Retinoblastoma is the most common intraocular malignancy in childhood, with an incidence of 11.8 per million children aged 0–4 years in the United States [1]. It results from inactivation of both alleles of the tumor suppressor RB1 gene. When the disease is familial (6–10 % of cases [2]) children inherit one RB1 gene mutation from a parent, and the other allele is lost somatically. The remaining cases of disease are considered sporadic. In 30 % of cases, the RB1 mutation in one allele occurs as a de novo mutation in parental germline cells (usually of paternal origin [3]) or happens in very early embryonic development, and the other allele is lost somatically sometime after conception. In both of the above instances, retinoblastoma typically presents bilaterally. In the remaining 60 % of cases, the disease results from two somatic alterations in a retinal cell at some point after conception; these cases present as unilateral disease.

Although little is known about the causes of sporadic retinoblastoma, there have been several studies that suggested a role for reproductive health factors. A meta-analysis of four studies reported an increased risk for retinoblastoma among children conceived via assisted reproductive technologies [4]; however, a recent study did not support this association [5]. Some, but not all, epidemiologic and laboratory studies have suggested a role for human papillomavirus (HPV) in retinoblastoma [6–15]. A study in Sweden found high but imprecisely estimated increased risks of retinoblastoma with longer breastfeeding, although a dose–response effect was not evident [16].

Parental prescription drug use in pregnancy has been of particular interest in cancer research ever since Herbst and Ulfelder reported an association between maternal use of diethylstilbestrol (DES) in pregnancy and clear cell adenocarcinoma of the vagina in young women [17]. A number of childhood cancer studies have examined the risk from maternal intake of medication during pregnancy [18–20], however studies of retinoblastoma are limited. While one study observed the prevalence of morning sickness to be the same among retinoblastoma cases and controls [21], the use of antinauseants was much higher among case mothers in a different study [7].

The goal of the present study was to examine maternal comorbidities, medication use during pregnancy, and reproductive health related factors in relation to retinoblastoma in offspring.

We conducted a multi-center case–control study of retinoblastoma across the USA and Canada from June 2006 to June 2011 [22]. Children with sporadic retinoblastoma were identified from Wills Eye Hospital in Philadelphia, or by the Children’s Oncology Group (COG), which includes over 200 institutions in the USA and Canada. Each participating COG institution, Wills Eye Institute, the University of Pennsylvania, and the University of California, Los Angeles approved the study. After a physician gave his or her approval to contact a patient, children were eligible for the present study if at least one biological parent consented to participate, if they resided in the continental U.S., Alaska, or Canada, if they had a telephone in their household, and if at least one parent spoke English or Spanish. Children conceived with a donor egg or sperm were eligible for the study. Participants provided written informed consent for biospecimen collection; because interviews took place via telephone, we collected verbal informed consent for the interview, which was documented on the questionnaire. In total, 282 cases of retinoblastoma (187 unilateral and 95 bilateral cases) were enrolled in the study.

The family of each case was asked to nominate one or more children of their friends or non-blood relatives under age 15 as potential controls. For bilateral cases, eligible control fathers were not a biological relative of the case’s father; for unilateral cases, eligible control mothers were not a biological relative of the case’s mother. We attempted to match controls to cases on the child’s age at the time of interview (0–1, 2–4, 5–9, 10–14 years old). If more than one control was nominated by the case parents, we attempted to recruit the eligible control closest in age to the case. If that control was not successfully recruited, we attempted to recruit the next control until we obtained a control or until we contacted all potential controls. When an “ideal control” (age-matched and not a biological relative) was unavailable, cases were asked to suggest another potential child as a control. Some families did not nominate any child as a control or the control’s family did not consent to enroll in the study, therefore for some cases there was no matched control. Overall, 155 controls were recruited.

The majority (>90 %) of sporadic bilateral cases are due to a de novo mutation in the father’s germline [3]. Therefore, for bilateral cases, we were most interested in paternal preconceptional exposures. In contrast, because unilateral cases derive from two RB1 mutations which occur during pregnancy, for unilateral cases we were most interested in examining pregnancy exposures. In a structured telephone interview with the parents, data were collected on demographic information, the mother’s medical conditions in pregnancy, her reproductive history, and other exposures. With regards to perinatal health conditions, mothers were asked several open-ended questions to prompt their memories about pregnancy-related and unrelated medical conditions which occurred in the month prior or during pregnancy, as well as conditions that had been diagnosed before the pregnancy but for which they had received treatment during the pregnancy. With regards to medications, mothers were asked “Did you…take prescription medicine for any condition, such as the flu, an infection, accident or injury, in the month before or during your pregnancy?” Over-the-counter medications were not ascertained. In total, 280 mothers of cases (185 unilateral and 95 bilateral) and 146 mothers of controls completed the interview. When one parent was unavailable, the interview was conducted with a proxy who was typically the other parent, with 16.5 % of paternal and 3 % of maternal interviews conducted by proxy.

