Iris Kanera

6 141 Long-term effects of the Kanker Nazorg Wijzer conducted to determine the 6-month effects of the KNW (Kanera et al., 2016). Dummy coding was used for categorical variables including more than two categories. The models were fitted by using the maximum likelihood procedure and an independent covariance structure was chosen (Twisk, 2006). Besides complete case analysis, intention-to-treat analysis (ITT) was also conducted (Montori & Guyatt, 2001). Missing data of the 6-month and 12-month measurement were imputed 20 times, and multiple imputation analyses were conducted including all variables of the fully adjusted MLA models, as described above (Enders, 2010). Due to the use of two outcome variables, the false discovery rate correcting procedure (FDR) was applied to account for multiple testing problems (Benjamini & Yekutieli, 2001). In order to evaluate whether module use, gender, age, and education possibly moderate the intervention effect, the fully adjusted three-level MLA model was adapted as follows: To evaluate module use as a moderator, intervention condition was categorized into three categories ( IC- module used , IC-module not used , control condition ). To explore moderating effects of gender, age, and education, interaction terms between intervention condition and gender, intervention condition and age, and intervention condition and educational level were also added to the model. The continuous variable age was centered and dummy coding was used for educational level (three categories), with educational level low as reference category. The moderator analyses were conducted by using the complete cases. Cohen’s d effect sizes were calculated for the main results and the results of the moderator analysis (Cohen, 1992). Additionally, Cohen’s f 2 was calculated in order to evaluate the local effect size within the context of the fully adjusted MLA model (Cohen, 1992; Selya et al., 2012). Due to differences in the datasets of the 6-month and the 12‑months follow-up (due to loss to follow-up) and the three-level data structure, the outcomes of the current paper varied slightly from the corresponding outcomes after 6 months (Kanera et al., 2016). All analyses were conducted using STATA version 13.1 and the calculations of the FDR corrections were conducted in R 3.2.3., base package (R Development Core Team, 2011). RESULTS Study population The flowdiagram in Figure 6.2 displays the study participation of the respondents during the study period. In total, 381 (82.5%) participants filled in the 12-month follow-up questionnaire and 81 (17.5%) were lost to follow-up since baseline. Dropout after 12 months was predicted by allocation to the IC ( B = 1.805, p = .000), gender (male) ( B  = -1.041, p = .046), (lower) high income ( B = -.866, p = .047), (lower) modal income ( B = -.992, p = .018), ‘other’ cancer treatment ( B = . 959, p = .020), (less) vigorous PA at baseline ( B = -.001, p = .022), (lower) vegetable intake at baseline ( B = -.005, p = .031), and (higher) fruit intake at baseline ( B =