Association of In Vitro Fertilization With Childhood Cancer in the United States

Abstract

Importance

In vitro fertilization (IVF) is associated with birth defects and imprinting disorders. Because these conditions are associated with an increased risk of childhood cancer, many of which originate in utero, descriptions of cancers among children conceived via IVF are imperative.

Objective

To compare the incidence of childhood cancers among children conceived in vitro with those conceived naturally.

Design, Setting, and Participants

A retrospective, population-based cohort study linking cycles reported to the Society for Assisted Reproductive Technology Clinical Outcomes Reporting System from January 1, 2004, to December 31, 2012, that resulted in live births from September 1, 2004, to December 31, 2013, to the birth and cancer registries of 14 states, comprising 66% of United States births and 75% of IVF-conceived births, with follow-up from September 1, 2004, to December 31, 2014. The study included 275 686 children conceived via IVF and a cohort of 2 266 847 children, in which 10 births were randomly selected for each IVF birth. Statistical analysis was performed from April 1, 2017, to October 1, 2018.

Exposure

In vitro fertilization.

Main Outcomes and Measures

Cancer diagnosed in the first decade of life.

Results

A total of 321 cancers were detected among the children conceived via IVF (49.1% girls and 50.9% boys; mean [SD] age, 4.6 [2.5] years for singleton births and 5.9 [2.4] years for multiple births), and a total of 2042 cancers were detected among the children not conceived via IVF (49.2% girls and 50.8% boys; mean [SD] age, 6.1 [2.6] years for singleton births and 4.7 [2.6] years for multiple births). The overall cancer rate (per 1 000 000 person-years) was 251.9 for the IVF group and 192.7 for the non-IVF group (hazard ratio, 1.17; 95% CI, 1.00-1.36). The rate of hepatic tumors was higher among the IVF group than the non-IVF group (hepatic tumor rate: 18.1 vs 5.7; hazard ratio, 2.46; 95% CI, 1.29-4.70); the rates of other cancers did not differ between the 2 groups. There were no associations with specific IVF treatment modalities or indication for IVF.

Conclusions and Relevance

This study found a small association of IVF with overall cancers of early childhood, but it did observe an increased rate of embryonal cancers, particularly hepatic tumors, that could not be attributed to IVF rather than to underlying infertility. Continued follow-up for cancer occurrence among children conceived via IVF is warranted.

This population-based cohort study compares the incidence of childhood cancers among children conceived in vitro with incidence of childhood cancers among children conceived naturally.

Key Points

Question

Is the incidence of childhood cancers among children conceived in vitro different than that among children conceived naturally?

Findings

In this population-based cohort study assembled by data linkage, the rate of hepatic tumors was significantly higher among children conceived via IVF than among children conceived naturally (hepatic tumor rate, 18.1 vs 5.7); the rates of other cancers did not differ between the 2 groups.

Meaning

An association of conception by IVF with childhood cancer is small and limited to rare tumors; however, continued follow-up for cancer occurrence among children conceived via IVF is warranted.

Introduction

In vitro fertilization (IVF) involves the ex vivo manipulation of both sexes’ gametes to achieve conception and accounted for 1.7% of US births in 2015.¹ It is well established that IVF is associated with adverse pregnancy outcomes, including greater risks for preterm birth, lower birth weight, structural birth defects, and imprinting disorders.^{2,3,4,5,6,7,8,9} Because the latter 2 conditions confer excess risk of childhood cancer^10,11,12,13 and because many childhood cancers originate in utero,¹⁴ describing the occurrence of cancer among children conceived by IVF is imperative.

We report the results of the linkage of the Society for Assisted Reproductive Technology Clinical Outcomes Reporting System (SART CORS) to the birth and cancer registries of 14 states to create the largest cohort study to date and the first in the United States, to our knowledge, of the association of IVF with childhood cancer.

