Data collection
When designing the EnCoRe study, a conceptual model was developed for studying lifestyle and HRQoL in CRC survivors [
21], based on the International Classification of Functioning, Disability and Health (ICF) of the World Health Organization [
22]. The ICF adopts a broad bio-psychosocial definition of human functioning as a multidimensional concept, which does not only incorporate physical health components (body perspective), but also an individual’s ability to perform his/her daily activities and societal roles (individual and societal perspectives) [
23]. Further, it enables identification of environmental and personal factors and the presence of health conditions that can influence functioning. The developed conceptual model [
21] was adapted for the current research question to identify relevant variables to be measured and included in our data analyses (Supplementary Fig.
1, Online Resource 1).
Sedentary and physical activity time
The triaxial MOX activity monitor (MMOXX1, upgraded version of the CAM monitor) was used for objective measurement of time spent in sedentary behavior, standing, and physical activity (Maastricht Instruments B.V., the Netherlands) [
8,
24]. The MOX has a high reproducibility and an excellent validity for estimating time spent in activities and postures in both a controlled laboratory setting (100 % accuracy and Cohen’s kappa of 0.99, compared with direct observation) and in free-living conditions (intraclass correlation coefficient of 0.98, compared with diary records) [
8]. The monitor was waterproofed in a finger cot (VWR International B.V., the Netherlands) and attached via hypoallergenic plaster (BSN Medical, the Netherlands) to the anterior thigh 10 cm above the knee. Participants were instructed to wear the monitor 24 h/day on seven consecutive days and to record sleep and any non-wear periods.
A customized MATLAB program (version R2012a, The MathWorks, Inc., USA) was used to classify each 1-second epoch of the data as sedentary (i.e., sitting/lying during waking hours with a low energy expenditure of ≤1.5 metabolic equivalents [METs] [
4]), standing (i.e., standing during waking hours with an energy expenditure ≤1.5 METs), or physical activity (i.e., all activities with an energy expenditure >1.5 METs). This classification was done using previously validated thresholds for parameters of motion intensity and orientation of the device [
24]. Time in physical activity was not further subdivided according to intensity level into LPA and MVPA, because of limited reproducibility of the monitor for estimating time in activities at a moderate-to-vigorous intensity [
8]. Self-reported non-wear and sleeping periods were checked by visualization of triaxial acceleration data, with non-wear time periods adjusted if necessary, and sleeping times determined if missing. Further processing of worn waking data was performed in SAS (version 9.3, SAS Institute Inc., USA). Monitor wear days with ≥10 h of waking wear time were considered valid; only participants with ≥4 valid days were included in the analyses [
25]. Sedentary, standing, and physical activity time (h/day) were calculated and averaged across valid measurement days.
HRQoL outcomes
Cancer-specific HRQoL was measured using the valid and reliable European Organization for the Research and Treatment of Cancer Quality of Life Questionnaire-Core 30 (EORTC QLQ-C30, version 3.0) [
26,
27]. For the subscales global quality of life and physical, role, and social functioning, 100-point scores were calculated [
28]. Disability was assessed by the 12-item version of the ICF-based World Health Organization Disability Assessment Schedule II [
29], which has good reliability and validity in different populations, including cancer survivors [
30,
31]. Fatigue was assessed through the 20-item Checklist Individual Strength, which was originally developed and validated in patients with chronic fatigue syndrome [
32,
33], but has also been applied in cancer survivors [
34]. The 14-item Hospital Anxiety and Depression Scale was used to determine levels of anxiety and depression [
35], which has adequate psychometric properties in cancer patients [
36]. By adding scores of individual items within the Checklist Individual Strength and Hospital Anxiety and Depression Scale, scores for fatigue (scale: 20–140), and depression (0–21) and anxiety (0–21) were calculated, with higher scores indicating higher levels of fatigue, depression, and anxiety.
Other factors
Socio-demographic characteristics (gender, age, education level, smoking status, paid employment) and the presence of a stoma were self-reported. Body height and weight were measured by trained personnel for calculation of body mass index (BMI, kg/m
2). The number of comorbidities was assessed using the 13-item Self-Administered Comorbidity Questionnaire [
37]. Perceived deficiency in social support (scale: 6–18) was measured by the six-item Dutch Social Support List (SSL-6) [
38]. Clinical characteristics (cancer stage, age at diagnosis, treatment, and tumor subsite) were collected through the Netherlands Cancer Registry.
