612 cases and 3366 controls previously genotyped in stage 1 by Allen et al.26 were used. Our secondary analysis is covered by the existing ethical approvals and informed consent reported for that study. Association with disease was adjusted for 10 principal components of ancestry but not for age or sex, allowing inclusion of 10 cases without data on age. Imputation was performed to the Haplotype Reference Consortium panel at the Michigan Imputation Server41. We retained variants with imputation R2 of 0.5, minor allele frequency >0.5%, Hardy–Weinberg equilibrium P >10–6, and at least five events for subjects with allele dosage >0.5. After harmonising the case/control and survival data, we analysed 7,983,997 variants.
