The Impact of Technology on the Diagnosis of Congenital Malformations

Abstract

As technology improves and becomes more widely accessible, more subclinical congenital malformations are being detected. Using a cohort of 1,780,156 pregnant women and their offspring nested in the 2000–2013 US Medicaid Analytic eXtract, we contrasted time trends in malformations which do not necessarily present with overt clinical symptoms early in life and are more likely to be diagnosed via imaging (secundum atrial septal defect, patent ductus arteriosus, ventricular septal defect, pulmonary artery anomalies, pulmonary valve stenosis, hydrocephalus) with trends in malformations that are unlikely to escape clinical diagnosis (tetralogy of Fallot, coarctation of the aorta, transposition of the great vessels, hypoplastic left heart syndrome, oral cleft, abdominal wall defect). Logistic regression was used to account for trends in risk factors while assessing the impact of increased screening intensity. Prevalence of the diagnosis of secundum atrial septal defect rose from 2.3‰ in 2000–2001 to 7.5‰ in 2012–2013, of patent ductus arteriosus from 1.9‰ to 4.1‰, and of ventricular septal defect from 3.6‰ to 4.5‰. Trends were not explained by changes in the prevalence of risk factors but were attenuated when accounting for screening tests. The other malformations showed no temporal trends. Findings suggest that increased screening partially explains the observed increase in diagnosis of milder cases of select common malformations.

The reported risk of congenital malformations varies dramatically between studies, with estimates ranging from around 2% to over 10% (1–7). Potential explanations for such variability include differences in the prevalence of risk factors in the study populations, different windows of ascertainment (e.g., follow-up from days to years after birth) (2), inclusion or exclusion of prenatal diagnoses and pregnancy losses (8–10), and the definition of congenital malformation itself (e.g., restriction to major malformations, exclusion of chromosomal anomalies, exclusion of prematurity-related conditions). An additional determinant of the risk estimate when ascertaining the presence of malformations is the quality of information and the need to use algorithms that might maximize specificity at the cost of sensitivity.

Another source of heterogeneity emerged in the last few decades with the advent of prenatal ultrasonography in the 1980s and echocardiography in the 1990s (8, 11). Prenatal and postnatal screening tests improve diagnosis but also increase the detection of anatomical variants with no clinical repercussions. For example, echocardiograms can identify features that are within the spectrum of normality, including small muscular ventricular septal defect (VSD), a common feature which usually closes spontaneously within the first year of life (12). However, excluding all muscular VSDs as a physiological feature would be inappropriate, since a large muscular VSD can produce congestive heart failure and therefore has medical importance. Overall, improving technology and increasing accessibility have continued to result in the detection of more defects (13, 14) and may contribute to geographical and temporal variations in the apparent prevalence of certain malformations. Thus, to allow appropriate interpretation and comparison of reported prevalence, it would be useful to determine which malformations are more prone to exhibiting changes in prevalence due to changes in methods of ascertainment (e.g., access to echocardiograms).

Using data from a large nationwide cohort of publicly insured pregnant women and their offspring, we evaluated time trends in the prevalence of specific congenital malformations and examined the impact of “technology-detected malformations”—that is, malformations that would be asymptomatic and therefore undiagnosed but become more frequently diagnosed because of technology.

METHODS

Data source and study cohort

We identified a cohort of 1,780,156 pregnant women and their liveborn infants from 46 US states and the District of Columbia nested in the 2000–2013 Medicaid Analytic eXtract. Medicaid is a joint state and federal health insurance program for low-income individuals and provides coverage for close to 50% of all pregnancies in the United States (15–17). The Medicaid Analytic eXtract is a nationwide health-care utilization database that includes information on demographic factors and Medicaid enrollment, inpatient and outpatient diagnostic and procedure claims for all hospitalizations and physician services, and filled outpatient prescriptions.