We used unconditional logistic regression to evaluate the risk of retinoblastoma. Given that a number of cases had no matched control, we chose to use unconditional logistic regression in order to improve statistical power. We reported odds ratios (ORs) and 95 % confidence intervals (CIs) adjusted for mother’s race/ethnicity (White non-Hispanic, Black non-Hispanic, Hispanic, other), maternal educational attainment (Less than high school, high school graduate, some college or other training, college graduate or more), household income (less than $35,000, $35,000 to 50,000, $50,000 to 75,000, more than $75,000), the mother’s age at child’s birth (continuous), and a behavioral indicator, the mother’s tobacco smoking in the month before or during pregnancy (Yes/No). We explored additional adjustment for marital status and child’s gender however they did not change the estimates by more than 10 %, and were not included in the final model. When statistical power allowed us to do so, we checked our results in the matched analysis using conditional logistic regression, adjusting for the same covariates other than child’s age (matching variable).

Many health conditions were identified by only a small number of mothers. We provide results for which there were at least 5 unilateral cases that reported having the medical condition; in addition, because of the prior reported associations between retinoblastoma and infertility treatment [4] as well as sexually transmitted diseases [6], we reported associations with any sexually transmitted disease and with the type of fertility treatment. The category “other viral infections” included hepatitis B and C, shingles, HPV, herpes, stomach virus, Murray infection, and Fifth disease. We defined hormonal birth control methods as oral contraceptive pills, injection, implant, skin patch, or vaginal ring. The category “other types of birth control” included diaphragm, cervical cap, sponge, IUD, Lea’s shield, other barrier method, vasectomy, tubal ligation, rhythm method, fertility awareness, and withdrawal.

We examined maternal and paternal weight prior to pregnancy and pregnancy weight gain. Based upon recommendations issued by the Institute of Medicine [23], we defined normal weight gain in pregnancy as 28–40 pounds for underweight women [body mass index (BMI) < 18.5], 25–35 pounds for normal women (BMI = 18.5–25), 15–25 pounds for overweight women (BMI = 25–30) and 11–20 pounds for obese women (BMI > 30).

In models which evaluated retinoblastoma’s association with birth control use, we adjusted for the same variables above except the mother’s age at child’s birth, because it did not change the point estimate by more than 10 %. For analyses of fertility treatment, parity and breast feeding, we utilized the same covariates in models except for mother’s age at child’s birth and mother’s tobacco smoking, because they did not change point estimates by more than 10 %.

Because information for some covariates was missing (primarily with regards to family income and maternal smoking status during pregnancy) we conducted sensitivity analyses in which we used multiple imputation methods (“PROC MI and PROC MIANALYZE”) in SAS 9.2 to compensate for missing data. In addition, due to differences between case and control groups, we then analyzed our imputed dataset using propensity score techniques where scores for all exposures were calculated from a logistic regression model with each exposure from tables 2, 3 and 4 set as the dependent variable and all covariates set as the independent variable. Results from the multiple imputation/propensity score analyses did not differ substantially from the main results, with most point estimates and confidence intervals changing by <0.1–0.2. Thus, we present results from logistic regression in main tables; multiple imputation/propensity score analyses are presented in Additional file 1.

We additionally conducted sensitivity analyses to examine whether results changed when we excluded the 3 % of maternal interviews that were conducted by proxy.

In sensitivity analyses which excluded proxy interviews, we observed similar results to those seen in the overall study.

This study is one of only a small number of investigations into maternal health and reproductive health related factors in relation to retinoblastoma. We observed a negative association for retinoblastoma with condom use in the year before pregnancy and an increased risk with maternal underweight at the start of pregnancy and when the mother took pain medication during pregnancy. For many self-reported health conditions, there were only a small number of mothers who indicated that they had had the condition in the perinatal period, thus we were underpowered in many analyses. Hence, results must be interpreted with caution. Although the small sample size led to wide confidence intervals, we observed trends towards increased risks with several maternal conditions during pregnancy, including viral infections, diabetes, depression or anxiety, and anemia. Other studies examining these conditions have found conflicting results [6, 7]. These findings require replication elsewhere, preferably in studies which utilize medical record review.

Maternal underweight prior to pregnancy is a well-recognized risk factor for a variety of adverse fetal outcomes such as intrauterine growth restriction [24], however there are only a few studies which examined it in relation to childhood cancers. Maternal underweight (BMI <20) was associated with a 40 % increase in hepatoblastoma risk in one population-based study [25]. Maternal undernutrition can alter fetal programming and potentially change specific cell groups or organogenesis including the developing retina [26]. Both animal and human studies have shown that fetal growth restriction can cause changes in retinal structure [27, 28]. A role for specific nutrients in retinoblastoma prevention has been suggested, particularly for folate, an important methyl donor [21, 29].