Methods

Study Data and Oversight

The SART CORS contains comprehensive data from more than 83% of all clinics providing IVF and 91% of all IVF cycles performed in the United States. Data were collected and verified by SART and reported to the Centers for Disease Control and Prevention in compliance with the Fertility Clinic Success Rate and Certification Act of 1992 (Public Law 102-493).¹⁵ The society makes data available for research purposes to entities that have agreed to comply with SART research guidelines. Data are submitted by individual clinics and vouched for by the medical director of each clinic. Approximately 10% of the clinics are audited each year to validate the accuracy of reported data.¹⁶ The study was approved by the institutional review boards at Michigan State University, the University of Minnesota, the University of Michigan, and each participating state department of health (California, Colorado, Connecticut, Florida, Illinois, Massachusetts, Michigan, New Jersey, New York, North Carolina, Ohio, Pennsylvania, Texas, and Virginia). Patients undergoing assisted reproductive technology at SART-associated clinics sign clinical consent forms that include permission to use their deidentified data for research.

Data Linkage and Comparison Birth Selection

The states in this study had at least 1000 IVF births annually. States that agreed to participate and contributed data were California, Colorado, Connecticut, Florida, Illinois, Massachusetts, Michigan, New Jersey, New York, North Carolina, Ohio, Pennsylvania, Texas, and Virginia.

The SART CORS data are sufficient to link to state birth registries, which contain both maternal and child information, but not to cancer registries, which contain only child information. Therefore, the first linkage was between SART CORS data and birth registries. To maintain data quality and to ensure anonymity of the final linked file provided to the investigators, the SART CORS data were sent directly from the contract organization charged with maintaining these data (ie, Redshift Technologies Inc) to each participating state. Participating states were not provided with data on IVF treatment but only with a unique identifier created by Redshift Technologies Inc.

The encrypted, password-protected linkage file provided by SART CORS contained the names, dates of birth, social security numbers (when present), zip codes of residence, and approximate dates of delivery for all women who resided in each respective state with records of IVF cycles from January 1, 2004, to December 31, 2012, that resulted in a live birth. Because only mothers’ names, but not offspring’s names, were available in the SART CORS, cycles were linked to birth records based on mothers’ information by the health departments of the participating states using probabilistic record linkage. To achieve uniform results, states were encouraged to use Link Plus software published by the Centers for Disease Control and Prevention’s National Program of Cancer Registries; however, some states opted to use their existing proprietary code. Across all participating states, 91.8% of SART CORS records of live birth after IVF matched to a birth record. All live-born children from each birth, including multiple births, identified by the linkage were included in the resulting files (IVF births). Ten births resulting in live-born children were randomly selected from among those that did not link to SART CORS records (non-IVF births) to serve as comparison for each IVF birth.

The IVF birth files and non-IVF birth files were next concatenated for subsequent linkage to each state’s cancer registry. Children’s names, dates of birth, and sex obtained from birth registries were used to identify those with a diagnosis of malignant neoplasm during the follow-up period. Because the latest available years of cancer registry data varied, as did the timing of linkage, states had different years of follow-up from December 31, 2010, to December 31, 2014.

The final file was stripped of identifying information before being provided to the investigators. It contained the unique SART CORS woman and cycle identifiers, birth certificate variables (birth weight, gestational age, plurality, race/ethnicity, parental ages, and parental educational level), and cancer variables (text description of cancer, morphology and topography codes, age at diagnosis in months, staging information, and laterality). It further contained age at death if it occurred during infancy because infant deaths are routinely linked to births by vital statistics departments.¹⁷ These files were then merged by the investigators with the SART CORS data using the unique identifiers.

Outcomes

Cancers were categorized according to the International Classification of Childhood Cancers, third edition.¹⁸ All models were repeated for combined childhood cancers and for individual cancers with a minimum of 5 cases among the IVF group and non-IVF group, each with multiple and singleton births combined. Embryonal cancers, including neuroblastoma, retinoblastoma, Wilms tumor (nephroblastoma), hepatoblastoma, embryonal rhabdomyosarcoma, pulmonary and pleuropulmonary blastoma, medulloblastoma, primitive neuroectodermal tumors, medulloepithelioma, and atypical teratoid rhabdoid tumors, were also examined as a single tumor class. Intracranial embryonal tumors (medulloblastoma, primitive neuroectodermal tumors, and medulloepithelioma) were also included in the combined central nervous system tumors. When fewer than 10 cases of an individual tumor type occurred among the IVF group, we reported rates and relative risks for the general tumor class.

