Radiation Exposure From Pediatric CT Scans and Subsequent Cancer Risk in the Netherlands

Methods

Results

We collected data on 262 227 pediatric CT scans performed on 168 394 children in 42 hospitals between 1979 and 2014. After exclusions due to foreign addresses, age restrictions, incompleteness or data errors, and follow-up of less than two years (leukemia) or five years (solid tumors) after the first scan, we included 140 612 patients in the leukemia analyses and 106 530 patients in solid tumor analyses. Four percent of individuals alive two years after their first CT had CT scans in multiple participating hospitals. Patient characteristics are presented in Table 1.

Among the 208 709 CT scans eligible for leukemia analyses, 65.5% were of the head, 8.3% of the extremities, 8.1% of the chest, and 7.6% of the abdomen. For brain tumor analyses (n = 158 843 CT scans due to the longer exclusion period), the distribution across body parts was similar (Table 2). The mean cumulative RBM dose at the end of follow-up was 9.5 mGy, and for 75% of patients it was less than 12 mGy. The mean brain dose was 38.5 mGy (Table 1).

Overall cancer incidence (starting five years after the first CT) was 1.5 times higher than expected (SIR = 1.47, 95% CI = 1.34 to 1.61, 454 observed cases) including malignant CNS tumors (SIR = 2.05, 95% CI = 1.48 to 2.83, 37 observed cases) and hematolymphoproliferative cancers (HLP; SIR = 1.39, 95% CI = 1.13 to 1.70, 93 observed cases) (Table 3). Statistically significant excesses were also noted for cancers of the colon, bone, large bowel, soft tissues, thyroid, and for nonmelanoma skin cancer.

Risk for all leukemia combined (44 cases) was not associated with bone marrow dose (ERR/100 mGy = 0.21, 95% CI = –0.12 to 2.40, P = .68) (Table 4), including all leukemia and MDS (ERR/100 mGy = 0.04, 95% CI = –0.12 to 1.61, P = .92). For all brain tumors combined, and for malignant and nonmalignant brain tumors separately, relative risks increased to between two and four for the highest dose category (≥120 mGy), with ERRs of 0.86 (95% CI = 0.20 to 2.22, 84 cases), 1.02 (95% CI = 0.01 to 4.30, 37 cases), and 0.78 (95% CI = 0.07 to 2.58, 47 cases) per 100 mGy and no evidence of nonlinearity (P = .72, .45, .92) (Table 5). The dose–response for brain tumors and for leukemia did not vary according to sex, number of CT scans, years since exposure, or age at exposure (Table 6).

Several factors potentially biasing risk estimates were assessed in sensitivity analyses, including—sometimes substantially—reduced numbers of patients, CT scans, and/or cancer cases. Adjustment for household income and exclusion of 82 patients with confirmed TSC did not materially affect brain tumor risks (Supplementary Table 5, available online) whereas they were attenuated with a 10-year lag instead of five years (Supplementary Table 6, available online). Reducing the exclusion period from five to two years did not change results, whereas increasing it to eight years somewhat attenuated radiation risks. Consistent results were obtained for nonmalignant brain tumors in analyses limited to the post-2001 period with complete registration of nonmalignant brain tumors. Further, results were generally similar (with reduced precision) when excluding subjects with a first CT before 1989 or born before 1989 (Supplementary Table 6, available online). Incidence of CNS cancers vs the general population remained elevated 10 or more years after the first (head) CT (Supplementary Tables 7 and 8, Supplementary Figures 1 and 2, available online). Leukemia results did not change substantially during comparable sensitivity analyses (Supplementary Tables 9 and 10, available online). Excluding patients scanned in hospitals near the German or Belgian border did not change the results (not shown). A threshold at 20 mGy resulted in the best goodness of fit for all brain tumors combined, although the deviance was only marginally (and statistically nonsignificantly) reduced compared with the no-threshold model and any other threshold level between 5 and 120 mGy (not shown). Excluding the 2% person-years with the highest cumulative brain dose (exceeding 220 mGy) resulted in a similar and significant ERR for all brain tumors as in the main analysis (based on 78 cases).

Discussion

In the Dutch Pediatric CT study, we observed a statistically significant dose–effect relationship for brain tumors and found no association between the estimated cumulative RBM dose and incidence of leukemia.

Many published studies with empirical data have noted a positive association between radiation dose from CT scans and brain tumor (5,7,12) and leukemia (excluding MDS) risks (8,12). In these studies, radiation dose estimates were based on 1) external estimates stratified by sex, age, calendar period and body part (7–9), 2) number of CT scans (8,10,12), or 3) protocol-based doses (11). None of these studies estimated doses based on PACS data for a sizeable subset of CT scans from the cohort and predicted doses for all CT scans in the cohort, as done here.

