Cohort Profile: the EPI-CT study: a European pooled epidemiological study to quantify the risk of radiation-induced cancer from paediatric CT

Medical diagnostic examinations, although delivering low doses of ionizing radiation, are the main man-made source of ionizing radiation exposure for the general population. The number of procedures performed has grown dramatically in high-income countries in recent decades.1 Among these examinations, the use of computed tomography (CT), a highly informative medical imaging technique, has dramatically increased, partly as a result of the ease and speed of image acquisition improvements over the years. These trends are also observed in paediatric CT, which currently represents approximately 11% of all CT examinations.1 CT results in much higher doses of ionizing radiation than conventional radiography. CT represents 5% to 10% of all imaging procedures, but 40% to 70% of the collective medical dose.1,2 There are concerns about potential health impacts of the radiation exposure from these procedures, particularly cancer in exposed paediatric patients. According to the epidemiological results from A-bomb survivors and from patients undergoing radiotherapy and/or radiological examinations, exposure to ionizing radiation at a young age is associated with a higher relative risk of several forms of cancer than exposure later in life.3 Moreover, children have a longer life expectancy than adults, and hence a longer time in which to express radiation-induced cancers. Increased cancer risk has been reported after medical diagnostic procedures associated with much higher doses than those reported nowadays.4,5 However, recent epidemiological studies, focusing on CT exposure during childhood or early adulthood, have also reported increased risks of central nervous system (CNS) tumours, leukaemia and other cancer types in relation to CT doses.6–12 The causal interpretation of some of these results has been questioned because of the lack of individual dose reconstruction, small sample size and potential methodological biases linked to confounding by indication and reverse causation.13,14 Because the risk associated with low doses is estimated to be small,1 only large studies can achieve adequate statistical power to quantify this risk accurately. The international ‘EPIdemiological study to quantify risks for paediatric Computerized Tomography and to optimize doses’ (EPI-CT) was set up in 2011 to provide an estimation of the radiation-related risks of cancer after CT exposure in childhood and adolescence. The study is coordinated by the International Agency for Research on Cancer (IARC) and partially funded by the European Community. This European collaborative study pools nine European national cohorts. It takes advantage of pre-existing cohorts from three countries (France, Germany and the UK, all of which were extended as part of this study) and of new studies in six countries, based on a common core protocol15 including a specific effort to provide individual organ dose estimation for the subjects in the cohort.16 The EPI-CT study comprises four main parts: an epidemiological cohort study assessing cancer risks following radiation exposure from CT; a dose-reconstruction model to estimate organ doses with associated uncertainties for each individual in the cohort; a pilot study regarding biological mechanisms involved17; and recommendations for optimization of paediatric CT protocols. EPI-CT is a retrospective European multinational cohort of children and young adults subjected to CT at least once before the age of 22 years, and who have not been diagnosed with cancer either before or at the time of the first recorded CT nor within 1 year after it. The study aimed to establish or expand existing cohorts in the following countries: Belgium, Denmark, France, Germany, The Netherlands, Norway, Spain, Sweden and the UK (Table 1). Patients were identified from the electronic records of the participating hospitals, which were mainly large paediatric hospitals or hospitals with a large paediatric patient population. Ethical agreements were obtained in each country before data collection. The study protocol15 and detailed procedures adopted by each country are described elsewhere.6,9,10,12,18 Description of the selection criteria and available data in the participating countries Initial number of patients included (before application of exclusion criteria). Final number of eligible patients. Errors in date of birth, unknown date of cancer diagnosis. In progress. Exclusion of patients with diagnosis of cancer within the first 6 months of follow-up at national level before sending the data for the pooled analysis. Description of the selection criteria and available data in the participating countries Initial number of patients included (before application of exclusion criteria). Final number of eligible patients. Errors in date of birth, unknown date of cancer diagnosis. In progress. Exclusion of patients with diagnosis of cancer within the first 6 months of follow-up at national level before sending the data for the pooled analysis. Of the 1 170 186 patients for whom information was collected, 948 174 (81%) had a follow-up of at least 1 year after the first recorded CT and had no previous cancer recorded in the cancer registry (Table 1). Table 2 summarizes the number and characteristics of patients included in EPI-CT by country. There were slightly more males (56%) than females in the cohort. The median age at the first recorded CT varied between countries, reflecting differences in the age range inclusion criteria. The main age range was 0–21 years, but the age range was restricted to 0–9 and 0–14 