The approach used for the development of our study cohort has previously been described in detail (17). We required women to be between 12 and 55 years of age at the date of delivery and to have been continuously eligible for Medicaid from 3 months prior to the date of the last menstrual period to 1 month after delivery. Infants were required to be eligible for the first 3 months after birth, unless they died sooner. The date of the last menstrual period was estimated using a previously validated algorithm based on the delivery date and on diagnostic codes for preterm birth (18).

Outcome definition

The presence of congenital malformations was defined on the basis of inpatient or outpatient diagnoses and procedure codes from the International Classification of Diseases, Ninth Revision, Clinical Modification (ICD-9-CM) and Current Procedural Terminology, Fourth Edition, in the maternal record during the first month after delivery or in the infant record during the first 3 months after birth. In sensitivity analyses, we further assessed the presence of malformations based on malformation codes recorded in infant records within 1 year after birth. We considered both infant and maternal records because infant claims are sometimes recorded under the mother’s identification number for the first few months after birth (19). We excluded minor malformations, expected features in preterm births (e.g., patent ductus arteriosus (PDA), patent foramen ovale), and malformations for which the etiology is known—that is, chromosomal, genetic, and positional malformations.

Based on validation studies (20–22), to increase specificity, we considered a malformation to be present if 1) an ICD-9-CM diagnosis was recorded for a specific type of malformation on at least 2 different dates, 2) a diagnosis was recorded on 1 date and a relevant therapeutic procedure or surgery code for this malformation was present, or 3) a diagnosis was recorded on 1 date and the infant died after birth. If a malformation of interest was identified through maternal claims only and was also present during the first 105 days of pregnancy, the outcome was excluded under the assumption that the malformation was maternal, since it would not have been diagnosed in the fetus that early.

Five cardiac malformations and 1 noncardiac malformation (“cases”) with a presumably high proportion of technology-detected cases (5, 23) and a high prevalence among neonates (24, 25) were selected: PDA, secundum atrial septal defect (ASD), ventricular septal defect (VSD), pulmonary artery anomalies, pulmonary valve stenosis, and congenital hydrocephalus. Since ICD-9-CM codes do not provide sufficient information to distinguish secundum ASD from patent foramen ovale, we excluded secundum ASD diagnoses in preterm infants, assuming that the majority of these are physiological patent foramina ovalia likely to close spontaneously. Similarly, we only considered PDA, pulmonary artery anomalies, and pulmonary valve stenosis if they were not prematurity-related.

As negative controls, we identified 4 cardiac defects and 2 noncardiac defects unlikely to escape clinical diagnosis (5, 23): tetralogy of Fallot, coarctation of the aorta, transposition of the great vessels, hypoplastic left heart syndrome, oral cleft, and abdominal wall defect.

A more detailed description of the outcome definitions is provided in Web Table 1 (available at https://academic.oup.com/aje).

Covariate assessment

We considered the following risk factors to potentially be associated with the prevalence at birth and temporal trends of some malformations: maternal age at delivery, racial/ethnic group, smoking status, alcohol abuse or dependence, obesity, pregestational diabetes, pregestational hypertension, and multiple gestation. We further assessed the frequency of preterm births over time, since preterm infants may receive more intense screening and fetuses with malformations are more often delivered prematurely (26, 27). Lastly, since detection of health outcomes may depend on proximity to health-care facilities, we distinguished between metropolitan and nonmetropolitan counties by merging zip codes with rural-urban continuum codes provided by the US Department of Agriculture (28). To assess trends in screening intensity, we considered the use of prenatal ultrasound, 3-dimensional imaging, fetal echocardiography, and postnatal imaging.

The presence of maternal coexisting conditions was assessed during the period from 3 months before pregnancy through the end of the first trimester. Prematurity was measured during the 30 days after birth and multiple gestation during pregnancy up to 60 days after birth. For zip code, the most recent value before the delivery date was used. The presence of screening tests was assessed during the test-specific performance window (e.g., codes for postnatal screening tests were assessed during the first 3 months after delivery). Many tests were considered as proxies for intensity of screening in the population, even though some cannot detect the malformation of interest (e.g., first-trimester ultrasound) and others might have been used to confirm them (e.g., postnatal cardiac imaging technologies). The recording of heart murmurs was assessed as a proxy for clinical exploration.