Statistical analyses
Descriptive statistics were calculated for socio-demographic and clinical factors in survivors included and not included in the analyses and for accelerometer-derived characteristics and HRQoL outcomes in included individuals stratified by gender. Multivariable linear regression models were used to analyze associations of standing and physical activity time (h/day) with HRQoL outcomes. Unstandardized regression coefficients (β) with 95 % confidence intervals (CIs) were calculated, representing differences in mean HRQoL scores per additional 1 h/day of standing or physical activity, which was similar to one standard deviation (SD) of these variables within the sample. Potential confounding factors included in multivariable models were selected a priori from our ICF-based conceptual model (Supplementary Fig. 1, Online Resource 1). These were either adjusted for in all models (age, gender, number of comorbidities, years since diagnosis, cancer stage, smoking status, and BMI) or, only when retained via backward elimination using
p > 0.2 as a cutoff for exclusion [
39] (education level, paid employment, having a partner, the presence of a stoma, radiotherapy and/or chemotherapy treatment, tumor subsite, and perceived deficiency in social support). None of the models showed evidence for multicollinearity (variance inflation factors ≤5 [
40]).
Three types of regression models were fitted to analyze associations of standing and physical activity with selected HRQoL outcomes [
10]. All models were adjusted for a similar confounder set, but differed with regard to the inclusion of activity variables. First, single-variable models were conducted, which included only one activity variable (sedentary, standing, or physical activity time), thereby estimating
overall associations of these activity categories with HRQoL outcomes separately. Secondly, partition models were fitted which included all activity variables (i.e., sedentary, standing, and physical activity time) in one model, to assess
independent associations of each activity category with the outcome, while keeping time in other activity categories constant.
Third, isotemporal substitution models were fitted for estimating associations with HRQoL of replacing (substituting) time in one category (e.g., sedentary time) with equal time in another category (e.g., standing), while keeping total time and time in the remaining category (e.g., physical activity) constant. Detailed explanation of these models has been published previously [
10]. To address our main research question of estimating associations of substituting sedentary time with standing or physical activity, standing and physical activity time were included in the isotemporal model and sedentary time was left out, while the model was adjusted for total waking wear time (i.e., total time was held constant). By constraining the total amount of time, an increase of 1 h/day in standing time implies substitution of the left-out variable (i.e., 1 h/day less sedentary time) with standing, while holding physical activity time constant. As a result, βs from the standing and physical activity time variables represent differences in mean HRQoL scores associated with substituting 1 h/day of sedentary time with equal time in standing or physical activity, respectively. These isotemporal substitution models were considered our main analyses. Similarly, as an additional analysis, we also assessed isotemporal associations of substituting standing time with physical activity, by including sedentary and physical activity time in the models and leaving out standing time.
Minimum differences of interest were defined and based on minimally important differences for the HRQoL outcomes, i.e., published “medium” differences for the EORTC QLQ-C30 subscales [
41], and 0.5 times the SD of the score for other outcomes [
42] (disability, fatigue, depression, and anxiety). We defined the association to be “meaningful” if the difference in HRQoL outcome associated with a difference of two SDs in the substituted activity variable (i.e., sedentary time or standing time) exceeded these minimum important differences. Otherwise, the association was described as “small.” As the regression coefficients represented the difference in HRQoL outcome score per 1 h/day of the substituted variable, we rescaled the minimum important differences into cutoffs that could be directly compared with the regression coefficients reported, based on this definition. This was done by dividing each of the minimum important differences by two SDs of the substituted activity variable. The cutoffs that were calculated accordingly are shown in Supplementary Table 1 (Online Resource 1). Potential effect modification by gender, age (<70 vs ≥70 years), number of comorbidities (≥2 vs <2), BMI (<30 vs ≥30 kg/m
2), and perceived deficiency in social support (no deficiency [six-item Social Support List score = 6] vs deficiency [score > 6] [
38]) was explored by performing subgroup analyses. To avoid over-interpretation of spurious findings, results were reported if a meaningful and significant association in a certain direction was observed in multiple HRQoL outcomes in one subgroup, but not in the other subgroup.
As HRQoL outcomes were not normally distributed, findings were verified in isotemporal logistic regression models with dichotomized outcomes using gender-specific medians as cutoff [
43]. All analyses were performed using IBM SPSS Statistics (version 22, IBM Corporation, USA), and
p < 0.05 (two-tailed) was considered statistically significant.