About 20%–30% of children with birth defects have multiple major birth defects in more than one organ system not known to be pathogenetically related, often referred to as multiple congenital anomalies or MCAs (Agopian, Evans, & Lupo, 2018; Calzolari et al., 2014; Garne et al., 2011). Evaluating the patterns of such MCAs can provide clues to the underlying causes, pathogenic mechanisms, and developmental pathways (Friedman, 1992; Kallen & Winberg, 1969). For example, many syndromes (disorders due to a specific factor, typically but not exclusively genetic or chromosomal) are preferentially associated with certain patterns of MCAs. In addition, because several known teratogens, including some infections (e.g., rubella) and medications (e.g., retinoic acid) produce recognizable patterns of MCAs, public health surveillance of cases with MCAs has been proposed as a tool for early detection of new teratogens, which may be missed or delayed by surveillance of isolated birth defects alone (Calzolari et al., 2014; Khoury, Adams, Rhodes, & Erickson, 1987; Khoury et al., 1993, 1994; Mastroiacovo, 1991).

Several challenges make the study of MCAs particularly difficult. MCAs are relatively rare and parsing them into specific combinations of birth defects generates groups that are typically too small for meaningful statistical analysis. Analytically, identifying new MCAs relies not only on statistical analysis but also on clinical evaluation and biological understanding. Several statistical methods for evaluating patterns of MCAs have been proposed and used, but no single approach has been shown to be ideal (Agopian et al., 2018; Beaty, Yang, Khoury, Harris, & Liang, 1991; Kallen et al., 1999; Khoury et al., 1993; Khoury, James, & Erickson, 1990).

To attempt to address these challenges, we sought to explore the patterns of MCAs within the National Birth Defects Prevention Study (NBDPS), a large, population-based, case–control study of major structural birth defects (Rasmussen et al., 2003; Reefhuis et al., 2015). Its large sample size, extensive clinical information on case children, exclusion of known genetic and chromosomal syndromes, and systematic review of records by a central team of clinical geneticists improves upon some of the limitations of previous studies. By describing the selected patterns of MCAs among those eligible for NBDPS, we hope to generate hypotheses for future research in identifying causes of birth defects.

During the study period, 50,186 case children were eligible for the NBDPS (Figure 1). Of these, 7,484 (14.9%) were classified as having multiple major birth defects. We excluded 4,750 case children classified as having multiple defects who had only one NBDPS-eligible defect; Table S1 includes counts of the eligible defects among these excluded cases. In our analysis, 2,734 case children had at least two NBDPS-eligible defects, 463 had at least 3 NBDPS defects, 95 had at least 4 NBDPS defects, 23 had at least 5 NBDPS defects, 5 had at least 6 NBDPS defects, and 2 had 7 NBDPS defects (Figure 1). The birth defect group with the largest proportion of cases with MCAs in our analysis was sacral agenesis, with 69.5% of sacral agenesis cases having at least 2 NBDPS- defects, whereas only 3.2% of gastroschisis cases had at least 2 NBDPS defects (Table 2). Among the cases with MCAs in our analysis, 63.1% had a CHD, 23.1% had an oral cleft, and 20.8% had anorectal atresia/stenosis.

The 2,734 children with MCAs (defined as two eligible NBDPS defects) comprised 209 distinct 2-way combinations of birth defects. Of these, 86 patterns had 5 or more observed cases and an adjusted O:E ratio greater than 1 (Figure 1). The 463 case children with MCAs with at least 3 NBDPS defects resulted in 297 distinct 3-way MCA patterns, of which 60 patterns had 5 or more observed cases and an adjusted O:E ratio greater than 1. The 95 case children with MCAs with at least 4 NBDPS defects resulted in 179 distinct 4-way MCA patterns, of which 5 patterns had 5 or more observed cases and an adjusted O:E ratio >1. While there were case children with 5, 6, and 7 NBDPS defects, no patterns had 5 or more cases.

The MCA patterns with adjusted O:E ratios in the top 20% are presented in Table 3; Table S2 includes adjusted O: E ratios for all MCA patterns with 5 or more affected cases. The largest adjusted O:E ratio was 106.41 for the 4-way pattern that included anorectal atresia/stenosis, esophageal atresia, limb deficiency, and sacral agenesis, meaning the prevalence of this combination was 106-fold higher than would be expected if the occurrence of these birth defects were independent. Of the 31 MCA patterns with O:E ratios in the top 20%, 21 were comprised of birth defects that seemed consistent with the VATER/VACTERL association. In our sensitivity analysis of 7 MCA patterns with more than 2 birth defects (Table 4), 18 of the 42 cases were thought to be syndromic on re-review. Excluding these individual cases resulted in consistently lower adjusted O:E ratios compared to the main analysis.

