Study Design

We completed a retrospective population-based cohort study of women who delivered a liveborn infant in Ontario between April 1, 1992 and March 31, 2008. The study was approved by the Ethics Review Board of Sunnybrook Health Sciences Centre.

Participants

We included singleton and multifetal pregnancies, and for the latter we included the first infant birth record to avoid multiple counts per pregnancy. We included maternal-infant pairs comprising term liveborn infants who survived more than 30 d after date of birth. For three reasons, we restricted the sample to term infants ≥37 wk gestation, and who weighed at least 2,500 g at birth: First, preterm infants are commonly exposed to radiodiagnostic imaging procedures in hospital [21],[22], which could potentially contaminate the studied effect of prenatal exposure. Second, since term infants represent 94% of all livebirths [23], our findings would remain applicable to most pregnancies. We defined term infants by actual gestational age from 2002 onward, the time in which it was captured in our dataset; before 2002, term status was based on the absence of any diagnostic codes for prematurity according to the International Classification of Diseases, 9th (ICD-9) and 10th (ICD-10CA) revisions (Table S1). Thus, limiting our cohort to term infants enabled us to better estimate the gestational age of fetal exposure to a major radiodiagnostic test prior to the year 2002.

Exposure and Outcome

The study exposure was CT or radionuclide imaging in an index pregnancy, up to 2 d before the day of delivery (to maximize the likelihood that the fetus was in utero). Prior to the year 2002, gestational age at birth was not recorded in our databases, so the estimated gestational age at exposure was calculated by subtracting 37 wk from the delivery date. This approach has a reported sensitivity of 90% and specificity of 99% [24]. From 2002 onward, we used the recorded gestational age at delivery to determine the timing of exposure to a radiodiagnostic test.

The main study outcome was a confirmed diagnosis of any pediatric malignancy in a child born in the index pregnancy at any time 30 d or more after his/her birth. This 30-d interval was used to exclude infants who died from major morbidity, such as neonatal sepsis or severe anomalies.

Database Sources

We used anonymized databases for the entire province of Ontario, where universal health care, including prenatal care and radiodiagnostic testing, is available to all residents. Databases were linked using unique encrypted identifiers, which enabled us to link radiation exposure in an index pregnancy to a diagnosed malignancy in a child.

Individual obstetrical deliveries were identified in the Canadian Institute for Health Information Discharge Abstract Database (DAD). The DAD contains age and sex, dates of admission, and up to 16 diagnoses coded by ICD-9 up to March 31, 2002, and up to 25 diagnoses coded by ICD-10-CA thereafter. Up to 2002, inpatient admission records for mothers and newborns are probabilistically linked using delivery/birth dates in the same hospital, same residential postal codes, and diagnostic information to create each mother-child pair. This approach leads to successful matching in 95% of cases. Validation studies of similar maternal-newborn linkages have shown that this approach yields high degrees of successful matching, with a sensitivity of about 95% and a specificity of nearly 90% [25]–[27]. In a Canadian study linking newborn birth records and infant death records, over 99% of records were successfully linked [28]. Starting in the fiscal year 2002/2003, a common identifying number between a delivering mother's chart number and her newborn's chart number permitted a deterministic linkage between the two from 2002 and onward. The DAD was also used to define study covariates (Table S1).

For the study exposure—a major radiodiagnostic test performed on the mother up to one day before her delivery date—we used the Ontario Health Insurance Plan (OHIP) database, which captures all billing information for physician services, including inpatient and outpatient major radiodiagnostic services [29],[30]. In an audit of a representative sample of 11,824 outpatient CT scans and 11,867 outpatient magnetic resonance imaging (MRI) scans identified from 29 randomly selected hospitals across Ontario, 95% of the procedures studied in the chart audit had a corresponding claim in the OHIP database, indicating that these administrative databases have high accuracy in identifying imaging procedures [30]. In the current study, major radiodiagnostic testing included all forms of CT and radionuclide imaging (Table S2). Because the OHIP database does not capture all billings for inpatient radionuclide imaging before 2006, the DAD was used to identify inpatient radionuclide tests performed throughout the study period (Table S2).

