Early-life cancer, infertility, and risk of adverse pregnancy outcomes: a registry linkage study in Massachusetts

Our study included 662,630 deliveries to Massachusetts (MA) resident women that took place between July 1, 2004 and December 31, 2013. Delivery information for women ≥18 years old from MA Registry of Vital Records and Statistics (RVRS) linked to hospital discharge records in the Pregnancy to Early Life Longitudinal (PELL) data system was additionally linked to ART cycles in the Society for Assisted Reproductive Technology Clinic Outcome Reporting System (SART CORS), and the MA Cancer Registry (MCR). The MCR was established in 1980 (MAs General Law, Chapter 111, Section 111B) and collects information on all cancer diagnoses for the state. The law requires hospitals, radiation centers, private physician offices, and other health care facilities to report newly diagnosed cancer or benign brain tumor cases diagnosed, evaluated, and/or treated at the facility/office to the MCR. The MCR has received a Gold Certification from the North American Association of Central Cancer Registries (NAACCR) every year from 1997–2016, indicating that they have achieved the highest standard for complete, accurate, and timely data collection. Additionally, the MCR has been recognized by Centers for Disease Control and Prevention (CDC) and the National Program for Cancer Registries (NPCR) for achievement of the NPCR standards for data completeness, timeliness, and quality data since 2002. Our analysis linked with a woman’s first reported occurrence of cancer and first utilization of cancer treatment type. Birth certificate data and hospital discharge data were obtained from the PELL data system. The PELL data system was originally developed as a collaborative effort between the MA Department of Public Health, the CDC, and Boston University School of Public Health and links birth certificates and fetal death records to their corresponding hospital discharges to create the core PELL. Then the core PELL is linked to (1) non-birth related hospital utilization overtime for both mother and child; (2) program participation data such as Early Intervention.; (3) linked infant death file; and (4) death certificates for all women to identify pregnancy-associated deaths. The PELL data system uses randomly generated unique identifiers for mothers and children to form a relational data system organized at the individual level. Over 98% of birth certificates have been linked to corresponding maternal or infant hospital discharge records. The result is the dyadic linkage of the birth and delivery records of the child and mother. The SART CORS database is used by SART to collect national ART data under the Fertility Clinic Success Rate and Certification Act of 1992 (Public Law 102–493) and to report these data to the Centers for Disease Control and Prevention (CDC). SART CORS data include patient demographic information, cycle-specific treatment data, and pregnancy outcome data. The data in SART CORS are validated annually with some clinics having on-site visits for chart review based on an algorithm for clinic selection. During each visit, data reported by the clinic are compared with information recorded in patients’ charts. In 2014, the 10 data fields selected for validation were found to have discrepancy rates of ≤6%[ 30 ]. The MA Outcome Study of Assisted Reproductive Technology (MOSART) is a study linking data from the SART CORS and PELL with the goal of evaluating pregnancy, child health, and women’s health outcomes on a population basis. Before performing the linkage, a Memorandum of Understanding was executed among SART, the Principal Investigators on the grant that funds the research, and the entities that participate in the PELL project. Human subjects approval was obtained from all entities and participating universities. The study also had the approval of the SART Research Committee. A detailed description of the linkage methodology between PELL and SART CORS has been described previously[ 31 ]. Briefly, pregnancies resulting in deliveries between July 1, 2004, and December 31, 2013, were linked to SART CORS deliveries using mother’s first and last name, father’s last name, mother’s date of birth, and date of delivery. When in doubt, baby’s sex, birthweight, and mother’s zip code were used to adjudicate potential links. Linkage proportions were 90.2% overall and 94.6% for deliveries in which both ART cycle patient zip code