Related Experiment Videos
Identifying diagnostic studies in MEDLINE: reducing the number needed to read
Lucas M Bachmann1, Reto Coray, Pius Estermann
1University of Zürich, Zürich, Switzerland. lucas.bachmann@evimed.ch
Journal of the American Medical Informatics Association : JAMIA
|October 19, 2002
Summary
This study developed a more precise PubMed search filter for diagnostic studies, improving information retrieval for evidence-based practice. The new filter demonstrated better precision and comparable sensitivity to existing methods.
Area of Science:
- Medical Informatics
- Information Retrieval
- Biomedical Literature Search
Background:
- PubMed's search filters are crucial for evidence-based practice, but the filter for diagnostic studies lacks precision.
- Existing search strategies for diagnostic studies in PubMed often yield low precision, hindering efficient information retrieval.
- Improving the accuracy of diagnostic study identification in biomedical databases is essential for evidence-based practice.
Purpose of the Study:
- To construct and validate improved search strategies for identifying diagnostic articles in MEDLINE.
- To enhance the precision of PubMed searches for diagnostic studies without significantly compromising sensitivity.
- To develop a more satisfactory filter for diagnostic studies in PubMed, addressing limitations of current tools.
Main Methods:
- A comparative, retrospective analysis using hand-searched diagnostic studies from 1989, 1994, and 1999 as gold standards.
- Word frequency analysis of abstracts to identify candidate search terms, followed by independent search strategy testing.
- Calculation of sensitivity, precision, and number needed to read (NNR) for candidate terms and developed filters.
Main Results:
- A new search strategy combining truncated terms (diagnos*, predict*, accura*) with MeSH term 'SENSITIVITY AND SPECIFICITY' achieved 98.1% sensitivity and an NNR of 8.3.
- Head-to-head comparisons showed the new filter achieved better precision than the current PubMed filter in 1994 (12.0% vs. 8.2%) and 1999 (5.0% vs. 4.3%).
- The new filter demonstrated slightly higher sensitivity (98.1% and 96.1%) compared to the existing filter (95.1% and 88.8%) in validation subsets.
Conclusions:
- The performance of the current PubMed filter for diagnostic studies may be overestimated, with precision being particularly unstable.
- The developed search filter offers improved precision and comparable sensitivity for identifying diagnostic studies in MEDLINE.
- Further research is needed to confirm the practical benefits of the enhanced filter in real-world search scenarios.