Leisure time television watching, computer use and risks of breast, colorectal and prostate cancer: A Mendelian randomisation analysis

Abstract Background Sedentary behaviours have been associated with increased risks of some common cancers in epidemiological studies; however, it is unclear if these associations are causal. Methods We used univariable and multivariable two‐sample Mendelian randomisation (MR) to examine potential causal relationships between sedentary behaviours and risks of breast, colorectal and prostate cancer. Genetic variants associated with self‐reported leisure television watching and computer use were identified from a recent genome‐wide association study (GWAS). Data related to cancer risk were obtained from cancer GWAS consortia. A series of sensitivity analyses were applied to examine the robustness of the results to the presence of confounding. Results A 1‐standard deviation (SD: 1.5 h/day) increment in hours of television watching increased risk of breast cancer (OR per 1‐SD: 1.15, 95% confidence interval [CI]: 1.05–1.26) and colorectal cancer (OR per 1‐SD: 1.32, 95% CI: 1.16–1.49) while there was little evidence of an association for prostate cancer risk (OR per 1‐SD: 0.94, 95% CI: 0.84–1.06). After adjusting for years of education, the effect estimates for television watching were attenuated (breast cancer, OR per 1‐SD: 1.08, 95% CI: 0.92–1.27; colorectal cancer, OR per 1‐SD: 1.08, 95% CI: 0.90–1.31). Post hoc analyses showed that years of education might have a possible confounding and mediating role in the association between television watching with breast and colorectal cancer. Consistent results were observed for each cancer site according to sex (colorectal cancer), anatomical subsites and cancer subtypes. There was little evidence of associations between genetically predicted computer use and cancer risk. Conclusions Our univariable analysis identified some positive associations between hours of television watching and risks of breast and colorectal cancer. However, further adjustment for additional lifestyle factors especially years of education attenuated these results. Future studies using objective measures of exposure can provide new insights into the possible role of sedentary behaviour in cancer development.


| INTRODUCTION
Breast, colorectal and prostate cancer are three of the most common malignancies collectively accounting for an estimated 29% of new cancer cases in 2020. 1 Sedentary behaviour is defined as any waking behaviour characterised by energy expenditure ≤1.5 metabolic equivalents while in a sitting, reclining or lying posture. 2 The most common sedentary activities are television watching and computer use; these are more accurately recalled than total sedentary time and are therefore commonly used as surrogates of sedentary behaviour. 3A recent US study reported that approximately two-thirds of adults spent two or more hours each day watching television and around 50% spend more than 1 h using their computer outside work. 4Studies in the United Kingdom and in the United States estimated that adults on average spend 5-6 h per day sitting. 4,5Given such a high prevalence, sedentary behaviours represent an important public health challenge as they have been linked with multiple adverse health outcomes. 6,7umerous observational studies have examined the associations between sedentary behaviours and the risks of breast, colorectal and prostate cancer. 8A meta-analysis of case-control and cohort studies reported that sedentary behaviour was not associated with colorectal cancer risk. 8More recently, however, a UK Biobank analysis found that greater volumes of television watching were associated with elevated colon cancer risk. 9The aforementioned meta-analysis did not observe any significant associations between sedentary behaviour and risk of prostate cancer. 8For breast cancer, when the meta-analysis included cohort studies only, sedentary behaviour was associated with a higher breast cancer risk. 81][12] Mendelian randomisation (MR) is an alternative way to investigate potential causal associations.MR uses germline genetic variants as proxies (or instrumental variables) for exposures of interest to make causal inferences between an exposure and an outcome. 13Unlike traditional observational epidemiology, if all underlying assumptions are satisfied, MR can reduce conventional confounding owing to the random independent assignment of alleles during meiosis. 14In addition, multivariable MR methods have been developed to adjust for confounding if found to be present or for possible pleiotropy bias due to horizontal pleiotropy of a specific effect.MR studies should be less prone to reverse causation, as germline genetic variants are fixed at conception and are consequently unaffected by the disease process. 14Recent MR analyses reported a positive effect estimate for television watching and overall sedentary time with breast cancer risk. 15,16However, these analyses either relied on a small number of instruments or were not very detailed in terms of cancer subtype.Furthermore, there is less evidence for colorectal and prostate cancers. 15e used a two-sample MR framework to examine potential causal associations between self-reported sedentary behaviours and risks of breast, colorectal and prostate cancer.Genetic variants associated with leisure television watching and computer use were identified from a recent subtypes.There was little evidence of associations between genetically predicted computer use and cancer risk.