and treatment clinic were located in MA. The linkage yielded deliveries identified for this study as deliveries conceived by ART. The linkage identifies live births and fetal deaths at 20 weeks or greater or 350 grams or more, but could not identify pregnancies for which there was no delivery. The data were matched with MCR data using Match*Pro, a software developed by Information Management Services, Inc. (IMS) to conduct probabilistic record linkages. First, MCR data from 1995 to 2013 were linked to PELL women who delivered between 1998 and 2013, using first name, last name, date of birth, and zip code. Second, through PELL-MOSART linkage identifiers, the results were further linked to MOSART to generate the base MOSART-MCR file. Deliveries with first diagnosis of cancer prior to the date of conception were classified as deliveries with a history of cancers ( Figure 1 ). Outcomes of interest included maternal morbidity (gestational hypertension/ preeclampsia/eclampsia and gestational diabetes), delivery complications (primary cesarean delivery), and birth outcomes (preterm birth, low birthweight, small for gestational age, neonatal mortality, neonatal prolonged hospital stay, and conditions of the infectious disease, cardiovascular, respiratory, gastrointestinal/nutritional, neurologic, and hematologic systems). Hypertension during pregnancy and gestational diabetes were identified in PELL from either the birth certificate or the hospital discharge delivery record (ICD-9 codes of 642 for pregnancy-related hypertension; 648.8 for gestational diabetes). Chronic diabetes and hypertension were determined based on birth certificate and delivery hospitalization records. Women with chronic diabetes or hypertension were excluded from pregnancy-related diabetes or hypertension analyses. Length of gestation was based on clinical estimates of first-trimester ultrasound or last menstrual period and was derived from the birth certificate. Deliveries were classified as premature if length of gestation was <37 weeks. Birthweight was obtained from the birth certificates. Low birthweight was defined as <2500 grams. A standard deviation (SD) score for birthweights (birthweight Z-score) was calculated as the SD score of the value for each individual from the mean value of the MA reference population divided by the SD for that reference population [ 32 ]. The birthweight Z-score evaluates the adequacy of weight for gestational age using population-based standards as recommended by Land [ 33 ], and these are modeled as continuous and categorical variables. We generated gender-, race/ethnicity-, and gestation length-specific birthweight means and SDs using MA data for all live births from 1998 to 2013. Any live-born infant with Z-scores below the 10th percentile (% 1.28 SD) was considered small for gestational age. Information on neonatal mortality was obtained from the linked infant birth and death data and prolonged neonatal hospital stay was determined from delivery birth records and infant hospital discharge data. Prolonged stay for infants was defined as >3 days for vaginal delivery or >5 days for cesarean section and was limited to deliveries in which gestational age was ≥ 35 weeks, and with known data on mode of delivery. Neonatal death was defined as mortality occurring 0 to 27 days after birth to live born infants. As described in more detail in prior analyses [ 28 ], conditions grouped by system were also analyzed based on ICD-9 codes: infectious disease, cardiovascular, respiratory, gastrointestinal and/or nutritional, neurologic, and hematologic. These analyses were restricted to live born infants. The analysis of prolonged hospital stay and small for gestational age were restricted to livebirths. The main exposure of interest was history of cancer compared to women without a history of cancer. Maternal demographic characteristics of age, race and ethnicity, gravidity, parity and education were obtained from the birth certificate in PELL. Preexisting maternal conditions (diabetes mellitus and chronic hypertension) were determined from either the birth certificates or hospital discharges (ICD-9 codes of 648.0 or 250 for diabetes mellitus; 401, 402, 403, 404, or 405 for chronic hypertension). Deliveries were defined as having used ART if the index delivery was linked to SART CORS. In our analysis, deliveries were considered as “subfertile” as described previously [ 34 ] if i) the birth certificate indicated fertility