Statistical Analysis

Statistical analysis was performed from April 1, 2017, to October 1, 2018. Data from each state were processed to generate a common data set. The only exclusions were as follows: a mother or father who was younger than 18 years or a nonviable birth (gestational age <22 weeks or birth weight <300 g even if indicated as a live birth). Because most independent variables were categorized, missing values were included as a separate category. The final population included 275 686 children in the IVF group (1 274 070 person-years; 209 586 births) and 2 266 847 children in the non-IVF group (10 596 144 person-years; 2 230 378 births). Person-years at risk were calculated as the time from birth to the earliest age at cancer diagnosis, age at death (if during infancy), or December 31 of the last year of complete cancer registry follow-up. We received dates as month and year. Therefore, we treated all observations as arriving at the middle of the month (0.5 months). When the month was not received (1 state), we treated the observation as arriving at the middle of the year (0.5 years).

We compared the incidence of childhood cancer between the IVF and non-IVF groups overall; in addition, we compared IVF treatment modalities (donor source, embryo state, use of intracytoplasmic sperm injection, assisted hatching, and day of transfer) and indication for IVF (male factor, diminished ovarian reserve, ovulation disorders, endometriosis, uterine factors, tubal factors, other, or unexplained) within the IVF group. Adjusted hazard ratios (HRs) and 2-sided 95% CIs were calculated by Cox proportional hazards regression using SAS, version 9.4 (SAS Institute Inc). P < .05 was considered significant. Covariates were selected a priori for inclusion based on established associations with childhood cancers^{19,20,21,22,23} and/or IVF^24,25: sex, plurality (singleton or multiple births), maternal educational level (high school or less, some college, college graduate, or unknown), maternal race/ethnicity (white, black, native American, Asian, or unknown; Hispanic or not Hispanic), maternal age, and state of birth. Birth weight and gestational age were not adjusted for because they may lie on the causal pathway.

In a second analysis, we selected a 1:1 match between children in the IVF group and those in the non-IVF group who matched on sex, plurality (singleton or multiple birth), race/ethnicity, maternal age categories, maternal educational level, and state of birth. This analysis (141 959 singleton births and 48 821 children from multiple births in each group) is presented in eTable 5 in the Supplement.

We could not properly account for correlation within twin pairs because data on twinship were inconsistently provided by the states; hence, we reported all HRs among singleton births and multiple births separately as well as combined. The analyses, including multiple births, should be considered as providing valid estimates of the rates, while the 95% CIs are very slightly narrower than if we could account for twinship.

Results

Characteristics of the Study Population

The characteristics of the IVF and non-IVF groups by plurality are presented in Table 1. Because of the large sample size, nearly all comparisons were significant, even if differences were small. The most notable differences reflected well-known attributes of IVF pregnancies.²⁶ In particular, mothers whose children were conceived by IVF were substantially older (singleton births: mean [SD] age, 35.9 [5.0] vs 28.7 [5.9] years; multiple births: mean [SD] age, 35.2 [5.1] vs 30.3 [5.9] years), more educated (singleton births: college or postgraduate education, 73.4% vs 32.9%; multiple births: college or postgraduate education, 71.6% vs 40.0%), and more often white (singleton births, 82.5% vs 76.0%; multiple births, 84.2% vs 75.9%) than those of children conceived naturally; among both singletons and multiple births, children conceived by IVF had a mean (SD) lower birth weight (singleton births, 3264 [614] vs 3315 [555] g; multiple births, 2314 [621] vs 2335 [628] g) and shorter mean (SD) gestation (singleton births, 38.4 [2.3] vs 38.7 [2.0] weeks; multiple births, 35.1 [3.1] vs 35.2 [3.2] weeks) than those conceived naturally. The mean (SD) length of follow-up was 4.5 (2.5) years for singleton births conceived via IVF, 4.7 (2.5) years for singleton births conceived naturally, 4.7 (2.54) years for multiple births conceived via IVF, and 4.6 (2.6) years for multiple births conceived naturally.

