The association between recreational sedentary behaviour and cancer risk: Findings from the UK Biobank, a large prospective cohort study

Background Evidence is suggestive of sedentary behaviour being associated with an increased risk of endometrial cancer, but the evidence base is too limited to draw any conclusions for other cancers. The aim of the study was to investigate the association between recreational sedentary behaviour and site-specific, and total, cancer incidence. Methods We analysed data from the prospective UK Biobank cohort study. Cox proportional hazards models were used to estimate hazard ratios (HRs) and 95% confidence intervals (CIs) between recreational sedentary behaviour (including television (TV) viewing time, computer use time and daily total screen time) and cancer risk. Partition models and isotemporal substitution models investigated the impact of substituting recreational sedentary behaviour with physical activity. Results During a mean follow-up of 7.6 years, 28,992 incident cancers were identified among 470,578 adults. A 1-hour increase in daily TV viewing time was associated with higher risks of oropharyngeal, oesophago-gastric and colon cancer in fully adjusted models. Participants who reported ≤1, compared with 1-≤3, hours/day of TV viewing time had lower risks of lung, breast, and oesophago-gastric cancer. Findings were inconsistent for computer use and daily total screen time. The majority of observed associations were small, and were attenuated after excluding cancers diagnosed within the first two years of follow-up, except for oesophago-gastric and colon cancers (HR 1.05, 95% CI: 1.01-1.10; and HR 1.04, 95% CI: 1.01-1.07 per 1-hour increase in daily TV viewing time, respectively). However, isotemporal substitution models showed reduced risk of total cancer (HR 0.97, 95% CI: 0.95, 0.99) and some site-specific (oropharyngeal, lung, breast and colorectal) cancers when replacing 1-hour/day of TV viewing with 1-hour of moderate-intensity physical activity or walking. Conclusions Our findings show that recreational sedentary behaviours, particularly TV viewing, were associated with small increased risks of oesophago-gastric and colon cancer. Replacing 1-hour per day of TV viewing with 1-hour


Introduction
Research in sedentary behaviours has grown rapidly over recent years (1). Such behaviours are seen as distinct from physical inactivity or sleep, and have been defined as "any waking behaviour characterised by an energy expenditure ≤1.5 metabolic equivalents (METs), while in a sitting, reclining or lying posture" (1,2). This definition is typically operationalised as self-reported sitting Evidence demonstrates that prolonged sedentary time is associated with increased risk of noncommunicable diseases (NCDs). Mechanistically, sedentary behaviour is thought to impact particularly on cardio-metabolic diseases through adverse effects on lipid and glucose metabolism (7,8). Recent evidence from a meta-analysis has demonstrated a significant direct association between 6-8 hours daily sedentary time and increased all-cause mortality, cardiovascular disease mortality and Type 2 Diabetes Mellitus risk (9). Prolonged sedentary behaviour is therefore a significant burden on our healthcare systems. In 2016-2017, for example, it was estimated to cost the UK National Health Service £0.8 billion (10).
However, much less is known about sedentary behaviour and cancer, and known biological mechanisms are less well understood (11). The World Cancer Research Fund/American Institute for Cancer Research (WCRF/AICR) global report in 2018 stated that evidence on sedentary behaviours is limited but is suggestive as being associated with an increased risk of endometrial cancer (pooled risk estimate from three studies comparing the highest versus lowest levels of sitting time was 1.46, 95% CI: 1.21, 1.76, cases=1579) (11-14). The evidence base was deemed too limited to draw any conclusions for other cancers (11). However, in a 2018 meta-analysis, Patterson et al. demonstrated significant linear associations of TV viewing with cancer mortality (N=4 studies; relative risk [RR] 1.02, 95% CI: 1.01, 1.03 per 1-hour increase in TV viewing per day) (9).
More recent evidence from analyses of the large, prospective UK Biobank cohort shows mixed evidence for an association between sedentary behaviour and cancer outcomes (15). Celis-Morales et al (2018) found significant associations of discretionary (or recreational) screen-time (time spent in TV viewing or computer screen use during leisure time) exposure and all-cause mortality (hazard ratio [HR] 1.06, 95% CI: 1.05, 1.07), and cancer incidence (HR 1.04, 95% CI: 1.03, 1.04). This study also found that these results were substantially attenuated by physical activity, cardiorespiratory fitness and grip strength (15). Our research group have previously found no association between nonoccupational screen-based sedentary behaviour levels and oesophago-gastric cancer risk within the UK Biobank cohort (16). In contrast, higher levels of TV viewing time were associated with a greater risk of colon cancer in the same study population (HR for ≥5 hours per day vs ≤ 1 hour per day = 1.32, 95% CI: 1.04, 1.68) (17), although time spent using computers (excluding using a computer at work) was not associated with colorectal cancer risk in the UK Biobank cohort (17). The findings of a 2017 meta-analysis including six studies also demonstrated significant associations between the highest compared with the lowest levels of occupational sedentary behaviour, and risk of colon cancer (pooled RRs 1.44, 95% CI: 1.28, 1.62) (18). On the other hand, there was little evidence of an association between sedentary behaviour and rectal cancer risk (18).
Many of the previous studies investigating the association between sedentary behaviour and health outcomes have attempted to adjust for physical activity levels in their analysis. A recent US Government report has highlighted limited evidence on the role of physical activity in displacing the mortality risks associated with sedentary behaviour (6). Previous research has demonstrated that high levels of physical activity can attenuate the risks associated with sedentary behaviour (19,20).
An improved understanding of these interactive effects would enable more specific recommendations to be made regarding quantifying prolonged sedentary time. Much of the previous research has modelled the joint effects of physical activity and sedentary behaviour (20). However, other analytical approaches such as isotemporal substitution and partition models (19), enable us to model replacing sedentary behaviour with physical activity which may be a more time efficient method of promoting healthy behaviour. Analytical techniques such as partition models and isotemporal substitution models (21) could help to model such predictions, but have yet to be extensively applied in large cohort analyses.
Therefore, this study aimed to add to the relatively scant evidence base (11) by investigating recreational sedentary behaviour (including TV viewing, computer use and total screen-use) in relation to the risk of site-specific and total cancer risk in the large UK Biobank cohort study. Partition and isotemporal substitution models were also used to investigate the impact of substituting recreational sedentary behaviour with physical activity, in relation to cancer risk.

