Association Between Nonsyndromic Congenital Heart Disease and Childhood Cancer in a Population‐Based, Retrospective Cohort of 11 Million Children

Abstract

Background

Epidemiologic studies have demonstrated associations between congenital heart disease (CHD) and childhood cancer. However, these assessments have not fully accounted for the pathogenetic and clinical heterogeneity of CHD, limiting their capacity to inform pathogenetic studies and the clinical management of children with CHD.

Methods

To achieve a more detailed understanding of cancer risk in children with nonsyndromic CHD, we evaluated 60 cardiac lesions in 7 categories in a cohort of 11 million children from Massachusetts, North Carolina, Oklahoma, and Texas. We used Cox proportional hazards regression to estimate the hazard ratios (HRs) and 95% CIs of cancer in children with each lesion relative to children without congenital anomalies when there were ≥3 co‐occurring cases.

Results

We identified 74 755 children with CHD and observed increased hazards for several malignancies. Notable associations included aortic arch obstruction–lymphoma (HR, 12.7 [95% CI, 4.9–31.0]); single‐ventricle disease–lymphoma (HR, 9.7 [95% CI, 3.8–24.4]); complex atrioventricular septal defect–leukemia (HR, 8.05 [95% CI, 2.8–23.0]); ventricular septal defect–hepatoblastoma (HR, 11.2 [95% CI, 5.9–21.1]); and conotruncal defects–neuroblastoma (HR, 6.4 [95% CI, 1.9–22.2]). The cumulative incidence of cancer was <1% for all CHD groups evaluated.

Conclusions

Cancer incidence in children with nonsyndromic CHD was generally low. However, we identified >5‐fold increases in the hazard of hepatoblastoma, neuroblastoma, leukemia, and lymphoma in children with certain lesions. Our findings may assist in generating hypotheses about the causes of cancer in children with CHD.

Nonstandard Abbreviations and Acronyms

AVSD

atrioventricular septal defect

CCLOC

Core Cardiac Lesion Outcome Classification

GOBACK

Genetic Overlap Between Anomalies and Cancer in Kids

HLHS

hypoplastic left heart syndrome

NHL

non‐Hodgkin lymphoma

PA‐VSD

pulmonary atresia with ventricular septal defect

UTAZIA

Utah, Arizona, and Iowa

Research Perspective.

What Is New?

• This study demonstrates an increased hazard of cancer in children with several specific congenital heart defects.

What Question Should Be Addressed Next?

• Additional research is warranted to identify the causes of cancer in children with congenital heart disease and to identify which, if any, subpopulations may benefit from genetic/genomic testing or cancer surveillance.

Approximately 1% of births in the United States are affected by congenital heart disease (CHD), making it collectively the most common congenital anomaly. ¹ As survival rates for affected infants have improved, there is an urgent need to understand health outcomes in children with CHD. Notably, children and adolescents with CHD are at increased cancer risk relative to those without anomalies ² , ³ , ⁴ , ⁵ , ⁶ , ⁷ , ⁸ : Large, nonoverlapping US registry linkage studies reported a 2.9‐fold increase in the incidence rate ratio (IRR) ² and 2.5‐fold increase in the hazard ratio (HR) ³ of cancer among children with CHD, respectively. Similarly, population‐based studies from Canada and Sweden observed increased risk of hematologic malignancies in children and adults with nonsyndromic CHD. ⁴ , ⁶

However, CHD describes a heterogeneous set of diagnoses that are distinct with respect to their pathogenesis, risk factors, and clinical presentation (ie, they impact different aspects of the heart and circulation), and many studies have aggregated CHDs into broad groups (eg, septal defects) or combined distinct entities (eg, pulmonary atresia and pulmonary stenosis). This approach may obscure CHD–cancer associations and provides limited insight into the underlying mechanisms. This, in turn, makes it difficult to identify children at the greatest risk of specific malignancies and to develop effective risk prediction and surveillance strategies.

We sought to address these limitations by applying a sophisticated CHD classification system (the Core Cardiac Lesion Outcome Classification [CCLOC] algorithm) that incorporates genetic and embryologic knowledge of CHD as opposed to the morphology‐based classifications used in previous epidemiologic assessments to 11 million children in the GOBACK (Genetic Overlap Between Anomalies and Cancer in Kids) Registry Linkage Cohort. ⁹ , ¹⁰ Our objective was to generate hypotheses about the causes of cancer in children with CHD by evaluating the associations of these biologically and clinically distinct nonsyndromic CHD phenotypes with childhood cancer.

Methods

One author (J.M.S.) had full access to the data in the study and takes responsibility for its integrity and the data analysis. Aggregated data are available from the corresponding author upon reasonable request. Restrictions apply to individual‐level data, which were used with permission of the participating states. Individual‐level data are available by application to Massachusetts Department of Health, North Carolina Department of Public Health, Oklahoma State Department of Health, and Texas Department of State Health Services.

