Evaluating the proportion of isolated cases among a spectrum of birth defects in a population-based registry

Abstract

Introduction:

Because the etiology and outcomes of birth defects may differ by the presence vs. absence of co-occurring anomalies, epidemiologic studies often attempt to classify cases into isolated versus non-isolated groupings. This report describes a computer algorithm for such classification and presents results using data from the Texas Birth Defects Registry (TBDR).

Methods:

Each of the 1,041 birth defects coded by the TBDR was classified as chromosomal, syndromic, minor, or “needs review” by a group of three clinical geneticists. A SAS program applied those classifications to each birth defect in a case (child/fetus), and then hierarchically combined them to obtain one summary classification for each case, adding isolated and multiple defect categories. The program was applied to 136,121 cases delivered in 2012–2017.

Results:

Of total cases, 49% were classified by the platform as isolated (having only one major birth defect). This varied widely by birth defect; of those examined, the highest proportion classified as isolated was found in pyloric stenosis (87.6%), whereas several cardiovascular malformations had low proportions, including tricuspid valve atresia/stenosis (2.3%).

Discussion:

This is one of the first and largest attempts to identify the proportion of isolated cases across a broad spectrum of birth defects, which can inform future epidemiologic and genomic studies of these phenotypes. Our approach is designed for easy modification for use with any birth defects coding system and category definitions, allowing scalability for different studies or birth defects registries, which often do not have resources for individual clinical review of all case records.

1 |. INTRODUCTION

Birth defects with surgical, medical, or serious cosmetic importance are usually considered to be “major” defects, whereas other defects are classified as “minor” defects. Because the etiology of birth defects may be heterogeneous (e.g., Ferencz, 1993; Holmes, Driscoll, & Atkins, 1976; Jones, 1988), many studies attempt to clinically classify cases (infants/fetuses) into having “isolated” vs. “multiple” major birth defects. Specifically, defects are isolated when no additional major defect is present (Tinker et al., 2015), whereas the co-occurrence of two or more major defects represents multiple defects. Classification may be extended to include other categories. For example, the National Birth Defects Prevention Study (NBDPS) excluded cases with chromosomal abnormalities or genetic syndromes, and clinical geneticists classified remaining cases as having birth defects that were isolated, multiple, teratogen-related, or defined associations (Rasmussen et al., 2003). The NBDPS has applied additional manual review to ensure accurate diagnoses and further categorize certain defects (e.g., cardiovascular malformations [Botto et al., 2007]).

In an ideal setting, case classification would be done at the level of the individual by clinical geneticists; however, that is not feasible for many birth defect registries due to limited resources, clinician availability, or sheer volume. The European Surveillance of Congenital Anomalies and Twins (EUROCAT), a consortium of birth defect surveillance systems, developed a computer algorithm to classify cases (Garne et al., 2011). We used a similar approach on data from the Texas Birth Defects Registry (TBDR) to identify the proportion of isolated cases by type of birth defect.

2 |. METHODS

Data on all cases with birth defects delivered in Texas in 2012 through 2017 were obtained from the TBDR. Details of the TBDR have been published previously (Texas Birth Defects Epidemiology and Surveillance Branch, 2020). The algorithm for defining isolated cases involved the following steps.

1. Classifying Birth Defects: In order to define major birth defects, the 1,041 birth defect codes used by the TBDR were entered into an Excel spreadsheet (one row for each six-digit birth defect code). TBDR uses the British Pediatric Association (BPA) codes as modified by the CDC. Columns were created for the following mutually exclusive categories: review (needing review by a clinical geneticist [CG] to determine classification), chromosomal, syndromic, and minor. Thus, defects not assigned to any of these categories were considered nonsyndromic major defects. Each of three CGs working with the TBDR was asked to indicate which category should be applied to each birth defect. In a modified Delphi technique, the results were collated and birth defects with disagreements were again sent out to the CGs, who could change their classification upon seeing the results from the other CGs, until agreement was reached and a single list was finalized.

