P1-30 Confounded clusters: an assessment of the adequacy of measures to minimise confounding in cluster RCTS

IntroductionCluster randomised controlled trials (CRCTs), where the unit of allocation is a group, are particularly useful in Public Health, when interventions are targeted at populations. Randomisation is the ‘gold standard’ for dealing with both known and unknown confounders as these are balanced between comparison groups with sufficient numbers randomised. However, CRCTs may have few units randomised, increasing the likelihood of confounding. Confounding can be minimised by design (eg, minimisation) or analysis. We investigated methods used to reduce confounding and their adequacy in a sample of alcohol prevention CRCTs.MethodsExisting systematic reviews published in the Cochrane database assessing efficacy/effectiveness of primary/secondary prevention methods in young people to reduce alcohol use/harms were identified. Reporting of the following for all included CRCTs was established: baseline cluster characteristics, type of randomisation, participation rate, attrition rate and measures reducing confounding at analysis. All data were collected using standardised forms by two independent reviewers with discrepancies resolved with a third reviewer.ResultsTwo Cochrane systematic reviews were identified yielding a total of 30 CRCTs. Baseline measures at the cluster level were not always reported and the potential for confounding was often not considered in design, analysis and interpretation of results.ConclusionTo our knowledge, this is the first study that investigates whether researchers adequately consider confounding at a cluster-level in CRCTs. We recommend greater adherence to the CONSORT extension for CRCTs with explicit consideration of confounders during all stages of research. This helps ensure misleading effectiveness estimates are not used for Public Health decisions.