For the study outcome—pediatric malignancy—each child record was linked to the Ontario Cancer Registry (OCR), a computerized database of information on all Ontario residents who have been newly diagnosed with cancer [31]. All new cases of cancer, except for nonmelanoma skin cancer, are registered, and about 80% have the tissue pathological diagnosis recorded within the OCR. Validation studies have shown the OCR to be effective in ascertaining cancer cases in the province, with a sensitivity of 98% [32]. Close to 400,000 records are submitted to the OCR annually, and coded using ICD-9 diagnostic codes (Table S1). Incident cases of pediatric malignancy were captured according to a date of diagnosis between May 1, 1991 (a minimum of 30 d after the earliest birth in the study cohort) and March 31, 2009 (a maximum of 18 y after the earliest birth in the cohort).

Mortality data were retrieved from the Registered Persons Database (RPDB), which contains demographic information for all OHIP-eligible individuals. Income quintile and rurality were defined using Statistics Canada census data. The OHIP database was also used to identify outpatient claims for prenatal ultrasonography at any point in pregnancy (Table S2) [33].

Statistical Analysis

We measured the absolute number of radiodiagnostic tests performed in each livebirth pregnancy in a given year and calculated the annual test rate per 1,000 livebirths, from 1992 to 2008. Changes over time in the rates for radionuclide tests, CT, and both were analyzed using a Cochran-Armitage test for trend.

Time-to-event analyses were conducted using Cox proportional hazards regression models, to derive a hazard ratio (HR) and 95% confidence interval (CI) for pediatric malignancy among children born to mothers exposed versus not exposed to major radiodiagnostic imaging in pregnancy.

The period of observation for each child started 31 d after birth, to ensure that he or she survived long enough to develop, or to be diagnosed with, a malignancy. A child was censored (i.e., determined to have not had a study outcome event) at the point in time in which he or she either reached the end of the period of March 31, 2009 without having experienced a study outcome event or died during the period of study. While emigration from the province could not be ascertained, the annual rate is less than 1% (http://www.statcan.gc.ca/pub/75-001-x/2008110/t/10711/5800473-eng.htm); such persons were classified as being event-free up to March 31, 2009.

The HR was adjusted for maternal age at delivery (in years), income quintile, urban status, and maternal cancer diagnosed in the index pregnancy or up to 6 mo thereafter, as well as infant sex, a chromosomal or congenital anomaly documented at the time of birth or up to 365 d thereafter (present or absent), and exposure to CT or radionuclide imaging (as present or absent) starting 31 d after birth and up to 365 d before either the date that he/she experienced a study outcome event or the date that he/she was censored. The latter considered postnatal exposure to the same ionizing radiation sources as in pregnancy. Information about maternal parity, prior pregnancy loss, smoking, and medication use were not available.

All p-values were two-sided, at a significance level of 0.05. All statistical analyses were performed using SAS for UNIX, Version 9.1 (SAS Institute).

Study Strengths and Limitations

Among those exposed to major radiodiagnostic testing, totaling nearly 5,600 pregnancies and 35,000 persons-years of follow-up, just four cases of childhood malignancy were identified. The incidence rate of 1.56 cancers per 10,000 person-years, based on 2,539 events arising among 1,829,995 unexposed pregnancies, is highly concordant with population studies from Canada, the United States, Ireland, and Europe [1]–[3]. Post hoc, considering the number of participants enrolled in the current study, at a conventional p-value of 0.05, we had just under 20% statistical power to detect the observed difference of 0.40 per 10,000 person-years in the incidence rate of childhood malignancy between exposed and unexposed mothers. To achieve a power of 80%, approximately 60 and 19,613 cancers would be needed in the exposed and unexposed groups, respectively—about eight times higher than the current number of events. That so few cases of childhood malignancy were identified among the exposed is a testament to the difficulty of conducting research in this area, and why previous data may have been lacking. Nonetheless, the wide confidence limits of our relative risk estimates do not exclude the possibility that fetal exposure to CT or radionuclide imaging in pregnancy is carcinogenic, ranging from as low as one-quarter, to as high as 1.8 times, that of an unexposed pregnancy.