2. Application of Algorithm: A SAS (version 9.4) program was developed for case classification. It first identified all definitely diagnosed birth defects; such birth defects lack diagnostic qualifiers such as “rule out” or “suggestive of” in the medical record and comprise roughly 95% of diagnoses. Each birth defect recorded for a case in the TBDR was assigned its relevant case classification as listed in the spreadsheet. The program then applied logic to examine all birth defects within each case. If one or more birth defects was classified as “review,” the case was assigned “review,” regardless of how any of the other co-occurring defects were classified. Taking the remaining cases, if one or more defects was classified as chromosomal, then that case was assigned “chromosomal.” This was done next for the remaining cases with one or more syndromic defects. After that, if the remaining cases had two or more major defects, they were assigned “multiple”; if they had only one major defect (with or without minor defects), they were assigned “isolated.” The remaining cases were assigned “other,” which were cases that had only one or more minor defects; this results from the TBDR trying to collect diagnosis data as comprehensively as possible. Given the lack of systematically defined birth defect sequences, the computer program did not differentiate between multiple defects as opposed to cases with isolated sequences (e.g., spina bifida sequence) or cases with only potentially related birth defects within the same organ or organ system (e.g., multiple heart defects). However, the program could be modified by the user for these purposes.

Once all cases were classified into the above mutually exclusive groups, the proportions of isolated cases out of total cases were calculated, both overall and for select common definitely diagnosed birth defects.

3 |. RESULTS

The SAS program can be found as Table S1; the classification spreadsheet is available upon request to the corresponding author. To demonstrate its utility, the algorithm was applied to data from the TBDR. There were 136,121 TBDR cases delivered in 2012 through 2017 and submitted to the computer algorithm. It took less than 30 seconds to run, including reading the data and completing the classification. Almost half of the cases (49.0%) were classified as isolated, having only one major birth defect (Table 1). That was followed by 25.0% having two or more major birth defects (multiple). Chromosomal and syndromic cases each made up less than 6.0%. If the “Other” category was excluded, the percentage in the remaining categories increased proportionally, with isolated cases now comprising 59.4%.

TABLE 1.

Classification of cases in Texas, 2012–2017

Classificationa	Number of cases	% of all cases	% of cases excluding “other”b
Total	136,121	100.0	100.0
Needs review by a clinician	870	0.6	0.8
Chromosomal	7,532	5.5	6.7
Syndromic	3,118	2.3	2.8
Multiple (2 or more major birth defects)	34,076	25.0	30.3
Isolated (only 1 major birth defect)	66,753	49.0	59.4
Other (only minor birth defects)	23,772	17.5	(excluded)

^aCategories presented in order of assignment by the computer algorithm.

^bBased on 112,349 cases.

The percentage of cases classified as isolated was highest among cases with pyloric stenosis (87.6%), hypospadias (76.4%), and epispadias (74.6%) (Figure 1). Cases with trisomies and achondroplasia were included for completeness of presentation of commonly reported birth defects but were assigned to the chromosomal or syndromic categories; thus, by our operational definition, they had 0% isolated cases. Several cardiovascular malformations had low proportions of isolated cases, including tricuspid valve atresia/stenosis (2.3%) and common truncus (5.4%). Non-cardiovascular malformations that had low proportions of isolated cases included bladder exstrophy (5.7%) and anophthalmia (6.3%).

FIGURE 1.

Commonly reported birth defects by descending percentage of isolated cases, Texas 2012–2017

4 |. DISCUSSION

While several studies have evaluated the proportion of isolated cases for selected birth defects, there have been few to explore co-occurring anomalies across the spectrum of birth defects; this can provide insight into relative differences by type of birth defect, as well as new etiologic insights. Overall, and as expected, the proportion of isolated cases identified in this study varied by type of birth defect. Additionally, because of the size of our assessment, we were able to evaluate the proportion of isolated cases among less frequent birth defects, including epispadias and biliary atresia, which have few reports of co-occurring anomalies.

