104 Chapter 5 Sample size and power The justification of sample size is calculated based on the primary hypothesis that the BEAR intervention will improve the child’s total joint engagement (measured on a 7-point Likert scale) at the end of treatment. In a comparable study by Kasari et al. (2014) with joint engagement as the primary outcome as measured in total seconds spent in joint engagement, a Cohen’s d effect size of 0.21 was found in difference in joint engagement at endpoint. We assume our effect size to be slightly lower than in the study of Kasari et al., due to a shorter duration of the BEAR intervention and the use of an ordinal variable instead of an continuous one. Based on the variance of the treatment effect, a power analysis on an alpha level of 0.05 (two sided) lead to an estimated sample size of 40 participants per group, with a power of .88. To allow for 10% drop out 44 participants per group (total N=88) will be recruited. Data collection and management Data collection follows the European General Data Protection Regulation. After inclusion, participants will receive an unique code number in order to anonymize all data. Nonanonymous data, for example informed consent forms, are locked in a closed cabinet and saved in protected folders, only accessible by appointed members of the research team. In order to collect and store data from the questionnaires, Castor Electronic Data Capture (Castor EDC, 2021) and QuestManager (version 5.6) software are used. Video-data will be anonymously saved on an external hard disk, which is password protected and saved in a locker. Parents of participants are asked to complete a number of online questionnaires (see Table 2) through the QuestManager software. Parents are only able to continue to a next questionnaire if all items are answered. To prevent missing data, a researcher will call parents as a reminder and offer support if parents encountered any problems, if they fail to complete the measurements on time. Parents may withdraw from the study at any time for any reason. If this happens, researchers will try to make a final appointment with parents in order to collect the primary outcome. Statistical analysis Data will be analyzed and reported in accordance with CONSORT guidelines. Baseline demographics (IQ, gender, age) and clinical characteristics (severity of symptoms as measured by the CoSoS) of the BEAR and CAU groups will be compared by chi-squared tests to check whether cluster randomization has led to two even groups. For the cluster RCT, all primary analyses will be intention to treat (ITT) using (generalized) linear mixed effects model for repeated measures with a random effect for cluster. Correlation of measurements (T1, T2, T3) within subject will primarily be modelled by a random effect of subject nested within clusters unless this results in insufficient fit, in which case other covariance structures
RkJQdWJsaXNoZXIy MjY0ODMw