Ietje Perfors

143 Linear mixed regression analyses were used for continuous outcome variables adjusted for baseline variables (if measured at baseline) and treating hospital and cancer type. In these longitudinal analyses, the statistic model accounts for missing data based on the observed data. 25 The questionnaires addressing satisfaction with the specialist, GP and nurse at T3 and T5, were only offered to the patients who had visited the corresponding healthcare workers. Besides, both the T3 and T5 assessments evaluated the period from inclusion to the assessment. So, the study population differs in size and the assessed period overlaps. Therefore, T3 and T5 were analysed separately using an ANOVA adjusted for treating hospital and cancer type (breast/lung/ colorectal/gynaecologic/melanoma). The difference in healthcare utilisation for categorical data (i.e., paramedical care) was calculated with a Pearson Chi-Square or a Fisher exact test and for count data with a log-binomial regression. Themajority of patients had no emergency department (ED) visits and/or emergency hospitalization. Therefore, we dichotomized the outcomes (i.e., no versus ≥1 ED visit). Healthcare utilisation outcomes were adjusted for treating hospital and cancer type. Additionally, because of group imbalances, we adjusted for co-morbidity (none/≥1) as sensitivity analysis. Pre-specified subgroup analyses (co-morbidity (none/≥1), type of cancer (breast/colorectal/other), sex (male/female), age (≤65/> 65 year), baseline levels of the outcomes of interest) 12 were performed to investigate differential intervention effects on patient satisfaction by adding interaction terms to the regression model. Sample size A medium effect size (0.5) was assumed to be a clinically relevant difference in patients’ satisfaction between the two study groups. Using a power of 0.8 and an alpha less than 0.05, at least 64 patients per study group were required. Accounting for an estimated dropout of 15%, 75 participants in each group were needed. 6