Study design
Between 2006 and 2010, UK Biobank recruited a cohort of 502 619 adults (5.5% response rate) aged 40-69 years from the general population (22,23). Approximately 9.2 million invitations were mailed to potential participants who were registered with the National Health Service (NHS) and living within a 25-mile radius of one of the 22 assessment centres across England, Scotland and Wales.
From this overall cohort, we excluded participants if: (1) they had been diagnosed with malignant cancer (excluding non-melanoma skin cancer) at baseline (n=26 868); and (2) they did not complete the self-report assessments of their TV viewing time (n=5078), computer time (n=8000) or total screen time (n=11232); (3) they requested to be removed from the UK Biobank dataset as per General Data Protection Regulation (GDPR) (n=95). This resulted in 470 578 participants being included in the analysis for TV viewing time, 467 656 participants being included in the analysis for computer use time and 464 424 participants being included in the analysis for total screen time. All participants provided informed consent.

Screen time assessment
We have used TV viewing time as our primary exposure. Firstly, TV viewing time was almost three times more prevalent as a recreational sedentary behaviour than computer use within this UK population. Secondly, we were concerned that using total screen time as the primary analysis may overestimate total screen time through double counting (if participants watched TV and used computers at the same time). Therefore, we did not feel that total screen time was an appropriate focus for our analysis primary analysis. Thirdly, as acknowledged in a recent study (24), a large body of research has used TV viewing as a measure of sedentary behaviour, demonstrating consistent associations, in particular with CVD risk, in population cohort studies akin to the UK Biobank.
Relevant screen-time exposure variables were assessed by self-reported time spent watching TV, time spent using the computer outside of work, which were used to derive total screen time. Selfreported TV viewing time was assessed for all participants by asking the following question: "In a typical DAY, how many hours do you spend watching TV? (Put 0 if you do not spend any time doing it)?" Self-reported computer use time was assessed for all participants by asking the following question: "In a typical DAY, how many hours do you spend using the computer? (Do not include using a computer at work; put 0 if you do not spend any time doing it)." Durations of <0 hours were set to missing, as were responses of "Do not know" or "Prefer not to answer". If the respondent replied "Less than an hour a day", this was recoded to 0.5 hours. Total screen time was then computed as the sum of hours spent watching TV and hours spent using the computer. If the summation of total hours spent watching TV and hours spent using the computer was greater than 24, this was set to missing (n=35).

