Creating an Indexing Scheme for Case Series Articles

Area of Science:

• Medical Informatics
• Bibliometrics
• Biomedical Literature Analysis

Background:

• Case series are a significant part of biomedical literature but are not indexed as a distinct publication type by the National Library of Medicine.
• This lack of specific indexing hinders the retrieval and analysis of evidence from case series by clinicians and researchers.

Purpose of the Study:

• To develop and evaluate an automated method for indexing case series articles.
• To improve the discoverability and medical value of case series in the biomedical literature.

Main Methods:

• A corpus of PubMed articles mentioning "case series" was curated, excluding those better classified as other publication types or meta-analyses.
• A transformer-based machine learning model was trained and tested on this corpus.
• Manual evaluation of a sample of articles confirmed the accuracy of the automated indexing.

Main Results:

• The automated indexing model demonstrated excellent performance on hold-out data (precision = 0.887, recall = 0.952, F1 = 0.918).
• Manual review of 100 articles tagged as "case series" by the model showed that 88% met a formal definition of case series.
• The study confirmed the feasibility of automatically indexing case series.

Conclusions:

• Automated indexing of case series is feasible and effective.
• Enhanced discoverability of case series will increase their utility for evidence synthesis and general biomedical literature users.