Further strengths of our study include the representativeness and completeness of the cumulative exposure estimates. The coverage of the previously published cohorts varies from a few hospitals to a nationwide health care system. Our study includes the majority of all pediatric CT scans conducted annually in the Netherlands (42 of 60 eligible hospitals performing pediatric CT scans and 84% of all Dutch pediatric CT scans for the period 1990–2012) (3). Moreover, we had the ability to combine CT scans of the same patients in multiple hospitals for complete cumulative exposures. The average follow-up time in our study was about eight years (maximum = 21 years), which is close to the largest studies published to date (7,9).

The absence of a statistically significant association between radiation dose and leukemia in our data is not inconsistent with other studies. After exclusion of MDS, the ERR from the UK-NCI study was no longer statistically significantly elevated (ERR/100 mGy: 1.9, 95% CI = –1.2 to 7.9) (7). Smaller studies from France and Germany also observed statistically nonsignificantly elevated risks (11,12). We chose not to include MDS in the main leukemia outcome definition because only 3%–7% of MDS progress to acute myeloid leukemia (28).

The ERR/100 mGy of 1.02 (95% CI = 0.01 to 4.30) for brain cancer observed in our data is compatible with the estimate for a comparable sample of atomic bomb survivors (<20 years of age at exposure, during ≤20 years of follow-up: 0.61, 95% CI = 0.012 to 6.39) (29), and also similar to the corrected estimate from the UK-NCI study (1.3 per 100 mGy) (8).

Our study has some limitations. Most importantly, we did not have information on the (clinical) indications for the CT scans to, for example, limit analyses to children with trauma.

Although pediatric CT scans are not performed outside hospitals in the Netherlands, we might have missed some CT scans, for example, CT scans in one of the 18 nonparticipating hospitals, or CT scans in radiology practices or hospitals across the border in Germany or Belgium, or CT scans in participating hospitals before electronic recording was introduced. However, excluding children with CT scans performed in Dutch border-region hospitals or excluding children born before electronic recording was available yielded similar results.

Socioeconomic status is associated with childhood leukemia incidence (30,31) and with the likelihood of having a CT scan (2,32), but it did not confound our results. Excluding TSC patients did also not change the results. For other syndromes, we were not able to obtain individual information. We showed in a theoretical assessment that CSS are unlikely to be major confounders for brain tumor and leukemia risks (16). The brain tumor dose–response in the UK-NCI study was attenuated after exclusion of subjects with a previous cancer diagnosis, that is, patients for whom the cancer ascertained in the study, in fact, was a subsequent primary tumor. After exclusion of subjects born before 1989, the first year that solid cancer registration was complete in the Netherlands, our results were based on a substantially smaller database and were attenuated, although still generally consistent with the results of the main analyses.

It has been questioned whether epidemiological cohort studies of children who received CT scans for medical reasons can provide unbiased insights into the low excess risks expected from the relatively low CT-related radiation doses, particularly if the reasons for the CT scans are unknown (13,14,33). These concerns are more relevant for brain tumors than for leukemia because CT scans are not routinely used for diagnosing leukemia, and the onset of predominantly acute leukemias among children and young adults is generally rapid. For brain tumors, concerns raised include slow-growing tumors in the brain leading to symptoms for which CT scans are used long before clinical diagnosis, or accidents leading to head CT scans while brain trauma might increase long-term brain tumor risk (34). The critics did not provide any data, however, to support their claims. To address the issues, we, and others, started follow-up five years after the first CT scan and included only children without a cancer diagnosis at that time (two years for leukemia). Results changed relatively little when the period was extended. However, the interpretation of small radiation-related excess risks is complicated by two observations. Brain cancer incidence in the study population is twofold higher than expected, even 10+ years after a first head CT, although the numbers are small. Also, the absence of clearly increased risks for leukemias is puzzling because these malignancies originate in the RBM, a tissue more radiosensitive than the brain. At the same time, the risk estimates for leukemias are likely less affected by indication bias and reverse causation. These considerations necessitate a cautious interpretation of the brain tumor findings. The observed general excess cancer risk (SIR = 1.47) is unlikely to be radiation-related but, rather, due to an over-representation of young patients who had diagnostic/screening CT scans for disorders related to genetic syndromes (eg, bone/soft tissue sarcoma, CNS tumors, nonmelanoma skin cancer) or other underlying medical conditions (eg, colon carcinoid detection during abdominal imaging for other reasons).

In summary, our results indicate that the relatively high brain doses from head CT scans (20–50 mGy) may increase brain tumor risk, whereas RBM doses generally under 10 mGy per CT and often under 5 mGy do not lead to an observable risk increase (Supplementary Tables 1 and 2, available online, for CT scans after 2001). The brain tumor findings may reflect a causal relationship because 1) our study is large and has almost nationwide coverage, 2) the exposure assessment is based on a large internal sample of CT scans with dosimetric information, 3) cancer registration in our study goes back to 1973 (leukemia) and 1989 (solid cancer) with high completeness, 4) the brain cancer results are very similar to those in the Life Span Study, 5) we could rule out confounding by TSC, the most common brain tumor susceptibility syndrome for which children have been exposed to CT scans, and 6) adjustment for socioeconomic status in our study is one of the most detailed ones so far. Nevertheless, the results need to be interpreted with caution because 1) the RBM is known to be more radiosensitive than the brain, 2) leukemia risk estimates are less prone to reverse causation and indication bias in CT studies than brain tumor risk estimates, 3) the number of cases in our study is too small for subgroup analyses, and 4) the increased brain tumor risk in our study is observed earlier after low-dose radiation exposure than current knowledge on latency for solid tumors suggests.