Physical activity assessment
Self-report physical activity was assessed for all participants using the validated short-form International Physical Activity Questionnaire (IPAQ) (25) on which participants reported the frequency (i.e. days/week) and duration (i.e. minutes/day) of walking, moderate-and vigorous-intensity physical activity in the past seven days. For each domain (walking, moderate, vigorous), durations of <10 minutes/day were recoded to 0 and durations of >180 minutes were truncated at 180 minutes/day in line with IPAQ processing rules. This was used to derive hours/day spent in walking, moderate-and vigorous-intensity physical activity. All data processing was carried out according to official IPAQ rules (26).

Assessment of covariates
Height (m), weight (kg), and waist and hip circumference (cm) were measured by staff at the UK Biobank study centre. Body mass index (BMI) was then calculated from the weight and height measurements (kg/m 2 ). Waist circumference measurements were taken from the level of the umbilicus and regarded as a measure of central obesity, using official cut-off values established by the International Diabetes Federation (>94cm in men and >80cm in women) (27). Age, sex and postcodes were acquired from a central registry for all participants and updated by the participant.
Participants also self-reported their ethnicity, educational attainment, lifestyle behaviours (smoking status, alcohol consumption, dietary intake and sunscreen/ultraviolet (UV) protection use) and medical history using electronic questionnaires. Townsend deprivation scores were derived from postcodes (28). Core confounders for all models included socio-demographic factors (i.e. age, sex, ethnicity, educational attainment and deprivation index), smoking status, alcohol consumption, fruit and vegetable consumption, BMI, height and waist-hip ratio. Cancer site-specific confounders included use of sun/UV protection (melanoma), self-reported oesophageal reflux (oesophagus cancer), diabetes at baseline (pancreatic and colorectal cancers), aspirin use (colorectal cancers), red and processed meat intake (colorectal cancers), hormone replacement therapy (HRT) use (breast, uterus and colorectal cancers), oral contraceptive use (breast and uterus cancers), number of live births (breast and uterus cancers), age at menarche (breast and uterus cancers), age at menopause (breast and uterus cancers), hysterectomy status (breast and uterus cancers) and self-reported family history of cancer (total cancer, lung, prostate, and breast cancers), based on known aetiological risk factors for these tumours.

Cancer ascertainment
For the present analysis, the main outcomes were incident site-specific cancers and total cancer (excluding non-melanoma skin cancer). Incident cancers for participants in the UK Biobank cohort were identified through records maintained at national cancer registries (Health and Social Care Information Centre and the NHS Central Register) and identified from the International Classification of Diseases, 9 th and 10 th revisions (ICD-9 and ICD-10 (29)). Cancer outcomes were coded according to ICD-9 and ICD-10 as follows: all cancers excluding non-melanoma skin cancer (ICD-10: C00-C97

Statistical analyses
Our statistical analyses addressed the following research questions:

1.
What is the association between recreational sedentary behaviours (i.e. TV viewing, computer use and total screen time) and site-specific cancers (including endometrial, colorectal, pre-and post-menopausal breast, prostate, lung, and other cancers)? intake. For analyses including gender-specific covariates (e.g. incident total cancers, colorectal cancer, colon cancer and rectum cancer), separate models were run for males and females and HRs were combined using inverse variance meta-analysis and a fixed-effects model (30-32). Participants were excluded from the analysis if they did not have the complete exposure and covariate data required for each model. We did not adjust for total dietary energy intake as the large amount of missing data (for 57.6% of participants) made this unfeasible. Further analyses were conducted to investigate the role of anthropometric factors by running all models with and without adjustment for waist-hip ratio, as a proxy for central adiposity. Models including incident total cancers, breast cancer, prostate cancer and lung cancer were run with and without adjustment for self-report family history (mother, father, siblings). The oesophageal cancer model was also run with and without adjustment for self-reported gastro-oesophageal reflux disease (GORD).
These analyses were repeated separately to investigate the relationship between a 1-hour increase/day in (1) time spent using the computer; (2) total screen time; and cancer risk. Since the association between time spent in recreational sedentary behaviour and cancer risk may not be linear (9,33), we repeated these analyses with categorised independent variables as follows: daily TV The proportional hazards assumption was tested for each model formally using Schoenfeld residuals (p<0.05 indicated potential violation of the proportional hazards assumption), and by visual inspection of scaled Schoenfeld residual plots (41) and log-log plots (parallel curves indicated that there was no evidence for violation of the proportional hazards assumption). Analyses were carried out using Stata 13 (42).