years, respectively, in France and Germany because only paediatric and adolescent cancer registries are available at the national level in these countries. Accordingly, the median age at first exposure was lower, 2.9 years and 6.8, respectively, in France and Germany, compared with 10.7 years for the whole cohort. Distribution of study subjects by sex, country and age at the first CT examination Distribution of study subjects by sex, country and age at the first CT examination Follow-up started at the date of the first recorded CT and ended at the earliest of: date of death; date of cancer diagnosis; or end of follow-up at the regional/national level. Cancer diagnoses and deaths were obtained through linkage with national or regional cancer registries, mortality registries and other available national/regional registries, depending on the country (Table 1). The follow-up period for cancer incidence ends usually 1 to 3 years before mortality follow-up, due to the delay in reporting cancer cases in cancer registries. In two countries, vital status was partially (France) or completely (Germany) unavailable. The earliest year of first recorded CT varied between 1977 in Sweden and 2001 in Belgium, and depended on the availability of data, i.e. complete cancer registration to exclude ineligible patients and electronic radiological information. Cancer incidence follow-up ended in 2014 in The Netherlands and Sweden, and between 2010 and 2013 in all other countries (Table 3). Median duration of incidence follow-up was 7.8 years for the whole cohort, ranging from 4.1 years in Belgium to 11.3 years in the UK (Table 3). The total incidence follow-up accounted for more than 8.7 million person-years (PY). The largest cohorts in terms of PYs were the UK (3.7 million PY), the Dutch (1.5 million PY) and the Swedish cohorts (1.4 million PY), due to the large number of children included but also due to the long follow-up. Only 1.3% of the participants had died by the end of follow-up. The median age at the end of follow-up ranged from 9.4 years in France to 24 years in the UK. Follow-up of the cohort Shorter follow-up period for cancer incidence compared with mortality follow-up period is linked to the delay of cancer registries in reporting cancer cases (availability of cancer cases 1 to 3 years after the cancer diagnosis date). Follow-up of the cohort Shorter follow-up period for cancer incidence compared with mortality follow-up period is linked to the delay of cancer registries in reporting cancer cases (availability of cancer cases 1 to 3 years after the cancer diagnosis date). For each individual CT, all computerized data were retrieved from the Radiological Information System (RIS) of the radiology department in participating hospitals. This includes patient-identifying information, patient sex, date of birth and basic variables about the examination (body part scanned, examination date and, in certain instances, indication for CT and referring hospital service). For more recent time periods, estimation of doses took advantage of data from the Picture Archiving and Communication System (PACS) (a system for storage, retrieval and distribution of images). The time periods when PACS data were available, together with the percentage of CT dosimetric data extracted from PACS for each country, are presented in Table 4. The percentage of data originating from the PACS system, extracted through a dedicated software tool (PerMoS,19 used to automatically collect technical parameters of each scan from the header of the image in PACS), differed between the countries since the system was only implemented relatively recently in hospitals and not all participating hospitals were willing to query their PACS in addition to RIS. Number of CT scans received by child Proportion of CTs extracted from PACS of all CTs. Number of CT scans received by child Proportion of CTs extracted from PACS of all CTs. From a combination of PACS data and information on radiological protocols used in the participating hospitals, individual doses to specific relevant organs, including red bone marrow and brain for respectively leukaemia and CNS cancer risk estimation, were then estimated for each CT scan for each child, using National Cancer Institute Dosimetry System for CT (NCICT) software20 and taking into account uncertainties in dose estimates.16 The dose reconstruction strategy which has been implemented allows accounting for missing data. Doses were therefore reconstructed for all examinations, but estimated doses were associated with large uncertainties when examinations were poorly characterized in the RIS data file, including missing information on the anatomical zone scanned. The total number of CT scans collected within the study was 1 430 454. The mean number of CT scans per patient was 1.5. The majority of patients (75%) received a single recorded CT scan (Table 4), and only 0.2% of them received more than 10 scans. The percentages of scans recorded before and after the year 2000 were 22% and 78%, respectively. Table 5 reports the distribution of CT according to the anatomical area explored. The main areas examined were ‘head and neck’, ranging from 49.5% in Denmark to 72.1% in Belgium, followed by ‘chest’. accounting for 8.4% in Sweden to 19.3% in France. Scans of multiple body parts represented 4.9% of the total number of CT scans. Those without anatomical area mentioned represented 2.9% of all collected CT, with large discrepancies between countries reflecting the variability of data storage at hospital level. Distribution of CT scans according to age, calendar period and scanned region by countries Distribution of CT scans according to age, calendar period and