Conclusions:
Our univariable analysis identified some positive associations between hours of television watching and risks of breast and colorectal cancer.However, further adjustment for additional lifestyle factors especially years of education attenuated these results.Future studies using objective measures of exposure can provide new insights into the possible role of sedentary behaviour in cancer development.[20]

| MATERIALS AND METHODS
2.1 | Data on leisure sedentary behaviours Summary-level data on duration of leisure sedentary behaviours for men and women combined were obtained from a recently published GWAS conducted in 408,815 participants of European ancestry from the UK Biobank using BOLT-LMM v2.3beta2, using a mixed linear model correcting for population structure and cryptic relatedness. 17To ascertain the duration of the sedentary behaviours, participants within the UK Biobank were asked three questions, 'In a typical DAY, how many hours do you spend watching television?', 'In a typical DAY, how many hours do you spend using the computer?(Do not include using a computer at work)' and 'In a typical DAY, how many hours do you spend driving?'. 17This GWAS identified 209 and 52 genome-wide-significant single nucleotide polymorphisms (SNPs) (p < 5 × 10 −8 ) for leisure television watching and computer use, respectively, using a linkage disequilibrium (LD) of R 2 < 0.005 within a five megabase window (Tables S1 and S2).The GWAS also identified five genetic variants associated with driving; however, we did not include these instruments in our MR analyses due to low statistical power (see Statistical power, below).The 261 SNPs included in both instruments were identified in 204 loci demonstrating a partial overlap between the two phenotypes with 22 common loci.The selected SNPs explained approximately 2% and 0.5% of the variability in television watching and computer use, respectively.

| Data on breast, colorectal and prostate cancer
Summary data for the associations of the above genetic variants with breast cancer were obtained from a GWAS of 247,173 women (133,384 breast cancer cases and 113,789 controls) of European ancestry from the Breast Cancer Association Consortium. 20We included six related outcomes in our analyses (overall, luminal A, luminal B, luminal B HER2 negative, HER2 enriched and triple negative breast cancer).
For colorectal cancer, summary data from 98,715 participants (52,775 colorectal cancer cases and 45,940 controls) were drawn from a meta-analysis within the ColoRectal Transdisciplinary Study, the Colon Cancer Family Registry, and the Genetics and Epidemiology of Colorectal Cancer consortia. 18We included five outcomes in our analyses (overall colorectal cancer, colorectal cancer for men, colorectal cancer for women, colon cancer and rectal cancer).The summary statistics did not include UK Biobank study to avoid potential overlap with the leisure sedentary behaviours GWAS.
For prostate cancer, summary data from a meta-analysis of 140,254 (79,148 prostate cancer cases and 61,106 controls) men of European ancestry in the Prostate Cancer Association Group to Investigate Cancer-Associated Alterations in the Genome and the Genetic Associations and Mechanisms in Oncology/Elucidating Loci Involved in Prostate Cancer Susceptibility consortia. 19The same consortia also conducted a GWAS of aggressive prostate cancer involving 15,167 cases and 58,308 controls, in which cancer cases were defined as aggressive based on the following characteristics: Gleason score ≥8, Prostate-Specific Antigen >100 ng/ mL, metastatic disease (M1) or death from prostate cancer. 19ll cancer estimates for the two exposures of interest are provided in Tables S3-S8.All participants provided written informed consent.Ethics were approved by respective institutional review boards. 17-20