Results
Participant characteristics according to total TV viewing time are shown in Table 1. Among the 470,578 participants included in this analysis, 53.8% were women and the mean age was 56.3 years.
Most participants reported that they spent between 2 and 8 hours/day watching TV or using the computer. During a mean follow-up time of 7.6 (SD 1.4) years (median 7.8 years, interquartile range 7.0-8.5), 28,992 incident cancers were identified.

Association of cancer risk and daily TV viewing time
After excluding cancers diagnosed within the first two years following baseline, all associations were attenuated except those for oesophagus and stomach cancers, and colon cancers (Table 3). Whilst the results of the Schoenfeld residual tests indicated that some of our models may not have been in line with the proportional hazards assumption, our visual inspection of log-log plots and Schoenfeld residual plots showed no serious violations. Therefore, we proceeded with the analyses as planned.

Results of partition models and isotemporal substitution models
Partition models showed there was an association between a 1-hour increase in daily TV viewing time when holding time spent in moderate-and vigorous-intensity physical activity constant (Table 4).
(Insert Table 4 here) (Insert Table 5 here) compared to participants who reported 1-≤4 hours of daily total screen time.

Association of cancer risk and daily total screen time
(Insert Table 6 here)

Overview of key findings
This large, prospective cohort study indicates that recreational sedentary behaviours were associated with some site-specific cancers (notably oropharyngeal, oesophagus and stomach, colon and lung cancer), particularly for TV viewing time, albeit mainly small associations. Results for oesophagus and stomach cancers, and colon cancers were robust to the omission of cancers occurring within the first two years of follow-up. However, for many of the other cancer sites the associations were attenuated after eliminating cancers diagnosed within two years, suggesting reverse causation.
Our study provides no evidence for an association between recreational sedentary behaviour and total cancer risk. However, the results of our isotemporal substitution models revealed a benefit in terms of reduced risk of several site-specific cancers and reduced total cancer risk when replacing 1hour of TV viewing per day with 1-hour of moderate-intensity physical activity or walking. Results were less consistent for time spent on computer and daily total screen time, and were often in the opposite direction to TV viewing time. This may suggest that the mechanism of action is more nuanced and complex than the act of being sedentary, but the specific activity that is being there is also a strong evidence base associating being overweight or obese to increased cancer risk (7,51). However we adjusted for BMI in our models to try to account for this. Known mechanisms associated with body fatness, such as sex hormones, insulin, and inflammation, may explain part of the association between recreational sedentary behaviours and cancer risk. The association between prolonged TV viewing time and lower levels of vitamin D have also been hypothesised as a possible mechanistic pathway (7,11). However, the association between TV viewing and cancer risk may also be explained by unmeasured confounders as people who do not watch TV are likely to be different from the broader population in a number of ways.

Computer use and cancer risk
The mean computer use time was 1.1 hours/day, which is almost three times less prevalent as a recreational sedentary behaviour than daily TV viewing time within this UK population. Paradoxically, our findings showed that a 1-hour/day increase in computer use was associated with lower risk of oropharyngeal cancer and the results of the categorical analysis showed that 0 hours/day of computer use was associated with higher risk of oropharyngeal and ovary cancers compared with ≤1 hour/day.
Reporting >3 hours/day of computer use was also associated with increased risk of lung cancer. It is difficult to compare the findings for computer use with other literature given the explicit exclusion of 'using a computer at work' from our measure. Most of the previous literature is focused on occupational sedentary time which largely encompasses computer use (17).

