Search

Background: PubMed translations of OvidSP Medline search filters offer searchers improved ease of access. They may also facilitate access to PubMed's unique content, including citations for the most recently published biomedical evidence. Retrieving this content requires a search strategy comprising natural language terms ('textwords'), rather than Medical Subject Headings (MeSH). We describe a reproducible methodology that uses a validated PubMed search filter translation to create a textword-only strategy to extend retrieval to PubMed's unique heart failure literature. Methods: We translated an OvidSP Medline heart failure search filter for PubMed and established version equivalence in terms of indexed literature retrieval. The PubMed version was then run within PubMed to identify citations retrieved by the filter's MeSH terms (Heart failure, Left ventricular dysfunction, and Cardiomyopathy). It was then rerun with the same MeSH terms restricted to searching on title and abstract fields (i.e. as 'textwords'). Citations retrieved by the MeSH search but not the textword search were isolated. Frequency analysis of their titles/ abstracts identified natural language alternatives for those MeSH terms that performed less effectively as textwords. These terms were tested in combination to determine the best performing search string for reclaiming this 'lost set'. This string, restricted to searching on PubMed's unique content, was then combined with the validated PubMed translation to extend the filter's performance in this database. Results: The PubMed heart failure filter retrieved 6829 citations. Of these, 834 (12%) failed to be retrieved when MeSH terms were converted to textwords. Frequency analysis of the 834 citations identified five high frequency natural language alternatives that could improve retrieval of this set (cardiac failure, cardiac resynchronization, left ventricular systolic dysfunction, left ventricular diastolic dysfunction, and LV dysfunction). Together these terms reclaimed 157/834 (18.8%) of lost citations. Conclusions: MeSH terms facilitate precise searching in PubMed's indexed subset. They may, however, work less effectively as search terms prior to subject indexing. A validated PubMed search filter can be used to develop a supplementary textword-only search strategy to extend retrieval to PubMed's unique content. A PubMed heart failure search filter is available on the CareSearch website (www.caresearch.com.au) providing access to both indexed and non-indexed heart failure evidence. [ABSTRACT FROM AUTHOR]

Background: Heart failure is a highly debilitating syndrome with a poor prognosis primarily affecting the elderly. Clinicians wanting timely access to heart failure evidence to provide optimal patient care can face many challenges in locating this evidence. This study developed and validated a search filter of high clinical utility for the retrieval of heart failure articles in OvidSP Medline. Methods: A Clinical Advisory Group was established to advise study investigators. The study set of 876 relevant articles from four heart failure clinical practice guidelines was divided into three datasets: a Term Identification Set, a Filter Development Set, and a Filter Validation Set. A further validation set (the Cochrane Validation Set) was formed using studies included in Cochrane heart failure systematic reviews. Candidate search terms were identified via word frequency analysis. The filter was developed by creating combinations of terms and recording their performance in retrieving items from the Filter Development Set. The filter's recall was then validated in both the Filter Validation Set and the Cochrane Validation Set. A precision estimate was obtained post-hoc by running the filter in Medline and screening the first 200 retrievals for relevance to heart failure. Results: The four-term filter achieved a recall of 96.9% in the Filter Development Set; 98.2% in the Filter Validation Set; and 97.8% in the Cochrane Validation Set. Of the first 200 references retrieved by the filter when run in Medline, 150 were deemed relevant and 50 irrelevant. The post-hoc precision estimate was therefore 75%. Conclusions: This study describes an objective method for developing a validated heart failure filter of high recall performance and then testing its precision post-hoc. Clinical practice guidelines were found to be a feasible alternative to hand searching in creating a gold standard for filter development. Guidelines may be especially appropriate given their clinical utility. A validated heart failure filter is now available to support health professionals seeking reliable and efficient access to the heart failure literature. [ABSTRACT FROM AUTHOR]