Statistical analysis

We calculated the overall and calendar-year–stratified prevalence at birth (and 95% confidence interval) for each malformation of interest. Malformations with prevalence trends over time were further classified on the basis of clinical severity and persistence—for example, with a code for congestive heart failure during the first 3 months after birth or a diagnosis that persisted for ≥6 weeks after birth (see Web Table 1). We assessed time trends for those risk factors that were associated with the malformations that showed temporal variations. We further looked at trends in each of the screening and imaging tests. Since the first (2000) and last (2013) study years only contributed 2.5% of all included deliveries, we combined deliveries from 2000 and 2001 (referred to hereafter as 2000 for brevity), as well as those from 2012 and 2013 (referred to hereafter as 2013 for brevity), when assessing time trends.

To evaluate the extent to which temporal trends in malformation prevalence are explained by changes in the prevalence of risk factors and changes in screening coverage, we fitted multivariable logistic regression models, adjusting for variables in a stepwise manner: 1) unadjusted (with year of delivery as the continuous exposure variable), 2) risk factors that show a time trend, 3) prenatal screening tests, and 4) postnatal imaging. We illustrate the predicted prevalence of malformations over time with each level of adjustment, standardizing to the characteristics of those women who delivered in 2000/2001.

All analyses were conducted using SAS 9.4 (SAS Institute, Inc., Cary, North Carolina). This study was approved by the Institutional Review Board of Brigham and Women’s Hospital (Boston, Massachusetts), which waived the need for informed consent.

RESULTS

Our cohort consisted of 1,448,703 women who contributed 1,780,156 pregnancies: 1,185,013 (81.8%) of these women contributed 1 pregnancy to our study cohort, 210,721 (14.5%) women contributed 2 pregnancies, and 52,969 (3.7%) contributed more than 2 pregnancies.

As shown in Figure 1A, none of the 6 negative controls showed temporal variations. Among the 6 selected malformations with a presumably high proportion of potentially technology-detected cases, the prevalence of secundum ASD, PDA, and VSD increased over time: from 2.3‰ in 2000 to 7.5‰ in 2013 for ASD, from 1.9‰ to 4.1‰ for PDA, and from 3.6‰ to 4.5‰ for VSD (Figure 1B; Table 1). Pulmonary artery anomalies, pulmonary valve stenosis, and congenital hydrocephalus did not show any trends over time. Web Table 2 provides more details on the overall and calendar-year–stratified prevalence of each malformation of interest.

Figure 1.

Trends in the prevalence (number of cases per 1,000 infants) of selected congenital malformations unlikely to escape clinical diagnosis (“negative controls”) (A) and comprising a presumably high proportion of technology-detected cases (“case malformations”) (B), per year of delivery, among 1,780,156 mother-infant pairs in a Medicaid Analytic eXtract pregnancy cohort, 2000–2013. Bars, 95% confidence intervals.

Table 1.

Prevalence (Number of Cases per 1,000 Infants) of Selected Congenital Malformations Among 1,780,156 Mother-Infant Pairs in a Medicaid Analytic eXtract Pregnancy Cohort, 2000–2013

