Data solely extracted from publications

Where analyses are to be based on information taken from published papers or reports, articles must be read in meticulous detail to ensure that the appropriate information is used. Data should be abstracted in a structured way, either onto standard paper forms (Box 11.6) or directly into an electronic database. There is usually no contact with the authors and consequently no opportunity for those responsible for the trials to verify that the data used are accurate and appropriate. It is therefore considered good practice for data extraction to be performed independently by two reviewers. Any discrepancies are resolved by discussion or by adjudication by a third party. Identifying multiple publications originating from the same trial can be particularly problematic. Often, it is not straightforward to determine whether multiple papers have reported the same trial, because authors and patient numbers may change between reports. Additional individuals may have been accrued to trials that are reported whilst ongoing, or different subsets of patients described. Ultimately the use of unique identifying numbers will distinguish multiple publications of the same report (see Box 11.2). In the meantime, it would be beneficial if reviewers tried to contact authors in cases where there is doubt about duplication to ensure that data from any one trial are represented only once in the meta-analysis.

0 0

Post a comment