| Statistical power
The statistical power was calculated a priori using an online tool at http:// cnsge nomics.com/ shiny/ mRnd/ . 21nder the scenario of a type 1 error of 5%, for leisure television use an expected OR per 1 standard deviation (SD) ≥ 1.09, ≥ 1.14 and ≥1.11 was needed to have adequate statistical power (>80%) for overall breast, colorectal and prostate cancer, respectively.Table S9 presents the power estimates for the three exposures by breast, colorectal and prostate cancer.

| Statistical analysis
A two-sample MR approach using summary data and the fixed-effect IVW method was implemented.All results correspond to an OR per 1-SD increment in genetically predicted hours of leisure sedentary behaviour (television watching: 1.5 h/day; computer use: 1.2 h/day).The heterogeneity of the causal estimates by cancer subtype (breast cancer), subsite (colorectal cancer) and sex (colorectal cancer only) was investigated by calculating the I 2 metric using a fixed-effect meta-analysis model. 22][25] For BMI, summary data from a GWAS meta-analysis of about 700,000 participants of European descent within the Genetic Investigation of ANthropometric Traits (GIANT) consortium and UK Biobank were obtained. 26For years of educational attainment, we obtained summary-level data from a published GWAS of 1.1 million participants of European descent within the Social Science Genetic Association Consortium and which measured the number of completed years of schooling among those individuals. 27Data on alcohol consumption (drinks per week) were drawn from a GWAS of 1.2 million individuals. 28The data for lifetime smoking were obtained from a recent GWAS and MR study on causal effects of lifetime smoking on risk for depression and schizophrenia. 29In the current analysis, we used data from 766,345 participants which were publicly available.All relevant summary statistics for the multivariable MR analyses is given in Tables S10-S25.MR studies have three main assumptions that must be satisfied in order for their causal estimates to be valid, which in the context of this study are as follows: (1) The genetic instrument is strongly associated with the levels of exposure (sedentary behaviour); (2) the genetic instrument is not associated with any potential confounder of the exposure (sedentary behaviour)-outcome (cancer) association; and (3) the genetic instrument does not affect the outcome (cancer) independently of the exposure (sedentary behaviour) (i.e.exclusion of horizontal pleiotropy).The strength of each genetic instrument can be evaluated through the F-statistic (provided by the initial GWAS). 17For multivariable MR, we also calculated the conditional F statistics which can be used to examine whether the genetic variants strongly predict each of the main (sedentary behaviours) and secondary exposures (e.g. years of education) conditional on the other exposure in the model; similar to univariable MR, F values over 10 suggest little evidence of weak instrument bias. 30