scanned region by countries Further information on underlying diseases was also collected from various sources including hospital diagnostic or discharge databases and rare diseases registries (in France, The Netherlands, Norway, Spain and Sweden). Socioeconomic status (SES), a potential confounder, was obtained from national census data by postal code or census track for Belgium, France, The Netherlands, Spain and the UK and from the National Education registry in Denmark, Norway and Sweden. Standardized Mortality Ratios (SMRs), as the ratio of observed and expected number of deaths based on national reference rates, were calculated for the eight countries for which vital status was available, using a 1-year exclusion period. Results are presented (Table 6) for two distinct periods of follow-up after the first CT: years 2 to 5 and 5 years and beyond. The percentage of deaths varied from 0.5% in Belgium to 2.6% in the UK for the whole study period. Even though the death rates were low, we observed strongly elevated SMRs in all countries, especially in the first 5 years following the first CT (Table 6). The SMRs for all-cause mortality during the 5 years following the first CT were statistically significant and greater than 1 in all countries, varying from 1.9 in Belgium to 4.9 in Denmark, France and the UK. For the time period greater than 5 years after the first CT, the SMR decreased but remained significantly raised, in all countries except Belgium and Spain. Cancer mortality decreased to the level of the general population when time since first exposure exceeded 5 years, except in Sweden [SMR = 1.3, 95% confidence interval (CI) 1.1, 1.6). However, non-cancer SMRs remained significantly increased when considering time since exposure greater than 5 years, except in France and Spain. These results illustrate the fact that, as expected, our study population was less healthy than the general population. Indeed, children undergoing CT, particularly those with repeated scans, are likely to suffer conditions that could be associated with increased mortality. The observed decrease of SMR for cancer when considering time since first CT greater than 5 years confirms the need to apply exclusion periods in the statistical analyses to avoid reverse causation (i.e. the CT was performed because of a suspicion of cancer). Various exclusion periods will be examined in the main risk analyses. Detailed dose-response analyses for cancer incidence will be presented in later papers. SMRs of all cause, cancer and non-cancer mortality according to the time since the first CT (<5 years; >= 5 years) PYs, person-years of follow-up; O, observed cases within the cohort; SMR, standardized mortality ratio; 95% CI, 95% confidence interval; -, cause of death unknown. SMRs of all cause, cancer and non-cancer mortality according to the time since the first CT (<5 years; >= 5 years) PYs, person-years of follow-up; O, observed cases within the cohort; SMR, standardized mortality ratio; 95% CI, 95% confidence interval; -, cause of death unknown. A summary of published results from national EPI-CT cohorts is provided in Table 7. Four national cohorts (the UK, France, Germany and The Netherlands) have published analyses of the relationship between CT exposure and cancer incidence. The British study reported a dose-response relationship between CT-related dose and CNS tumours and leukaemia in exposed children and young adults,6,21 but not for Hodgkin lymphoma.22 The German study reported a significantly increased incidence of cancer and lymphoma in exposed children compared with the general population.9 Both the French and the German studies, based on small numbers of cases, reported a dose-related increase for leukaemia and CNS tumours, though it was statistically significant only for CNS tumours in Germany.9–11 The Dutch study reported a dose-response relationship for cranial CNS tumours and found no association with leukaemia.12 Results from EPI-CT national cohorts CNS, central nervous system; PF, predisposing factor; RBM, red bone marrow; ERR, excess relative risk; RR, relative risk; HR, hazard ratio; mGy, milligray. Follow-up period until 2005 only. Exclusion period 5 years instead of 2 years. Hodgkin lymphoma only. Results from EPI-CT national cohorts CNS, central nervous system; PF, predisposing factor; RBM, red bone marrow; ERR, excess relative risk; RR, relative risk; HR, hazard ratio; mGy, milligray. Follow-up period until 2005 only. Exclusion period 5 years instead of 2 years. Hodgkin lymphoma only. Reverse causation bias has been considered by applying various exclusion periods for cancer risk analyses. Exclusion periods allowed accounting for individuals who were potentially scanned because of a suspicion of leukaemia (although CT generally is not required for the initial diagnosis of leukaemia) or CNS tumours. Extending the exclusion period from 5 to 10 years for CNS tumours in the British study did not decrease the dose-risk estimates, as would be expected in case of reverse causation bias.6 Potential confounding by indication for CT, meaning that children requiring CT may be at risk of cancer because of an underlying condition, has been handled in different manners in these studies. In the British study, an analysis published in 2016 took into account clinical information available in the RIS for 40% of the cohort and death certificates, to evaluate this potential bias.21 This resulted in a slightly decreased, but still significant, dose-related increased risk for CNS tumours. Exclusion of previously unreported cancers reduced the Excess Relative Risk (ERR) per mGy by 15% from 0.036 to 0.033 for leukaemia and by 30% from 0.023 to 0.016 for CNS