Study Population

This study used data from the GOBACK Registry Linkage Cohort ¹⁰ and included all live births in Massachusetts (2000–2017), North Carolina (2003–2018), Oklahoma (1997–2012), and Texas (1999–2018). These states were selected on the basis of their similar CHD case finding and classification methods (described below). In each state, data from population‐based surveillance systems were linked to birth certificates to identify children diagnosed with congenital anomalies or cancer and to obtain demographic and maternal information. Deidentified, linked data sets were then securely transferred to investigators at Baylor College of Medicine for harmonization and statistical analysis. The study complied with the Principles of the Declaration of Helsinki and was approved by the institutional review boards of Baylor College of Medicine. The requirement for written informed consent was waived because the study involved secondary analysis of deidentified data.

Ascertainment and Classification of CHD (Exposure)

Participating birth defect registries were population‐based, used active case finding, and classified CHD using the Centers for Disease Control–modified British Pediatric Association coding system. Using these codes and the CCLOC algorithm, ⁹ we identified cases with 60 mutually exclusive CHD phenotypes belonging to 7 groups: left‐sided lesions, right‐sided lesions, conotruncal defects, endocardial defects, looping/arterial malposition defects, double‐outlet right ventricle, and other CHD types (complex pulmonary atresia, aortopulmonary window, partial and total anomalous pulmonary venous return, and ventricular septal defect [VSD]). The 7 larger groups are formed by presumed shared embryologic alterations. For example, tetralogy of Fallot, pulmonary atresia with VSD, and truncus arteriosus are distinct CHD lesions but are grouped together in the conotruncal lesion group, as they are often seen together in families and in similar genetic conditions and are all thought to be due to altered second heart field/neural crest development. ¹¹ , ¹² Children were classified according to the most severe CHD diagnosis, such that, for example, a child with coarctation of the aorta and hypoplastic left heart syndrome (HLHS) would be included in the HLHS but not the coarctation group. CCLOC provides a more detailed and clinically meaningful classification than traditional methods that may not account for the full pathogenetic and clinical diversity of CHDs. Although CHD physiology can change over time with interventions and aging, by grouping by presumed shared pathogenetic pathways, CCLOC allows for stable fundamental classification that does not change over the life span. In addition to morphology, this new classification system uses embryology and genetics and was shown to more accurately classify HLHS, a critical congenital heart defect, relative to other approaches. ⁹ We also examined single‐ventricle disease, which comprised severe lesions across several CCLOC groups (HLHS, pulmonary atresia with intact ventricular septum, tricuspid atresia, and double‐inlet left ventricle), as some studies have reported associations between CHD complexity and cancer. ¹³ We excluded children diagnosed with chromosomal abnormalities, pathogenic copy number variants, or single‐gene conditions (N=34 208), who were identified using a set of diagnostic codes selected by a clinical geneticist (S.A.R.) (Table S1). ¹⁴ Down syndrome was the most common diagnosis in this excluded group, and we have previously reported on the associations between congenital anomalies and cancer in the Down syndrome population. ¹⁵ In the present analysis, our objective was to characterize cancer risk in children with CHD and no evident genetic condition. Children with extracardiac anomalies without a known syndrome were not excluded.

Ascertainment and Classification of Childhood Cancer (Outcome)

We classified all primary malignant neoplasms and benign brain tumors diagnosed among subjects ≤18 years of age per the International Classification of Childhood Cancer, Third Edition, ¹⁶ using International Classification of Diseases for Oncology, Third Edition (ICD‐O‐3) site, histology, and behavior codes population‐based cancer registries.

Covariates

Maternal race and ethnicity were categorized as Hispanic (any race), non‐Hispanic Asian or Pacific Islander, non‐Hispanic Black, and non‐Hispanic White; all other non‐Hispanic individuals were combined due to low numbers (“additional non‐Hispanic groups”). Maternal age was classified as <20, 20 to 24, 25 to 29, 30 to 34, 35 to 39, or ≥40 years at delivery. Maternal educational attainment was categorized as either no high school diploma, high school diploma or equivalent, or any post–high school education. Perinatal characteristics included state and year of birth, sex assigned at birth (male, female, unspecified), birth weight (<2500, 2500–3999, or ≥4000 g), gestational age at delivery (preterm, defined as <37 weeks, or term), and plurality (singleton or multiple).

Statistical Analysis

Statistical analyses were conducted in R version 4.3.1 (R Foundation for Statistical Computing, Vienna, Austria). We first summarized maternal and perinatal characteristics among children with CHD and children without congenital anomalies using counts and percentages. We reported the percentage of children with CHD who were diagnosed with cancer at age ≤18 years to those with no anomalies and compared these by the χ²e test. Then we compared the cumulative incidence of cancer among children with CHD and children without congenital anomalies using the Kaplan–Meier method and the log‐rank test. We then used Cox proportional hazards regression with age as the underlying time scale to assess the HRs and 95% CIs of cancer among children with each CHD phenotype when there were ≥3 cases with CHD and cancer. Birth was used as time 0, and subjects were followed until cancer diagnosis, right‐censoring at age 18 years, or the end of the study period. We applied Firth's penalized maximum likelihood bias reduction method due to the small number of events in some models using the coxphf package. ¹⁷ Because death was treated as a censoring event and not modeled as a competing risk, our estimates reflect cause‐specific hazards. While this approach may underestimate cumulative incidence in CHD phenotypes with higher mortality rates, cause‐specific HRs may be preferred to subdistribution HRs from competing risks models in studies focused on pathogenetic associations rather than prediction. ¹⁸ , ¹⁹ We computed both crude and adjusted models; the latter included state of birth, sex, maternal age at delivery, and maternal education as a priori covariates. Although the distribution of birth weight and gestational age differed between exposure groups, we considered these potential mediators on the CHD–cancer pathway and therefore did not adjust for them. We present results for all childhood cancers combined, groups of related cancers (eg, leukemias, central nervous system [CNS] tumors), and individual cancer subtypes (eg, acute lymphoblastic leukemia, ependymoma). We applied a 5% false discovery rate correction to account for multiple testing and focused our reporting on associations with false discovery rate–adjusted q values <0.05. ²⁰

