15250-m-cuypers

98 Chapter 5 range found for decisional conflict was between 0 and 0.02. Given the considerable variability in ICCs that is found in literature, ICC for the current trial is set conservative at 0.01. The attrition rate is set at 25% to compensate for non-response to the questionnaires. This rate is comparable to studies in similar populations and following the samemethods as this study does 65,66 . Calculations show that a design with 19 clusters of 20 patients (380 patients in total) achieves a power of .8. Taking into account a 25% attrition rate between the first and third questionnaire, the total sample size (rounded) will be set at 475 patients. This results in a recruitment of 25 patients per hospital. Statistical analysis All analyses will be conducted using SPSS version 19.0 (Statistical Package for Social Sciences, Chicago, IL, USA). A 0.05-significance level will be adopted in all statistical tests. We will perform a descriptive statistical analysis of organizational (hospitals) and socio- demographic (patients) characteristics in order to assure the comparability of the intervention and control groups. Baseline measures and changes in outcome variables over the study period for each study arm will be presented as means (± SD). The main outcome decisional conflict is measured at T1 and will be compared between both groups (intervention and control). Multilevel modelling will be carried to take the hierarchical structure of the data into account by specifying random effects at both hospital and patient level. The least square mean proportions will be estimated and compared to assess the effect of the DA on decisional conflict. The secondary outcomes will also be compared between both groups using multilevel modelling. Some of the secondary measures consist of repeated measures (for example HRQoL and decisional regret) and will be treated according the appropriate mixed- model approach, that is repeated measures anova/ancova will be used for outcomes with two time points (decisional regret, treatment satisfaction) and a random coefficient approach will be used for outcomes with three time points (HRQoL) 67 . Observed variation in treatment choice during the trial period will be compared between groups and at level of the individual hospital. For this second comparison each hospital’s particular historical treatment variation profile (2008 to 2012) will be obtained from the Netherlands Cancer Registry.