| Sensitivity analyses
Several sensitivity analyses were conducted to identify and correct for the presence of horizontal pleiotropy in the results from the main analysis.Cochran's Q was computed to quantify heterogeneity across the individual causal effects, with a p ≤ 0.05 indicating the presence of pleiotropy, and consequently, a random effects IVW MR analysis was used. 22,31MR-Egger regression provides valid MR estimates in the presence of horizontal pleiotropy when the pleiotropic effects of the genetic variants are independent from the genetic associations with the exposure. 32Large deviations from zero for the intercept test represent the presence of horizontal pleiotropic effects across the genetic variants.In such a case, the slope of the MR-Egger regression provides valid MR estimates when the pleiotropic effects of the genetic variants are independent from the genetic associations with the exposure. 32,33Moreover, causal estimates were also computed using the weighted-median method that can give valid MR estimates under the presence of horizontal pleiotropy when up to 50% of the included instruments are invalid. 34The MR pleiotropy residual sum and outlier test (MR-PRESSO) was also used to assess the presence of pleiotropy.The MR-PRESSO test relies on a regression framework to identify outlying genetic variants which may potentially be pleiotropic, we then reran the analysis after excluding these outlying variants. 35We also examined the selected genetic instruments and their proxies (r 2 > 0.8) and their associations with secondary phenotypes (p-value < 5 × 10 −8 ) in populations of European descent in Phenoscanner (http:// www.pheno scann er.medsc hl.cam.ac.uk/ ) to explore potential pleiotropy of the included SNPs.Finally, as a post hoc analysis based on the results from the multivariable MR and trying to understand the observed attenuation, we also conducted a bidirectional MR study to examine the associations between sedentary behaviours and the four secondary traits (BMI, years of education, alcohol consumption and lifetime smoking) (Tables S26-S33).
All the analyses were conducted using the MendelianRandomization and TwoSampleMR packages, while the LD clumping (LD < 0.001) in the multivariable MR analyses between SNPs of sedentary behaviour phenotypes with those for the secondary traits was done using the ieugwasr R package (https:// mrcieu.github.io/ ieugw asr/ ) and the R programming language (version 4.1.2). [36][37][38] Reporting guidelines for MR studies were followed. 39,40 3 |RESULTS

| Baseline characteristics
For the sedentary behaviour GWAS, the average age of the participants was 57.4 (SD: 8.0) years old, and 45.7% were men.Mean daily reported time of leisure television watching and leisure computer use was 2.8 (SD: 1.5) and 1.0 (SD: 1.2) h, respectively.The mean BMI was 27.4 kg/ m 2 , 55% were never smokers or quit >12 months ago, and 67% were physically active (i.e.≥150 min/week moderate or ≥75 min/week vigorous or 150 min/week mixed [moderate and vigorous] activity) behaviour. 17
The multivariable MR analysis adjusting for years of education led to the attenuation of all effect estimates between genetically predicted television watching and the risk of breast (OR per 1 SD: 1.08, 95% CI: 0.92-1.27)and colorectal cancer (OR per 1 SD: 1.08, 95% CI: 0.90-1.31)(Figure 1A, C, Table S22).Additional attenuations were observed for the models adjusting for lifetime smoking.For women, risk estimates for colorectal cancer were attenuated towards the null in all multivariable MR models adjusting for each of the four secondary traits (Figure 1C, Table S34).Finally, genetically predicted television watching was associated with HER2 negative, HER2 positive and triple negative breast cancer after adjusting for BMI in the multivariable MR models with effect sizes ranging from 1.32 to 1.46 per SD (Figure 1A).
Based on the Cochran's Q values, there was evidence of heterogeneity of SNP effects for most outcomes except for triple negative breast cancer (Tables 1-3).Scatter plots (with coloured lines representing the slopes of the different regression analyses) and funnel plots of the association between leisure television watching and the risk of breast, colorectal and prostate cancer risk are presented in Figures S1-S6.
In the multivariable MR analysis for triple negative breast cancer, after adjusting separately for years of education, alcohol and BMI the inverse effect estimates for genetically predicted computer use found in the univariable MR analysis were no longer statistically significant with the new attenuated effect sizes ranging from 0.73 to 1.06 per SD (Figure 1B,D, Table S34).Similarly, the inverse effect estimates for rectal cancer observed in the univariable analysis were attenuated after adjusting for years of education or alcohol consumption (Figure 1D, Table S34).
Based on Cochran's Q values, heterogeneity in SNP effects was found for overall breast cancer, luminal A breast cancer, luminal B breast cancer and colorectal cancer.Scatter plots (with coloured lines representing the slopes of the different regression analyses) and funnel plots of the association between leisure computer use and risks of breast, colorectal and prostate cancer are presented in Figures S7-S12.

