Background
Cluster randomised controlled trials (CRCTs) are increasingly used to evaluate the effectiveness of interventions for improving health. A key feature of CRCTs is that individuals in clusters are often more alike than individuals in different clusters, irrespective of treatment. This similarity within clusters needs to be taken into account when planning CRCTs to obtain adequate sample sizes, and when analysing clustered data to obtain correct estimates.
Methods
Nationally representative data from 15 to 16 year olds were analysed, from 21 of the 35 countries that participated in the 2007 European School Survey Project on Alcohol and Other Drugs. Within country school level intra-class correlation coefficients (ICCs) were calculated for substance use (self-reported alcohol use, regular alcohol use, binge drinking, any smoking, regular smoking, and illicit drug use) and psychosocial health (depressive mood and self-esteem). Unadjusted and adjusted ICCs are presented. ICCs are adjusted for student sex and socioeconomic status.
Results
ICCs ranged from 0.01 to 0.21, with the highest (0.21) reported for regular smoking. Within country school level ICCs varied substantially across health outcomes, and among countries for the same health outcomes. Estimated ICCs were consistently higher for substance use (range 0.01–0.21), than for psychosocial health (range 0.01–0.07). Within country ICCs for health outcomes varied by changes in the measurement of particular health outcomes, for example the ICCs for regular smoking (range 0.06–0.21) were higher than those for having smoked at all in the last month (range 0.03–0.17).
Conclusions
For school level ICCs to be effectively utilised in informing sample size requirements for CRCTs and adjusting estimates from meta-analyses, the school level ICCs need to be both country and outcome specific.
