Assessment of Birth Defects and Cancer Risk in Children Conceived via In Vitro Fertilization in the US

Key Points Question Is the incidence of birth defects and childhood cancer among children conceived via in vitro fertilization different from that among children conceived naturally? Findings In this population-based cohort study of 1 053 415 children in 4 states, the presence of a birth defect and the number of birth defects were associated with an increased risk of childhood cancer. The increased risk was 2-fold higher for children conceived via in vitro fertilization than for children conceived naturally. Meaning In this study, cancer risk increased in the presence of birth defects at a higher rate in children conceived via in vitro fertilization than in children conceived naturally; further study is warranted.


Introduction
Births conceived with in vitro fertilization (IVF), the ex vivo manipulation of both sexes' gametes to achieve conception, accounted for 2% of all US births in 2017. 1,2 Children born from IVF have been shown to be at greater risk for birth defects 3 and for childhood cancer 4,5 compared with children conceived their naturally. A growing body of literature reports an association between birth defects and the development of cancer, but this association has yet to be evaluated among children conceived via IVF. 6 In this study we present the results of a population-based linkage of IVF births, birth defects, and childhood cancer in 4 US states to estimate the risk of cancer among children with birth defects conceived through IVF vs those conceived naturally.

Methods
This study linked 2004-2016 birth certificates data to birth defects registries, cancer registries, and the national IVF database, the Society for Assisted Reproductive Technology Clinical Outcomes Reporting System (SART CORS), in 4 US states (New York, Texas, Massachusetts, and North Carolina). Data from birth certificates (2004)(2005)(2006)(2007)(2008)(2009)(2010)(2011)(2012)(2013) linked to the SART CORS and the cancer registries were collected in a study of the risk of childhood cancer and IVF. 5 The remaining data (linkages to the birth defects registries, linking or relinking to the cancer registries, and linking to death records) were obtained in the current study of the risk of birth defects in assisted reproductive technology. New York, Texas, Massachusetts, and North Carolina were chosen for the current study because they are large and ethnically diverse, with birth defect registries using the same case definitions and data collected. These 4 states ranked second, third, sixth, and twelfth in highest number of annual IVF births in the United States, respectively, in 2016, with IVF births accounting for 3.0%, 1.5%, 4.7%, and 1.4% of all births in each state. 1,7 This study was approved by the institutional review boards at Michigan State University with a waiver of informed consent as research not involving human participants, as well as the institutional review boards at the University of Michigan and the 4 study state departments of health. This study followed Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) reporting guideline for cohort studies. 9

SART CORS Data
The SART CORS database contains comprehensive information on IVF procedures from more than 83% of all clinics providing IVF and more than 92% of all IVF cycles in the US. Data were collected and verified by SART and reported to the Centers for Disease Control and Prevention (CDC) in compliance with the Fertility Clinic Success Rate and Certification Act of 1992 (Public Law 102-493). 2 The Society makes data available for research purposes to entities that have agreed to comply with SART research guidelines. Patients undergoing IVF at SART member clinics sign clinical consent forms that include permission to use their data for research with appropriate provisions for safeguarding confidentiality. Data are submitted by individual clinics and verified by the medical director of each clinic. Approximately 10% of the clinics are audited each year to validate the accuracy of reported data. 2 During each audit visit, data reported by the clinic are compared with information recorded in the medical record; most data fields have discrepancy rates less than 2%. IVF cycle data reported to the SART CORS is available within 24 months after the close of the cycle year. Ninety percent of SART CORS cycles resulting in a live birth to residents of the 4 study states during the study period linked to their respective birth certificates; all other live births were classified as fertile (non-IVF).

Cancer Data
Each of the 4 study states maintains a high-quality cancer registry with data going back at least as far North Carolina) or 2017 (New York). All of the cancer registries participating in this study follow the same standards in terms of 90% or greater ascertainment by 12 months and 95% or greater by 24 months of the close of the diagnosis year, which is when most registries construct analytic files and publish incidence rates. All study children were also linked to state death records.

Birth Defects
The 4 states participating in this project are current or former CDC Centers for Birth Defects Research and Prevention. Thus, they conduct enhanced birth defects surveillance in terms of scope and quality of data. Each state conducts active or a combination of active and passive population-based surveillance that includes the major birth defects. These states use standard case definitions as defined by the National Birth Defects Prevention Study and National Birth Defects Prevention Network (NBDPN) and code birth defects using the CDC coding system adapted from British Pediatric Association codes, which is more specific for birth defects than International Classification

of Diseases, Ninth Revision (ICD-9) or International Statistical Classification of Diseases and Related
Health Problems, Tenth Revision (ICD-10) coding (eTable in the Supplement). 6 They use multiple quality assurance procedures including validity checks, double-checking of assigned codes, clinical review of at least a subset of cases, and comparison or verification between multiple data sources.
Birth defect data are available within 24 months after the close of the birth year. For this study, we analyzed major birth defects diagnosed within the first year of life, as defined in eTable in the Supplement, and subdivided children with major birth defects into any chromosomal birth defect (with or without another birth defect) or nonchromosomal birth defect (without any chromosomal birth defect).

Race and Ethnicity
Maternal race and ethnicity were obtained from the birth certificate; maternal race and ethnicity were also the assigned race of the infant, a rule that was initiated in 1989 by the National Center for Health Statistics. Classification of race and ethnicity was either self-reported by the mother after delivery or by the birth registrar in the birthing facility and reported to the state vital records, as per the local and state policy. Race and ethnicity were included as a factor in this study because of known associations with birth defects and cancer.