| Evaluation of assumptions and sensitivity analyses
The strength of the genetic instruments according to the F-statistic was ≥10 for both exposures of interest and ranged between 23 and 164 (Tables S1-S3).In the multivariable MR framework, the conditional F statistics were mainly above 10 (indicating little evidence of weak instrument bias) for both our exposures of interest and the adjusting factors.For models including television watching and years of education, conditional F statistics for both variables were below 10.Also, adjusting for BMI or years of education resulted in low F statistics (<10) for computer use.Little evidence of directional pleiotropy was observed based on the MR-Egger's test (MR-Egger intercept p > 0.05) (Tables 1-3).The effect estimates from MR Egger regression models were generally in the same direction with those from the main analysis but with wider confidence intervals (Tables 1-3).Similarly, the weighted-median approach effect estimates were consistent in direction and magnitude to the IVW models (Tables 1-3).The MR-PRESSO analysis identified several (10 in total) outlying SNPs (Table S35); however, no major differences were observed when these outlying genetic variants were excluded from the analyses (Tables 1-3).After examining Phenoscanner, T A B L E 1 Mendelian randomisation estimates for sedentary behaviour and breast cancer risk.we found that several of the genetic variants were also associated with adiposity or education-related phenotypes, such as BMI and highest qualification (Table S36).

| MR estimates for the bidirectional MR
In post hoc analyses, inverse bidirectional associations were observed between the genetically predicted duration of leisure television watching and years of education.A one SD increase in genetically predicted duration of leisure television watching reduced years of education by 0.54 SD (95% CI: −0.58 to −0.49).Similarly, a one SD increase in genetically predicted years of education reduced duration of leisure television watching by 0.63 SD (95% CI: −0.66 to −0.59) (Figure 2, Tables S37 and S38).These observations taken together with the inverse effect estimate found for years of education with breast and colorectal cancer (Table S39) point to education having a complex dual confounding and mediating role in the association between television watching with breast and colorectal cancer risk.Contrary to this, positive bidirectional associations were observed for genetically predicted duration of leisure computer use (beta computer use→education : 0.59; 95% CI: 0.48-0.70 and beta education→computer use : 0.34; 95% CI: 0.30-0.37).Additionally, positive bidirectional associations were observed between the genetically predicted duration of leisure television watching with BMI and smoking status while inverse bidirectional associations were observed between the genetically predicted duration of leisure computer use and smoking status.Finally, alcohol consumption was inversely associated with computer use (Figure 2, Tables S37 and S38).