Increased medical or diagnostic scrutiny in children with cancer could lead to incidental discovery of CHD. To evaluate whether this could have biased our effect estimates, we performed a sensitivity analysis excluding children diagnosed with cancer before their first (MA, NC, and TX) or second (OK) birthday on the basis of the ages at which these birth defect registries accepted diagnoses. As in our primary analysis, we reported results only for CHD–cancer combinations with ≥3 co‐occurring cases.

Results

The cohort included 11 046 027 children born between 1997 and 2018, of whom 74 755 (0.66%) were diagnosed with nonsyndromic CHD (Table 1). Compared with those without congenital anomalies, children with CHD were more likely to be born preterm (<37 weeks) or have low birth weight (<2500 g). Women who delivered children with CHD were more likely to be aged ≥35 years at delivery. Median follow‐up was 7.9 years for children with CHD and 9.0 years for children without anomalies. Cancer was diagnosed in 150 children with CHD (0.2%) and 14 450 children without anomalies (0.1%) (P<0.001 by χ² test). Median age at cancer diagnosis was 2.7 years in the group with CHD and 3.9 years in the group without anomalies. At 18 years, the cumulative incidence of cancer was 0.37% in children with CHD and 0.24% in children without anomalies (P<0.001; Figure1).

Table 1.

Demographic and Maternal Characteristics of Subjects in the Genetic Overlap Between Anomalies and Cancer in Kids Registry Linkage Cohort by CHD Status

	No birth defect, n (%)	Congenital heart disease, n (%)
Total	10 971 272 (100.0)	74 755 (100.0)
State of birth
Massachusetts	1 320 010 (12.0)	4591 (6.1)
North Carolina	1 896 312 (17.3)	11 271 (15.1)
Oklahoma	773 412 (7.1)	4829 (6.5)
Texas	6 981 538 (63.6)	54 064 (72.3)
Sex
Male	5 569 182 (50.8)	36 642 (49.0)
Female	5 402 090 (49.2)	38 113 (51.0)
Birth year
1990–1999	475 516 (4.3)	2546 (3.4)
2000–2009	5 808 461 (52.9)	36 356 (48.6)
2010–2018	4 687 295 (42.7)	35 853 (48.0)
Birth weight, g
<2500	821 622 (7.5)	15 830 (21.2)
2500–3999	9 245 781 (84.3)	52 862 (70.7)
≥4000	793 837 (7.2)	4864 (6.5)
Unknown	106 515 (1.0)	1199 (1.6)
Gestational age, wks
<37 (preterm)	1 082 767 (9.9)	18 143 (24.3)
≥37 (term)	9 812 957 (89.4)	56 129 (75.1)
Unknown	75 548 (0.7)	483 (0.6)
Plurality
Multiple	346 610 (3.2)	4589 (6.1)
Singleton	10 624 196 (96.8)	70 165 (93.9)
Unknown	466 (0.0)	1 (0.0)
Maternal race and ethnicity
Hispanic	3 951 719 (36.0)	31 716 (42.4)
Non‐Hispanic Asian/Pacific Islander	471 940 (4.3)	2495 (3.3)
Non‐Hispanic Black	1 438 277 (13.1)	8686 (11.6)
Non‐Hispanic White	4 900 736 (44.7)	30 523 (40.8)
Additional non‐Hispanic groups	190 737 (1.7)	1239 (1.7)
Unknown	17 863 (0.2)	96 (0.1)
Maternal education
Less than high school	2 962 966 (27.0)	21 783 (29.1)
High school	2 868 444 (26.1)	19 531 (26.1)
Greater than high school	4 961 113 (45.2)	32 313 (43.2)
Unknown	178 749 (1.6)	1128 (1.5)
Maternal age, y
<20	1 202 443 (11.0)	7716 (10.3)
20–24	2 788 207 (25.4)	18 334 (24.5)
25–29	2 990 499 (27.3)	20 009 (26.8)
30–34	2 529 009 (23.1)	17 283 (23.1)
35–39	1 198 872 (10.9)	9075 (12.1)
≥40	260 182 (2.4)	2327 (3.1)
Cancer diagnosis	13 680 (0.1)	150 (0.2)

CHD indicates congenital heart disease.

Figure 1. Cumulative incidence of cancer in children with nonsyndromic CHD and children without congenital anomalies in the Genetic Overlap Between Anomalies and Cancer in Kids Registry Linkage Cohort.

CHD indicates congenital heart disease.

