Iris Kanera

116 Chapter 5 Other relevant measures Background information was collected at baseline using standard questions on age, gender, marital status ( with partner : married, cohabiting partners; without partner : single, divorced, widowed), education level ( low : lower vocational education, medium general secondary education; medium : secondary vocational education, higher general secondary education; high : higher vocational education, university education), income level ( below average : < €1800 per month; average : > €1800 and < €2200 per month; above average : > €2200 per month), employment status ( working : self-employed, in paid employment; not working : unemployed, retired, unable to work), type of cancer, type of treatment, time since completion of primary treatment, aftercare, comorbidities, length and weight (body mass index [BMI]). Although other variables were also assessed, these were not used for the current study. Following specific modules and the number of weeks since first login were derived from program logging data. Sample size Sample size calculation revealed that each intervention condition needed to contain 144 participants (effect size = .30; one sided α = 0.05; β = 0.2; power = 80%); intra-class correlation coefficient (ICC) = .005). With an expected dropout of some 20% - 23%, the required sample size was N = 376 (188 per condition) at baseline.  Statistical analyses Preparatory and descriptive analyses were conducted using SPSS 22, and for calculation of the intervention effects, STATA version 13.1 was applied. The dataset was assessed for outliers and aberrant measurement data. Baseline differences between IC and UC concerning lifestyle behaviors, demographic and cancer-related characteristics were examined using independent t‑tests and chi-square tests. Selective dropout was assessed by applying logistic regression analysis with dropout as outcome variable (0 = no; 1 = yes) and group assignment and baseline characteristics as predictive factors. In order to measure intervention effects at follow-up in PA and dietary behavior, multilevel linear regression analysis (MLA) was applied. A two-level data structure was used with individuals (level 1) nested within hospitals (level 2), taking the possible aftercare differences between hospitals into consideration because there might be interdependence between participants from the same hospital. Model testing proceeded in two phases, the crude and adjusted analyses, in line with Twisk (2006). The crude model included the dependent variable (behavior), the intervention condition (0 = UC, 1 = IC), and the baseline value of the outcome behavior as fixed intercepts with random slopes, and hospital as random