2.1 Data source and participants
This study used data collected from the UK Biobank, which is a national cohort study in the UK. Demographic and health-related data were collected from 500,000 participants (aged 40–69 years) between 2006 and 2010, with a reassessment of 20,000 participant between 2012 and 2013 (the most recent data were used, this is referred to as baseline data for the purpose of this current study). Between 2013 and 2015, participants were re-contacted and invited to wear an accelerometer (Axivity AX3 wrist-worn triaxial accelerometer) if they provided a valid email address at the baseline assessment. Of the total recruited 500,000 UK Biobank sample, 236,519 participants were asked to join the accelerometer study to obtain objectively measured PA data under free-living conditions. A total of 106,053 agreed to wear a PA monitor (response rate to invitations = 44.8%) and 103,720 participants returned data between 2013 and 2015 . Data from cancer and death registries were linked to the UK Biobank cohort to provide information on cancer diagnoses and death. The UK Biobank protocol was approved by the North West Multicenter Research Ethics Committee.
We included participants if they: 1) were diagnosed with primary lung cancer, breast cancer, colorectal cancer or prostate cancer after completing accelerometer data collection (see Supplemental Table 1 for ICD 9 and ICD 10 codes); 2) had valid accelerometer data (≥ 3 days of wear time) ; 3) had no missing values for socio-demographic and health-related variables (see below). A total of 2,662 participants were included (lung cancer = 248, breast cancer = 858, colorectal cancer = 451, prostate cancer = 1,105) (see Fig. 1 for details).
Physical activity was collected with the Axivity AX3 wrist worn triaxial accelerometer between 2013 and 2015. The participants were instructed to do the following: 1) start wearing the accelerometer device immediately after receiving it, 2) wear it for seven continuous days on their dominant wrist, 3) carry on with their normal activities, and 4) mail the device back to the research center, in a pre-paid envelope, after the seven-day monitoring period . The raw accelerometer data were calibrated, and wear-time periods were identified using the UK Biobank preprocessing methods described by Doherty et al. . Accelerometer-based summary measures in the dataset included the total mean acceleration/24 hours (vector magnitude in milligravity units = mg) and time spent in sedentary, light, and moderate-to-vigorous PA (MVPA). The proportion of time spent in moderate and vigorous PA was defined as the proportion of time spent in accelerations of 101–425 and > 425 milligravity, respectively [28–31].
Health-related factors were collected at baseline (2006–2010 and 2012–2013). (1) Self-reported overall health was rated as excellent, good, fair, or poor. (2) Self-reported comorbidities were measured using a 13-item comorbidity check list. For the purposes of this research, we analyzed data from patients with the most common cardiovascular and pulmonary comorbidities (heart attack, angina, stroke, hypertension, COPD, and asthma) and diabetes. The number of comorbidities ranged from 0 to 7. (3) Self-reported walking pace was measured using an item “How would you describe your usual walking pace?” with response options of slow, steady/average, or brisk. Participants could access further information which defined a slow pace as less than 3 miles per hour, a steady/average pace as between 3–4 miles per hour, and a brisk pace as more than 4 miles per hour. (4) Grip strength was assessed in each hand using a hydraulic hand dynamometer (Jamar J00105, Lafayette, IN, USA), which can measure isometric grip force up to 90 kilograms . Grip strength was measured in both hands and the highest value was used for analyses. (5) Self-reported anxiety and depression was measured using a short version of the Patient Health Questionnaire (PHQ-9) and Generalized Anxiety Disorder [33–35]. Participants were asked “How often have you felt down, depressed, or hopeless”, “How often have you had little interest or pleasure in doing things”, “How often have you felt tense, fidgety or restless” and “How often have you felt tired or had little energy” over the past two weeks, with response options of “not at all = 1”, “several days = 2”, “more than half the days = 3” and “nearly every day = 4”. Scores ranged from 4 to 16, in which higher scores indicated more severe symptoms.
Socio-demographic characteristics were collected at baseline (2006–2010 and 2012–2013), including age, sex, ethnicity (white/non-white), Townsend Index of deprivation (high scores indicated higher levels of socioeconomic deprivation) , body mass index (BMI, underweight/normal/overweight/obese), smoking status (never/previous/current smoker), and alcohol drinking frequency (≤ 1–3 times/month, 1–4 times/week, daily or almost daily).
Date of cancer diagnosis and death were linked to the UK Biobank dataset. The included participants were diagnosed with lung/breast/colorectal/prostate cancer between 2013 and 2020 (at 4 days-6.5 years after accelerometer data collection). We followed participants from their date of cancer diagnosis to their date of death as provided by UK Biobank’s linkage to death registration data or to the latest follow-up date for mortality data (2021/3/21) if they did not have a death record.
2.3 Data analysis
Stata SE 17.0 software was used for data analysis. Descriptive statistics (percentages for categorical variables and mean and standard deviation for continuous variables) were calculated for socio-demographic, health-related characteristics, and accelerometer-measured PA in each type of cancer, stratified by gender. We compared the socio-demographic, health-related characteristics, and accelerometer-measured PA among different cancer groups stratified by gender. Chi-square tests of independence were used for categorical variables and ANOVA tests were used for continuous variables. A P-value of less than 0.01 was considered statistically significant for all analyses.
To address the three study aims (see Introduction section), we used linear regressions and survival analyses. Aim 1: Linear regression was used to compare time spent in MVPA between patients with lung cancer and other types of cancer, stratified by gender (independent variable = type of cancer; dependent variable = MVPA). The linear regression models included both unadjusted and adjusted estimates that control for socio-demographic characteristics. The unadjusted and adjusted coefficients and 95% confidence intervals (95% CI) were reported.
Aim 2: Linear regressions were used to examine the correlates of PA for each type of cancer (independent variable = gender, age, race, Townsend Index of Deprivation, BMI, smoking status, Alcohol drinking frequency, overall health rating, number of comorbidities, walking pace, grip strength, and anxiety and depression; dependent variable = MVPA). The linear regression models included both unadjusted and adjusted estimates that control for other socio-demographic and health-related characteristics. The unadjusted coefficients, adjusted coefficients, 95% CI, and adjusted standardized coefficient and were reported.
Aim 3: Survival analyses (Cox regressions) were used to assess the potential impact of time spent in MVPA on all-cause mortality. We used Cox regressions to model time-to-death as a function of time spent in MVPA per day and controlled for socio-demographics, cancer types and comorbidities. The unadjusted and adjusted Hazard Ratio (HR) and 95% CI for all-cause mortality were reported. Interaction analysis was performed to explore whether cancer types modified the association between MPVA and all-cause mortality.