Premium
Designing psycho‐oncology randomised trials and cluster randomised trials: variance components and intra‐cluster correlation of commonly used psychosocial measures
Author(s) -
Bell Melanie L.,
McKenzie Joanne E.
Publication year - 2013
Publication title -
psycho‐oncology
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 1.41
H-Index - 137
eISSN - 1099-1611
pISSN - 1057-9249
DOI - 10.1002/pon.3205
Subject(s) - quartile , psychosocial , sample size determination , observational study , medicine , cluster (spacecraft) , correlation , variance (accounting) , clinical trial , randomized controlled trial , analysis of variance , repeated measures design , cluster randomised controlled trial , statistics , confidence interval , mathematics , computer science , psychiatry , geometry , accounting , business , programming language
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. Copyright © 2012 John Wiley & Sons, Ltd.