We did not include women who had a radiodiagnostic test in a pregnancy that ended in a spontaneous or therapeutic termination before 20-wk gestation. However, this study's focus was on childhood malignancy after 30 d of life, rather than on pregnancy loss, and it is unlikely that including that group of women would have systematically biased the association between the study exposure and outcome. Moreover, the rates of stillbirth after 20 wk among excluded women exposed (0.65%) and not exposed (0.84%) to major radiodiagnostic testing in pregnancy did not differ significantly. By limiting our sample to term deliveries between 37 and 43 wk gestation and restricting the period of exposure up to 2 d before a woman's date of delivery, we improved the accuracy of our determination of the timing of exposure to radiodiagnostic testing. At the same time, our findings may not apply to preterm infants, who both have higher rates of congenital and chromosomal anomalies, as well as major radiodiagnostic testing soon after birth (Table S3) [21],[22].

We did not have on record whether a woman's pregnancy was known at the time of exposure, or if she received lead apron shielding of the abdominal-pelvic area. Assuming that shielding occurred in most cases, as is routine practice, then this may partly explain the low rate of childhood malignancy in the exposed group. Moreover, two-thirds of CT scans were of an extremity or head, with the remaining sites in closer proximity to the fetus, but only 23% were of the abdomen, spine, or pelvis. Hence, much of our CT data reflects “low-risk” scans, in terms of fetal radiation exposure.

We also do not possess information about the generation(s) of CT scans that were used, or the measured dose of ionizing radiation that a woman and her fetus were exposed to. The cited average effective maternal radiation dose of CT of the head or neck is about 2 millisievert (mSv); for a CT of the chest, spine, abdomen, or pelvis it ranges from 6 to 8 mSv [9],[34]; and for a ventilation-perfusion lung scan it is about 2.5 mSv [34]. While these doses are between 10 to 100 times more than that of an equivalent single plain X-ray [34], the effective radiation dose to the fetus remains lower than that to the mother [18],[35]. The rate of major radiodiagnostic testing in pregnancy in our study is certainly lower than when plain X-rays were performed in pregnancy for maternal pelvimetry [36]. In the era of the current study, pelvimetry was not routine, but we would have missed plain X-rays performed for other reasons.

Mothers, as well as clinicians of all types—internists, surgeons, radiologists, obstetricians and midwives, pediatricians, and family and emergency medicine physicians—need data to better estimate the risk of exposure to radiodiagnostic testing in pregnancy. Our findings, though statistically underpowered, inform this process in a manner that has been lacking to date, especially using population-based data over a long period of follow-up, and within a health care system that provides universal coverage of radiological testing and health care.

Clinical Implications

Our findings can help inform clinicians and mothers about the risk of childhood malignancy following major radiodiagnostic testing in pregnancy. First, the absolute risk appears to be low (about 0.07% or 1 per 10,000 person-years), while the relative risk is not statistically significantly higher than unexposed controls, notwithstanding the wide confidence limits. However potentially small that risk may be, beta hCG testing should continue to be done in all potentially pregnant women before undergoing major radiodiagnostic testing, and lead apron shielding used in all women of reproductive age, whether or not pregnant [6],[37]. Furthermore, nonradiation-emitting imaging (e.g., MRI and ultrasonography) should be considered first, when appropriate.

Some pregnant women will nevertheless be faced with the decision to undergo CT or nuclear imaging because the test is clinically warranted. Unfortunately, some health care providers may be unwilling to counsel a woman about the related fetal risks or may provide misinformation [37],[38]. This issue seems more pressing than ever, given that we and others [9] have noted that CT testing is on the rise, and has become the mainstay for the investigation of pulmonary embolism [39], stroke [40], appendicitis [41], and other common conditions encountered in emergency situations [10]. Both physiological and anatomical changes in pregnancy may obscure the clinical diagnosis of pulmonary embolism [42], for example, such that CT or radionuclide testing may be especially warranted. Delaying the diagnosis of such conditions may postpone therapy, in turn, jeopardizing both mother and child [43].

When indicated, major radiodiagnostic testing in pregnancy should be carried out, along with brief counseling [37]. The latter will hopefully lessen the level of anxiety experienced by an expectant mother (and her family), not only at the time of illness, but after her child is born.