports allows for the creation of "in silico" cohorts. This enables researchers to build virtual patient groups for rare diseases that would otherwise be impossible to study due to low patient numbers at individual institutions.
Analysis of the dataset revealed that sections such as "Diagnostic Techniques" and "Signs and Symptoms" possess the highest information entropy. This indicates these sections are linguistically complex and contain the most unique data, making them high-value targets for future Natural Language Processing (NLP) mining. Furthermore, the MACCRs set provides an average of 10+ new clinical details per report that are entirely missed by standard PubMed indexing.
In the mitochondrial disease subset, the dataset successfully characterized 500 unique ICD-10 codes. This revealed common but often under-indexed symptoms, such as acidosis and cardiomyopathy, providing a more holistic view of these complex conditions.
Conclusion and Future Directions
While manual curation is labor-intensive, the MACCRs dataset serves as a vital training set for the next generation of machine learning and NLP models. By providing a high-quality, structured reference, this work paves the way for automated clinical extraction tools.
Future efforts will focus on scaling these pipelines and encouraging clinicians to adopt the MACCRs template during the initial writing of reports. Addressing geographic biases—currently skewed toward the US, Japan, and China—remains a priority to ensure the dataset reflects global clinical diversity.
Summary updated