Another important component of this report is the algorithm used to classify cases of birth defects. A computer algorithm has also been used to classify all 17,733 cases with birth defects in 2004 in the EUROCAT database, collected from 25 registries (Garne et al., 2011). It differed from our approach in that birth defects were coded using ICD-10 instead of BPA codes, which are based on ICD-9. Their algorithm classified 76% of all cases as having isolated congenital anomalies (higher than our value of 49%, but closer to our value of 59.4% excluding the other category). The EUROCAT approach also classified 11% with multiple anomalies but 7% after manual review by clinicians (which they did but we did not; their final value was lower than our value of 25.0% or 30.3% excluding other), 15% as having chromosomal anomalies (higher), and 2% as monogenic syndromes (similar). There are several possible reasons for inconsistencies. If we combined our isolated category (one major defect) with other (only one or more minor defects), the resulting 66.5% is closer to the EUROCAT value of 76%; alternatively, excluding other resulted in 59.4%, almost as close. That suggests that most of the inconsistency might relate to what is called major vs. minor by different systems, as well as what is defined as syndromic or chromosomal. By including sequences and defects within the same organ system as isolated (which we did not do), the EUROCAT percentage of isolated cases would be higher. Differences in scope and granularity of surveillance/ascertainment would also play a large role. Because EUROCAT prioritizes cases with multiple birth defects, extra care including manual review was undertaken to be accurate and specific, possibly leading to a lower proportion with multiple defects than ours. The higher EUROCAT percentage of chromosomal anomalies might also have been related to older mothers, more complete ascertainment of terminated pregnancies, and more easily available lab results among terminated pregnancies. Note, however, that the order of classes was similar: most in isolated, followed by multiple, chromosomal, and syndromic in our study, with multiple and chromosomal switching in the EUROCAT study.

The term “minor” defects can be used in different ways. (A) Birth defects without surgical, medical, or serious cosmetic importance are minor with respect to severity; however, this can be problematic with using a code-based system to classify defects such as atrial septal defect which can have a range of severity. (B) Birth defects which co-occur with only one major birth defect, would typically result in that child being considered an isolated case; that was the usage in this paper. We initially considered other terms instead of “minor” such as “non-multiple,” but felt that was too awkward and confusing.

Our use of a code-based rather than a comment- or text-based approach allowed for objective and systematic classification. However, it would miss several syndromes as well as major birth defects that are within heterogeneous codes such as 756.080 “Other specified skull and face-bone anomalies”. As mentioned above, by not considering whether birth defects co-occurred within the same organ system (such as heart defects) as is done by some other investigators, the current version of the algorithm undercounted isolated cases. The same is true of sequences; for example, a child with only spina bifida and club foot would be assigned multiple. Those enhancements are planned for future versions. In the absence of systematic genetic laboratory diagnoses to indicate the presence of a chromosomal abnormality or genetic disorder, there may be a tendency for cases with those disorders to be classified instead as multiple. An automated approach has inherent limitations, and should only be used to enhance and not replace work by human clinical geneticists. For example, it could be used as an initial screening tool to identify cases that are easily identified as isolated so that fewer cases can be sent for manual clinical review. Finally, the appropriate classification depends on the goals of the data users. Etiological or outcomes research, which may want to identify a case with spina bifida and club foot as isolated, has different needs than studies of the burden on communities or health care which may prefer to treat that as multiple. But different goals illustrate the benefits of a flexible classification approach such as presented in this paper.

A strength of our algorithm is that it can easily be enhanced in several ways, particularly taking advantage of the flexibility offered by the separate spreadsheet. (1) The rows (birth defects) in the spreadsheet can be modified to fit the defects and codes used by any registry (e.g., ICD-9-CM, ICD-10, and CDC-BPA). (2) The columns (categories) or algorithm can also be modified; different registry systems may identify different defects as being minor/major, chromosomal anomalies, or syndromic. This could also help for example, as more chromosomal causes for specific birth defects are identified. (3) The efficiency of this approach means it could easily be run multiple times to achieve different goals if desired. For example, a first run could produce the original six categories mentioned above; then by using a modified spreadsheet, the user could classify cases ignoring chromosomal/syndromic status. (4) The SAS program might be modified to consider sequences or whether multiple birth defects occur in the same or different body systems (for example, to better account for heart defect complexity). Because much of the logic is handled by the spreadsheet, the resulting SAS program is quite simple and could easily be converted to other software such as R or Stata.

Another major strength of this approach is its efficiency as demonstrated by the number of cases that were included in this report (N = 136,121); it is one of the largest population-based assessments of case classification among a range of birth defects. Because of that, we were also able to evaluate less frequent birth defects. Finally, by automatically classifying the majority of cases, this approach could dramatically increase the efficiency of CGs by flagging only those cases most needing their review.

Abbreviations:

CG

clinical geneticist

EUROCAT

European Surveillance of Congenital Anomalies and Twins

NBDPN

National Birth Defects Prevention Network

NBDPS

National Birth Defects Prevention Study

TBDR

Texas Birth Defects Registry

DATA AVAILABILITY STATEMENT

The data that support the findings of this study are available from the corresponding author upon reasonable request, and may require TXDSHS IRB review.