As recommended by UK Biobank, we analysed UK Biobank samples by adjusting for genotyping array because a subset of (extreme smokers in) UK Biobank were genotyped on a different array (UK BiLEVE). However, this adjustment could potentially introduce collider bias in analyses of smoking traits. Given that the UK BiLEVE study is relatively small compared to the full study, and the genetic effect sizes for smoking-associated variants are small, we expect the influence of collider bias to be small [61]. Nevertheless, we performed sensitivity analyses to assess the impact of collider bias. Firstly, we performed a meta-analysis excluding the UK BiLEVE samples, and secondly, we re-analysed UK Biobank without adjusting for genotype array. As expected, the estimated genetic effects from these additional analyses were very similar to our reported results suggesting collider bias is not a concern (Suppl. Table 8).
