Article type
Year
Abstract
Background: Searching electronic databases such as MEDLINE is a standard way to identify randomized controlled trials (RCTs). However, prior studies have shown that MEDLINE fails to recognize 30% of RCTs, primarily due to miscoding by National Library of Medicine coders and unclear description of methods by authors. We compared RCTs identified by MEDLINE with those identified through a hand search of articles for the journal AIDS.
Methods: We hand searched all issues of AIDS from January 1987 through December 1999 (including supplements) to identify all published RCTs. We searched for studies that compared two or more interventions in humans, were prospective and assigned participants to a particular intervention by some type of randomized allocation. All hand searching was done by one of two investigators; to assess quality of the results, a second investigator searched selected issues. Inter-rater reliability was high and a third investigator adjudicated all discrepancies. We compared hand-search results with two types of MEDLINE searches for RCTs - one using the "publication type" icon and the second using the MeSH keyword "randomized controlled trial".
Results: We reviewed 165 issues of AIDS and identified 172 RCTs in our handsearch. Our MEDLINE "publication type" search identified 170 studies as RCTs; 15 (9%) of these of these citations did not meet our definition of RCTs. Thus, the "publication type" search identified 155 (90%) of the 172 RCTs. The MEDLINE MeSH keyword search identified 171 studies as RCTs; 16 (9%) were miscoded as RCTs. Thus, the MeSH keyword search identified 155 (90%) of the 172 RCTs. The number of RCTs per issue increased over time and the percentage of RCTs identified by PT and MeSH improved over time. The 15 RCTs that were not identified in the "publication type" search were the same 15 trials missed in MeSH search. Thus the sensitivity of the combined searches remains at 90%. Years # Issues # RCTs #RCT per Issue #RCTs Identified by PT (%) MeSH (%) 1995-1999 82 123 1.50 113 (92%) 113 (92%) 1994-1990 61 43 .70 39 (91%) 39 (91%) <1990 22 6 .27 3 (50%) 3 (50%) Total 165 172 1.04 155 (90%) 155 (90%)
Conclusions: The number of RCTs being published in a leading AIDS specialty journal is increasing. To identify these trials, searching MEDLINE by "publication type" or by MeSH keyword "randomized controlled trial" produces virtually the same results yet each failed to identify 10% of RCTs. To improve indexing of articles we suggest that authors explicitly suggest RCT as a keyword or use the term "randomized controlled trial" in their title and/or abstract.
Methods: We hand searched all issues of AIDS from January 1987 through December 1999 (including supplements) to identify all published RCTs. We searched for studies that compared two or more interventions in humans, were prospective and assigned participants to a particular intervention by some type of randomized allocation. All hand searching was done by one of two investigators; to assess quality of the results, a second investigator searched selected issues. Inter-rater reliability was high and a third investigator adjudicated all discrepancies. We compared hand-search results with two types of MEDLINE searches for RCTs - one using the "publication type" icon and the second using the MeSH keyword "randomized controlled trial".
Results: We reviewed 165 issues of AIDS and identified 172 RCTs in our handsearch. Our MEDLINE "publication type" search identified 170 studies as RCTs; 15 (9%) of these of these citations did not meet our definition of RCTs. Thus, the "publication type" search identified 155 (90%) of the 172 RCTs. The MEDLINE MeSH keyword search identified 171 studies as RCTs; 16 (9%) were miscoded as RCTs. Thus, the MeSH keyword search identified 155 (90%) of the 172 RCTs. The number of RCTs per issue increased over time and the percentage of RCTs identified by PT and MeSH improved over time. The 15 RCTs that were not identified in the "publication type" search were the same 15 trials missed in MeSH search. Thus the sensitivity of the combined searches remains at 90%. Years # Issues # RCTs #RCT per Issue #RCTs Identified by PT (%) MeSH (%) 1995-1999 82 123 1.50 113 (92%) 113 (92%) 1994-1990 61 43 .70 39 (91%) 39 (91%) <1990 22 6 .27 3 (50%) 3 (50%) Total 165 172 1.04 155 (90%) 155 (90%)
Conclusions: The number of RCTs being published in a leading AIDS specialty journal is increasing. To identify these trials, searching MEDLINE by "publication type" or by MeSH keyword "randomized controlled trial" produces virtually the same results yet each failed to identify 10% of RCTs. To improve indexing of articles we suggest that authors explicitly suggest RCT as a keyword or use the term "randomized controlled trial" in their title and/or abstract.