Malformation	No. of Affected Infants	Total		2000–2001		2012–2013
		‰	95% CI	‰	95% CI	‰	95% CI
Selected cases
Secundum atrial septal defect	7,945	4.46	4.37, 4.56	2.34	2.02, 2.65	7.50	7.11, 7.90
Severe	3,290	1.85	1.79, 1.91	1.15	0.92, 1.37	2.97	2.73, 3.22
Less severe/nonsevere	4,655	2.61	2.54, 2.69	1.20	0.97, 1.43	4.56	4.26, 4.87
Patent ductus arteriosus	5,179	2.91	2.83, 2.99	1.93	1.64, 2.22	4.07	3.78, 4.35
Severe	1,687	0.95	0.90, 0.99	0.76	0.58, 0.95	1.30	1.13, 1.46
Less severe/nonsevere	3,492	1.96	1.90, 2.03	1.19	0.96, 1.42	2.78	2.54, 3.02
Ventricular septal defect	7,057	3.96	3.87, 4.06	3.58	3.19, 3.98	4.49	4.19, 4.79
Severe	1,365	0.77	0.73, 0.81	0.99	0.78, 1.20	0.74	0.62, 0.87
Less severe/nonsevere	5,692	3.20	3.11, 3.28	2.60	2.26, 2.93	3.77	3.49, 4.04
Pulmonary valve stenosis	780	0.44	0.41, 0.47	0.42	0.28, 0.55	0.41	0.32, 0.50
Anomalies of pulmonary artery	1,239	0.70	0.66, 0.73	0.72	0.54, 0.90	0.76	0.64, 0.89
Hydrocephalus	1,234	0.69	0.65, 0.73	0.63	0.46, 0.79	0.56	0.45, 0.66
Selected negative controls
Tetralogy of Fallot	626	0.35	0.32, 0.38	0.40	0.27, 0.54	0.38	0.29, 0.47
Transposition of great vessels	708	0.40	0.37, 0.43	0.39	0.26, 0.52	0.36	0.28, 0.45
Hypoplastic left heart syndrome	522	0.29	0.27, 0.32	0.22	0.13, 0.32	0.29	0.22, 0.37
Coarctation of aorta	919	0.52	0.48, 0.55	0.46	0.32, 0.60	0.55	0.44, 0.65
Anomalies of abdominal wall	1,454	0.82	0.77, 0.86	0.71	0.53, 0.88	0.85	0.72, 0.98
Oral cleft	1,346	0.76	0.72, 0.80	0.83	0.64, 1.02	0.75	0.63, 0.87

Abbreviation: CI, confidence interval.

Severe or persistent cases of secundum ASD, PDA, and VSD accounted for a modest proportion of the cases of each of these malformations (41.4% for secundum ASD, 32.6% for PDA, and 19.3% for VSD), and their temporal trend was more moderate than that of the less severe cases (see Figure 2).

Figure 2.

Trends in the prevalence (number of cases per 1,000 infants) of secundum atrial septal defect (A), patent ductus arteriosus (B), and ventricular septal defect (C) per year of delivery, overall and by severity, among 1,780,156 mother-infant pairs in a Medicaid Analytic eXtract pregnancy cohort, 2000–2013. Bars, 95% confidence intervals.

Among the risk factors associated with the selected outcomes (Web Table 3), the prevalence of tobacco use, overweight or obesity, diabetes, hypertension, and advanced maternal age (≥35 years vs. <35 years) increased from 2000 to 2013 (e.g., diabetes from 1.6% to 2.5%, hypertension from 1.4% to 3.2%), whereas the frequency of multiple birth remained constant at approximately 2% and the frequency of preterm births at approximately 10%. We further observed changes in the racial/ethnic composition of women in our cohort, with the proportion of black women decreasing from 39.6% in 2000 to 29.4% in 2013. The relative frequency of women living in metropolitan areas versus urban or rural areas increased slightly in more recent years (from 82.1% to 87.7%). For more details, see Table 2 and Web Table 4.

Table 2.

Prevalence (Number With Characteristic per 100 Women) of Selected Maternal Characteristics Among 1,780,156 Mother-Infant Pairs in a Medicaid Analytic eXtract Pregnancy Cohort, 2000–2013

