42 Chapter 2 calculating the effect size for cluster-randomised trials, we therefore reduced the sample size to its ‘effective sample size’ (Rao & Scott, 1992), by dividing the original sample size of each trial arm by a quantity called the ‘design effect’. The design effect was calculated with the following formula: 1 + (M – 1) * ICC, where M is the average size of each cluster and ICC is the intracluster/intraclass correlation coefficient describing the ‘similarity’ of individuals within the same cluster (Eldridge et al., 2009). The average cluster size was directly extracted, or calculated based on the number of clusters/groups randomised to each intervention arm and the total number of participants in the study. The ICC was directly extracted from the paper or requested from the authors, or an external estimate obtained from another study examining the same or a similar outcome was used (Higgins, Eldridge et al., 2022). Note that in the calculation of the effect size, only the sample size was reduced; means and standard deviations remained unchanged. To verify the results, we compared the magnitude, significance and direction of all computed effect sizes with the findings presented in the original paper (McKenzie et al., 2022). Cohen’s d effect sizes and 95% confidence intervals will be presented in forest plots that were made in Excel (Lajeunesse, 2021). For consistency within and across plots, the unadjusted effect size for cluster-randomised trials will be included in the plot and the adjusted effect size will be reported in a note. If the original paper reported Cohen’s d as well, this will be included as a note to the plot. It should be noted, however, that these values may deviate due to the specific calculation method followed. All data included in the forest plots are stored on a repository (Radboud Data Repository, https://doi.org/10.34973/t6df-m766). RESULTS The literature search resulted in 2835 hits. After removing 387 duplicates, the title and abstract of 2448 studies were screened. During the screening phase, 250 papers were initially included. After reading the full-texts of these papers, 92 were included. An additional 53 papers were identified through snowballing. In total, the current review includes 145 papers. See Figure A.1 for the flow diagram (Page, McKenzie et al., 2021). Endnote libraries with screened, included and excluded papers are available upon request. Based on the target group and outcome measures, included papers were grouped into two main categories: papers including a (sub)clinical participant sample (n = 75) and papers including a (mentally) healthy participant sample
RkJQdWJsaXNoZXIy MTk4NDMw