Algorithmic indexing in MEDLINE frequently overlooks important concepts and may compromise literature search results

Abstract

Objective: To evaluate the appropriateness of indexing of algorithmically-indexed MEDLINE records.

Methods: We assessed the conceptual appropriateness of Medical Subject Headings (MeSH) used to index a sample of MEDLINE records from February and March 2023. Indexing was performed by the Medical Text Indexer-Auto (MTIA) algorithm. The primary outcome measure is the number of records for which the MTIA algorithm assigned subject headings that represented the main concepts of the publication. 

Results: Fifty-three percent of screened records had indexing that represented the main concepts discussed in the article; 47% had inadequacies in the indexing which could impact their retrieval. Three main issues with algorithmically-indexed records were identified: 1) inappropriate MeSH assigned due to acronyms, evocative language, exclusions of populations, or related records; 2) concepts represented by more general MeSH while a more precise MeSH is available; and 3) a significant concept not represented in the indexing at all. We also noted records with inappropriate combinations of headings and subheadings, even when the headings and subheadings on their own were appropriate.

Conclusions: The indexing performed by the February-March 2023 calibration of the MTIA algorithm, as well as older calibrations, frequently applied irrelevant or imprecise terms to publications while neglecting to apply relevant terms. As a consequence, relevant publications may be omitted from search results and irrelevant ones may be retrieved. Evaluations and revisions of indexing algorithms should strive to ensure that relevant, accurate and precise MeSH terms are applied to MEDLINE records.