Maternal Covariate	No. of Women	Total		2000–2001		2012–2013
		%	95% CI	%	95% CI	%	95% CI
Maternal age ≥35 years	116,504	6.54	6.51, 6.58	5.75	5.59, 5.90	8.25	8.13, 8.38
Race/ethnicity
White	700,442	39.35	39.28, 39.42	36.99	36.67, 37.30	39.83	39.61, 40.06
Black	587,489	33.00	32.93, 33.07	39.60	39.28, 39.92	29.39	29.18, 29.59
Hispanic	345,341	19.40	19.34, 19.46	15.75	15.51, 15.99	11.59	11.45, 11.74
Other or unknown	146,884	8.25	8.21, 8.29	7.66	7.48, 7.83	19.19	19.01, 19.37
Maternal comorbidity
Smoking	62,584	3.52	3.49, 3.54	1.44	1.37, 1.52	5.61	5.51, 5.72
Overweight or obesity	39,195	2.20	2.18, 2.22	0.92	0.86, 0.98	4.23	4.14, 4.32
Diabetes	36,606	2.06	2.04, 2.08	1.59	1.51, 1.68	2.54	2.47, 2.61
Hypertension	42,479	2.39	2.36, 2.41	1.44	1.36, 1.52	3.20	3.12, 3.28
Pregnancy conditions
Preterm birth	188,266	10.58	10.53, 10.62	10.55	10.35, 10.76	10.24	10.11, 10.38
Multiple gestation	33,400	1.88	1.86, 1.90	2.06	1.97, 2.15	1.86	1.80, 1.92
Residence in metropolitan area	1,516,112	85.17	85.12, 85.22	82.14	81.89, 82.39	87.72	87.57, 87.86

Abbreviation: CI, confidence interval.

We further observed increasing trends in pre- and postnatal imaging tests (Web Table 4): The prevalence of first-trimester ultrasound increased from 12.1% of all pregnancies in 2000 to 59.9% in 2013; during that same period, the prevalence of second- or third-trimester ultrasound increased from 79.5% to 90.0%, the prevalence of prenatal echocardiography increased from 2.8% to 7.1%, and the prevalence of postnatal cardiac imaging increased from 4.1% to 5.5%. The prevalence of 3-dimensional ultrasound increased from 0.3% in 2006, when codes for 3-dimensional screening became available, to 0.7% in 2013.

Results from the logistic regression analyses indicated that the observed prevalence increase of secundum ASD, PDA, and VSD could not be attributed to changes in the distribution of measured risk factors over time (Figure 3). However, accounting for increased use of prenatal screening and postnatal cardiac imaging tests did attenuate the predicted prevalence increase of secundum ASD and PDA and resulted in a stable prevalence of VSD over time.

Figure 3.

Trends in the predicted prevalence (number of cases per 1,000 infants) of secundum atrial septal defect (ASD) overall (A), severe secundum ASD (B), less severe/nonsevere secundum ASD (C), patent ductus arteriosus (PDA) overall (D), severe PDA (E), less severe/nonsevere PDA (F), ventricular septal defect (VSD) overall (G), severe VSD (H), and less severe/nonsevere VSD (I), per year of delivery, among 1,780,156 mother-infant pairs in a Medicaid Analytic eXtract pregnancy cohort, 2000–2013. Results were standardized to the population of mother-infant pairs with deliveries in 2000–2001, with stepwise adjustment for risk factors (age at delivery, race/ethnicity, smoking status, overweight/obesity, pregestational diabetes, pregestational hypertension), prenatal screening tests (first-trimester ultrasound, second-/third-trimester ultrasound, 3-dimensional ultrasound, fetal echocardiography), and postnatal cardiac imaging.

The trends were attenuated for severe or persistent cases of secundum ASD, PDA, and VSD after accounting for pre- and postnatal imaging. Extending the malformation assessment window to 1 year after birth led to similar trends, with a slight increase in the prevalence of secundum ASD and VSD (on the order of 1 per 1,000) but not of PDA (see Web Figure 1). The prevalence of heart murmurs increased from 32.4‰ in 2000 to 44.4‰ in 2013 (see Web Figure 2).

DISCUSSION

Key findings

Our findings suggest that it was mainly the diagnosis of cardiac malformations that may not present with overt symptoms that increased between 2000 and 2013. The diagnosis of nonsevere secundum ASD, PDA, and VSD—3 of the most common cardiac malformations (24, 29, 30)—strongly increased over time, whereas none of the negative controls showed any trends. We did not find a trend for less common malformations (pulmonary artery anomalies, pulmonary valve stenosis, and hydrocephalus) (30–32), which we had identified as potentially sensitive to screening detection, since they might already be detectable with older routine screening and some might be difficult to diagnose despite more and better screening (e.g., mild pulmonary valve stenosis).

