General Practice and Primary Care - Research Publications
Permanent URI for this collection
Search Results
1 - 1 of 1
-
ItemThe design and analysis of matched-pair cluster randomised trials( 2012)Cluster randomised trials (CRTs) are studies in which groups (clusters) of individuals, such as a group of patients that see the same general practitioner (GP), are randomised to one of two (or more) treatment conditions that are being compared, rather than the individuals themselves. CRTs have certain design and analytical challenges not found in trials that randomise individuals. One design issue is related to the greater risk of imbalance of important prognostic factors between trial arms, especially when the number of clusters randomised is small. In order to avoid chance imbalance on important factors, a matched-pair design is often employed, where clusters are paired according to a number of different characteristics, such as size of the practice,geographical location or baseline score on the outcome of interest, and then one cluster within each pair is randomised to the intervention arm and the other to the control arm. To date, the literature is unclear regarding when it is sensible to match clusters in pairs in the design of a CRT in order to provide greater assurance that known prognostic factors are balanced between the trial arms, especially when the number of clusters is small. Furthermore, there are a number of unique design and analytical challenges associated with the matched-pair design related to having only one cluster in each combination of trial arm and pair of clusters. This research aimed to provide guidance to assist researchers and applied statisticians in deciding when to use a matched-pair CRT and how best to analyse data from such studies. The research aim was addressed with four major components of original research. Firstly, a systematic review was performed to investigate recent practice in primary care research use of the matched-pair design and it sets the context for the research. Three extensive simulation studies were performed to address some of the gaps identified in the literature. The first two simulation studies evaluated the performance of selected analytical methods that can be applied to the matched-pair CRT for continuous and binary outcomes, respectively. The third simulation study compared the relative efficiency of the matched-pair design, in terms of gains in precision for estimation of the intervention effect, to the stratified and completely randomised designs. This research led to the development of a set of practical recommendations for the design and analysis of CRTs within primary care research. Choosing the best randomisation approach for CRTs is not straightforward and the design of each study needs to be considered carefully. Even when the matched-pair design minimises the chance of imbalance of prognostic factors and provides the greatest statistical efficiency, the practical and analytical limitations of using a matched-pair design may ultimately determine the final choice of study design. It is hoped the practical recommendations offered will assist researchers and applied statisticians in choosing the best design for future proposed CRTs.