Children with MCAs differed from children with isolated defects by several factors (Table 5). Compared to children with isolated defects, cases with MCAs were more frequently stillborn or terminated. Among livebirths, case children with MCAs were more frequently preterm and weighed less than 2,500 g at birth than isolated case children. Compared to isolated case children, case children with MCAs were more frequently from multiple gestations and more frequently reported a race/ethnicity other than White. There were also differences by site and birth year.

We report on the MCA patterns among the 2,734 case children within the NBDPS who had at least two NBDPS-eligible defects in different organ systems. As reported in previous studies, children with MCAs were more often stillborn or terminated compared to children with isolated defects, and liveborn children with MCAs were more often born preterm and of low birthweight compared to liveborn children with isolated defects (Calzolari et al., 2014; Toxværd & Garne, 2021). Similar to data from European registries, we did not find differences by child sex or maternal age at delivery between cases with MCAs and isolated cases (Calzolari et al., 2014). Additionally, we observed significant variations in the prevalence of MCAs by site ranging from 5% to 15%. This may be attributable to differences between sites in factors such as genetic testing, ascertainment of stillbirths and pregnancy terminations (NY and NJ had the lowest percentage of cases with MCAs and only ascertained livebirths in the beginning of the NBDPS), or racial-ethnic distribution (e.g., CA and TX have the highest percentage of cases with MCAs and the highest proportion of Hispanic cases, in whom terminations of pregnancy are less common; Calzolari et al., 2014; de Graaf, Buckley, & Skotko, 2016).

We used the strength of association (measured as adjusted O:E ratios) among individual birth defects to identify and rank specific MCA patterns. Using this approach, we found that most (21 out of the 31) MCA patterns included in the highest 20% of adjusted O:E ratios were those that included typical components of the VATER/VACTERL phenotype. This finding demonstrates that our method of comparing adjusted O:E ratios among cases with MCAs was effective in identifying clinically established birth defect associations and is similar to other studies (Benjamin et al., 2021). Also, most (15 of the 21) MCA patterns in Table 3 that included components of the VATER/VACTERL phenotype included sacral agenesis. Although it is the only eligible vertebral birth defect within NBDPS, sacral agenesis is not a typical component of the standard VATER/VACTERL phenotype (Solomon, 2011). Both sacral agenesis and certain VATER/VACTERL defects (alone or in combination), including anorectal atresia, are commonly reported in diabetic embryopathy. A recent NBDPS analysis reported that while pregestational diabetes was associated with a markedly increased risk for many specific birth defects, the strongest association was observed for sacral agenesis, with an odds ratio of 67.8 (95% confidence interval 37.0, 124.2) (Tinker et al., 2020). Among the interviewed case children with MCAs in our analysis who had sacral agenesis, 23% of mothers reported pregestational diabetes. We speculate that the VATER/VACTERL cases with sacral agenesis may represent a subset with shared etiology or another pattern entirely. Our clinical geneticist review of the two MCA patterns in Table 4 that included anorectal atresia and sacral agenesis as component features showed that the case descriptions indicated either caudal dysplasia, a mixed phenotype (caudal dysplasia and VATER), or an atypical phenotype for caudal dysplasia or VATER/VACTERL. These finding suggest an early insult to the mesoderm may have played a role in the patterns that included anorectal atresia and sacral agenesis.

Additionally, three of the seven MCA patterns in Table 4 were strongly suggestive of CHARGE syndrome (e.g., eye anomaly, congenital heart disease, choanal atresia; van Ravenswaaij-Arts et al., 2006). We observed six case children with the 4-way pattern of CHD, orofacial cleft, limb deficiency, and anotia/microtia (Table 4), of which three were judged to likely be either VATER/VACTERL association or CHARGE syndrome. Interestingly, a possible known association was suggested for only one of the five case children with the final MCA pattern (orofacial cleft, esophageal atresia, and hydrocephaly) listed in Table 4. Thus, most but not all case children with MCA patterns having the highest adjusted O:E ratios may be instances of known syndromes or associations. Those for which known syndromes or associations were not identified may represent syndromes or associations not previously recognized or variations of known patterns. Table S2 contains all observed MCA patterns involving at least two NBDPS birth defects, which may include unrecognized MCA patterns worth further exploration in other large surveillance data sets.

