Developing efficient search strategies to identify papers on adverse events. B: using statistical analysis

Article type
Authors
Golder S, Duffy S, Glanville J, McIntosh H
Abstract
Background: Cochrane Reviews are increasingly focusing not just on the effectiveness of health care interventions, but also on adverse events. However, reviewers have generally restricted their review of adverse events to information from RCTs meeting the criteria for inclusion in the review of effectiveness. As most trials are short term, have a relatively small sample size, and often fail to report adverse events there is a risk that adverse consequences are not being fully identified.

The issues about how and when to carry out additional specific searches for adverse events are now being debated. However, the best way to search for this type of evidence is not well established.

Objectives: To assess the performance (in terms of precision and sensitivity) of different approaches to searching MEDLINE and EMBASE to identify studies of adverse events, for a review. Methods: A systematic review of the effectiveness and adverse effects of seven new anti-epileptic drugs was used as a case study. This review included a thorough search for studies of effectiveness and additional searches for adverse events.

5 different approaches to searching for adverse events evidence were identified from a search of the literature [1, 2] and from scanning technology appraisals produced at the Centre for Reviews and Dissemination (CRD). We searched MEDLINE and EMBASE using each of the 5 approaches and augmented the combined search results with additional studies identified from reference lists, experts, submissions from drug companies, and a search on TOXLINE. This combined set of records was used as a quasi gold standard against which the sensitivity and precision of the 5 individual search approaches were tested.

Results: The adverse events searches provided 19 relevant papers that were not found by the effectiveness searches, even though the effectiveness searches were not restricted by study design. Comparisons of the sensitivity and precision of the different search strategies suggest that thorough searching for adverse events requires a combination of different approaches and that the indexing of EMBASE records enables more precise searching. A simple search strategy on EMBASE using the drug name indexing terms with the subheadings "adverse events" or "drug toxicity" retrieved 81 percent of the 73 relevant papers identified on EMBASE.

Conclusions: The searches for adverse events provided relevant papers in addition to those found via the effectiveness searches. This case study indicates that a range of databases, in addition to submissions from drug companies, may be appropriate for locating reports of adverse events. EMBASE may also provide a better starting point than MEDLINE for those with more limited resources searching for adverse events of drugs.

References : 1. Bagett R, Chiquette E, Anagnostelis B, Mulrow C. Locating reports of serious adverse drug reactions. Proceedings of the 7th Annual Cochrane Colloquium; 1999 Oct 6-9; Rome. 2. Derry S, Loke YK, Aronson JK. A. Incomplete evidence: the inadeqacy of databases in tracing published adverse drug reactions in clinical trials. BMC Med Res Methodol [serial on the internet]. 2001 [cited 2004 Mar 29];1[1]: [about 7 p.]. Available from: http://www.biomedcentral.com/1471-2288/1/7