High dropout rates in trials included in Cochrane reviews

Article type
Authors
Biester K, Lange S, Kaiser T, Potthast R
Abstract
Background: High dropout rates (HDRs) in clinical trials may lead to attrition bias, resulting in a systematic distortion of study results and the loss of internal validity. Guidelines provide insufficient advice on the adequate handling of HDRs in Cochrane reviews. Consideration of HDRs is complicated by the inconsistent use of the terms 'dropouts', 'losses to follow up', and 'withdrawals'.

Objectives: The aim of our study was to assess whether the issue of HDRs is addressed in Cochrane reviews, and if so, how HDRs are considered.

Methods: A random sample of 100 Cochrane reviews was drawn from The Cochrane Library, 25 per year for 2002 to 2005. To be included, the reviews had to refer to therapeutic interventions, a primary outcome had to be defined, and the reviews had to summarise the results of at least five single trials. Protocols or updates of Cochrane reviews were excluded. Two reviewers independently assessed: (i) whether the Cochrane reviews reported HDRs (defined as a dropout rate of > 15%); (ii) whether HDRs were considered in Cochrane reviews; and (iii) whether their consideration affected the Cochrane review results and conclusions.

Results: (i) Seventy-three Cochrane reviews reported dropout rates; 60 (82%) of these reported HDRs (HDR range: > 15-79%). (ii)Twenty-two (37%) of the 60 Cochrane reviews stated that HDRs were considered; 19 (86%) of these stated that this was planned a priori. However, only 17 (77%) applied the described methods, such as defining cut-offs (HDR of >15->50%) that led to exclusion of the data, the conduct of worst/best case scenarios, other sensitivity analyses, heterogeneity analyses or meta-regressions. (iii) Results were in some cases not robust, independent of the respective approach.

Conclusions: Cochrane reviews inadequately consider high dropout rates of clinical trials. Furthermore, the methods applied to consider high dropout rates are inconsistent.