From a clinical perspective, we observed an excess absolute risk for all brain tumors of 1.3 (95% CI = 0.4 to 2.2) per 100 000 person-years per 20 mGy, the average brain dose among head CT scans during 2012–2014 in our study. This means that during the decade after the first head CT scan, one excess case per 10 000 head CT scans is estimated to occur. We estimate the number of annual head CT scans among Dutch children to be around 10 000 (3), leading to one brain tumor case annually attributed to radiation. In the Dutch population, nearly 120 pediatric brain tumors are diagnosed annually (35).

CT scans (for children) represent a potentially life-saving and quality of life–improving technique for many patients. In addition, the tumors evaluated here are associated with small absolute excess risks. Nonetheless, careful justification of pediatric CT scans and dose optimization, as are customary in many hospitals, are essential to minimize risks.

Funding

This work was supported by the European Community Seventh Framework Programme (grant number FP7/2007-2013) under grant agreement number 269912-EPI-CT (“Epidemiological study to quantify risks for paediatric computerized tomography and to optimize doses”) and by Worldwide Cancer Research, formerly known as the Association for International Cancer Research (grant number 12-1155). Dr. Cecile Ronckers is supported by the Dutch Cancer Society.

Notes

Affiliations of authors: Department of Epidemiology and Biostatistics, Netherlands Cancer Institute, Amsterdam, the Netherlands (JMM, PG, FEvL, MH); Department of Paediatric Oncology, Emma Children’s Hospital (CMR, JHM), and Department of Radiology (AMJBS), Academic Medical Center Amsterdam, Amsterdam, the Netherlands; Dutch Childhood Oncology Group, the Hague, the Netherlands (JHM); Department of Radiology (RAJN) and Department of Internal Medicine (BZ), University Medical Center Utrecht, the Netherlands; Division of Cancer Epidemiology and Genetics, National Cancer Institute, NIH, Rockville, MD (CL); Luxembourg Institute of Science and Technology (LIST), Esch-sur-Alzette, Luxembourg (AJ); Department of Radiology and Nuclear Medicine, Erasmus MC Rotterdam, the Netherlands (MvS); Department of Neurology and Paediatric Neurology, Erasmus MC, Sophia Children's Hospital, Rotterdam, the Netherlands (MCYdW); Department of Radiology and Nuclear Medicine, Radboud University Medical Center, Nijmegen, the Netherlands (WMK); Department of Registration, Netherlands Comprehensive Cancer Organisation, Utrecht, the Netherlands (OV).

The funders had no involvement in the study design, data collection, analysis and interpretation, writing of the report, or the decision to submit the paper for publication.

We sincerely thank the staff members, clinicians, and boards of directors of all hospitals who provided data for our study (AMC Amsterdam, Amphia Ziekenhuis, Albert Schweitzer Ziekenhuis, Bethesda Ziekenhuis, Canisius-Wilhelmina Ziekenhuis, Diaconessenhuis Leiden, Isala Diaconessenhuis Meppel, Diakonessenhuis Utrecht, Elkerliek Ziekenhuis, Erasmus MC, Flevoziekenhuis, Groene Hart Ziekenhuis, Medisch Centrum Haaglanden, HagaZiekenhuis, Ikazia Ziekenhuis, Isala Ziekenhuis, Jeroen Bosch Ziekenhuis, Kennemer Gasthuis, Streekziekenhuis Koningin Beatrix, Leids Universitair Medisch Centrum, Medisch Centrum Alkmaar, Medisch Centrum Leeuwaarden, Meander Medisch Centrum, Medisch Spectrum Twente, Onze Lieve Vrouwe Gasthuis, Radboudumc, Rijnland Ziekenhuis, Rijnstate, Rivas Beatrix Ziekenhuis, Saxenburgh groep Röpcke-Zweers Ziekenhuis, Sint Franciscus Gasthuis, Slotervaartziekenhuis, St. Antonius Ziekenhuis, St. Elisabeth Ziekenhuis, Sint Maartenskliniek, Maastricht UMC, UMC Groningen, UMC Utrecht, Het Van Weel-Bethesda Ziekenhuis, VUMC Amsterdam, Ziekenhuisgroep Twente, ZorgSaam Zeeuws-Vlaanderen). This study would not have been possible without the dedicated support of numerous professionals in the participating radiology departments.

Supplementary Material