Developing and Validating a PubMed Infant Hedge: An MLA Pediatrics Librarians Caucus Initiative

Objective: MLA Pediatrics Librarians Caucus members expressed a need for updated pediatric hedges because of revisions to PubMed's search algorithm and librarians’ growing collaboration on systematic reviews. Therefore, to create modular hedges based on age groups, the caucus started with an initiative to develop and validate five infant search hedges for use in PubMed. Four hedges were developed by the caucus, based on a combination of personal experience and existing published hedges. These hedges are sensitive and intended for systematic reviews and other broad searches. The fifth hedge, for comparison purposes, is PubMed's infant filter.

Methods: We selected five infant hedges to test and validate. Partnering with a statistician, we conducted a pilot test on a subset of 50 articles, then expanded to the full set of 1000 articles. We identified five search topics that retrieved references on adult and infant populations. We gathered 200 references from each of the five topics, giving us 1000 total. Various date ranges ensured inclusion of indexed and non-indexed references. References were exported into Covidence for screening. Five reviewers, with two votes needed per reference, screened each title and abstract, indicating if the reference included the infant population (human infants, birth-23 months). A third reviewer resolved conflicts. Full-text articles were screened using the same process. The final selected references were the true positive reference set which was used to calculate interrater reliability, sensitivity and specificity of each hedge.

Results: Interrater reliability was 53% for title/abstract screening, and 80% for full-text screening. Hedge 1 (no field tags) has the highest sensitivity (88%). Hedge 2 (text word field tags), hedge 3 (title/abstract field tags) and hedge 4 (super simple keywords) have the same sensitivity (87.4%). Hedge 4 (super simple keywords) has high sensitivity (87.4%) and specificity (89.3%). Hedge 5 (PubMed infant filter) has the highest specificity (96.2%).

Conclusion: MLA Pediatrics Librarians Caucus members developed and validated five sensitive search hedges for use in PubMed. This project provides transparent statistics on the sensitivity and specificity of each infant hedge so librarians can select the hedge that best suits their needs. Hedge 1 may be most appropriate for systematic and scoping reviews, while hedge 5 may be most appropriate for bedside searches. We intend to use the same process to develop additional age-specific hedges for use in PubMed, then translate these hedges for use in other databases.