Table 1. Descriptive Characteristics of the IVF and Non-IVF Study Populations^a.

Characteristic	Singleton Births		Multiple Births
	IVF	Non-IVF	IVF	Non-IVF
Mothers, No.	146 875	2 194 854	62 711	35 524
Children, No.	146 875	2 194 854	128 811	71 993
Follow-up, mean (SD), y	4.5 (2.5)	4.7 (2.5)	4.7 (2.5)	4.6 (2.6)
Maternal age, mean (SD), y	35.9 (5.0)	28.7 (5.9)	35.2 (5.1)	30.3 (5.9)
Age range of mothers, %, y
≤24	0.5	26.9	0.7	18.3
25-29	8.5	27.8	10.9	25.7
30-34	31.0	27.4	35.6	31.7
35-37	23.3	10.5	22.7	13.7
38-40	18.9	5.3	15.2	7.1
41-43	10.9	1.8	7.9	2.3
>43	6.8	0.3	7.0	1.2
Maternal educational level, %
≤High school graduate or GED	10.2	42.6	10.9	35.7
Some college	16.4	24.5	17.5	24.3
College graduate or postgraduate education	73.4	32.9	71.6	40.0
Maternal race, %
White	82.5	76.0	84.2	75.9
Black	5.1	14.7	4.9	17.0
American Indian or Alaska Native	0.3	0.5	0.3	0.5
Asian or Pacific Islander	12.1	8.8	10.6	6.6
Maternal ethnicity, % Hispanic	8.1	24.3	8.7	17.3
Male infant, %	50.9	51.2	50.8	50.2
Plurality, %
Twins	NA	NA	94.7	97.5
Triplets	NA	NA	5.1	2.4
≥Quadruplets	NA	NA	0.1	0.1
Length of gestation, mean (SD), wk	38.4 (2.3)	38.7 (2.0)	35.1 (3.1)	35.2 (3.2)
Length of gestation, %, wk
22-27	0.8	0.5	3.5	3.9
28-32	1.9	1.1	12.3	11.4
33-36	9.5	6.5	45.6	43.6
≥37	87.9	91.9	38.6	41.1
Birth weight, mean (SD), g	3264 (614)	3315 (555)	2314 (621)	2335 (628)
Weight range of infants at birth, %, g
<1000	0.8	0.5	3.9	4.2
1000-1499	0.9	0.5	6.8	6.2
1500-2499	7.1	5.0	47.5	45.7
≥2500	91.2	94.0	41.8	43.9
Birth weight z score, mean (SD)	0.03 (1.00)	0.01 (1.01)	−0.60 (0.92)	−0.62 (0.93)
Small for gestation (z score ≤–1.28), %	8.4	8.4	20.8	21.7
Appropriate for gestation (z score –1.27 to 1.27), %	81.4	82.0	77.5	76.4
Large for gestation (z score ≥1.28), %	10.2	9.6	1.7	1.9

Abbreviations: GED, General Education Development; IVF, in vitro fertilization; NA, not applicable.

^aComparisons of children conceived via IVF with children conceived naturally are all statistically significant at P < .001 except for the percentage of male infants, which was P = .03 for singleton births and P = .009 for multiple births. Missing values for singleton and multiple births: maternal age (0.012% and 0.013%), race (4.4% and 3.2%), educational level (2.2% and 2.6%), length of gestation (1.0% and 1.4%), birth weight (0.4% and 0.7%), and birth weight z score (1.2% and 1.7%).

Risk of Cancer in IVF and Non-IVF Groups

There were 321 cancers detected among the IVF group and 2042 cancers among the non-IVF group (Table 2). The overall cancer rate (per 1 000 000 person-years) was 251.9 in the IVF group and 192.7 in the non-IVF group (HR, 1.17; 95% CI, 1.00-1.36). The rate of embryonal tumors was slightly higher among the IVF group (102.8 vs 70.4; HR, 1.28; 95% CI, 1.01-1.63), which seemed to be driven mainly by an elevated rate of hepatic tumors (18.1 vs 5.7; HR, 2.46; 95% CI, 1.29-4.70), most of which were hepatoblastoma. The results for singleton births and multiple births separately are presented in eTable 1 in the Supplement. The magnitude and direction of HRs seen in the matched subset are similar to those in the unmatched sample (eTables 4 and 5 in the Supplement); however, the 95% CIs are wider because of the smaller sample size.