Effect estimates from crude and adjusted Cox models were similar (Table S2); we present adjusted results in the main manuscript. We observed 2‐ to 4‐fold increases in the hazard of any cancer among children with left‐sided lesions, right‐sided lesions, endocardial defects, and looping/arterial malposition defects (Table 2). Considering specific anomalies, HRs were increased in children with dextro‐transposition of the great arteries with and without VSD, aortic arch obstruction, pulmonary atresia with VSD, complex atrioventricular septal defect, VSD, and membranous VSD specifically. Neither single‐ventricle defects as a group nor hypoplastic left heart syndrome were associated with overall higher cancer risk. We were unable to evaluate double‐outlet right ventricle (CCLOC group 6) as we did not observe the requisite number of co‐occurring cases.

Table 2.

HR and 95% CI of Cancer Among Children With Nonsyndromic CHD, Genetic Overlap Between Anomalies and Cancer in Kids Registry Linkage Cohort

	Diagnosed with CHD, N	Diagnosed with CHD and cancer, n (%)1	Adjusted HR (95% CI)*	q value†
Left‐sided lesions	9900	26 (0.26)	2.45 (1.67–3.59)	2.8E‐05
Aortic arch obstruction	3155	14 (0.44)	4.40 (2.63–7.36)	9.4E‐07
Aortic stenosis	1081	3 (0.28)	2.65 (0.93–7.57)	0.10
Hypoplastic left heart syndrome	1643	3 (0.18)	2.01 (0.71–5.74)	0.23
Right‐sided lesions	8070	18 (0.22)	1.97 (1.24–3.16)	0.01
Pulmonary stenosis	5332	10 (0.19)	1.64 (0.87–3.10)	0.17
Other right‐sided lesions	973	4 (0.41)	3.41 (1.58–10.05)	0.008
Conotruncal defects	4292	7 (0.16)	1.34 (0.62–2.90)	0.49
Pulmonary atresia with VSD	575	3 (0.52)	5.13 (1.80–14.63)	0.006
Endocardial defects	1875	8 (0.43)	4.06 (2.07–7.96)	2.3E‐04
AVSD, complex	1153	5 (0.43)	4.25 (1.84–9.81)	0.002
Looping/arterial malposition defects	3052	9 (0.29)	2.30 (1.18–4.52)	0.03
Dextro‐transposition of the great arteries	1999	8 (0.40)	3.12 (1.53–6.39)	0.005
Dextro‐transposition of the great arteries with VSD	832	4 (0.48)	3.46 (1.21–9.85)	0.03
Dextro‐transposition of the great arteries without VSD	1045	4 (0.38)	3.58 (1.42–9.03)	0.01
Other CHD
VSD	45 143	81 (0.18)	1.53 (1.22–1.91)	7.1E‐04
VSD, muscular	12 557	13 (0.10)	1.86 (1.04–3.32)	0.06
VSD, membranous	3637	8 (0.22)	2.82 (1.31–6.08)	0.02
VSD, other	28 834	60 (0.21)	1.44 (1.12–1.85)	0.01
Single‐ventricle disease (including pulmonary atresia with intact ventricular septum)	3634	7 (0.19)	1.88 (0.92–3.85)	0.11
Single‐ventricle disease (excluding pulmonary atresia with intact ventricular septum)	3151	5 (0.16)	1.60 (0.69–3.68)	0.27

CHD indicates congenital heart disease; HR, hazard ratio; and VSD, ventricular septal defect.

^*Adjusted for state of birth, sex assigned at birth, maternal age at delivery, and maternal education. Counts were suppressed and HRs were not estimated when there were <3 exposed cases.

^† Adjusted P values based on a 5% false discovery rate correction.

We found that leukemia was associated with complex atrioventricular septal defect (AVSD; HR, 8.1 [95% CI, 2.8–23.0]) and dextro‐transposition of the great arteries (HR, 4.4 [95% CI, 1.5–12.5]) and that acute myeloid leukemia was associated with muscular VSD (HR, 13.1 [95% CI, 4.5–37.5]; Table 3). Lymphoma was associated with aortic arch obstruction (HR, 12.3 [95% CI, 4.9–31.0]) and single‐ventricle disease; the latter finding was driven by non‐Hodgkin lymphoma (NHL; HR, 28.8 [95% CI, 11.4–73.0]). We observed no statistically significant associations between nonsyndromic CHD lesions and acute lymphoblastic leukemia.

Table 3.

HR and 95% CI of Specific Cancers in Children With Nonsyndromic CHD, Genetic Overlap Between Anomalies and Cancer in Kids Registry Linkage Cohort