| DISCUSSION
The univariable MR analyses showed that a high level of genetically predicted television watching increased risks of breast and colorectal cancer but after multivariable MR adjustment for years of education, the positive effects were attenuated.Our post hoc analyses further suggested that education has a complex dual confounding and mediating role in the association between television watching with these cancers.The effect estimates for television watching were robust according to most of the univariable sensitivity analyses conducted to assess the influence of pleiotropy.We found little evidence that genetically predicted leisure computer use was associated with breast, colorectal and prostate cancer.
Inconsistent results have been reported in prospective cohort studies that have examined the association between sedentary behaviours and breast cancer risk.A recent meta-analysis reported a statistically significant 10% higher risk for the highest sedentary behaviour group when compared with the lowest group (relative risk: 1.10, 95% CI: 1.02-1.18). 8However, a recent study in UK Biobank found little evidence of any association between hours spent watching television and the risk of breast cancer (OR per 1 h increase: 1.01, 95% CI: 0.99-1.03). 9In our analysis, we initially observed positive associations between hours of television watching and the risk of breast cancer.However, these positive effect estimates were attenuated towards the null in our multivariable MR models adjusting for other risk factors, particularly years of education.
Numerous observational studies have investigated the associations between sedentary behaviours and colorectal cancer risk.Results from the most recent meta-analysis of case-control and cohort studies reported a non-significant 10% risk increase for colorectal cancer for the highest sedentary behaviour group when compared with the lowest group (RR = 1.10, 95% CI: 0.96-1.26). 8Television viewing time has been the most investigated sedentary behaviour trait, and positive associations have been found with colon cancer. 9,41A recent UK Biobank analysis reported that higher levels of television watching time were associated with greater colon cancer risk (HR per 1-hour increase, 1.04, 95% CI: 1.01-1.07;p-value = 0.016), but not rectal cancer. 9The same UK Biobank study found no association between leisure computer use and colorectal cancer risk. 9esults from our univariable MR analyses were generally  consistent with this prior observational evidence, with positive effect estimates found for television watching, and little evidence of an association between computer use and colorectal cancer risk, except of rectal cancer.However, these positive associations attenuated towards the null in multivariable MR models adjusted for years of education and smoking (colorectal; television watching) or alcohol (rectal; computer use).
We found little evidence of any associations between sedentary behaviours and prostate cancer risk, consistent with prior observational evidence. 9,41The null effects we found were similar for overall and aggressive prostate cancer risk.
Recently, two MR studies investigated the associations between sedentary behaviours and the risks of breast, colon and prostate cancer. 15,16The first included six SNPs associated with the probability of engaging in sedentary behaviours and found that longer genetically predicted sedentary time was associated with higher hormone-receptor-negative breast cancer risk (OR per-SD [~7% time spent sedentary] = 1.77; 95% CI: 1.07-2.92)with an attenuated effect observed for overall breast cancer (OR per-SD =1.20; 95% CI: 0.93-1.55). 16These results are in general agreement with our study in which we observed positive effects estimates for both HER2 enriched and triple negative breast cancers.However, the earlier MR analysis did not include multivariable analyses to adjust for other risk factors.The second MR study used the same instruments as our study and similarly identified the positive effects of television watching with overall breast cancer and similarly observed an attenuation of the estimates after adjusting for years of education. 15However, no positive effects were observed for television watching and colon cancer in this study, most likely due to the small number of colon cancer cases included (n = 2437).
Current literature suggests that the mechanisms connecting sedentary behaviours with cancer risk overlap at least partially with those underpinning the physical activity relationship and include interrelated pathways such as excess adiposity, metabolic dysfunction and alterations in sex hormone and inflammatory pathways. 8,16trong genetic correlations have been reported between television watching (inverse) and computer use (positive) with years of education (r TV g = −0.79 and r PC g = 0.53). 17The low conditional F statistics in our multivariable models including the sedentary behaviour traits with years of education provided a further indicator of strong correlations.A recent MR study reported an inverse association between years of education and breast (OR: 0.89, 95% CI: 0.83-0.96;p-value = 0.001) and a positive association for prostate cancer (OR: 1.10, 95% CI: 1.01-1.21;p-value = 0.035). 