Background Interventions to prevent or manage sexually transmitted infections (STI) are often evaluated at the clinic or community level. Cluster randomised controlled trials (cluster-RCT) need to take into account similarities between characteristics within clusters, increasing the required sample size. However, information about intracluster correlation coefficients (ICC) is rarely known at the design stage. We estimated ICCs for four STI and sexual behaviour variables at the levels of clinic and postcode.
Methods Data were collected during the Australian Chlamydia Control Effectiveness Pilot (ACCEPt), a cluster-RCT of a chlamydia testing intervention in women and men aged 16–29 years attending general practice. ICCs were calculated for: chlamydia prevalence, proportion with a chlamydia test in the last 12 months, condom use last sex and concurrent sex partners. Population-averaged unadjusted and covariate-adjusted logistic regression models with exchangeable correlation matrix were fitted to the clustered data, estimated by generalised estimating equations. ICCs were calculated separately at the levels of clinic and postcode.
Results The trial was conducted in 130 clinics in 54 Australian postcodes. For the prevalence outcome, the median cluster size was 25 for clinic and 74 for postcode. ICCs were larger at clinic than postcode level for all outcomes. ICC at the clinic and postcode level were, respectively: chlamydia prevalence 0.0044 and 0.0026; chlamydia testing 0.0105 and 0.0074; condom use last sex 0.0032 and 0.0010; and concurrent partners 0.0007 and 0.0006. In general, adjustment for individual- and postcode-level characteristics reduced ICCs. The design effect for chlamydia prevalence accounting for clustering was 1.35 and 1.21 using the clinic or postcode cluster level respectively.
Conclusion For STI and sexual behavioural outcomes in ACCEPt, the size of the ICC depended on the level of cluster randomisation. By publishing these ICC estimates, STI researchers can undertake more robust sample size calculations for future cluster-RCTs.
Disclosure No significant relationships.
Statistics from Altmetric.com
If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.