Daily total screen time and cancer risk
The mean daily total screen time was almost 4 hours/day, reflecting combined TV viewing and computer use time. The most notable associations were observed for an increased risk of lung cancer in both continuous and categorical analysis. Previous literature has demonstrated that household air pollution exposure from solid fuel is associated with high rates of lung cancer, especially in low-and middle-income countries, such as China (52). However, this seems an unlikely mechanistic pathway in the UK. It is plausible that indoor sedentary behaviour may be linked to increased residential radon exposure which is known to be associated with an increased risk of lung cancer, particularly in European populations (53). Results were somewhat mixed for other cancers which may be due to the combined nature of essentially two different behaviours (i.e. TV viewing and computer use).

Findings in relation with other literature
Our observations are somewhat mixed to those previously reported for total cancer incidence (15), oesophago-gastric cancer risk (16) and colon cancer risk (17) in relation to sedentary behaviour.
However, it is difficult to draw direct comparisons between these studies and our current analysis, since each of those used the lowest category of screen-time exposure as their reference category.
Due to our a priori hypothesis that individuals with less than 1-hour of screen time may have different characteristics, we chose 1-3 hours of screen-time as our reference category. This revealed some novel associations not previously identified, such as protective associations for oesophageal and stomach, cancers in individuals with the lowest screen-time exposure.

Strengths and limitations
This study provides a comprehensive overview of recreational sedentary behaviours for site-specific cancers and total cancer risk. By investigating all cancer sites within the same analytical population, The analysis uses self-report recreational sedentary behaviour data, which may be subject to social desirability and recall bias, and the measure has not been investigated for criterion validity (15).
However, the estimates are in line with previous population estimates (56, 57). Although the UK Biobank cohort does measure sedentary behaviour using accelerometers, we were unable to use this data to examine the association with cancer incidence as the follow-up time was too short (mean follow-up time 1.9 years). The nature of the observational study means that we cannot attribute causal interpretations to our results owing to the potential for residual confounding, particularly for alcohol and tobacco-related cancers. Finally, some associations were attenuated when excluding cancers diagnosed within the first two years of follow-up, suggesting that our results could have been affected by a possible reverse causation bias.

Future research
Due

Conclusions
In summary, our findings show that recreational sedentary behaviours were associated with some

Availability of data and materials:
The data that support the findings of this study are available from UK Biobank but restrictions apply to the availability of these data, which were used under license for the current study, and so are not publicly available. Data are however available from the authors upon reasonable request and with permission of UK Biobank.

Competing interests:
The authors declare that they have no competing interests.  b Based on IDF criteria (waist circumference >94cm in men; >80cm in women).
c Diagnosed by doctor.
d Self-reported.
e Based on self-reported illnesses of father, mother and siblings.
f Regular use defined as most days of the week for the last 4 weeks.
g Female participants only. e Additional site-specific covariates in the final model include diabetes at baseline (yes/no), aspirin use (regular use/non-regular use or no use), HRT use (ever used/never used), red meat intake (portion/week), processed meat intake (portion/week). f Final model also adjusted for waist-hip ratio (>94cm in men, >80cm in women). g Results for males and females combined using meta-analysis as covariates are different. h Final model also adjusted for family history of cancer (mother/father/sibling had cancer, no family history). **Schoenfeld test indicated potential violation of the proportional hazards assumption (p<0.05). Table 3. Results of Cox proportional hazards analyses investigating the association between self-report TV viewing time and cancer incidence (excluding cancers diagnosed within the first 2 years following baseline). 1   e Additional site-specific covariates in the final model include diabetes at baseline (yes/no), aspirin use (regular use/non-regular use or no use), HRT use (ever used/never used), red meat intake (portion/week), processed meat intake (portion/week). f Final model also adjusted for waist-hip ratio (>94cm in men, >80cm in women).
g Results for males and females combined using meta-analysis as covariates are different.
h Final model also adjusted for family history of cancer (mother/father/sibling had cancer, no family history). **Schoenfeld test indicated potential violation of the proportional hazards assumption (p<0.05). e Additional site-specific covariates in the final model include diabetes at baseline (yes/no), aspirin use (regular use/non-regular use or no use), HRT use (ever used/never used), red meat intake (portion/week), processed meat intake (portion/week). f Final model also adjusted for waist-hip ratio (>94cm in men, >80cm in women).
g Results for males and females combined using meta-analysis as covariates are different.
h Final model also adjusted for family history of cancer (mother/father/sibling had cancer, no family history).
**Schoenfeld test indicated potential violation of the proportional hazards assumption (p<0.05).