Table 2. Data on the Rates of Cancer by Study Group for All Children (Singleton and Multiple Births Combined)^a.

Cancer	Cases, No.		Cancer Rate/1 000 000 Person-Years		HR (95% CI)b
	IVF	Non-IVF	IVF	Non-IVF
Any cancer	321	2042	251.9	192.7	1.17 (1.00-1.36)
Leukemia	93	707	73.0	66.7	0.93 (0.70-1.22)
ALL	72	534	56.5	50.4	0.96 (0.71-1.32)
AML	13	108	10.2	10.2	0.74 (0.35-1.53)
Lymphoma	22	139	17.3	13.1	1.00 (0.56-1.80)
CNS cancer	59	383	46.3	36.1	1.26 (0.89-1.79)
Astrocytoma	34	197	26.7	18.6	1.50 (0.95-2.36)
Ependymoma	5	48	3.9	4.5	0.53 (0.16-1.72)
Intracranial embryonal tumors	13	83	10.2	7.8	1.41 (0.67-2.95)
Neuroblastoma	47	260	36.9	24.5	1.10 (0.74-1.65)
Retinoblastoma	14	127	11.0	12.0	1.11 (0.57-2.18)
Renal cancer	28	186	22.0	17.6	1.10 (0.66-1.84)
Hepatic cancer	23	60	18.1	5.7	2.46 (1.29-4.70)
Soft-tissue sarcoma	18	97	14.1	9.2	1.50 (0.81-2.84)
Germ cell tumors	11	43	8.6	4.1	2.13 (0.91-4.96)
Embryonal tumorsc	131	746	102.8	70.4	1.28 (1.01-1.63)

Abbreviations: ALL, acute lymphoblastic leukemia; AML, acute myeloid leukemia; CNS, central nervous system; HR, hazard ratio; IVF, in vitro fertilization.

^aThe final population included 275 686 children in the IVF group (1 274 070 person-years; 209 586 births) and 2 266 847 children in the non-IVF group (10 596 144 person-years; 2 230 378 births).

^bAdjusted for state of birth, maternal race and ethnicity, maternal educational level (college graduate vs less than college graduate), maternal age, and child’s sex; missing data for maternal race (4.3%) and educational level (2.2%) were replaced by a category labeled “missing” in each variable.

^cNeuroblastoma, retinoblastoma, nephroblastoma, hepatoblastoma, embryonal rhabdomyosarcoma, pulmonary and pleuropulmonary blastoma, medulloblastoma, primitive neuroectodermal tumor, medulloepithelioma, and atypical teratoid and rhabdoid tumor.

Risk of Cancer Comparing IVF Treatment Modalities and Indication for IVF

There were no significant differences in the rate of either all cancers or embryonal tumors comparing children conceived by donor egg vs autologous egg, frozen embryos vs fresh embryos, use of intracytoplasmic sperm injection vs none, assisted hatching vs none, and day 2 to 3 embryo transfer vs day 5 to 6 embryo transfer (eTable 2 in the Supplement). Similarly, there was no difference in the rate of either all cancers or embryonal tumors by indication for IVF (eTable 3 in the Supplement).

Discussion

These findings are from the first cohort study of the association between IVF and the risk of childhood cancer in the United States, and it is the largest cohort published to date. We observed little evidence of excess risk of most cancers comparing children conceived by IVF with children conceived naturally. There was a significant 28% increase in the rate of embryonal tumors as a class, which seemed to be due to the elevated risk of hepatic tumors, most of which were hepatoblastoma. We estimated that an excess of 8 embryonal cancer cases would occur between birth and 5 years of age among the 68 000 children conceived via IVF who were born in 2013¹⁶ if the higher rates applied. However, because we were not able to compare the risk of cancer in children conceived via IVF with the risk in children of subfertile women who did not undergo IVF, we cannot attribute the observed small elevated risks to IVF itself rather than the underlying causes of infertility.

