We used the Danish Cancer Registry, which has covered all incident cancers in Denmark since 1943, classified according to the International Classification of Diseases (ICD-7) (Storm et al, 1997
), to trace all women with a diagnosis of HD (ICD-7 code 201). Information included the civil registration number of the woman, date of diagnosis, and radiation treatment administered within 4 months of diagnosis.
Since 1 January 1973, all births in Denmark have been registered in the Danish Medical Birth Registry (Knudsen and Olsen, 1998
). Data are obtained from birth notifications, which are completed by midwives (who attend all births, including home births, in Denmark). The main variables in the Birth Registry are gestational age, birth weight, parity, stillbirth, place of birth, and the civil registration number of the mother and child (which encodes sex and date of birth and is assigned to all live-born children and new residents; Frank, 2000
Using the civil registration number, we linked the Cancer Registry data with the Birth Registry to establish a cohort of all Danish women with a diagnosis of HD in, 1970–2002, and who gave birth in 1973–2002. Women were included if they were diagnosed with HD before pregnancy, during the pregnancy, or until 2 years postpartum. We restricted all analyses to singleton births, since multiple births have been associated with an adverse birth outcome (Pinborg et al, 2004
For each birth by a woman with HD, 50 comparison births matched by month and year of the birth, by county of mother's residence, and born to 50 different women who were not diagnosed with any cancer before, during, or within 2 years after the pregnancy were selected from the Birth Registry. If fewer than 50 births fulfilled the matching criteria, we used all the available births. If more than 50 comparison births were eligible after matching, we selected a random subset of 50 births. On average, 48 comparison births were selected for each exposed birth.
The outcome data collected from the Birth Registry included preterm birth (birth before 37 completed weeks of pregnancy), low birth weight at term <2500
37 completed weeks), stillbirth (delivery of a dead foetus at
28 completed weeks of pregnancy), male proportion of newborns, and birth weight. The potential confounders included maternal age, parity, gestational age, and calendar period of the birth. For live-born children, data on congenital (including chromosomal) abnormalities, diagnosed during the first year of life were collected from the National Hospital Discharge Registry, covering all hospital discharge diagnoses since 1977 and outpatient visits since 1995 (Andersen et al, 1999
). Thus, data on congenital abnormalities applied to births from 1977 to 2002. The data include the civil registration number, dates of admission and discharge, and up to 20 discharge diagnoses, (ICD-8 before 1994 and ICD-10 from 1994 onwards; Andersen et al, 1999
). The codes for congenital (including chromosomal) abnormalities were 740.00–759.99 in ICD-8 and Q0.00 to Q99.9 in ICD-10. Diagnoses of congenital dislocation of the hip and undescended testis were excluded because of their poor validity (Larsen et al, 2003
g probably reflected coding errors and were excluded, as were births with a gestational age below 20 or over 44 weeks. Owing to a coding change in the Birth Registry in 1978, there were more missing data on gestational age for the years 1978–1981 than for other years (mean 22.6% missing for 1978–1981, compared with 0.8% in 1973–1977 and 1.2% in 1982–2002). Births without data on gestational age were excluded from the study (N
=20 in the exposed and 698 in the comparison cohort).
We classified the births of women with HD into three groups: group 1 included the first birth after an HD diagnosis (that is, women who were diagnosed before pregnancy). Group 2 included the births by women diagnosed with HD during pregnancy (that is, diagnosed between the first day in the last menstruation until the date of birth). Group 3 included births by women who were diagnosed with HD after delivery (that is, diagnosed between the day after the delivery until 2 years later). If a woman gave birth more than once in this 2-year period, only the last birth before the HD diagnosis was included based on the assumption that the preclinical cancer would be more likely to affect the birth closest to the time of diagnosis.
For all three groups, we computed the difference between proportions of male newborns of mothers with HD and comparison mothers.
We computed prevalence odds ratios (PORs) as estimates of the relative risks with associated 95% confidence intervals (95% CIs) for preterm birth, low birth weight at term, stillbirth, and congenital abnormalities. The PORs were controlled for month and year of birth and county of mother's residence by matching. We used unconditional logistic regression analysis to further adjust for maternal age and parity. We also included the calendar period of the birth (1973–1986, 1987–1994, and 1995–2002), as an independent variable in the model. Although there was no change in the risk estimates when calendar period of birth was included in the model, we kept the variable in the model. Stillborn children were excluded from the analyses of preterm birth, low birth weight at term, and congenital abnormalities.
To examine whether sex of the child or maternal radiotherapy modified the POR estimates for births in group 1, we repeated the analyses in strata of boys and girls and strata of births of women who were treated with radiotherapy and women who were not. Furthermore, to examine whether calendar period of HD diagnosis modified the POR estimates for births in group 1, we repeated the analyses in different calendar periods of HD diagnosis (1981–1990 and 1991–2000), using 1970–1980 as reference. We used the Wald test to evaluate the homogeneity of the POR estimates for congenital abnormalities in 1981–1990 and 1991–2000. The low count of outcome events in groups 2 and 3 precluded stratified analyses.
We used linear regression to estimate differences in mean birth weight, while controlling for maternal age, parity, gestational age, and calendar period of birth. Stillborn children were excluded from these analyses.
The study was approved by the Danish Data Protection Agency (record no. 2003-41-2833). All analyses used SAS software, version 8.2. The SAS procedures used were PROC FREQ, PROC MEANS, PROC LOGISTIC, and PROC GLM.