42n agreement with that, we observed inverse effect estimates for years of education in our multivariable models for breast and colorectal cancer.An additional MR study found that higher educational attainment levels were further inversely associated with smoking, BMI and sedentary F I G U R E 2 Bidirectional associations of leisure time television watching and computer use with the four secondary traits: ΒΜΙ, years of education, smoking and alcohol.The solid lines correspond to the effects of time television watching and computer use on the four secondary traits while the dashed lines correspond to the effects of the four secondary traits on time television watching and computer use.The black colour corresponds to statistically significant associations and the grey colour to non-significant.All the results, odds ratios and 95% confidence intervals correspond to a 1-SD change in the levels of the variables.BMI, body mass index.
behaviours, and positively with vigorous physical activity levels and alcohol consumption. 43Therefore, education may be a proxy for overall lifestyle, with higher educated individuals practising healthier lifestyle behaviours and actively participating in screening programmes that lower their risk of developing cancer. 42Additionally, traits like sedentary behaviours, education, smoking, alcohol consumption and obesity are correlated and it is therefore difficult to disentangle their complex interrelationships.As an example, in our post hoc analyses we found evidence of education having a dual confounding and mediating role in the association between television watching with breast and colorectal cancers.Previous studies and ours have shown that education plays an important role in cancer incidence of these three cancer types.However, the role of other lifestyle factors in these relationships is unclear, and further studies are needed to disentangle these complex interrelationships.
The main strength of the current study is the use of large-scale summary genetic data from consortia and the UK Biobank that allowed us to investigate the role of leisure sedentary behaviours on risk of developing breast, colorectal and prostate cancer.A limitation of our study is that leisure sedentary behaviours were derived from self-reported questionnaires that are prone to measurement error. 44,45An alternative approach is to use genetic instruments derived from objectively measured levels of physical activity using accelerometer data from the UK Biobank. 46,47However, a current limitation is that the number of genetic instruments is comparatively small as the GWAS on accelerometer data was analysed in a subset of 91,000 participants.Analysing two highly correlated phenotypes together, like sedentary behaviours and years of education may have introduced collinearity which leads to greater imprecision and possible bias.Furthermore, caution is needed regarding the results from the analyses for leisure computer use as the genetic instruments explained a small proportion of the phenotypic variance resulting in a lower powered analysis.Also, our analyses focused solely on leisure sedentary exposures so non-leisure sedentary behaviours were unaccounted for.The genetic correlation between television watching and objectively measured sedentary behaviour in UK Biobank was weak (r TV g = 0.14) while the correlation for computer use was higher (r PC g = 0.46). 17This can be at least partially explained from the fact that accelerometers measure total but not domain-specific sedentary time (e.g.television watching) that has been studied in previous observational studies. 3,48Therefore, our results cannot be generalised to overall sedentary behaviour.The genetic instruments were derived from UK Biobank which is not without limitations.For example, the average age of the participants in UK Biobank was 57 years, an age group that spends most time watching television. 17Consequently, the results cannot be generalised to younger ages as the habits of younger people are not included in the analysis and similarly the phenotype of leisure time computer use perhaps is not optimal to capture sedentary behaviours of this population.In addition, large biobanks like UK Biobank often suffer from participation bias since the participants are not representative of their target population and it has been shown to distort genome-wide findings and downstream analyses particularly for socio-behavioural traits. 49Furthermore, we cannot exclude the possibility of confounding due to population stratification in our dataset.The genetic instruments were derived from a sex combined population while some of the outcomes were sex specific which could introduce some bias in our results if the effects of the genetic instruments differ between two sexes.Additionally, we cannot exclude potential dynastic and assortative mating effects as it has been reported that the estimates of education could be attributed at least partially to parental effects to the child's characteristics. 43Moreover, parents do not mate randomly but assort on characteristics such as educational level. 43These cross-generational effects could also have biased our results. 50,51Finally, the results cannot be generalised to diverse populations due to the lack of ancestral diversity in UK Biobank.

