Data sources and single-nucleotide polymorphism selection

Genetic association estimates with the outcomes were obtained from the UK Biobank.13 During 2006 and 2010, about 500 000 community-dwelling adults, aged 40–69 years, across the UK were enrolled into the cohort (https://www.ukbiobank.ac.uk/).13 To reduce confounding by ancestry, we restricted the analytic cohort to individuals of White-British descent. After exclusion of related individuals (third-degree relatives or closer), low call rate (3 or more standard deviations from the mean), and excess heterozygosity, 367 703 individuals remained for analysis. We used follow-up data until 17 February 2016, and defined the outcomes based on electronic health, hospital procedure codes, and self-reported information confirmed by interview with a nurse (Supplementary material online, Table S1). To be able to determine differential cardiovascular risk associated with adiposity, we analysed a broad range of CVDs, including cerebrovascular diseases (ischaemic stroke, transient ischaemic attack, intracerebral haemorrhage, and subarachnoid haemorrhage), aneurysms (abdominal and thoracic aortic aneurysm), thrombo-embolic diseases (deep vein thrombosis and pulmonary embolism), and other CVDs (coronary artery disease, aortic valve stenosis, atrial fibrillation, heart failure, and peripheral vascular disease) as well as arterial hypertension. Logistic regression, adjusted for 10 genetic principal components, was applied to estimate the genetic associations with the outcomes. The North West Multicenter Research Ethics Committee approved the UK Biobank study. All participants provided written informed consent to participate in the study.

As instrumental variables we used SNPs associated with BMI at the genome-wide significance threshold (P < 5 × 10⁻⁸) in a meta-analysis of genome-wide association studies, including 339 224 individuals.14 One of the BMI-associated SNPs was unavailable in UK Biobank, leaving 96 SNPs as instrumental variables (Supplementary material online, Table S2). For fat mass and fat-free mass (assessed using bioelectrical impedance technique), we considered the 98 SNPs associated with body composition at genome-wide significance among 362 499 UK Biobank participants.15 Of those SNPs, we included 82 that had an imputation quality score greater than 0.8 and were in Hardy–Weinberg equilibrium (Supplementary material online, Table S3). We calculated fat mass index analogously to BMI as fat mass divided by height squared, and fat-free mass similarly. Genetic associations with BMI were taken from the published meta-analysis14 and do not include UK Biobank participants, whereas genetic associations with fat mass and fat-free mass indices were estimated in UK Biobank by linear regression with adjustment for 10 genetic principal components.