tumours, but these ERRs remained statistically significantly elevated. In the French cohort, using reliable information on predisposing factors (PF) for cancer from hospital discharge registries, 3% of the subjects were found to have a PF for CNS tumours and/or leukaemia.10 This small percentage of individuals with PF was nevertheless much larger than expected in the general population.23 Separate analyses showed a dose-related increase of leukaemia and CNS tumours in children with no PFs, whereas there was no evidence of an increase in those with PFs.11 The difference in radiation-related risks observed according to the presence of PFs might be explained by much higher mortality risks in patients having PFs compared to children without PFs. Furthermore, Meulepas et al. calculated the magnitude of predisposing syndrome-related confounding of relative risk (RR) estimates for leukaemia and CNS tumours after diagnostic CT, under various assumptions for the association between predisposing syndromes and the frequency of CT. that these syndromes were to cause confounding as were and CT frequency was only elevated these about children included in the study, it has been that the EPI-CT study has statistical power to small excess cancer risks in this first follow-up at least for leukaemia and CNS It will provide new into the potential cancer risk from CT exposure during the study of specific as of age at sex, exclusion period and cancer particularly as the population and follow-up is The coordinated international analyses in EPI-CT not only increase statistical but also to and results from different countries methodological to a common were to the estimation of organ doses for each are provided in the the reconstruction for the data were extracted from the RIS and, for more recent time periods, from the PACS of participating hospitals, dose The implemented allows of uncertainties in doses due to missing data and a range of potential doses for each CT each set for use in a dose-response as a of the missing is represented by a of the of for the time period. For each of the cohort dose of parameters are from the between all examination types into common and of organ doses using the software with for paediatric age allowed a of the from the use of doses for the main risk EPI-CT also the for a large and of doses from CT in children in previously only information was A of the study is the to million individuals to collect more information on potential as underlying predisposing other medical radiation procedures performed in the hospital procedures, other or the participating hospitals. for CT is also not and countries have electronic medical that could provide this information for a retrospective cohort of this studies, based on of expected range of potential based on information available in some of the participating countries or will be performed to provide information on the of missing data in the risks analyses will be performed on cohorts with information on potential to provide a for risk studies are also in several countries to of data on other radiation procedures, missing scans, predisposing factors and other potential of the relation between CT dose and leukaemia and CNS of the EPI-CT is that the cohort is still relatively whereas cancers non-cancer are more at Further follow-up of this cohort will provide information and will a more of the of exposure on different as as on the of age at exposure and age on estimated be provided to the follow-up. follow-up is currently under in of the largest countries with from the European within the will be in other countries to the follow-up of the other cohorts in the years. data are not available because of and data The data are at the and be the for in analyses of the data be to and will be by the EPI-CT The multinational EPI-CT study was set up in 2011 to provide estimates of risk of tumours and leukaemia children and young adults who had computed tomography (CT), and to the for optimization of paediatric CT protocols and patient a common cohort studies were in Belgium, Denmark, France, Germany, The Netherlands, Norway, Spain, Sweden and the UK, coordinated by the International Agency for Research on Cancer The study a total of about patients who had CT at least once before the age of 22 years. A total of 8.7 million person-years of incidence follow-up were between 1977 and were followed up through linkage with cancer and mortality registries. A was to individual organ doses and estimate associated using data available in electronic of the radiology of participating hospitals. Description of the cohort and analysis of mortality risk are presented for in analyses of the data be to and will be by the EPI-CT The UK study was partially by the National Cancer Institute Research of the UK of Cancer Research UK and Research in and and The French study received from the French National Cancer Institute and from The Dutch study was partially by Cancer as the for International Cancer Research The German study was partially funded by the German of Education and Research under numbers and The cohort was funded by the Research of The study was partially by a from the was the of a of the for a at the national cohorts the were funded also by the European for and under The are to the and of participating hospitals, the national registries of cancers and the national mortality and death registries that provided exposure and clinical data. to all the in the of the more and for the British cohort, and for the German cohort, and of the radiology in the French cohort. and are with the National Institute for Research Research in and and at in with The are those of the and not those of the the the of or of