Case-control comparisons of low FEV1 versus high FEV1, low FEV1 versus average FEV1, and high FEV1 versus average FEV1 were done within each of the heavy and never smokers subsets separately (appendix p 17). To identify whether any individual variants had a significantly different effect on the risk of airflow obstruction in heavy smokers compared with never smokers, we tested for interaction with smoking (appendix p 17). We calculated the proportion of the variance in FEV1 explained by genetic variants (appendix p 17). We compared heavy versus never smokers to identify loci associated with smoking behaviour. Association testing was done using a Score test (and Firth test for variants with minor allele count <400)32 with imputed marker doses, adjusting for pack-years in smokers and ten principal components. Full genome-wide association results are available via UK Biobank (appendix p 17). For genome-wide association analyses, we set genome-wide significance as p<5 × 10−8 and suggestive significance as 5 × 10−8<p<5 × 10−7. For other analyses, we used a Bonferroni correction for multiple testing. The appendix (p 17) describes quality control after association
