Article type
Year
Abstract
Background: To reduce the impact of selection bias in meta-analyses and increase the power of their results the identification of as many trials as possible is essential. Electronic databases are effective and timesaving tools to identify clinical trials. . The identification of Non-English-language trials is important to reduce the impact of the foreign language bias, one important source of selection bias in meta-analyses. The objective of the presented project was to determine the value of the two largest international medical databases (MEDLINE and EMBASE) for the identification of clinical trials published in German general health care journals
Methods: Six German general health care journals were manually searched for randomized controlled trials (RCT) from 1948 to 1998. For each identified trial a search in MEDLINE and EMBASE was performed to check if the trial was registered in one or both databases. Additionally electronic searches using the controlled vocabulary of the databases (e.g. 'Randomized Controlled Trial' as publication type in MEDLINE) were performed to compare these results with the result of the manual search.
Results: The manual search identified 1690 RCT. 937 (55%) were registered in MEDLINE and 877 (52%) in EMBASE. 1232 (72%) of the 1690 RCT were registered in at least one of the databases. The remaining 458 RCT (28%) were not identifiable in these databases. Recall and precision of electronic searches using the controlled vocabulary were low. Compared with the result of the handsearch (as gold standard) the recall of the publication type 'Randomized Controlled Trial' in MEDLINE was 33% (95% CI: 30;36) and the precision was 79% (75;83). The search with the controlled term 'controlled study' in EMBASE reached a recall of 42% (38;45) and a precision of 17% (14;17).
Discussion: The identification of clinical trials in electronic databases should not only be based on the controlled vocabulary because the indexing with these terms is incomplete. A carefully designed free text search can lead to a higher recall (e.g. the 'highly sensitive search strategy' of the Cochrane Collaboration which uses free text and controlled terms). But even the use of sensitive search strategies in these two great international databases can only identify about 70% of the trials published in German general health care journals. To increase this proportion the use of additional databases like the Cochrane Controlled Trials Register is necessary.
Methods: Six German general health care journals were manually searched for randomized controlled trials (RCT) from 1948 to 1998. For each identified trial a search in MEDLINE and EMBASE was performed to check if the trial was registered in one or both databases. Additionally electronic searches using the controlled vocabulary of the databases (e.g. 'Randomized Controlled Trial' as publication type in MEDLINE) were performed to compare these results with the result of the manual search.
Results: The manual search identified 1690 RCT. 937 (55%) were registered in MEDLINE and 877 (52%) in EMBASE. 1232 (72%) of the 1690 RCT were registered in at least one of the databases. The remaining 458 RCT (28%) were not identifiable in these databases. Recall and precision of electronic searches using the controlled vocabulary were low. Compared with the result of the handsearch (as gold standard) the recall of the publication type 'Randomized Controlled Trial' in MEDLINE was 33% (95% CI: 30;36) and the precision was 79% (75;83). The search with the controlled term 'controlled study' in EMBASE reached a recall of 42% (38;45) and a precision of 17% (14;17).
Discussion: The identification of clinical trials in electronic databases should not only be based on the controlled vocabulary because the indexing with these terms is incomplete. A carefully designed free text search can lead to a higher recall (e.g. the 'highly sensitive search strategy' of the Cochrane Collaboration which uses free text and controlled terms). But even the use of sensitive search strategies in these two great international databases can only identify about 70% of the trials published in German general health care journals. To increase this proportion the use of additional databases like the Cochrane Controlled Trials Register is necessary.