Cancer diagnosis	CHD diagnosis	Diagnosed with CHD, N	Diagnosed with CHD and cancer, n (%)*	Adjusted HR (95% CI)†	q value‡
Leukemia	Left‐sided lesions	9900	5 (0.05)	1.51 (0.65–3.48)	0.37
	Right‐sided lesions	8070	5 (0.06)	1.84 (0.80–4.26)	0.19
	Pulmonary stenosis	5332	5 (0.09)	2.81 (1.22–6.49)	0.03
	Endocardial defects	1875	4 (0.21)	6.43 (2.55–16.21)	3.6E‐04
	AVSD, complex	1153	3 (0.26)	8.05 (2.82–22.97)	4.1E‐04
	Looping/arterial malposition defects	3052	3 (0.10)	2.84 (1.00–8.10)	0.08
	Dextro‐transposition of the great arteries	1999	3 (0.15)	4.38 (1.53–12.48)	0.01
	VSD	45 143	23 (0.05)	1.37 (0.91–2.05)	0.17
	VSD, muscular	12 557	3 (0.02)	1.73 (0.61–4.93)	0.34
	VSD, other	28 834	18 (0.06)	1.29 (0.82–2.04)	0.33
Acute lymphoblastic leukemia	Left‐sided lesions	9900	4 (0.04)	1.60 (0.63–4.04)	0.36
	VSD	45 143	13 (0.03)	1.03 (0.60–1.75)	0.93
	VSD, other	28 834	12 (0.04)	1.14 (0.65–1.98)	0.69
Acute myeloid leukemia	VSD	45 143	6 (0.01)	3.22 (1.48–7.00)	0.008
	VSD, muscular	12 557	3 (0.02)	13.05 (4.54–37.49)	1.4E‐05
	VSD, other	28 834	3 (0.01)	2.11 (0.74–6.05)	0.21
Lymphoma	Left‐sided lesions	9900	7 (0.07)	6.16 (3.01–12.64)	6.2E‐06
	Aortic arch obstruction	3155	4 (0.13)	12.27 (4.86–30.97)	1.3E‐06
	Right‐sided lesions	8070	3 (0.04)	3.67 (1.29–10.48)	0.03
	Single‐ventricle disease (including pulmonary atresia with intact ventricular septum)	3634	4 (0.11)	9.65 (3.82–24.36)	1.3E‐05
	Single‐ventricle disease (excluding pulmonary atresia with intact ventricular septum)	3151	3 (0.10)	8.68 (3.04–24.80)	2.6E‐04
	VSD	45 143	4 (0.01)	0.83 (0.33–2.10)	0.73
	VSD, other	28 834	4 (0.01)	0.96 (0.38–2.44)	0.94
Non‐Hodgkin lymphoma	Left‐sided lesions	9900	4 (0.04)	11.03 (4.36–27.95)	4.2E‐06
	Right‐sided lesions	8070	3 (0.04)	10.93 (3.82–31.33)	5.3E‐05
	Single‐ventricle disease (including pulmonary atresia with intact ventricular septum)	3634	4 (0.11)	28.81 (11.37–72.99)	5.6E‐11
	Single‐ventricle disease (excluding pulmonary atresia with intact ventricular septum)	3151	3 (0.09)	25.92 (9.04–74.30)	2.8E‐08
Central nervous system tumor	Left‐sided lesions	9900	6 (0.06)	2.98 (1.38–6.43)	0.01
	VSD	45 143	20 (0.04)	2.01 (1.30–3.11)	0.005
	VSD, muscular	12 557	3 (0.02)	2.95 (1.03–8.43)	0.07
	VSD, other	28 834	15 (0.05)	1.83 (1.11–3.01)	0.03
Astrocytoma	VSD	45 143	6 (0.01)	1.34 (0.62–2.90)	0.49
	VSD, other	28 834	4 (0.01)	1.12 (0.44–2.82)	0.83
Ependymoma	Ventricular septal defect	45 143	3 (0.01)	3.51 (1.22–10.12)	0.03
	VSD, other	28 834	3 (0.01)	4.28 (1.48–12.36)	0.03
Neuroblastoma	Right‐sided lesions	8070	3 (0.04)	4.89 (1.71–13.96)	0.008
	Conotruncal defects	4292	3 (0.07)	6.40 (1.85–22.15)	0.008
	Ventricular septal defect	45 143	6 (0.01)	0.85 (0.30–2.43)	0.79
Hepatoblastoma	Left‐sided lesions	9900	3 (0.03)	18.53 (6.42–53.53)	9.4E‐07
	VSD	45 143	10 (0.02)	11.30 (5.88–21.75)	3.1E‐11
	VSD, other	28 834	7 (0.02)	11.65 (5.59–24.30)	1.6E‐09
Nephroblastoma	VSD	45 143	4 (0.01)	1.60 (0.63–4.05)	0.36
Rhabdomyosarcoma	VSD	45 143	4 (0.01)	3.11 (1.23–7.91)	0.03
	VSD, other	28 834	4 (0.01)	3.76 (1.48–9.57)	0.01
Soft‐tissue sarcoma	VSD	45 143	9 (0.02)	3.48 (1.83–6.62)	5.5E‐04
	VSD, other	28 834	8 (0.03)	3.73 (1.89–7.35)	5.5E‐04
Germ cell tumors	VSD	45 143	3 (0.01)	2.14 (0.74–6.13)	0.20

AVSD indicates atrioventricular septal defect; CHD, congenital heart disease; HR, hazard ratio; and VSD, ventricular septal defect.

^*Proportion of children diagnosed with each congenital heart defect who developed the index cancer before age 18 y.

^† Adjusted for state of birth, sex assigned at birth, maternal age at delivery, and maternal education. Counts were suppressed and HRs were not estimated when there were <3 exposed cases.

^‡ Adjusted P values based on a 5% false discovery rate correction.

The hazard of CNS tumors was increased 3.0‐fold (95% CI, 1.4–6.4) among children with left‐sided lesions, but sample size prohibited us from evaluating specific CNS tumors. Other and unspecified VSD was the most common subtype among children with CNS tumors and was associated with CNS tumors overall (HR, 1.8 [95% CI, 1.1–3.0]) and ependymoma (HR, 4.3 [95% CI, 1.5–12.4]) but not astrocytoma.