Our findings on condom use reinforce the observation of an earlier study which reported that the use of barrier conception in the year prior to pregnancy was associated with lower risk of retinoblastoma [7]. Given that the use of condoms not only can prevent incident HPV infections but also may promote clearance of HPV in women with a previously positive HPV test [30, 31], our results support a possible role for HPV in retinoblastoma development. Vertical transmission of HPV would presumably occur during passage of the fetus through the birth canal [32]. In a previous study, we observed a weak lowered risk of unilateral retinoblastoma with Cesarean Section (OR=0.8) [6]. In some but not all studies, HPV DNA has been isolated from retinoblastoma tumor tissue [8–15]. Contradictory findings across studies may be due to variation in the underlying prevalence of HPV across regions, cofactors (dietary or behavioral) that may affect HPV behavior, or specimen contamination. The biological plausibility of this association is supported by the well-known binding of HPV protein E7 to the tumor suppressor protein pRb, rendering it inactive [33].

In the United States, 12 % of women ages 15–44 receive infertility services at some point in their lives, and 1 % of children born each year are conceived through assisted reproductive technologies [34]. Although only a small number of our study participants reported having treatment for infertility, we observed a trend towards increased risk for retinoblastoma with infertility treatment. In our California studies, we observed increased cancer risk among the offspring of multiple births, particularly among embryonal tumors; this may support associations with infertility treatment [6, 35–37]. Although the rarity of childhood cancer has limited research in this area, studies suggest an increased risk for childhood cancers among children conceived after infertility treatments, including a meta-analysis of four studies which observed increases in risk for retinoblastoma (RR = 1.62) [4, 38]. Possible mechanisms that may explain this increased cancer risk include epigenetic changes such as altered DNA methylation and changes in chromatin structure, which may cause imprinting disorders and modified gene expression [39]. However, there is debate as to whether the source of any increased risk for childhood cancer may be the infertility treatment itself or rather due to factors related to the underlying fertility problems.

While breastfeeding confers a number of health benefits to infants, it may also expose the newborn to elevated levels of varying exogenous chemicals that can be found in milk, such as pharmaceuticals taken by the mother, heavy metals, or volatile organic compounds [40]. Our finding of a negative association with breastfeeding was not confirmed in dose–response analyses, which examined duration of breastfeeding. Because we did not collect the reason that parents ceased breastfeeding, it is not known if case parents’ weaning was a result of the retinoblastoma diagnosis.

We additionally observed an increased risk for retinoblastoma with maternal intake of prescription pain medication during pregnancy. The majority of mothers who had taken medication reported intake of acetaminophen (paracetamol). However, misclassification of exposure is a concern given that there were likely a considerable number of mothers in our study who took over-the-counter acetaminophen during their pregnancies. In a pooled analysis of two large studies, 56 % of US women reported taking acetaminophen while pregnant [41]. Despite that acetaminophen is a commonly recommended pain medication for use in pregnancy, there are reports of increased risk of adverse childhood health conditions such as preterm birth, asthma, and hyperkinetic disorders with acetaminophen use [42–44].

A limitation of our study is the possibility of recall errors. In retrospective studies in which mothers are queried about pregnancy health, mothers do best at remembering serious health conditions or major events. For example, hypertension, diabetes, and previous stillbirths tend to be recalled quite accurately, while recall of anemia is more moderate [45, 46]. Mothers tend to remember medications taken for chronic conditions more accurately than those taken for only a short time period [47], and accuracy in recall of short-term medication may differ between cases and controls [48]. However, maternal health conditions were reported in our study at a similar prevalence to rates observed in recent large-scale epidemiologic studies in the US and elsewhere. Pregnancy prevalence of chronic hypertension is 0.5–3 % and gestational hypertension, 4–14 % [49, 50]; chronic diabetes is 1–2 % and gestational diabetes is 8 % [51]; asthma is seen in 4–9 % of US pregnant women [52] and hypothyroidism in 2–3 % [53]. In contrast, the prevalence of depression was much lower in our study than has been seen in other US surveys (10–12 % [54, 55]), potentially due to parental reluctance to disclose mental health problems to interviewers.

Another limitation in this study is the possibility of selection bias in the form of overmatching. Due to the use of friend controls, it is likely that maternal characteristics of cases and controls would be similar. However, differing response rates resulted in controls who tended to be White, more educated, and of higher SES. In matched/conditional analyses, cases without matched controls would drop out of the analysis, while unconditional analyses would need to be controlled for these factors to mitigate the effect of bias due to overmatching. In our analyses, we attempted to control for race, education, and SES, though it may not have completely controlled for all bias due to selection. As a consequence, the forced similarities between cases and controls are likely to bias results to the null.