Validation of five search filters for retrieving clinical guidelines showed low precision

Article type
Authors
Lunny C1, Salzwedel D1, Gerrish S1, Ramasubbu C2, Lui T2, Puil L1, Wright JM1
1Cochrane Hyptertension Review Group, Therapeutics Initiative, University of British Columbia
2Faculty of Pharmaceutical Sciences, University of British Columbia
Abstract
Background: clinical practice guidelines (CPGs) are up-to-date evidence-based summaries and consensus-based recommendations that provide a means of answering questions during patient care. CPGs are often included as eligible studies in systematic reviews of clinical interventions and management strategies. Therefore, CPG search filters are needed for comprehensive search strategies during systematic reviews, as well as to develop and update clinical guidelines. Search filters for databases such as MEDLINE provide an efficient way to retrieve CPGs. To our knowledge, this is the first study to validate search filters for CPGs.

Objectives: to validate search filters for retrieving CPGs in MEDLINE and Embase using the Ovid and PubMed interfaces.

Methods: a search for filters for identifying CPGs was conducted in Google and on the InterTASC Information Specialists' Sub-Group website. To retrieve a random sample of CPGs to test sensitivity and precision of the filters, we used the TRIP and Epistemonikos databases. Two researchers independently screened citations. The filters were translated from their original databases and platforms into Ovid MEDLINE, Ovid Embase, and PubMed syntax (e.g. MeSH terms, adjacency operators) by an expert librarian (DS). The Ovid MEDLINE filters were entered into MEDLINE Transpose software to translate to PubMed format. Each line of the filters was entered into the database interface to check for syntax errors. When errors were encountered, the librarian checked the lines and translated them as closely as possible to the original filter line. The PubMed ID numbers (PMIDs) or digital object identifiers (DOIs) were used to identify the citations in the test set in the databases. Sensitivity and precision were calculated with corresponding confidence intervals. Confidence intervals of proportions for binomial data were calculated using Stata, version 13 (using the command cii [confidence interval immediate]). Data were displayed in tables categorized by database and search filter.

Results: five search filters were retrieved: two from the Canadian Agency for Drugs and Technologies in Health (CADTH), two from the University of Texas School of Public Health, and one from the MD Anderson Cancer Center Library. A total of 478 records were screened to identify 109 CPGs, which comprised the sample for testing sensitivity and precision. Sensitivity ranged from 87% to 98% for the five search filters with very low precision (< 1%) across all filters. The CADTH broad search filter had the highest sensitivity across all databases.

Conclusions: this study provides new comparative information on the performance of various search filters for retrieval of CPGs. Our analysis shows that it remains difficult to efficiently identify CPGs due to low precision of five search filters. Future research should focus on developing a search filter that balances high sensitivity with moderate precision.