Concerning non‐CNS solid tumors, hepatoblastoma was associated with left‐sided lesions (HR, 18.5 [95% CI, 6.4–53.5]) and VSD (HR, 11.3 [95% CI, 5.9–21.8]). VSD was additionally associated with soft‐tissue sarcomas and rhabdomyosarcoma but not nephroblastoma or germ cell tumors. Finally, neuroblastoma was associated with conotruncal defects (HR, 6.4 [95% CI, 1.9–22.2]) and right‐sided lesions (HR, 4.9 [95% CI, 1.7–14.0]).

We were able to perform sensitivity analyses excluding cancers diagnosed during the first 1 to 2 years of life for most CHD–cancer combinations reported in the primary analysis (Table S3). Our effect estimates were largely unchanged, with a mean decrease of 3.9% compared with those based on all cancer cases. To evaluate potential geographic effects, we estimated the association between VSD and cancer (the most frequent CHD–cancer combination) separately among children in each participating state and found that HRs were similar across states (Table S4).

Discussion

We applied a detailed CHD classification algorithm in a large, population‐based cohort with active CHD surveillance to perform the most comprehensive cardiac lesion–based analysis of cancer in children with CHD to date and improve our understanding of cancer risk in affected children and adolescents. Our findings demonstrate that CHD is associated with pediatric cancer in a lesion‐specific manner. We observed several novel associations such as those between muscular VSD and acute myeloid leukemia, left‐sided lesions and hepatoblastoma, and aortic arch obstruction and lymphoma. These findings may assist in generating hypotheses about the causes of cancer in children with CHD. They also emphasize the need for careful phenotyping of CHD in epidemiological studies.

As discussed, several population‐based studies have identified associations between CHD and cancer in children. In a pooled analysis of data from >190 000 Utah, Arizona, and Iowa children (UTAZIA), the incidence rate of cancer was 2.9 times greater in those with CHD compared with those without anomalies. ² In our prior report from the GOBACK Registry Linkage Cohort, we observed a 2.5‐fold increase in the hazard of cancer in children with CHD compared with those without anomalies. ³ A nationwide Swedish cohort study including 21 982 children and young adults with CHD reported a similar association (HR, 2.2 [95% CI, 2.01–2.5]). ⁴ More recently, a cohort study using nationwide Canadian databases reported a standardized incidence ratio of 3.2 for hematologic malignancies in children with nonsyndromic CHD,⁶ and a similar analysis of South Korean data found a 65% (95% CI, 44%–89%) increase in the HR of cancer. ⁸ However, data on specific CHD phenotypes have been limited.

In the present study, left‐sided lesions (aortic arch obstruction; aortic, mitral, or supravalvular aortic stenosis; HLHS; cor triatriatum) were associated with cancer, especially hepatoblastoma and NHL. A study of 3 million Texas children reported increased cancer incidence in those with left ventricular outflow tract defects (IRR, 4.2 [95% CI, 3.2–5.5]). ⁵ Mandalenakis et al found that the risk of cancer was increased in people with coarctation of the aorta (HR, 2.00 [95% CI, 1.3–3.2]). ⁴ The UTAZIA study reported increased cancer incidence in association with aortic valve stenosis (IRR, 4.5 [95% CI, 1.1–18.1]) but not coarctation of the aorta or HLHS. ² In our own assessment, aortic arch obstruction appeared to be more strongly associated with lymphoma than left‐sided lesions overall, which has not been reported previously.

VSD was the most prevalent lesion among cases with cancer in our population, but the HR of cancer overall was only modestly increased in affected children. Our effect estimate is nearly identical to that from our earlier evaluation, conducted in a partially overlapping cohort (HR, 1.5 [95% CI, 0.7–3.0]) ³ but is somewhat weaker than others reported in the literature. Mandalenakis et al reported a 2‐fold increase in the hazard of cancer (95% CI, 1.6–2.6) ⁴ and Altmann et al reported a 4.4‐fold increase in the odds ratio (95% CI, 0.9–22.3) among children with VSD. ²¹ Carozza et al found a 3‐fold increase in the IRR of cancer (95% CI, 2.3–3.9) for children with septal defects including atrial septal defect and AVSD. ⁵ The UTAZIA investigators reported increased cancer risk in those with membranous VSD (IRR, 2.5 [95% CI, 1.0–6.0]) and VSD not otherwise specified (IRR, 2.2 [95% CI, 0.8–6.1]). ² These differences may have arisen because we classified cases with CHD by their most severe phenotype, so our VSD group primarily included isolated cases. In evaluating VSD subtypes, we found that muscular VSD was associated with acute myeloid leukemia, while VSD not otherwise specified was associated with hepatoblastoma, rhabdomyosarcoma, and other tumors. While membranous VSD was more strongly associated with cancer overall than the other subtypes included in our assessment, we were unable to evaluate specific cancers due to the number of co‐occurring cases. Collectively, these data suggest that even isolated VSD is associated with increased cancer risk.