Our results both contrast and concur with those of several relevant studies. A meta-analysis that compiled 9 studies of IVF and childhood cancer published through 2012 reported a significant increased risk of overall cancer (relative risk, 1.33; 95% CI, 1.02-1.98), including leukemia, central nervous system cancers, neuroblastoma, and other solid cancers.²⁷ We did not detect associations as pervasive or as large, nor did 2 prospective studies conducted in the United Kingdom²⁸ and Nordic countries.²⁹ Because the meta-analysis included mainly case-control studies with few exposed cases, prospective studies with few exposed children, and several hypothetical cohorts, it must be considered that their results may be due to systematic bias.

The UK and Nordic studies were conducted with methods more comparable to our study. Williams et al²⁸ linked 106 381 records of nondonor IVF identified by the UK Human Fertilisation and Embryology Authority between 1992 and 2008, inclusive, to the National Registry of Childhood Tumours. Observed cancers were compared vs the number expected based on population rates to produce standardized incidence ratios. An important difference between the study by Williams et al²⁸ and our study is our inclusion of a comparison cohort of children not conceived via IVF with covariates available for statistical adjustment. Two cancers in the UK study had a significantly higher incidence than in each study’s reference population: hepatoblastoma (standardized incidence ratio, 3.27; 95% CI, 1.20-7.12), which appeared to be associated specifically with low birth weight (≤2500 g, a known, strong risk factor for this tumor³⁰), and rhabdomyosarcoma (standardized incidence ratio, 2.62; 95% CI, 1.26-4.82).²⁸ We similarly detected increased rates of hepatic tumors, which are composed largely of hepatoblastoma, and soft tissue sarcomas, which are composed largely of rhabdomyosarcoma. The study by Williams et al²⁸ included cancers diagnosed in individuals up to 15 years of age vs our study, in which the longest follow-up attained was 8 years. Another difference between studies is the possibility of changing clinical practices in IVF treatment during the 17-year inclusion period in the study by Williams et al.²⁸

The Nordic study linked 91 796 children conceived via IVF and 358 419 children conceived naturally with records in the population-based birth and cancer registries of Denmark, Finland, Norway, and Sweden during the period from 1982 to 2007.²⁹ Similar to our study, statistical adjustment was made for covariates, but the Nordic study covered a 25-year time span and had a longer mean follow-up of 9.5 years, as well as an unambiguous length of follow-up. In vitro fertilization was not associated with either overall cancers (HR, 1.08; 95% CI, 0.91-1.27) or most specific cancers; however, central nervous system tumors (HR, 1.44; 95% CI, 1.01-2.05) and combined carcinomas (HR, 2.03; 95% CI, 1.06-3.89) were significantly more prevalent.

In our study, we observed an increased risk mainly of embryonal cancers, which was a hypothesis at the outset of the study because IVF has previously been associated with epigenetic disruption. Large offspring syndrome occurs after in vitro embryo culture of ruminant animals³¹ and is characterized by multilocus loss of imprinting.³² Beckwith-Wiedemann syndrome is a human analogue of large-offspring syndrome³³ that occurs more frequently in children conceived by intracytoplasmic sperm injection^34,35 and that predisposes children especially to embryonal tumors, including hepatoblastoma.^13,36 Excess occurrence of Beckwith-Wiedemann syndrome among children conceived via IVF could be one mechanism to explain a portion of the excess risk of embryonal tumors in our study. However, no information on Beckwith-Wiedemann syndrome among children conceived via IVF was available to assess this possibility.

This data set presents a strong foundation in which to examine the risk of childhood cancer after IVF. The SART CORS represents the single largest source of data available for linkage in the United States. Participating state departments of health matched approximately 85% of the records supplied by the SART CORS, and participating states represented 66% of US births and 75% of IVF-conceived births during the study period. Critical covariates recorded with high validity by birth registries³⁷ were available for multivariate analysis.

