Data collection and follow-up
Patients with newly diagnosed CRC are identified by their treating clinician during their hospital stay and are interviewed in the hospital or contacted by mail shortly after their discharge by clinicians or clinical cancer registries. At baseline, sociodemographic information, medical, and lifestyle history (including PA) are obtained by trained interviewers using a standardized questionnaire. Three years after diagnosis, detailed information about treatment, other diseases, and recurrence is collected from attending physicians, using a standardized questionnaire. In order to obtain follow-up data including changes in lifestyle (including PA), medical, or recurrence history, and fatigue, CRC patients are sent a questionnaire by mail 5 years after diagnosis. Information about recurrence, other diseases, and new cancers is verified by the patients’ physicians. Patients’ vital status is regularly checked through population registries.
Assessment of physical activity
At baseline, information on retrospective PA was collected by trained interviewers in a personal interview for each age decade between 20 and 80 years, depending on participant’s age at diagnosis. Patients were asked for the hours per week they had engaged in different activities. One question was asked to estimate the amount of time spent on hard work-related PA (e.g. in agriculture, as health care worker or in the military), one question on light work-related PA (housework, gardening, as sales person, hairdresser), one question on walking (e.g. going for walks, going shopping, walking to and/or home from work), one question on cycling (e.g. means of transportation in everyday life, using the bike to and/or home from work), and one question on sports (e.g. soccer, swimming, skiing, mountain climbing, jogging). These retrospective data have been used to address the prognostic impact of PA in recent papers [
11,
32]. Five years after CRC diagnosis, information on average PA during the past week was assessed with a mailed questionnaire that included the short-form of the International Physical Activity Questionnaire (IPAQ). The questionnaire asks for the number of days and minutes per week spent with vigorous PA e.g. jogging, moderate PA e.g. swimming, walking, and sitting.
Based on activity-specific metabolic equivalent (MET) score values described by Craig et al. [
33], MET hours per week (MET-h/wk) were calculated according to activities performed at baseline and at 5YFU. The following task-specific MET-h/wk score values were used at baseline: hard work = 8 MET-h/wk, light work = 2.5 MET-h/wk, walking = 3.3 MET-h/wk, cycling = 6 MET-h/wk, sports = 8 MET-h/wk; and at 5YFU: vigorous PA = 8 MET-h/wk, moderate PA = 4 MET-h/wk, and moderate walking = 3.3 MET-h/wk While from both assessment methods these MET-h/wk can be derived, the wider range of PA domains assessed at baseline compared to the 5YFU and the difference in the assessment methods (personal interview and mail) might hamper the comparability of the obtained METs from baseline and 5YFU and should be kept in mind.
From the baseline assessment, activity-specific lifetime MET-h/wk were derived from the MET-h/wk spent at ages 20, 30, 40, 50, 60, 70, and 80 (assessed at baseline), considering the current age at diagnosis of the patient and the years spent in each decade. Information from the age decade preceding the patients’ current age at diagnosis was used to calculate the activity-specific MET-h/wk for the last age decade (e.g. PA at diagnosis age 60 for participants in the age group 60–69). The activity-specific MET-h/wk were summed up to create the variables baseline PA lifetime and last decade.
In subgroup analyses, baseline PA was categorized into different PA domains (leisure time PA [walking, cycling, sports] and work-related PA [light work, hard work]) and intensities (light PA [light work], moderate PA [walking], and vigorous PA [cycling, sports, hard work]). Physical activity was classified according to the second version of the Physical Activity Guidelines for Americans [
34]: light-intensity PA = 1.1–2.9 METs, moderate PA = 3–5.9 METs, and vigorous PA = ≥6 METs.
From the 5YFU, the MET-h/wk of the last week were calculated for each of the specific activity types and then summed up to obtain the 5YFU PA.