Collectively, right‐sided lesions (Ebstein anomaly, pulmonary stenosis, pulmonary atresia with intact ventricular septum, tricuspid atresia) were associated with NHL and neuroblastoma; there was also evidence that pulmonary stenosis was associated with leukemia. Most epidemiologic studies have combined pulmonary atresia and stenosis as a single entity, ignoring clinical differences between these entities and their subtypes (pulmonary atresia with intact ventricular septum but not pulmonary atresia with ventricular septal defect [PA‐VSD] is classified as a critical congenital heart defect). The UTAZIA investigators did not identify any malignancies among children with pulmonary valve atresia (N=182), tricuspid atresia or stenosis (N=153), or Ebstein anomaly (N=141). ² In contrast, pulmonary valve stenosis was associated with a non–statistically significant increase in cancer incidence (IRR, 2.2 [95% CI, 0.5–8.9]). In our own prior assessment, children and adolescents with pulmonary atresia or stenosis were at increased risk for neuroblastoma (HR, 7.6 [95% CI, 3.8–15.3]). ³ While we did confirm an association between right‐sided lesions and neuroblastoma, we observed a stronger association for conotruncal defects, a group that includes PA‐VSD, suggesting that PA‐VSD specifically may be linked to neuroblastoma.

In our assessment, AVSD and dextro‐transposition of the great arteries, regardless of subtype, were each associated with cancer overall, and complex AVSD was associated with leukemia. In the UTAZIA study, AVSD was associated with cancer (IRR, 9.1 [95% CI, 2.2–36.6]). ² Regarding other endocardial and looping/arterial malposition defects, the UTAZIA and Swedish studies found that atrial septal defect was associated with cancer in children; the UTAZIA study also suggested transposition of the great arteries was associated with cancer (IRR, 3.7 [95% CI, 0.9–14.9]). ² , ⁴ Atrial septal defect and AVSD are prevalent in Down syndrome, ²² which could potentially explain their associations with cancer in general and leukemia in particular in these studies. Notwithstanding the possibility of children with undiagnosed Down syndrome in the cohort, our finding suggests that nonsyndromic AVSD also confers increased risk of leukemia.

Some studies suggest that complex heart disease is associated with cancer, particularly lymphoma. ² , ⁴ , ⁷ , ⁸ , ¹³ In our assessment, there was a strong association between single‐ventricle disease and NHL. Single‐ventricle disease is a leading indication for pediatric heart transplant ²³ and transplant recipients have an increased risk of NHL. ²⁴ , ²⁵ , ²⁶ While we lacked data with which to investigate the possible role of heart transplant in our population, we speculate that this contributed to our finding. Intriguingly, an epigenome‐wide association study identified hypomethylation at cg25574765 in PPFIA1, ²⁷ a proposed marker of maternal preeclampsia, among 8 subjects with complex CHD and lymphoma. This raises the possibility that epigenetic changes linked to in utero exposures may also be implicated.

Based on the observed risk difference at 18 years of 0.13% and the commonly applied but perhaps conservative estimate that 1% of live births are affected by CHD, ²⁸ we estimate that ≈45 childhood cancer cases attributable to nonsyndromic CHD will occur among the 3.6 million US children delivered annually. This estimate should be interpreted cautiously, as it does not account for cases with syndromic CHD and may be biased by our inability to fully account for death and migration in our cohort. In particular, because the mortality rate is higher in children with more complex CHD, failing to model death as a competing risk could lead to biased estimates of cumulative incidence, particularly for phenotypes with high mortality rates. However, it illustrates 2 broad points. First, while many of the relative measures presented in this and other published studies are qualitatively “large” (eg, HRs, IRRs, odds ratios ≥5), childhood cancer is rare. We estimate that the absolute risk in those with CHD is <0.5%, which would generally not warrant screening or surveillance. That said, knowledge of the genetic architectures of CHD and cancer are advancing rapidly. Our work may inform focused investigations of pathogenesis and perhaps therapeutics. Second, at the population level, hundreds of excess childhood cancers may occur in the US over a period of several years to a decade, causing substantial morbidity and death.

In trying to understand the biological mechanism of the connection between CHD and cancer, 1 possibility is that there are shared genetic pathways. Different cardiac lesions have distinct genetic architectures, some involving genes implicated in pediatric malignancies. For example, perturbations in Notch signaling have been linked to neuroblastoma dedifferentiation and, separately, to tetralogy of Fallot and truncus arteriosus (conotruncal lesions) as well as left‐sided lesions, ²⁹ , ³⁰ , ³¹ , ³² which were associated with a 6‐fold increase in the HR of neuroblastoma in our assessment. Since differentiation status influences neuroblastoma prognosis, ³³ it is possible that left‐sided lesions and conotruncal defects are associated with neuroblastoma outcomes, which would have implications for the management of affected children. Similarly, we observed a nearly 10‐fold increased hazard of NHL in children with left‐sided lesions that may be attributable in part to shared genetic risk. One case series identified pathogenic KMT2D variants in 27 of 28 children with Kabuki syndrome, of whom 19 had CHD (most commonly left‐sided obstructive lesions). ³⁴ Somatic loss of KMT2D has been described in NHL. ³⁵ , ³⁶ While cancer predisposition is not a feature of Kabuki syndrome, there is considerable phenotypic variation associated with KMT2D alterations that may be incompletely understood. ³⁷ , ³⁸ Other possibilities include shared maternal or environmental risk factors, iatrogenic exposures like ionizing radiation from recurrent catheterization and computed tomography imaging, and altered physiology resulting either from CHD or the procedures performed to repair it. ³⁹ , ⁴⁰ These findings highlight the utility of our approach for generating hypotheses for future genetic epidemiology and clinical studies.

