Developing and validating PubMed infant hedges for PubMed and Ovid MEDLINE: a Medical Library Association pediatrics librarians caucus initiative

Abstract

Background: To support evidence synthesis and clinical searching, a team of librarians developed and validated infant age (birth to 23 months) search hedges for PubMed (National Library of Medicine) and Medline (OVID).

Methods: We developed four sensitive hedges by selecting terms that refer to infants. Three of the hedges had identical MeSH terms and keywords but used different field tags, and the fourth was a simple keyword hedge. We compared our hedges to the built-in MeSH-based infant filter. We used relative recall calculations to validate each hedge’s performance against a gold standard reference set.

Results: In PubMed the similarly structured hedges performed in a range of 83.2%-83.8% sensitivity and 88.2%-89.7% specificity. The simple keyword hedge performed with an 83.5% sensitivity and 89.7% specificity. The filter generated a 70.1% sensitivity and 96.2% specificity. Similarly, in Ovid Medline, the set of similar hedges performed in a range of 82.9%-83.6% sensitivity and 88.1%-89.4% specificity. The simple keyword hedge performed with an 82.9% sensitivity and 90.8% specificity. The filter generated a 69.6% sensitivity and 96.2% specificity.

Discussion: The variation in field tags did not provide a significant difference in the areas of sensitivity and specificity. The filter performed as expected with higher specificity rather than sensitivity. The simple keyword hedge performed better than anticipated with comparable sensitivity and specificity of the more complex hedges. When searching for infant population articles, the simple keyword search and filter work well for quick, clinical searching. For evidence synthesis, we recommend using one of the more sensitive infant hedges.