Linking Procedure
The SART CORS database, linkage of data from the SART CORS database to the birth certificate, cancer registries, and death certificates, and transfer of the deidentified data to the investigators has been described previously. 5 Briefly, IVF cycles to residents of the study states resulting in a live birth during the study period were linked to their birth certificates, and non-IVF births were selected in a 10:1 ratio and within the same 4-to 6-week period as the IVF births. For this study additionally the IVF-birth certificate data and the fertile-birth certificate data was linked to each state cancer registry and birth defects registry through 1 year of age; study children were relinked to their state's cancer registry in later study years to identify cancers diagnosed after infancy. All data were deidentified before transfer to the investigators.

Data
Birth records with gestational age less than 22 weeks or birth weight less than 300 g were excluded because such births are considered nonviable. Because IVF is rare for a mother younger than 18 years of age, we did not request to include parents aged less than 18 years in the study; therefore, those with ages less than 18 years were excluded; in addition, mothers whose ages were not specified were also excluded. Among the IVF births, only those conceived with autologous oocytes (the mother's own oocytes, not donor oocytes) and fresh embryos (not thawed) were included because these IVF conditions are most parallel to the conditions of the fertile births. The few infants where sex was not specified were also excluded. Births were categorized as fertile and IVF. The number of defects in a child was classified as none, 1, 2, or 3 or more. The analyses in this study are limited to singleton births because there were fewer twin births and higher multiple births and, as a result, too few infants with major malformations to provide stable estimates of cancer incidence.

Statistical Analysis
Cox proportional hazards regression models were used to generate hazard ratios (HRs) and 95% CIs for childhood cancer risk separately in IVF and fertile groups. We also fit models to all participants defects. There are 2 models with 6 covariates each. All but 1 test was not significant; 1 for sex of the child had a P = .04. The probability of 1 or more tests will be significant when 12 random tests are conducted is 0.46. In addition, we repeated the models without covariates and obtained similar results. All analyses were performed using SAS version 9.4 (SAS Institute). Associations were statistically significant at P < .05, and all P values were 2-sided.

Results
The study population included 1 000 639 children born to fertile women ( We then fitted a model that included both the fertile and IVF groups with an interaction term between group and major defect; the interaction was significant (χ 2 1 =5.52; P = .02). This significance indicates that the difference in HRs between cancer with a major defect and cancer without a major defect is associated with the group; ie, the 2 groups are not parallel. Therefore, we fitted a model with 4 levels defined by the combinations of groups (fertile or IVF) and the presence or absence of a major defect ( Table 3) with children of fertile women without a major birth defect as the reference group.
The risk of cancer was not significantly increased among children conceived via IVF without a major birth defect (HR, 1.20; 95% CI, 0.93-1.53), but was significantly increased for children with a major birth defect (fertile group: HR. 3 [11.6] in singletons, respectively).

Discussion
More than 80 studies have reported an association between birth defects and cancer, based on a variety of study designs, sample sizes, and data sources. 10 The 2.07-fold increase in cancer risk among children with major nonchromosomal birth defects in the fertile group that we report is consistent with the other US population-based cohort studies that linked to state birth defect and cancer registries, with increases ranging from 1.58-fold 11 to 2.0-fold 12, 2.5-fold, 13,14 to 2.86-fold. 15 The finding of a 15.45-fold cancer risk among children with chromosomal birth defects in the fertile group is also consistent with the point estimates and 95% CIs of previous reports, ranging from 11.6 (95%   21 This finding provides evidence of the role of epigenetic alternations on the co-occurrence of birth defects and cancer among children conceived via IVF. 22 Another potential genetic phenomenon that could explain the role of IVF in the association between birth defects and childhood cancer is germline de novo variation. De novo genetic variation has been demonstrated in IVF-conceived embryos, and whole-genome sequencing has been proposed as a method for detecting these events in embryos for preimplantation genetic diagnosis. 23 Germline de novo variants have been reported to cause several genetic diseases, including birth defects (eg, congenital heart disease) and some childhood cancers, including neuroblastoma. [24][25][26] While more work is warranted to understand the role of de novo germline genetic variants in the interface between birth defects and childhood cancer, this remains a plausible mechanism as these events are also more likely in children conceived via IVF. 27

Limitations and Strengths
This study has limitations. The primary limitation of this study is that we were not able to explore specific birth defect-cancer combinations due to small sample sizes of children conceived via IVF with major birth defects and cancer. Another limitation is that the follow-up period averaged 5.8 years, when the incidence of childhood cancer is at its lowest rate. 28 Additional years of follow-up, through ages 10 to 14 years and 15 to 19 years, when cancer rates increase to the high levels observed during infancy, may offer a broader picture of this association. The strengths of this study include its population-based design and contemporary time period, inclusive of the years when IVF has been widely used in the US. The 4 study states include racially and ethnically diverse populations, with high linkage rates, and birth defects and cancer registries that use similar case definitions. The data on infertility, birth defects, and cancer were independently collected, minimizing ascertainment bias. In addition, we limited our analyses to singletons because of the known higher risk of birth defects in twin births and higher order multiple births.

Conclusions
This study found a stronger association between birth defects and cancer in children conceived via IVF vs children conceived naturally. With IVF births rising worldwide, further investigations into these associations are warranted.