We noted little evidence of non-linearity of the dose-response, using either linear-quadratic or linear-exponential forms of departure from linearity (leukaemia exponential p=0·2672 and quadratic p=0·4683, brain tumour exponential p=0·9203 and quadratic p=0·8993). In sensitivity analyses in which all scans 10 years before brain tumour diagnosis were excluded, the magnitude of the dose-responses was increased rather than decreased as might be expected if the association was driven by bias from CT scans related to the diagnosis (appendix). When follow-up for leukaemia was restricted to 2004, the dose-response also increased, which was as expected given the short latency period for leukaemia and early peak in excess risk reported in previous studies.10, 16 To assess whether the missing exposure data after age 22 years resulted in underestimation of doses and hence overestimation of the relative risks, we restricted follow-up to individuals younger than 28 years for brain tumours and individuals younger than 25 years for leukaemia, but this did not change the dose-response estimates.