Based on sample distribution, quartiles (Q) for PA at baseline for the last age decade (Q1 = < 74.7 MET-h/wk, Q2 74.7- < 118.3 MET-h/wk; Q3 118.3- < 183.0 MET-h/wk; ≥183.0 MET-h/wk) and 5YFU (Q1 = < 11.6, Q2 = 11.6- < 34.1, Q3 = 34.1- < 79.0, Q4 = ≥79.0) were calculated. Patients in Q1 were defined as physically inactive whereas patients in Q2-Q4 were defined as physically active. To assess associations of different PA levels with fatigue, the lowest quartile was used as the reference category. Further, these quartiles were used to classify survivors in four groups: active maintainers (active at baseline and at 5YFU), increasers (inactive at baseline, active at 5YFU), decreasers (active at baseline, inactive at 5YFU), and inactive maintainers (inactive at baseline and at 5YFU).
For the main analyses, baseline PA information of the last decade was used and defined as pre-diagnosis PA whereas PA at 5YFU was defined as post-diagnosis PA.
Assessment of fatigue
At 5YFU, fatigue was measured using the Fatigue Assessment Questionnaire (FAQ) developed by Glaus et al. [
35], and the Quality of Life Questionnaire-Core 30 (QLQ-C30) [
36] which was developed by the European Organization for Research and Treatment of Cancer (EORTC). The FAQ assesses the dimensions physical, cognitive, and affective fatigue. Since in the DACHS study, only the cognitive (3 items) and affective (5 items) questions of the FAQ were assessed, the fatigue scale of the QLQ-C30 (3 items) was included to additionally assess the physical aspect of fatigue [
37,
38]. Scoring was performed according to the FAQ and the QLQ-C30 scoring manuals [
35,
39]. Cognitive scores were linearly transformed to a 0–9 point scale, affective scores to a 0–15 point scale, and physical fatigue to a 0–100 point scale. Lower scores on cognitive, affective, and physical fatigue imply less fatigue.
Statistical analysis
To estimate the ordinal association between pre- and post-diagnosis PA, Kendall rank correlations were calculated. Adjusted means were computed using multivariable linear regression models to explore the association of pre-diagnosis PA quartiles with fatigue. Comprehensive covariate adjustment included baseline variables such as age, sex, marital status, residential area, education, comorbidities, alcohol intake, smoking, body mass index (BMI), cancer site, cancer stage, radiotherapy, chemotherapy, and stoma.
Multivariable linear regression analyses were repeated, calculating beta values (ß) with 95% confidence intervals (CI) and modeling pre-diagnosis PA as a continuous variable (per 100 MET-h/wk) for different domains (leisure time vs. work-related) and intensities of PA (low vs. moderate vs. vigorous) with fatigue. In order to assess the independent association of the PA domains with fatigue, the multivariable models were additionally mutually adjusted for the other domain. The same procedure was implemented for the intensities of PA.
Additionally, multivariable linear regression models were calculated to explore the association between post-diagnosis PA quartiles and fatigue. Covariate adjustment included the same covariates (updated with information at 5YFU) as used in the analysis of pre-diagnosis PA and fatigue. In sensitivity analyses, pre-diagnosis PA was added to the model, and in a second step CRC recurrence. Since the results did not substantially change using the additional covariate adjustments, only results of the first covariate adjustment are reported. Moreover, partial r2-values were calculated to assess the independent proportion of the explained variance of fatigue by pre- and post-diagnosis PA after adjustment for potential confounders.
Multiple linear regression models were repeated for the association between changes in PA and fatigue, using the same covariates (updated with information at 5YFU) as used in the analysis of pre-diagnosis PA and fatigue.
Complete case analyses were performed since the proportion of missing values was generally low. Information regarding fatigue at 5YFU was missing in less than 2.5% of all cases. No adjustment for multiple testing was performed, given the exploratory nature of the analysis. The statistical software SAS 9.4 (SAS Institute) was used to perform all data analyses. All statistically significant results mentioned in this study refer to a p-value < 0.05 in two-sided testing.