The increase in measured clinical risk factors for malformations over time did not explain the observed malformation trends. Intensified pre- and postnatal imaging appeared to contribute in part to the observed trends. Models indicated that of all cases of secundum ASD, PDA, and VSD diagnosed in 2013, only 75% would have been detected if the imaging intensity had stayed the same since 2000. Although prenatal imaging cannot diagnose PDAs (as the ductus arteriosus is physiological prenatally) and might not detect all ASDs and VSDs (33–35), prenatal screening intensity is probably a proxy for neonatal examination intensity, including cardiac imaging, which would lead to a higher detection of these specific malformations.

A slight increase in secundum ASD and PDA remained after accounting for screening and diagnostic tests. While this could theoretically be ascribed to time trends in unmeasured risk factors for cardiac malformations, the lack of a time trend in the negative control malformations suggests this is unlikely. Residual trends may be due to improvements in the imaging technology resolution, better training of technicians, and increased clinical emphasis on detection of cardiac defects in newborns, as suggested by the increased recording of heart murmurs over time (13, 31, 36).

When extending the assessment period of ASD, VSD, and PDA to 1 year after birth, prevalence trends were very similar to what was observed when identifying cases up to 3 months after birth, and there was no decline in the number of malformations detected after 3 months. This suggests that the additional diagnoses in recent years were not malformations that were going to be diagnosed in the following months anyway but rather malformations that were going to close spontaneously or at least remain asymptomatic during the first year of life.

Findings from other studies

In 1979, the Centers for Disease Control and Prevention reported a 2-fold increase in the incidence of VSD between 1970 and 1977 (37, 38). At that time, it could not be determined whether this was caused by an increase in biological risk factors or an improvement in detection or reporting over time. Layde et al. (39) concluded that the increase could not simply be explained by increasing assessment of milder cases of VSD because there was no change in the spontaneous closure rate of VSD by the age of 1 year during that period. However, subsequent studies suggested that the observed increase in malformations was more likely to be due to better ascertainment (e.g., through better clinical training and increasing availability of echocardiography) than to a real increase in VSDs (40, 41). In a literature review, Hoffman and Kaplan (41) reported that the incidence of cardiac malformations increased from 4‰–5‰ in the 1950s to 12‰–14‰ in the late 1990s, and that the increase was driven primarily by the number of small VSDs and other nonsevere defects included. Using a more recent population-based registry in metropolitan Atlanta, Georgia, Bjornard et al. (24) found that the prevalence of major cardiac malformations at birth had increased from 5.0‰ in 1978–1983 to 8.6‰ in 2000–2005. VSDs, secundum ASDs, and pulmonary valve stenosis accounted for most of this increase, whereas the prevalence of more severe defects like tetralogy of Fallot, hypoplastic left heart syndrome, and coarctation of the aorta did not change over time.

Limitations and strengths

The strengths of this study include the large cohort size, with almost 1.8 million liveborn infants linked to their mothers, and the availability of detailed clinical and demographic patient-level information in the Medicaid Analytic eXtract. This allowed us to assess trends in both common and uncommon malformations while controlling for trends in relevant risk factors. Furthermore, the assessment of data from a 14-year period (2000–2013) allowed longitudinal evaluation during a time of important technological advances (13).

Nevertheless, our study was also subject to certain limitations. Diagnoses based on administrative insurance claims are prone to outcome misclassification. However, we previously showed that algorithms that prioritize specificity for cardiac malformations overall, as well as specific malformations such as VSD, have an acceptable positive predictive value (78% for cardiac malformations overall and 76% for VSD) (21). Furthermore, for outcome misclassification to affect the interpretation of the findings, the positive predictive value would have had to change over the course of the study period, which is unlikely.

