Sir,

In response to our publication in the BJC (Journy et al, 2015), Dr. Colin R. Muirhead gave insightful comments for the interpretation of the potential impact of predisposing factors (PF) for cancer in estimating radiation-related cancer risks from CT scans (Muirhead, 2015). He pointed out that the possibility of an effect modification by the presence of PF, which was reported in the published study, should be considered for providing relevant CT-related risk estimates.

The paper’s results indicated that PFs (i.e., some genetic disorders and immune deficiencies) might be a confounding factor (Journy et al, 2015). In the cohort, PFs were, as expected, associated with high relative cancer risks, but also with specific patterns of CT exposures. However, as underlined by Dr. Muihead, the excess relative risks (ERRs) related to CT exposure differed in individuals with or without PF. In particular, CT exposure was associated with reduced cancer risks in children with PFs, and the risk estimates in patients without PF were equal to or greater than unadjusted ERRs in the overall cohort, for each of the three cancer sites of interest.

Biological processes, leading to reduced radiation sensitivity in presence of genetic disorders and/or immune deficiencies, are not likely to have been involved in such an effect modification observed with various PFs. From further analyses conducted in the cohort (Journy, 2014), the reduced radiation-related risks in children with PFs might rather be explained by competing events initiated or promoted by PFs, that is, cancer or death. Finally, we agree that the decrease in ERRs with adjustment for PFs reflected, at least partly, an effect modification by PFs.

From our paper’s results, Dr. Muihead stated that risk estimates adjusted for the presence of PFs – expressing averaged risks in a population of patients with or without PF – are not relevant for public health purposes, as they are driven by the ERRs in predisposed individuals. Indeed, adjusted ERRs in all exposed individuals might be appropriate to correct the estimations for a potential confounding bias, provided that CT-related risks are homogeneous in the studied population. In the cohort, however, adjusted risk coefficients would represent underestimated risk estimates for children without PFs who accounted for the great majority of patients exposed to CT scans (97% of the cohort). Joining Dr. Muihead’s conclusion, these results thus suggest that the most relevant risk coefficients for radiation protection concerns are estimates excluding patients with PFs.

In epidemiological studies on cancer risk after CT scans, in which information on PF is most often inaccessible, a central question is to determine to which extent risk estimates without considering PFs at all might be biased or not. In our study, the results suggested an effect modification without totally excluding the possibility of a confounding bias. It should be noted that issues on reverse causation (Walsh et al, 2014) might also differ according to PFs, with enhanced medical surveillance for cancer and early cancer detection in predisposed patients. Our results should nevertheless be interpreted with much caution owing to the small numbers of cases, especially in the subgroup analyses. Indeed, the estimated ERRs were imprecise, and not interpretable for leukemia in children without PF. The duration of follow-up was another major limitation given the latency time between radiation exposure and stochastic health effects. Longer follow-up of this cohort, as well as of other studies that benefit from clinical information (Meulepas et al, 2014; Krille et al, 2015), will allow a better assessment of the impact of PFs on CT-related risk estimates.