treatment but there was no linkage to SART CORS, or ii) they had indication of hospitalization before delivery with ICD code for infertility (ICD9 628.9; V23), or iii) they had a prior delivery within the past 5 years with SART CORS linkage or notation on the birth certificate indicating fertility treatment was used. We defined pregnancies as having experienced “Subfertility/fertility treatment” if they either fit the definition of subfertile (above) or had undergone ART treatment to conceive the pregnancy that resulted in that delivery. We modeled the risks for adverse pregnancy outcomes using log-Poisson regression to yield relative risks and 95% confidence intervals. Generalized estimating equations were used to take into account multiple pregnancies per woman. Models were adjusted a priori for potential confounding variables, maternal age (18–29, 30–34, 35–37, ≥ 38), race (Non-Hispanic white, other groups), education (≤ high school, some college, ≥college, missing), plurality (singleton, multiple) and birth year (2004, 2005, 2006, 2007, 2008, 2009, 2010, 2011, 2012, 2013). Analyses of neonatal mortality and prolonged neonatal hospital stay were additionally adjusted for gestational age (continuous variable 17–44 weeks). Sensitivity analyses of low birthweight were restricted to full-term deliveries. Given the increased risk associated with infertility/fertility treatment and adverse pregnancy outcomes, analyses were stratified by history of subfertility/fertility treatment. We tested for differences in the association between cancer history and adverse pregnancy outcomes between women who experienced subfertility/fertility treatment and women who were fertile by comparing a model with the interaction term to a model without the interaction term using the Likelihood Ratio Test. In supplemental descriptive analyses, among women with a history of cancer, we stratified outcomes by history of the five most common cancer types (breast, leukemia/lymphoma, thyroid, melanoma, and cancers of the female genital organs).

Of the 662,630 deliveries, 2,983 (0.5%) had a history of early-life cancer ( Table 1 ). Deliveries with a history of cancer had an older maternal age delivery (mean:33.6 vs. 29.9 years). Deliveries with a history of cancer were more likely to be non-Hispanic white women (80.2% vs. 65.8%), be a college graduate (53.9% vs. 43.3%), have private insurance (69.8% vs. 55.8%), and to have utilized ART for the index pregnancy (7.4% vs. 3.2%), or to have experienced any subfertility (12.2% vs. 5.3%) ( Table 1 ). The majority of malignancies were diagnosed after age 20 (88.6%). The most common cancer types were breast (10.4%), lymphoma (9.2%), thyroid (23.4%), melanoma (18.8%), and those involving female genital organs (uterus, cervix, ovaries, vagina, vulva) (20.8%). The majority of cancers were initially treated with surgery (84.4%), with 20.9% of tumors being initially treated with radiation and 17.5% being initially treated with chemotherapy ( Table 2 ). Women with a history of cancer contributed a mean (SD) of 1.27 (0.51) deliveries to this analysis where women without a history of cancer contributed on average 1.39 (0.61) deliveries. Among maternal morbidities, women with a history of cancer were at greater risk for gestational diabetes (risk ratio (RR):1.29 95% confidence interval (CI):1.13– 1.48) and cesarean section (RR:1.17 95% CI:1.11–1.23) compared to those never diagnosed with cancer in crude models ( Table 3 ). These associations attenuated after adjustment for maternal age, race, education, plurality, and birth year (RR GDM:1.08 95% CI:0.94–1.23; RR Cesarean:1.05 95% CI:1.00–1.11), with maternal age as the primary covariate driving the attenuation. We saw no association between overall cancer history and hypertensive disorders of pregnancy in either crude or multivariable adjusted models (RR:1.05 95% CI:0.92–1.19). For birth outcomes, infants born to women with a history of cancer did not have a higher risk for SGA compared to women never diagnosed with cancer ( Table 3 ). However, the association between cancer history and SGA differed by history of subfertility/fertility treatment (P-value, test for heterogeneity: 0.02) ( Table 4 ). Specifically, among pregnancies with subfertility/fertility treatment, births to women with a history of cancer were associated with a higher risk of SGA (Absolute risk: 13.3%, RR:1.36 95% CI:1.02–1.83). However, among fertile pregnancies, there was no