Genetic susceptibility, screen-based sedentary activities and incidence of coronary heart disease

We used data from a large-scale prospective cohort study, UK Biobank, which includes over half a million UK adults 40–69 years of age at recruitment [19]. The key eligibility criteria in the UK Biobank study included living in a place <25 miles away from one of 22 assessment centres in the UK and is registered within the National Health Service database. Between 2006 and 2010, the baseline measurement was carried out collecting information on a wide variety of variables including genotype data, demographic indicators, body composition and lifestyle behavioural outcomes. The present analysis was based on 373,026 participants who were considered white British (based on self-reported ethnicity and principal component analysis of genotype data), had no prevalence of CHD/stroke (based on self-report and hospital admission records) and had no missing data for any covariates (Additional file 1: Fig. S1). The UK Biobank study protocol was approved by the Northwest Multi-Centre Research Ethics Committee (11/NW/0382). Before participation, participants provided signed informed written consent.

Measured data of participants in UK Biobank were linked with their national death registry and hospital admission records. CHD incidence in this study was ascertained according to a series of algorithms based on both death registry and hospital admission data [24]. Codes of International Classification of Diseases (ICD) were used to adjudicate CHD cases (ICD-9: 410-411,412.X, ICD-10: I21-I24, I25.2) accrued until October 31, 2021, for individuals in England and Wales and November 12, 2021, for individuals in Scotland. Incident CHD was defined as the first observation of CHD events that occurred over a 12.6-year median follow-up (interquartile range: 12.0–13.4years), resulting in a total of 9185 incident CHD cases.

The following variables that may confound the associations between genetic risk, TV viewing/computer use and CHD were included as confounders [25]: age (underlying timescale), sex, body mass index (BMI) (i.e. higher BMI associated with higher sedentary time including TV viewing [26, 27] and higher CHD risk [28], but not acting as a mediator [29]), smoking status (never, previous, current), employment (unemployed, employed), Townsend Deprivation Index (a numerical deprivation score generated based on employment, car ownership, home ownership and household overcrowding according to postcode of participants’ home address), alcohol consumption (never, previous, currently <3times/week, currently ≥3times/week), salt-adding behaviour (never/rarely, sometimes, usually, always), oily fish consumption (never, <once/week, once/week, >once/week), coffee intake (cups/day), fruit and vegetable intake (a composite score generated based on intake of fresh/dried fruit and intake of raw/cooked vegetable ranging from 0 to 4), processed/red meat intake (days/week), hypertension medication use, cholesterol-lowering medication use, glucose-lowering medication use, sleep (≤5, 6, 7, 8, ≥9hours/day) and moderate-to-vigorous physical activity (minutes/day; calculated by summing up moderate activity [frequency×duration] and vigorous activity time [frequency×duration] (multiplied by 2) [30] performed in a typical week; questions derived from the modified International Physical Activity Questionnaire-Short Form). The genotyping array type (UK Biobank Axiom Array, UK BiLEVE Axiom Array) and the first ten principal components of ancestry (to control for population stratification [31]) were adjusted for in models for the polygenic risk score.

Cox regression models (with age as the underlying timescale) using either TV viewing or computer use as the main exposure were established by adjusting for no confounders (Model 1), and confounders (Model 2) with an additional adjustment for the polygenic risk score (as a continuous variable) and mutual adjustment of TV viewing and computer use (Model 3), after excluding the first 2 years of follow-up. Models using the polygenic risk score as the main exposure were also fit with adjustment for sex, the genotyping array type and the first ten principal components of ancestry. Multiplicative interactions between three categories of TV viewing or computer use and polygenic risk scores were tested in models adjusted for confounders. Models were also fit to estimate the joint associations of TV viewing or computer use and genetic risk with incident CHD, with low genetic risk combined with ≤1h of TV viewing or computer use as the reference group; these models did not include the polygenic risk score variable as a potential confounder. The cumulative hazards of CHD across three categories of TV viewing, computer use and polygenic risk score were plotted. Population attributable fractions (PAFs) were calculated to estimate proportional risk reductions in CHD that would occur if ≥2h/day of TV viewing and computer use were reduced to ≤1h/day of TV viewing and computer use, respectively, assuming causality [10, 32, 33]. The calculation of confidence intervals for the PAFs assumed asymptotic normality of the estimates (‘punafcc’ command in Stata/MP Version 16.0) [34, 35]. All models were fit using age as the underlying timescale and adjusting for the 2nd-degree genetic relatedness (kinship coefficients between 0.0442 and 0.0884) [36] using cluster-robust standard errors [20]. Log-log plots supported the proportional hazards assumption for each exposure. Nine sensitivity analyses were performed: (1) excluding an additional 2 years of follow-up (4 years in total) to address reverse causality, (2) excluding individuals with poor self-reported health status (i.e. based on the 4-level self-reported health ratings; poor [excluded], fair, good, excellent) to address reverse causality, (3) excluding individuals with the 2nd-degree genetic relatedness, (4) using a weighted polygenic risk score calculated based only on 46 lead SNPs (from 46 loci) which were genome-wide significant at a p value of 5×10^-8 and in low linkage disequilibrium defined according to r²<0.001 (Additional file 1: Table S1 and Fig. S4), 5) using values imputed for the covariates missing (using multiple imputation with changed equations), assuming data missing at random, (6) including prevalence of type 2 diabetes and renal dysfunction as confounders, (7) excluding body mass index as a confounder, (8) with CHD follow-up censored on January 1, 2020, to account for potential CHD cases not captured due to participants’ fear of visiting clinics during the COVID-19 pandemic, and (9) with adjustment for education, household income and occupation as individual-level indicators of socio-economic status as opposed to area-level socio-economic status, Townsend Deprivation Index. All data files are available from the UK Biobank database (https://​www.​ukbiobank.​ac.​uk/​). Statistical analyses were performed using Stata/MP Version 16.0 (StataCorp LP, College Station, TX).

This study has several limitations. First, due to the observational nature of this study, conclusions about causality cannot be drawn. As such, the PAF analyses were grounded upon an assumption that the associations of TV viewing and computer use with CHD risk are causal [10], which has yet to be fully determined in the current literature. Moreover, there may be a measurement error due to recall bias in assessing TV viewing and computer use through questionnaires. Larger measurement errors in TV viewing and computer use may have led to attenuated associations with CHD [50]. There is also potential for residual confounding due to unmeasured confounders (e.g. unhealthy eating behaviours while watching TV) or measurement error in the self-reported confounders (e.g. smoking, employment, socioeconomic status, alcohol consumption, food intake variables, medication use, sleep and moderate-to-vigorous physical activity). Furthermore, we included only two prevalent forms of screen-based sedentary activities, due to the limited information on other types of sedentary behaviour available in the UK Biobank database. Hence, our findings may not be applicable to overall sedentary/sitting time as well as other types of sedentary behaviour [10]. Given that this study is of European descendants, findings may not be generalisable to individuals of other ethnicities. Moreover, UK Biobank participants tend to have more favourable health profiles than the general UK population [51], so generalizing our findings to average UK adults or individuals with sub-clinical symptoms should be made with caution. While we excluded the first 2 years and 4 years of follow-up in the main and sensitivity analyses, respectively, there may still be potential for reverse causality in the associations identified herein.