| Conclusions
In conclusion, after adjusting for lifestyle factors, especially years of education, leisure time television watching no longer increased the risks of breast and colorectal cancer and demonstrated how highly intercorrelated these exposures are.These multivariable results should be interpreted cautiously as we detected evidence of education having a dual confounding and mediating role in the associations between television watching with risks of breast and colorectal cancer.Future analyses utilising objective measures of exposure (e.g.accelerometers) and novel analytic frameworks (e.g.target trial emulation) are required to provide new insights into the possible role of sedentary behaviour in cancer development.

a
The estimates correspond to a standard deviation increase in duration of sedentary activity.b p-value or pleiotropy based on MR-Egger intercept.c p-value for heterogeneity based on Q statistic.

a
The estimates correspond to a standard deviation increase in duration of sedentary activity.b p-value or pleiotropy based on MR-Egger intercept.c p-value for heterogeneity based on Q statistic.F I G U R E 1 Associations of leisure time television watching and computer use with breast and colorectal cancer after adjusting for the four secondary traits.The black dot corresponds to the 1-SD odds ratio and the corresponding error bar to the 95% confidence interval.Each error bar refers to the associations of leisure time television watching or computer use with breast or colorectal cancer after adjusting for the specific trait on the left side of the graph.(A) leisure time television watching-breast cancer associations; (B) leisure time computer usebreast cancer associations; (C) leisure time television watching-colorectal cancer associations; (D) leisure time computer use-colorectal cancer associations.BMI, body mass index; IVW, inverse-variance weighting.
Mendelian randomisation estimates for sedentary behaviour and colorectal cancer risk.
T A B L E 2Abbreviations: CI, confidence interval; MR, Mendelian randomisation; OR, odds ratio; MR-PRESSO, MR pleiotropy residual sum and outlier test.aTheestimates correspond to a standard deviation increase in duration of sedentary activity.b p-value or pleiotropy based on MR-Egger intercept.c p-value for heterogeneity based on Q statistic.
Mendelian randomisation estimates for sedentary behaviour and prostate cancer risk.: CI, confidence intervals; MR, Mendelian randomisation; OR, odds ratio; MR-PRESSO, MR pleiotropy residual sum and outlier test.
T A B L E 3Abbreviations Figueireido/Colditz, 2P30CA015704-40 (Gilliland), R01 CA207371 (Ulrich/Li)), the Matthias Lackas-Foundation, the German Consortium for Translational Cancer Research and the EU TRANSCAN initiative.The Colon Cancer Family Registry (CCFR, www.colon cfr.org) is supported in part by funding from the National Cancer Institute (NCI), National Institutes of Health (NIH) (award U01 CA167551).Support for case ascertainment was provided in part from the Surveillance, Epidemiology and End Results (SEER) Programme and the following U.S. state cancer registries: AZ, CO, MN, NC and NH; and by the Victoria Cancer Registry (Australia) and Ontario Cancer Registry (Canada).The CCFR Set-1 (Illumina 1M/1M-Duo) was supported by NIH awards U01 CA122839 and R01 CA143247 (to GC).The CCFR Set-3 (Affymetrix Axiom CORECT Set array) was supported by NIH award U19 CA148107 and R01 CA81488 (to SBG).The CCFR Set-4 (Illumina OncoArray 600K SNP array) was supported by NIH award U19 CA148107 (to SBG) and by the Center for Inherited Disease Research (CIDR), which is funded by the NIH to the Johns Hopkins University, contract number HHSN268201200008I.Additional funding for the The content of this manuscript does not necessarily reflect the views or policies of the NCI, NIH or any of the collaborating centres in the Colon Cancer Family Registry (CCFR), nor does mention of trade names, commercial products or organisations imply endorsement by the US Government, any cancer registry, or the CCFR.COLON: The COLON study is sponsored by Wereld Kanker Onderzoek Fonds, including funds from grant 2014/1179 as part of the World Cancer Research Fund International Regular Grant Programme, by Alpe d'Huzes and the Dutch Cancer Society (UM 2012-5653, UW 2013-5927, UW2015-7946), and by TRANSCAN (JTC2012-MetaboCCC, JTC2013-FOCUS).The Nqplus study is sponsored by a ZonMW investment grant (98-10030); by PREVIEW, the project PREVention of diabetes through lifestyle intervention and population studies in Europe and around the World (PREVIEW) project which received funding from the European Union Seventh Framework Programme (FP7/2007-2013) under grant no.312057; by funds from TI Food and Nutrition (cardiovascular health theme), a public-private partnership on precompetitive research in food and nutrition; and by FOODBALL, the Food Biomarker Alliance, a project from JPI Healthy Diet for a Healthy Life.Colorectal Cancer Transdisciplinary (CORECT) Study: The CORECT Study was supported by the National Cancer Institute, National Institutes of Health (NCI/NIH), U.S. Department of Health and Human Services (grant numbers U19 CA148107, R01 CA81488, P30 CA014089, R01 CA197350; P01 CA196569; R01 CA201407) and