Data extraction problems associated with RCTS of cross-over design assessing the effects of medical treatment for menorrhagia

Article type
Authors
Lethaby AE, Farquhar C, Jepson R
Abstract
Introduction: A commonly used design for RCTs is the cross-over trial which allows each participant to serve as his own control. Although the cross-over design is generally considered highly efficient because of a reduction in variability compared to the parallel group design, a potential disadvantage is the possibility of residual or spill-over effects from the first treatment affecting the result of the second. The statistical package, MetaView, which is currently used to enter and pool data for Cochrane systematic reviews, is not able to check for period-treatment interaction, so reviewers are required to obtain data at the end of the first treatment period for the assessment of effects.

Objective: To examine the frequency of crossover design in RCTs meeting the criteria for inclusion in 2 systematic reviews on menorrhagia. To discuss the practical difficulties associated with retrieval of original data so that effects may be determined at the end of the first treatment period.

Methods: An extensive search of MEDLINE, Embase, Psychlit, Cinahl, Current Contents, Biological Abstracts and Social Sciences Index was performed to identify RCTs assessing the effects of 2 major medical therapies (antifibrinolytic drugs and nonsteroidal antiinflammatory drugs) on menorrhagia (heavy menstrual bleeding). The proportion of trials with cross-over design which met the criteria for inclusion was determined. The success of attempts to retrieve the original data from these trials will be discussed. A comparison will be made of treatment effects with and without the incision of cross-over trials.

Results: A significant proportion of RCTs for 2 systematic reviews on menorrhagia, 58% and 67% respectively, was of cross-over design. Very few of these papers provided sufficient information for the determination of effects at the end of the first treatment period. Attempts to retrieve this data from the authors were not often successful, especially for RCTs published more than 5 years earlier. As a consequence, final lists of RCTs for the menorrhagia reviews were biased towards recent studies and studies with a parallel group design.

Discussion: A method for using all the information from an RCT with cross-over design needs to be developed to assist reviewers in making full use of data to develop more comprehensive lists of RCTs for inclusion in systematic reviews. Furthermore, those researchers planning and publishing cross-over designed RCTs need to be encouraged to publish the data from both pre and post cross-over phases of the trials.