Iris Kanera

5 117 Lifestyle-related effects of the Kanker Nazorg Wijzer intercept. All random parameters were added with an independent data structure. Next, the crude model was adjusted for standard demographic and disease-related characteristics, significant variables from dropout analysis, and baseline differences: i.e., gender, age, marital status, education level, income level, employment status, BMI, type of cancer, having had cancer before, type of treatment, time since completion of primary cancer treatment, aftercare, comorbidities, vegetable, fruit, whole grain bread and fish intake at baseline. These variables were added as fixed intercepts and dummy-coding was used for categorical variables including more than two categories. For testing the effect of following a specific module, intervention condition was categorized into three categories (0 = UC, 1 = IC, specific module not followed, 2 = IC, specific module followed) in the fully adjusted MLA-models. Analyzing the intervention effect on smoking behavior after six months by using multilevel logistic regression analysis was not possible due to the small number of smokers. Chi-square tests were applied to assess differences between IC and UC at baseline and follow-up. Cohen’s d effect sizes were calculated for the main effect results on PA and dietary behavior by dividing the difference between the relevant two means of IC en UC at follow-up by the pooled standard deviations of those means (Cohen, 1992). For the sub-analysis of following modules ( yes , no ), Cohen’s d was adjusted for the baseline value by dividing the difference between the means of the relevant change scores by the pooled standard deviation of those means. Additionally, Cohen’s f 2 was calculated in order to evaluate the local effect size within the context of the fully adjusted MLA model with f 2  ≥ .02, f 2 ≥ .15, and f 2 ≥ .35 represent small, medium, and large effect sizes, respectively (Cohen, 1992; Selya, Rose, Dierker, Hedeker, & Mermelstein, 2012). To index the magnitude of the effect for smoking, as according to Durlak (2009), the odds ratios (OR) were calculated by comparing the odds of smoking cessation for the intervention group with the odds of smoking cessation for the control group. For generating computer tailored messages within the intervention, it was necessary that respondents filled out all questions of the baseline measurement. Consequently, only those respondents, who completed the baseline measurement without missing data were included in analyses. To assess the intervention effects among respondents who also participated during the follow-up measurement, only complete cases were analyzed. This means that cases with missing data at the follow-up measurement were excluded. Besides that, intention-to-treat analysis (ITT) has been conducted in order to additionally display unbiased estimates of the intervention effects (Montori & Guyatt, 2001). For PA and dietary behavior outcomes, multiple imputation analyses were conducted by including all variables of the fully adjusted MLA model into the multiple imputation process and using 20 imputed datasets. This is in accordance with the argumentation of Enders (2010). With regard to smoking outcomes, for ITT, participants who were identified as smokers at baseline were