Development of a subject search filter to find information relevant to palliative care in the general medical literature

Purpose: The research developed and validated palliative care search filters for use in the general biomedical literature. Methods: Four general medical journals were hand- searched to identify articles relevant to palliative care, forming a ʹʹgold standardʹʹ reference set. Searches comprising Medical Subject Headings (MeSH) and text-words were created for use in Ovid MEDLINE, and retrieved references were compared to the gold standard. Sensitivity, specificity, accuracy, and precision rates were calculated. Results: By hand-searching 20,501 articles published in the 4 journals during the 3-year study period (1999-2001), reviewers identified 773 items relevant to palliative care (3.8%). A master search combining 9 MeSH descriptors with 3 text-words achieved 45.4% sensitivity, 99.3% specificity, 73% precision, and 97.3% accuracy. Efforts to increase the sensitivity by modifying 3 relevant published but unvalidated searches did not improve the yield, except in 1 case which resulted in an improved sensitivity of 56.9% but was offset by reduced specificity (92.1%), precision (22%), and accuracy (90.8%). Conclusions: The study confirmed that literature relevant to palliative care is difficult to identify in general medical journals. While the filter developed in this research represents the best trade-off between sensitivity, specificity, accuracy, and precision, the sensitivity is unacceptably low. Further research, such as frequency analysis of text-words and MeSH terms, is required to increase the sensitivity of searching in this subset of the literature.