By defining lesions that are distinct in their diagnosis and clinical management, we also hope to identify potential iatrogenic risk factors for cancer, ⁴¹ , ⁴² , ⁴³ which could suggest refinements in clinical practice. Pediatric patients with CHD who undergo radiography, computed tomography scans, or cardiac catheterization are exposed to ionizing radiation, an established risk factor for cancer. ⁴⁴ , ⁴⁵ Increased rates of DNA double‐strand breaks have been described in these patients, indicating that biologically relevant effects occur at doses used in clinical practice. ⁴⁶ This may contribute to the observed association between PA‐VSD and cancer, for example, since children with PA‐VSD may undergo serial diagnostic interventions. ⁴⁷ The Fontan procedure has been linked to maladaptive lymphatic remodeling, leukopenia, and hepatocellular carcinoma. We did not observe an association between single‐ventricle disease and liver tumors, possibility because of their rarity. ⁴⁸ , ⁴⁹ Our finding that single‐ventricle disease was strongly and specifically associated with NHL is consistent with case reports of lymphoma in Fontan‐treated children, ⁵⁰ , ⁵¹ though some population‐based studies report that neither radiation exposure nor surgery are associated with cancer in children with CHD. ⁴ , ⁴² , ⁵²

We believe our results are primarily applicable to children with nonsyndromic CHD, as we excluded individuals with known chromosomal and genetic abnormalities. However, because of the lack of systematic genetic testing in this population, changes over time in testing modalities, and our present incomplete understanding of the genetic causes of CHD, we acknowledge that our sample likely includes children with genetic conditions associated with CHD. As an example, patients with Noonan syndrome, a genetic condition due to alterations in the renin–angiotensin system/mitogen‐activated protein kinase signaling pathway, commonly have pulmonary valve stenosis and are at increased risk of juvenile myelomonocytic and other myeloid leukemias. ⁵³ , ⁵⁴ Patients with Noonan syndrome who had not received a diagnosis might account for some of the increased leukemia risk that we observed among those with pulmonary stenosis. Should this be the case, it would suggest that some children with cancer predisposition syndromes do not receive a timely diagnosis, possibly resulting in suboptimal care and outcomes.

A key strength of our study is that we evaluated clinically and pathogenetically distinct CHD phenotypes that have been aggregated in most prior epidemiologic studies. In addition, information on CHD and cancer diagnoses were collected prospectively from population‐based registries covering a sizeable fraction of US births, resulting in a large and representative sample and minimizing exposure misclassification and bias. That said, our findings should be interpreted in the context of certain limitations. Our CHD classification system relied on prenatal and early life (<1–2 years) diagnosis and procedure codes. Therefore, it may not fully capture the complexity of CHD, which can change over time as lesions are repaired and new physiology is introduced. Studies leveraging longitudinal data sources to investigate treatment‐related effects will be useful for refining associations we report here. That said, we have demonstrated that our approach leads to improved classification of critical CHD relative to methods that have previously been applied to similar registry data. ⁹ We were unable to account for death due to CHD or other causes. This may have led us to overestimate survival times in children without cancer and attenuated CHD–cancer associations. The practical impact on our analysis is likely minimal for lesions with high survival rates (eg, VSD) but may be more pronounced for others like single ventricle. Similarly, we were unable to account for residential mobility and could not identify cancers diagnosed among children who moved outside their state of birth. Because childhood cancer is rare, few cases were observed among those with some of the specific CHD phenotypes in our assessment, reducing precision. Nonetheless, given the paucity of data on many of these associations, our findings can help generate hypotheses and identify priorities for future research. Finally, increased medical attention in children with CHD or cancer could lead to incidental discovery of the other. We evaluated the possibility that incidental discovery of CHD in children with cancer affected our reporting and found little evidence this was a major driver of the observed associations. However, it is also true that more frequent clinical encounters and imaging among children with CHD could increase the likelihood of cancer detection. While we could not assess this directly, the neoplasms included in our assessment were malignant, and we anticipate that these would eventually have been detected regardless of any increased surveillance (with the possible exception of neuroblastomas, which can spontaneously regress). ⁵⁵ Therefore, we do not believe that surveillance bias substantially affected our findings.

Conclusions

Through a detailed analysis of a curated set of distinct CHD phenotypes, we aim to generate hypotheses about the causes of cancer in children and move toward realizing the eventual goal of precision screening and risk reduction strategies for childhood cancer. Further research is warranted to confirm the novel CHD–cancer associations we report. If these are replicated, investigations should be performed to characterize the underlying mechanisms, and the potential of cancer screening and surveillance strategies to improve outcomes for individuals with CHD should be evaluated.

Sources of Funding

This research was supported by the Rally Foundation for Childhood Cancer Research (Career Development Award 23CND05 to J.M.S.), the National Cancer Institute (1R01CA284531 to C.D.H. and P.J.L.), and the Department of Defense (Impact Award CA190214 to P.J.L.). The funders had no role in the decision to publish.

Disclosures

None.

Supporting information

Tables S1–S4

STROBE Statement—Checklist