Background: Clinicians commonly search bibliographic databases such as Medline to find sound evidence to guide patient care. Unfortunately, this can be a frustrating experience because database searches often miss relevant articles. We addressed this problem for transplant professionals by developing kidney transplantation search filters for use in Medline through PubMed and Ovid Technologies, and Embase.
Methods: We began by reading the full-text versions of 22,992 articles from 39 journals published across 5 years. These articles were labeled relevant to kidney transplantation or not forming our "gold standard." We then developed close to five million kidney transplantation filters using different terms and their combinations. Afterward, these filters were applied to development and validation subsets of the articles to determine their accuracy and reliability in identifying articles with kidney transplantation content. The final kidney transplantation filters used multiple terms in combination.
Results: The best performing filters achieved 97.5% sensitivity (95% confidence interval, 96.4%-98.5%), and 98.0% specificity (95% confidence interval, 97.8%-98.3%). Similar high performance was achieved for filters developed for Ovid Medline and Embase. Proof-of-concept searches confirmed more relevant articles are retrieved using these filters.
Conclusions: These kidney transplantation filters can now be used in Medline and Embase databases to improve clinician searching.