association between cancer history and SGA (RR:0.95 95% CI:0.82–1.10). Infants born to women with a history of cancer were at higher risk of LBW (RR:1.19 95% CI:1.07–1.32) compared with those born to women never diagnosed with cancer ( Table 3 ). In sensitivity analyses restricted to full-term deliveries, no association between cancer history and LBW was seen (RR:1.05 95% CI:0.83–1.33) (Data not shown in table). Women with a history of cancer were at greater risk of preterm birth (RR:1.19 95% CI:1.07–1.32) and prolonged neonatal hospital stay (RR: 1.16 95% CI: 1.01–1.34) compared to women who had never been diagnosed with cancer ( Table 3 ). We observed an association between livebirths to women with a history of cancer and risk of adverse gastrointestinal and nutritional conditions (RR: 1.17 95% CI:1.02–1.35). We found no association between history of cancer and conditions of the cardiovascular, respiratory, neurologic, or hematologic systems, or infectious diseases. In supplemental, descriptive analyses stratified by cancer type, we saw that women with a history of cancer of the female genital organs had high incidence of gestational diabetes (9.4% vs. 5.9% among women with no history cancer), low birth weight (14.7% vs. 7.9% among women with no history cancer), and preterm birth (17% vs. 8.5% among women with no history of cancer). We found that women with a history of breast cancer had a high incidence of birth by cesarean section (48.4% vs. 32.4% among women with no history of cancer) and low birth weight births (13.9% vs. 7.9% among women with no history cancer).

Survival among children, adolescents, and young adults diagnosed with cancer has improved dramatically in the last 40 years, leading to increasing interest in quality of life and long-term health among survivors [ 1 , 2 ]. Specifically, there is substantial concern among survivors regarding the influence of their diagnosis and treatment on their future fertility and pregnancy [ 3 – 6 ]. Previous research has shown that early-life cancer survivors have nearly 40% fewer offspring than their peers with no history of cancer [ 7 ] and that those who successfully conceive, may be at increased risk for adverse pregnancy outcomes, such as preterm birth, small for gestational age, and other pregnancy and neonatal complications [ 8 , 7 , 9 ]. Cancer history and treatment regimens have the potential to influence female reproductive health through a variety of mechanisms [ 10 – 13 ]. Radiation to the abdomen and pelvis is hypothesized to harm the vasculature, plasticity of the uterus, and the ovaries [ 14 ], whereas total body radiation has been associated with decreased uterine volume [ 15 , 16 ]. Additionally, cranial radiation may have long-term effects on hypothalamic pituitary function [ 17 ]. Surgical treatment may involve removal of the fallopian tubes, ovaries, and uterus [ 11 ]. Moreover, cytotoxic chemotherapy regimens, most notably high-dose alkylating agents, may influence ovarian failure and germ-cell toxicity [ 8 ]. Among early-life female cancer survivors who achieve pregnancy, the literature has consistently shown a higher risk of preterm birth [ 18 – 20 , 10 , 21 , 8 ]. However, the relationship between history of cancer and other adverse pregnancy outcomes for the mother and baby, including preeclampsia [ 22 , 21 , 10 , 9 ], gestational diabetes [ 10 , 9 , 22 ], low birthweight [ 23 , 20 , 18 ], and neonatal mortality [ 24 , 21 ], have been mixed, with some studies suggesting an increased risk and some studies suggesting no risk among cancer survivors. Compared to women with no cancer history, those who are early-life cancer survivors are also more likely to experience infertility and ultimately utilize fertility treatment [ 7 , 25 , 21 ]. Independent of cancer history, women with subfertility and pregnancies conceived utilizing fertility treatment may also be at greater risk for some of the same adverse pregnancy outcomes, including preterm birth, low birthweight, and small for gestational age [ 26 – 29 ]. However, there is limited information on the role that subfertility or fertility treatment play on the adverse pregnancy outcomes that some cancer survivors experience. Thus, our goal was to 1) investigate the relationship between history of cancer and adverse pregnancy outcomes and 2) to determine whether subfertility, including receipt of fertility treatment such as assisted reproductive technology (ART), modified that relationship.

