The Cochrane Highly Sensitive Search Strategies for identifying randomized trials in MEDLINE

The first Cochrane Highly Sensitive Search Strategy for identifying  randomized trials in MEDLINE was designed by Carol Lefebvre and published in 1994 (Dickersin 1994). This strategy was subsequently published in the Handbook and has been adapted and updated as necessary over time. The Cochrane Highly Sensitive Search Strategies for MEDLINE in subsequent sections are adapted from strategies first published in 2006 as a result of a frequency analysis of MeSH terms and free-text terms occurring in the titles and abstracts of MEDLINE-indexed records of reports of randomized controlled trials (Glanville 2006), using methods of search strategy design first developed by the authors to identify systematic reviews in MEDLINE (White 2001).


Two strategies are offered: a sensitivity-maximizing version and a sensitivity- and precision-maximizing version. It is recommended that searches for trials for inclusion in Cochrane reviews begin with the sensitivity-maximizing version in combination with a highly sensitive subject search. If this retrieves an unmanageable number of references the sensitivity- and precision-maximizing version should be used instead. It should be borne in mind that MEDLINE abstracts can be read quite quickly as they are relatively short and, at a conservative estimate of 30 seconds per abstract, 1000 abstracts can be read in approximately 8 hours.


The strategies have been updated, after re-analysis of the data used to derive those strategies, to reflect changes in indexing policy introduced by the US National Library of Medicine since the original analysis and changes in search syntax. These changes include:

The strategies are given in Box 6.4.a and Box 6.4.b for PubMed, and in Box 6.4.c and Box 6.4.d for Ovid.


It must be borne in mind that the strategies below are based on data derived from MEDLINE-indexed records and were designed to be run in MEDLINE. These strategies are not designed to retrieve ‘in process’ and other records not indexed with MeSH terms. It is, therefore, recommended that these strategies are run in the MEDLINE-indexed versions of MEDLINE and separate searches for non-indexed records are run in the database containing the ‘in process’ and non-indexed records. For example, in Ovid the strategies below should be run and updated in databases such as ‘Ovid MEDLINE(R) 1950 to Month Week X 200X’ and non-indexed records should be searched for in 'Ovid MEDLINE(R) In-Process & Other Non-Indexed Citations Month X, 200X'. For identifying non-indexed records a range of truncated free-text terms would be required, such as random, placebo, trial, etc, and the search must not be limited to humans (as the records are not yet indexed as humans).


As discussed in Section, MEDLINE has been searched from 1966 to 2004 inclusive, using previous versions of the Cochrane Highly Sensitive Search Strategy for identifying  randomized trials, and records of reports of trials (on the basis of the titles and abstracts only) have been re-indexed in MEDLINE and included in CENTRAL. Refer to Sections and for further guidance as to the appropriate use of these Highly Sensitive Search Strategies.