A learning-based approach for performing an in-depth literature search using MEDLINE

Abstract

What is known and objective: Exhaustive literature searching is a core requirement for developing guidelines for evidence-based practice. MEDLINE is typically used. Searching requires the user to identify appropriate search terms, called Medical Subject Headings (MeSH) and refine the search to retrieve relevant articles. The objective of this study was to develop and test a learning algorithm for conducting a thorough literature search.

Methods: A learning algorithm to effectively utilize MeSH terms is presented. This algorithm creates combinations of available MeSH terms from which a search is conducted. The algorithm was applied to search MEDLINE (January 1950 to Janaury 2008) focusing on the impact of pharmaceutical care in HIV-infected patients. The number of relevant articles retrieved from the learning algorithm search was then compared against a static search with a fixed set of keywords implemented by an independent user.

Results and discussion: The learning algorithm retrieved 1670 articles with six relevant articles identified. The static search retrieved a total of 49 articles, with three being relevant. These three articles were also located from the learning algorithm-based search. WHAT IS KNOWN AND CONCLUSION: Performing a literature search for retrieving evidence-based studies can be a daunting and error-prone process. The introduction of automatic, learning tools for searching is desirable and we present a possible approach.