Designing psycho-oncology randomised trials and cluster randomised trials: variance components and intra-cluster correlation of commonly used psychosocial measures

Abstract

Objective: The study aims to provide information about variance components of psychosocial outcomes: within and between-participant variance, within-participant correlation and for cluster randomised trials, the intra-cluster correlation (ICC) and, also, to demonstrate how estimates of these variance components and ICCs can be used to design randomised trials and cluster randomised trials.

Method: Data from 15 longitudinal multi-centre psycho-oncology studies were analysed, and variance components including ICCs were estimated. Studies with psychosocial outcomes that had at least one measurement post-baseline including individual randomised controlled trials, cluster randomised trials and observational studies were included.

Results: Variance components and ICCs from 87 outcome measures were estimated. The unadjusted, single timepoint (first post-baseline) ICCs ranged from 0 to 0.16, with a median value of 0.022 and inter-quartile range 0 to 0.0605. The longitudinal ICCs ranged from 0 to 0.09 with a median value of 0.0007 and inter-quartile range 0 to 0.018.

Conclusions: Although the magnitude of variance components and ICCs used for sample-size calculation cannot be known in advance of the study, published estimates can help reduce the uncertainty in sample-size calculations. Psycho-oncology researchers should be conservative in their sample-size calculations and use approaches that improve efficiency in their design and analysis.

Keywords: ICC; cancer; design; longitudinal; oncology; statistical methodology.