Additionally, when using claims data, we often cannot differentiate between subtypes of a specific malformation. For instance, there may be patent foramina ovalia coded as secundum ASDs, we do not have information on the size of the ASDs or PDAs, and ICD-9-CM codes do not distinguish between subclinical muscular VSDs and more severe types such as perimembranous VSDs. Therefore, we had to use proxies for severity, such as cardiac surgery or congestive heart failure. The use of an imperfect measure of severity could explain why we still see a small residual increase in some of the more severe malformations (i.e., there might still be some subclinical patent foramina ovalia left among secundum ASDs classified as severe).

Lastly, when restricting the cohort to live births, severe congenital malformations that result in fetal death or pregnancy termination will be missed, and the observed malformation prevalence and prevalence trends in live births could therefore be misleading. Better detection of malformations through screening followed by termination would not affect PDA, since it is a condition that can only be diagnosed after birth, when the ductus arteriosus does not close spontaneously. Pregnancy terminations are further not expected to affect ASD or VSD, since these conditions are unlikely to be diagnosed prenatally and the majority of these defects are considered non–life-threatening and treatable (10, 33, 42, 43). However, terminations due to improvements in screening could affect more severe malformations, such as severe cardiac defects (e.g., tetralogy of Fallot, coarctation of aorta). The lack of trends in these severe malformations among liveborn infants could be the result of technological improvements, with more and earlier prenatal malformation detection leading to more therapeutic abortions. Yet, with constant improvements in surgical procedures, terminations due to prenatally diagnosed severe cardiac defects are becoming increasingly rare (44–49).

Implications

The increasing detection of small, often clinically asymptomatic defects can have clinical implications and can affect health-care costs, since more neonates may receive follow-up and interventions. However, while the newly diagnosed malformations may be mainly subclinical, a proportion may represent severe conditions, and thus better imaging might also be improving health outcomes and saving lives.

Improved detection can also have research implications, since inclusion of these subclinical malformations would increase estimates of the prevalence of birth defects in such studies and might even result in heterogeneity of effect estimates among studies evaluating teratogens. If malformations are a continuum and even very small defects can be affected by the exposure, then improvements in technology could help identify teratogens. On the other hand, if subclinical malformations such as small VSDs are just variations of the normal state, then inclusion would reduce the specificity of the outcome and impair the identification of teratogens. Thus, depending on the assumptions and aims, one may want to exclude or include these subclinical defects when studying risk factors for malformations.

Another important determinant potentially causing heterogeneous effect estimates is the definition of congenital malformation itself. Using ICD-9-CM codes 740.xx to 759.xx to establish the prevalence of malformations in newborns would result in estimates of up to 13% (7). However, further exclusion criteria are usually applied (5, 23). First, minor anomalies and birthmarks—which can represent 50%–70% of all malformations identified on the basis of ICD-9-CM codes (5, 7)—are typically excluded because of decreased clinical relevance (50). Second, chromosomal abnormalities or genetic syndromes are often excluded from etiological studies under the assumptions that their etiology is known and that teratogens only affect malformations of nongenetic origin. These malformations account for approximately 4% of all malformations (7, 51). A third group excluded because of their known cause are positional anomalies, which account for around 2.5% (1, 5, 7). Fourth, features considered physiological in preterm infants are often excluded because most will resolve spontaneously (7). These account for around 2% of the cases (5).

An additional and unrelated complication when ascertaining the presence of malformations is the quality of information. Many studies rely on coded diagnoses, either in medical records or in administrative insurance claims. Errors in coding and inclusion of rule-out diagnoses can account for approximately 8% of the recorded malformations identified on the basis of a single ICD-9-CM code (7). Validation studies have demonstrated that requiring diagnostic codes for the malformation of interest on at least 2 dates or 1 code and a surgical procedure increases the positive predictive value of the outcome-defining algorithm for a range of major malformations (20–22).

Thus, investigators need to pay attention to the inclusion or exclusion criteria used when assessing malformation prevalence before comparing results from different studies.

Abbreviations

ASD

atrial septal defect

ICD-9-CM

International Classification of Diseases, Ninth Revision, Clinical Modification

PDA

patent ductus arteriosus

VSD

ventricular septal defect