Limitations

Several weaknesses of the data set must also be considered. A common problem in observational studies is unmeasured confounders. As can be seen in Table 1, mothers who conceived via IVF were more likely to be white, non-Hispanic, more educated, and older than the mothers who conceived naturally. These differences may be indicative of unmeasured confounders, such as income, medical insurance, and prenatal care, which may affect the risk of adverse perinatal outcomes and the risk of childhood cancer. Although race, ethnicity, educational level, and age were included in the models, it is not possible to estimate the effect of unmeasured confounders on the HRs. However, the analysis in eTable 5 in the Supplement that matched on race, educational level, age, and state produced similar HRs but wider 95% CIs owing to the smaller sample size.

We could not account for non-IVF treatments for infertility, such as intrauterine insemination or ovulation induction, although their prevalence in the non-IVF cohort would be small (about 4% in a nationally representative survey³⁸). We also had no way to detect when children moved out of participating states or died after infancy, with the result that person-time in our study may be slightly overestimated. A small number of cancer cases may also have not been detected, although, because childhood cancer is very rare, this is likely not a large problem. Overestimation of person-time would lead to underestimation of absolute cancer rates, but so long as denominator error did not differ by group, it should not have biased the HRs. We have no data available to assess this contention but note that gross person-time misspecification particular to children conceived via IVF would have been required to produce effects of the size observed.

Although we controlled for several potential confounders, the IVF and non-IVF groups may have varied in unmeasured ways. We were unable to account for Beckwith-Wiedemann syndrome or structural birth defects, which were also associated with both IVF and an increased risk of childhood cancer.^10,11 It is common to all studies of sequelae of IVF that observed risks cannot easily be separated from the underlying causes of infertility.³⁹ A large Scandinavian study found about a 20% increased rate of childhood cancer among offspring of women who had ever been evaluated for infertility,⁴⁰ and the TP53 gene that, when mutated, causes the cancer-predisposing Li-Fraumeni syndrome has also been implicated in the regulation of fertility.⁴¹ Although our study had the largest number of children conceived via IVF followed up for cancer occurrence, to date, the absolute number of observed cancers was relatively small, and the association of IVF with especially rare types of cancer was not evaluable. Follow-up was shorter in our study than in 2 comparable studies^28,29; however, it was long enough to capture many of the cancers that occur prior to 5 years of age. We did not adjust for multiple comparisons.

Conclusions

We found, at most, a small, marginally significant association between IVF and overall cancer in childhood. We found no association of specific modes of IVF treatment or indication for IVF with overall cancer or embryonal tumors. Although we observed increased rates of embryonal cancers, particularly hepatic tumors, these cancers remained rare in absolute terms. We also could not attribute the increased rate to treatment per se rather than to underlying infertility. Despite the large size of the study, these results do not definitively establish an association between IVF and embryonal tumors. These results suggest that continued follow-up of children conceived via IVF for cancer occurrence is warranted, while other approaches to understanding potential mechanisms should also be pursued, for instance, by comparing the somatic epigenetic landscape of tumors in children conceived via IVF with children conceived naturally.

Supplement.

eTable 1. Hazard Ratios and 95% Confidence Intervals (CI) and Rates per 1 000 000 Person-Years by Study Group for All Children: Results for Singletons and Multiples

eTable 2. Hazard Ratios and 95% Confidence Intervals (CI) and Rates per 1 000 000 Person-Years Among IVF-Conceived Children by IVF Treatment Parameters

eTable 3. Hazard Ratios and 95% Confidence Intervals (CI) and Rates per 1 000 000 Person-Years Among IVF-Conceived Children by Indication for IVF

eTable 4. Descriptive Characteristics of the IVF and Non-IVF Study Populations Matched on Child’s Sex, State of Birth, Maternal Age, Race/Ethnicity, and Education

eTable 5. Hazard Ratios and 95% Confidence Intervals (CI) and Rates per 1 000 000 Person-Years by Study Group for the Matched Sample (eTable 4) of IVF and Non-IVF Children