Although the NBDPS aimed to exclude known syndromes, the ability to do so was dependent on knowledge and genetic testing practices during the study period. The evolution of testing and knowledge over the 15-year study period meant that some unrecognized genetic syndromes may have been inadvertently included, modifying the prevalence and perhaps the distribution of MCAs. Tests such as chromosome microarray, next generation sequencing gene panels, and whole exome sequencing became increasingly utilized diagnostic tests during the study period (Malinowski et al., 2020). The diagnostic yield of approximately 25%–40% for whole exome sequencing alone in the evaluation of fetuses or newborns with MCAs suggests that at least some cases not tested by this modality were actually caused by a genetic abnormality (Fu et al., 2018; Meng et al., 2017). The NBPDS changed eligibility criteria as genetic testing evolved. For example, at the start of NBDPS, cases with CHARGE were considered to be MCAs. Following the discovery of the causative gene, CHD7, CHARGE was redefined as a genetic syndrome (OMIM #214800) (Hale, Niederriter, Green, & Martin, 2016; Kancherla et al., 2014; Sanlaville & Verloes, 2007). In our sensitivity analysis, 18 of the 42 case children with MCA patterns with elevated adjusted O:E ratios that included birth defects not part of VATER/VACTERL were thought to be syndromic on re-review. In the current analysis, the inclusion of undiagnosed syndromes may have increased the adjusted O:E ratios for certain MCAs.

Our agnostic approach provides another example of computation-driven techniques for exploring MCAs involving a large number of combinations of birth defects. Although the value of data generated by computation-driven techniques will always be dependent upon the quality of birth defect surveillance data on which the computations are based, such methods can enhance traditional coding and analysis of individual birth defects and could be particularly useful for state and country birth defect surveillance systems lacking manual review for syndromic diagnosis identification. Designating cases as syndromic or as belonging to recognized patterns enables exclusion of those cases from some analyses and a focus on certain groups for other analytic goals. Including these potentially syndromic cases in an analysis may make it difficult to identify other MCAs with a potentially teratogenic cause, as they may have been masked by the stronger syndromic cases. Additionally, it may be that inclusion of undiagnosed genetic conditions attenuated associations of more traditional risk factor-birth defect analyses by including cases for whom environmental exposures would not play a role. The recognition of potential syndromes may in turn help generate future genetic research or improve understating of developmental pathways. The observed MCA patterns may help focus future genetic studies by generating case groups of higher yield for genetic or genomic analyses.

Our analysis was restricted to case children with MCAs defined as having at least two NBDPS-eligible birth defects. Children classified as having multiple birth defects where only one defect was eligible for NBDPS were excluded from our analysis, and these cases account for the majority (63%) of children with multiple birth defects. The prevalence of non-eligible birth defects would be unknown within the NBDPS study population, as only cases with at least one NBDPS defect were included in the study population. Although this limited our ability to explore all MCA patterns, we were able to leverage NBDPS to explore the co-occurrence of over 20 categories of major structural birth defects. There were differences in the proportion of cases with MCAs across the 10 NBDPS sites, which may suggest differences in genetic testing practices, insurance coverage, or parental consent for testing.

The NBDPS employed strict inclusion criteria and case classification by clinical geneticists, so we expect little misclassification of individual birth defects. By using all eligible birth defect cases, we were able to explore MCA patterns in a large, multisite, population-based sample, which limited the potential for selection bias. The large sample allowed us to assess high order combinations of birth defects and employ an analytic approach that accounted for nonspecific clustering of birth defects.

In combining our agnostic, computation-driven analysis with clinician review of certain MCA patterns, we identified several patterns of birth defects that occurred more frequently than would be expected. Whereas most of these combinations represent clinically recognized MCA associations (e.g., VATER/VACTERL) or syndromes (e.g., CHARGE syndrome), a few others may represent unrecognized recurrent MCA phenotypes that could benefit from further etiologic investigations. For example, we observed five cases with orofacial cleft, esophageal atresia, and hydrocephaly; a nearly 13-fold higher prevalence than would be expected. This combination of birth defects does not fit with known syndromes or associations. Other two-way, three-way, and four-way combinations of birth defects were observed and had more than five affected cases. Although we did not attempt to summarize and interpret these findings, genetic analysis of some of these patterns may uncover unknown genetic causes of birth defects. These findings may be useful to birth defects surveillance programs, which classify cases, especially for surveillance programs without a clinical geneticist reviewer.