Comment on ‘Are the studies on cancer risk from CT scans biased by indication? Elements of answer from a large-scale cohort study in France'—Evidence of confounding by predisposing factors unclear

Sir,

The paper by Journy et al (2015) presents the first results of a very carefully conducted cohort study of paediatric computerised tomography (CT) patients from France, part of the European collaborative study ‘EPI-CT' (Bosch de Basea et al, submitted). Because of criticisms raised about the results of previous studies of CT patients (Pearce et al, 2012; Mathews et al, 2013; Huang et al, 2014), the authors made particular efforts to collect information on potential factors which could invalidate estimates of radiation risks in these studies. The current paper emphasises, in particular, the potential impact of predisposing factors (PFs) for leukaemia, central nervous system (CNS) tumours and lymphoma, the outcomes under study in this paper. For this study, a list of PFs was developed by paediatric oncologists based on the literature, and hospitalised discharge records of cohort members were searched to identify cohort members with PFs. These included familial adenomatous polyposis, multiple endocrine neoplasia, retinocytoma, Fanconi anaemia, ataxia telangiectasia, neurofibromatosis, other phacomatoses, xeroderma pigmentosum, Down syndrome, Noonan syndrome, Klinefelter syndrome and Bloom syndrome as well as immune deficiencies (HIV/AIDS, severe combined immune deficiency, Wiskott–Aldrich syndrome, common variable immune deficiency and organ transplantation). The frequency of PFs for CNS tumours in the cohort was 0.54% it was 1.7% and 1.6%, respectively, for PFs of leukaemia and lymphoma. The most frequent PFs were organ transplantation (observed in 749 of the 67 274 members of the cohort −1.11%), HIV/AIDS (0.36%), Down syndrome (0.3%), neurofibromatosis types 1 and 2 (0.16%) and other phacomatoses (0.29%). These percentages, though low, are greater than in the general population and their presence appears to be related to a slightly increased frequency and slightly decreased age at CT examinations, thus potentially confounding the association between radiation from CTs and risks of the aforementioned neoplasms.

During the study period, 27 CNS tumours, 25 leukaemia and 21 lymphoma were observed in the cohort; of these 7, 5 and 7, respectively, had a PF for CNS, leukaemia or lymphoma. In Table 5 of their paper, the authors show that adjustment for PFs reduced the excess relative risk estimates related to cumulative doses from CT scans (Table 1). This led them to conclude ‘This study suggests that the indication for examinations, whether suspected cancer or PF management, should be considered to avoid overestimation of the cancer risks associated with CT scans'. Results shown in their Supplementary Table 6, however, focusing on the ERR/mGy among subjects with and without PF, challenge, in our opinion, this interpretation.

Table 1. Number of cases (N) and ERR per mGy for tumours of the CNS, leukaemia and lymphoma, crude or adjusted for the presence of PFs and by patient's characteristics regarding presence of factors predisposing specifically to cancer at the site specified (PF).

	All cases (2-year exclusion period)			Subgroups
		Unadjusted	Adjusted for PF	Without PF		With PF
	N	ERR/mGy (95% CI)	ERR/mGy (95% CI)	N	ERR/mGya	N	ERR/mGya
CNS tumours	22	0.022 (−0.016; 0.061)	0.012 (−0.013; 0.037)	15	0.028	7	−0.005
Leukaemia	17	0.057 (−0.079; 0.193)	0.047 (−0.065; 0.159)	12	0.187	5	−0.012
Lymphoma	19	0.018 (−0.068; 0.104)	0.008 (−0.057; 0.073)	12	0.025	7	−0.005

Abbreviations: CNS=central nervous system; ERR=excess relative risks; PF=predisposing factor.

^aConfidence intervals not provided.

Indeed, risk estimates among subjects with no PF are similar to—although slightly higher than—the unadjusted risk estimates for brain tumours and lymphoma (see Table 1). This observation suggests that PFs are not, in fact, confounders of the association between cumulative organ radiation dose from CT and risk of these tumours, but rather possible effect modifiers. Though the authors conducted tests of homogeneity of risks between subjects with and without PFs, they were based on small numbers of subjects and hence the power to formally identify effect modification was very limited. For leukaemia, the ERR/mGy among subjects without PF are substantially higher (but quite uncertain given the small number of cases) than the unadjusted estimates, again suggesting effect modification.

Numbers of cases with PFs are, unfortunately, too small to allow the study of the radiation effect associated with different types of PFs. For brain tumours, the majority of cases with PFs had neurofibromatosis; for lymphomas, organ transplantation, whereas for leukaemia there was a mixture of Down syndrome, primary immunodeficiency and organ transplantation (Journy, 2014). As the mechanism and the magnitude of the increased cancer risk differ for these different types of PFs, it is somewhat surprising that they would all have a similar effect on the risk estimates when adjustment is made for PFs in the analysis. The observation that, among subjects with PFs, the ERRs/mGy for all three outcomes were very close to 0, suggests instead that any effect of low doses of radiation would be too small to detect given the already very high cancer risk among these subjects in the absence of radiation. This would strengthen the argument that PFs are effect modifiers and not confounders of the association between CT radiation dose and risk of cancer.

This finding, if it can be replicated in other larger cohorts, is very important as information on PFs is not available in many cohorts and lack of information about predisposing factors is one of the main criticisms of published studies on the carcinogenic effect of radiation from CT scans in paediatric patients.

As the goal of EPI-CT and other similar studies is to estimate directly the risk of cancer associated with radiation exposure from CT scan examinations in the general paediatric population (where the proportion of PF is relatively low), the findings of Journy and collaborators suggest that the unadjusted ERR/mGy may be a reasonable (and unconfounded) estimate of the true risk, particularly since the frequency of PFs in this cohort is high, due to the inclusion in the study of a number of specialised referral hospitals (Journy, 2014).