Overall, we observed that deliveries to women with a history of cancer were at higher risk of low birthweight, prematurity, prolonged neonatal hospital stay, and gastrointestinal/nutritional conditions compared to deliveries to women who had never been diagnosed with cancer. For some outcomes, such as small for gestational age, low birthweight, and preterm birth, the relationships varied by history of subfertility/fertility treatment in the index pregnancy. Given the baseline increased risk of adverse pregnancy outcomes among women with a history of subfertility/fertility treatment, this heterogeneity by subfertility/fertility treatment history may help explain some of the seemingly inconsistent results in the prior literature. Reports of adverse maternal outcomes among cancer survivors have been mixed in previous studies, with some studies reporting a higher risk among specific groups [ 10 , 9 ]. In our analysis, adjustment for maternal age attenuated the observed increased risk of gestational diabetes in crude models, which supports the findings of Mueller et. al [ 22 ]. Cancer history may contribute to delayed attempts to conceive or longer time to pregnancy and thus result in older age at pregnancy among cancer survivors. Maternal age may be on the causal pathway between cancer experience and GDM and thus our findings that include adjustment for maternal age must be interpreted with this in mind. While some research has suggested a higher risk of hypertensive disorders of pregnancy among female cancer survivors [ 10 ] or Wilms tumor survivors [ 9 ], our findings support the overall conclusions from two prior studies [ 21 , 22 ], which found no overall association between cancer survival and hypertensive disorders of pregnancy or preeclampsia. The relationship between history of cancer and cesarean section has also yielded mixed results [ 9 , 10 ]; our findings, while modest (RR:1.05), did reach the threshold of statistical significance even after multivariable adjustment and support previous findings of similar magnitude [ 18 ] but ultimately suggest no meaningful clinical association between history of overall cancer and cesarean delivery [ 18 , 21 ]. However, women with a history of breast cancer appeared to be at meaningfully elevated risk of cesarean section (48.4%) compared to women without a history of cancer (32.4%). Among all women, we found no association between history of cancer and small for gestational age. However, this relationship was influenced by history of subfertility/fertility treatment for the index pregnancy. Among subfertile women, infants born to women with a history of cancer had a 30–40% greater risk of SGA compared to those born to women who had never been diagnosed with cancer – suggesting a synergy between cancer history and history of subfertility/fertility treatment. Prior research among cancer survivors has reported conflicting results for SGA, with some studies suggesting a lower risk [ 20 , 22 ] and others suggesting an elevated risk among cancer treatment groups [ 8 ]. History of subfertility/fertility treatment may represent a high-risk group for SGA as ART has been shown to be associated with higher risk for SGA in some studies [ 35 , 36 ]. Indeed, prior research from MOSART found that certain infertility diagnoses among those that utilized ART were at greater risk of SGA [ 34 ]. We also found that women with a history of cancer had a higher risk of low birthweight. However, there was no longer a statistically significant difference in sensitivity analyses restricting our population to full-term deliveries – supporting that there is a strong relationship between maternal early-life cancer and gestational age at delivery. Indeed, we found a 25% greater risk of preterm birth among early-life cancer survivors, which is consistent with prior research in the field [ 37 , 19 , 20 , 38 , 18 , 22 , 21 , 8 , 7 , 39 , 10 ], indicating that pregnancies to early-life cancer survivors may have a greater risk of preterm birth. Our descriptive sensitivity analyses also suggested that the risk of preterm delivery may vary by cancer type, as women with a history of cancer of the female genital organs and breast cancer had the highest risk of preterm birth compared to both other cancer types and also to women without a history of cancer. The relation between early-life cancer and neonatal mortality has been challenging to study given the rarity of the outcome and heterogeneity in definitions of mortality across studies [ 40 , 41 ]. Our data suggested an increased risk of neonatal mortality, although this association varied models of covariate adjustment and heterogeneity by history of subfertility could not be explored given the small sample size. Prior research among cancer survivors has suggested that there is indeed an increased risk of neonatal death in this population [ 40 , 38 , 21 , 24 , 42 ], however, differences have been reported across tumor sites and treatments, as well as differences in outcome definitions. Future research should focus on this potential for elevated risk and the underlying pathophysiology. Prolonged neonatal hospital stay may be a marker of adverse delivery complications. Our analyses found a modest (16%) risk of prolonged neonatal hospital stay among pregnancies in women with a history of cancer, even after adjustment for gestational age. Information on Apgar scores has not been routinely collected and we were not able to evaluate it in this analysis, but findings from Western Australia indicated an increased risk of neonatal distress (Apgar score <7), need for resuscitation, and need for special care nursery admission for deliveries to early-life cancer survivors [ 10 ]. When we looked specifically at neonatal disease conditions, we found that women with a history of cancer had infants with a higher risk of adverse gastrointestinal and nutritional conditions. Prior research in our cohort has suggested that live births among women who experienced any subfertility may also have infants with a greater risk for these disease conditions [ 28 ]. However, the biologic mechanisms that underlie this association are not known and requires further investigation. This study has many strengths including its large sample size, detailed hospitalization information and utilization of vital records, however, we must also recognize its limitations. As is true of all state-level linkage studies, information will be missing if women received cancer treatment, delivered, or received fertility treatment outside the state of MA. Additionally, given the nature of the data linkage performed, information will also be missing if women experienced cancer before 1995. We expect the number of women who delivered out of state and who access fertility treatment out of state to be minimal and that cancer survivors who moved out of state prior to delivery would not be different with respect to adverse pregnancy outcomes than the cancer survivors who stayed in MA. Additionally, we expect that the number of women with a history of cancer before 1995 will be small relative to the population of women without cancer. Thus, these limitations would likely attenuate our reported relationships. However, our findings may not be fully generalizable to the experience of all women with childhood cancer. Our analysis assumes that history of cancer proceeds infertility and fertility treatment utilization but there may be some women who experience infertility prior to cancer and index delivery. Future research should try to disentangle the temporality of infertility experience and cancer diagnosis, as infertility has also been shown to be associated with subsequent cancer risk [ 43 ]. Our data are based on vital records and hospital discharge information, thus information on some potentially important covariates (i.e. body mass index, details of medically assisted reproduction that is not IVF, fetal growth) are not available. These covariates are hypothesized to be either potential confounders [ 44 ] or potential mediators [ 45 ] in the relation between early-life cancer and adverse pregnancy outcomes and thus not taking this information into account is a limitation of this data source. Additionally, our linkage with the cancer registry is limited to a woman’s first reported occurrence of cancer and the first treatment utilized. Information on subsequent cancer treatments and cancer stage were not able to be evaluated in this analysis but should be investigated in future research as they may have important implications for pregnancy outcomes. Lastly, we were not able to link information on the cancer history of the male partner for pregnancies and this information may be influential and may be with the likelihood of having a female partner with or without a history of cancer. Overall, our findings support prior work indicating that early-life women cancer survivors are at increased risk for some adverse pregnancy outcomes, specifically preterm birth and low birthweight, and these risks may vary by whether these women experienced subfertility/fertility treatment, a high-risk group for many adverse outcomes. Thus, future research should continue to investigate this heterogeneity driven by subfertility and fertility treatment, as such research may lead to targeted screening or treatment recommendations during pregnancy for cancer survivors. Additionally, future research should more fully investigate differences by cancer type and the time interval between cancer diagnosis and pregnancy. Moreover, our research indicates that providers should take a comprehensive history of both cancer and subfertility experience as this information may be important when counseling cancer survivors.