124 Chapter 6 multifaceted factors potentially influencing our primary outcome measures at both the individual and cluster levels (see Snijder et al., 2022 for the original analysis plan). However, adjustments were necessary to the original analysis plan, due to uneven participant numbers across conditions and the addition of an individual arm in the study design (see Figure 1). Following an intention-to-treat approach, all randomized participants and available data at each time point were included in the analysis. Below, we outline the data analysis plan and the statistical analyses conducted. First, to ensure that (cluster) randomization resulted in comparable groups without significant differences in demographic characteristics and key baseline measures, independent sample t-tests and chi-square tests were performed. For ordinal-level variables, non-parametric correlation coefficients were calculated. Second, missing values for all outcome variables were calculated and assessed. For the primary outcome measure (total joint engagement), the percentage of missing values at endpoint (T2) was 17.5%, and at follow-up (T3), it was 35.1%. For secondary outcome measures, the percentage of missing values at T3 ranged from 38.5% to 50.9%. Missing values were imputed using the expectation-maximization method (Schafer & Graham, 2002). Due to the high level of missing data on the Mac-Arthur N-CDI (68.4%) and WEMWBS (65.5%), we opted not to impute missing values for these questionnaires. Little’s Missing Completely at Random (MCAR) test confirmed that missing data were completely at random [X2=330.864 (354), p = .806]. Attrition analyses were conducted to examine the association between participant characteristics at baseline (T1, as presented in Table 3) and drop-out at follow-up (T3). Third, to examine differences between groups (BEAR vs. CAU) over time (baseline, endpoint, follow-up), a Repeated Measures Analysis of Variance (ANOVA) was conducted. The interaction between time and group represents the intervention effect. No outliers were removed from the JERI outcomes data, since these higher scores were not deviant from a clinical perspective. There were no additional outliers with exception of the WEMWBS, where one outlier at T3 was identified and subsequently excluded from analyses. Given the relatively small sample size, confidence intervals around the estimated marginal means were calculated. Statistical assumptions inherent to the Repeated Measures ANOVA were met for most variables, except for coordinated joint engagement, symbol-infused joint engagement, and expressive language use as measured by the JERI. For these variables, the assumption of normally distributed data was violated. As a result, these variables were categorized into three groups: improvers, decreasers, and non-responders (comparing T1 to T2 and T1 to T3). Chi-square tests were then used to analyze differences between the treatment and care-asusual groups. Analyses were conducted on the imputed dataset.
RkJQdWJsaXNoZXIy MjY0ODMw