Development of search filters for retrieval of literature on the molecular epidemiology of cancer

ObjectiveThe available tools for searching literature in the area of molecular epidemiology of cancer, a relatively novel discipline, are largely unsatisfactory. The aim of this project was to develop two search filters in MEDLINE that would facilitate the retrieval of studies on cancer molecular epidemiology and to compare their performance.MethodsCitations of all articles published in 2007 in three top journals in this area were hand-screened to establish a gold standard reference set of relevant papers. Two filters were created with this set: the first was based on the frequency of descriptors listed in the National Library of Medicine (NLM) Medical Subject Headings Thesaurus (MeSH terms) only, the second was based on the frequency of words in titles and abstracts (text-words), besides MeSH terms.The sensitivity, specificity, precision and accuracy rates of the two filters were calculated as performance measures.ResultsOut of all 1602 published articles, 344 were found to be pertinent to cancer molecular epidemiology; these formed the gold standard set. The first filter was the most specific (99%) and showed a good precision (93.2%), the second showed a higher sensitivity (78.2%) and accuracy (91.6%).ConclusionThe two filters proposed here represent a balance between sensitivity and specificity. Redefinition and elaboration of more appropriate NLM MeSH terms could greatly facilitate the retrieval of studies on the molecular epidemiology of cancer. Authors themselves could facilitate this by inserting adequate words in title or abstract of their articles.