A systematic review of fracture-related adverse effects associated with the use of thiazolidinediones was used as a case study to calculate the sensitivity, precision and Number Needed to Read (NNR) of published adverse effects search filters in MEDLINE and Embase. 12 MEDLINE filters and 3 Embase filters were tested.
The results showed that 4 search filters in MEDLINE achieved high levels of sensitivity (95 or 100%) with improved levels of precision compared with searches without any adverse effects filters. The highest level of precision in MEDLINE (55%) was achieved with search filters that relied only on subject headings (MeSH). No search filter in Embase achieved sensitivity higher than 83% and precision remained low using any of the filters (all under 5%).
Limitations stated by the reviewer(s):
In addition to the limitation noted above by the authors, it should be noted that the number of included studies on which the search filters were tested was relatively small (19 records in MEDLINE and 24 records in Embase).
It should also be noted that the filters were tested as published. Testing of individual terms within the filters might have resulted in higher precision and better trade-off between sensitivity and precision than the full filters.
This publication is related to Su Golder’s PhD Thesis “Evaluating and Optimising the Retrieval of Research Evidence for Systematic Reviews of Adverse Drug Effects and Adverse Drug Reactions” from 2013. The thesis is available from http://etheses.whiterose.ac.uk/4749/