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

Area of Science:

• Medical Informatics
• Information Retrieval
• Evidence-Based Practice

Background:

• Exhaustive literature searching is crucial for evidence-based practice guidelines.
• Current methods often rely on manual identification of Medical Subject Headings (MeSH) in databases like MEDLINE, which can be time-consuming and error-prone.

Purpose of the Study:

• To develop and evaluate a novel learning algorithm designed to automate and optimize the process of conducting thorough literature searches.
• To compare the effectiveness of this algorithm against traditional static keyword searches.

Main Methods:

• A learning algorithm was created to generate and utilize combinations of MeSH terms for searching.
• The algorithm was applied to the MEDLINE database (1950-2008) to identify studies on pharmaceutical care in HIV-infected patients.
• Results were compared to a static search conducted by an independent user.

Main Results:

• The learning algorithm retrieved 1670 articles, identifying six relevant studies.
• The static search yielded 49 articles, with only three being relevant.
• All relevant articles found via the static search were also identified by the learning algorithm.

Conclusions:

• Automated, learning-based tools can enhance the efficiency and comprehensiveness of literature searches.
• The developed algorithm presents a promising approach to streamline the identification of evidence for clinical practice.
• This method addresses the challenges associated with manual literature searching.