Organic solvents are carbon-based chemicals that are widely used in occupational settings for dissolving or dispersing substances such as fats, oils or waxes, as well as in chemical manufacturing. Millions of workers in the United States have the potential for exposure to organic solvents used in such products as paints, varnishes, adhesives, and degreasing/cleaning agents, as well as in the production of dyes, polymers, plastics, textiles, printing inks, and agricultural products.¹ Most organic solvents are highly volatile, and inhalation is the most common occupational exposure route, although exposure can also occur dermally or orally.¹ Several organic solvents have been classified as probable reproductive hazards.²

Congenital heart defects (CHDs), with an overall birth prevalence of approximately 1%, are among the most common types of birth defects^3,4 and are a leading cause of birth defects-associated mortality, morbidity and costs.^5–7 Reported associations between CHDs and occupational solvent exposures have been inconsistent. Studies using linked data between the Finnish Register of Congenital Malformation and the Children’s Cardiac Register (1982–1984) identified occupational exposure to organic solvents as a risk factor for CHD subtypes, including ventricular septal defects (VSDs) ^8,9 and conotruncal defects (with exposure to “dyes, lacquers or paints”).¹⁰ Results from the Baltimore Washington Infant Study (BWIS; 1981–1989) indicated associations between organic solvents and all CHDs combined, as well as with left ventricular outflow tract (LVOT) obstruction defects (specifically, hypoplastic left heart syndrome and coarctation of the aorta), conotruncal defects (specifically, tetralogy of Fallot and transposition of the great arteries),^11,12 pulmonary stenosis,^11–13 total anomalous pulmonary venous return,¹⁴ and Ebstein’s malformation.¹⁵ Neither the Finnish registry studies nor the BWIS made use of industrial hygiene expertise to assess the potential for exposure to various solvents associated with jobs held by study participants during pregnancy and based their analyses entirely on maternally reported job titles and self-reports of occupational exposures. Other studies using similar exposure assessment methods relying exclusively on self-reported job titles and exposures have not suggested a role for maternal exposure to organic solvents in the etiology of CHDs.^16,17

The National Birth Defects Prevention Study (NBDPS), an on-going, multisite, population-based case control study exploring both genetic and non-genetic risk factors for birth defects–. As part of the study, industrial hygienists estimated maternal occupational exposures to organic solvents for study participants who delivered between 1997 and 2002. Funding for the occupational exposure assessment was limited, and therefore only completed for data from the first five study years. Given these exposure data and the detailed review and refined classification of all CHD cases in the study,^18,19 the NBDPS dataset provides a unique opportunity to investigate the association between maternal occupational exposure to organic solvents and CHDs. Specifically, we investigated possible associations between 15 categories of CHDs and classes of organic solvents for which exposures are likely to occur in occupational settings: chlorinated solvents, aromatic solvents, and a mixture of C₁₀ or higher hydrocarbons known as Stoddard solvent.

Overall, our results indicate that maternal occupational exposure to organic solvents during the period of one month before conception through the first trimester of pregnancy is a potential risk factor for some specific CHD phenotypes. The observed associations with individual CHDs may warrant further investigation, given that some of these associations have been previously reported in the literature. We observed associations with specific subtypes of LVOT obstruction defects, RVOT obstruction defects, conotruncal defects, and septal defects. Associations between solvents and LVOT obstruction defects were previously observed in the BWIS – specifically, with “degreasing agents” (which may include the solvents analyzed in the current study as well as others not analyzed here). BWIS data also showed an association with hypoplastic left heart syndrome, an LVOT obstruction defect subtype, for which we did not see any associations in our data. In addition, BWIS data showed an association between a high cumulative solvent level and a different LVOT obstruction defect subtype, coarctation of the aorta.^11,12 Our results for Stoddard solvent and RVOT obstruction defects, specifically pulmonary valve stenosis, are consistent with findings from the BWIS, which also reported an association between pulmonary valve stenosis and “degreasing agents”.¹¹ Our results for perimembranous VSD were previously suggested by Finnish data (though their results were reported for VSDs in the aggregate, not by VSD subtype)⁸, but not in data from the BWIS.

With respect to the similarity of associations across exposure assessment methods, arguably septal defects and RVOT obstruction defects were more consistently associated with solvents than LVOT obstruction defects or conotruncal defects. Given that the previous literature has reported associations across several subtypes, these results add to the literature suggesting that associations are present, but do not clarify which of these associations are more likely to be causal.

The exact biological mechanism whereby exposure to organic solvents during the periconceptional period could result in CHDs is unknown. One potential mechanism of interest is oxidative stress.³⁵ Several chlorinated solvents are known to cause oxidative stress in animal models. Additionally, there is evidence from animal models that maternal exposure to chlorinated solvents is associated with CHDs. For instance, toxicologic studies using both chick embryos and fetal rats have shown an increased risk for CHDs in animals exposed to the chlorinated solvent tricholoroethylene and its metabolites. No particular CHD subtypes have predominated - septal defects, as well as left-sided and right-sided obstructive defects have been reported.^36,37

The potential for exposure misclassification in one or both exposure assessment strategies is the most important potential limitation of the current study.³⁸

A recent analysis of the NBDPS expert consensus data found that inter-rater reliability of exposure status (yes/no) improved substantially when discordant ratings were resolved via consensus conference. Kappa coefficients increased from 0.59 to 0.81 for chlorinated solvents, and from 0.55 to 0.92 for Stoddard solvent when comparing estimates developed before the conference with a set of estimates developed after the conference on different jobs.²⁴ Exposure misclassification was therefore reduced to some degree by using the consensus assessments as the final estimated exposure status; however, some inaccuracy in true exposure assignment likely remains, resulting in the potential for residual exposure misclassification.

Both the consensus-based and the literature-based approach were subject to exposure misclassification due to the correlation of exposures within a solvent class. There was generally insufficient detail available from the maternal questionnaire responses for the industrial hygienists to identify specific chlorinated solvents, singly or in combination, to which the mother was exposed during her job, especially because several chlorinated solvents may be used interchangeably in some tasks and jobs, such as degreasing. This overlap would have made the interpretation of associations with individual solvents challenging and therefore, the decision was made to conduct an analysis based on solvent classes instead. Although this decision mitigated the problem of interpreting individual solvent results, it may have resulted in a dilution of any true associations with individual solvents, as was likely the situation in the expert consensus-based approach as well. In addition, both exposure assessment methods were indirect and retrospective, and yielded estimates of exposure with unknown sensitivity and specificity compared to true exposure status. Since there was no gold standard against which the exposure assessments could be compared, nor was there a biological marker to use for validation, it was not possible to determine the accuracy of one assessment method over another.

We expected that exposure misclassification in this study would dilute the dichotomous (yes/no during the periconceptional period) exposure contrast and, assuming there is an underlying positive association between maternal exposure to organic solvent(s) and one or more CHDs, the observed effect estimates would be biased toward the null. Our sensitivity analysis of the literature-based method was intended to sharpen the exposure contrast by limiting exposed jobs to those with estimated probability of exposure ≥50%. In this secondary analysis, two of the originally observed relationships were strengthened in magnitude (i.e. Stoddard solvent and RVOT obstruction defects and pulmonary valve stenosis), though the loss in sample size reduced precision. Several previously unseen associations were noted (e.g. Stoddard solvent and septal defects, and atrial septal defects), which may have been unobserved in the primary analysis due to exposure misclassification.

Several additional study limitations should be noted. First, our decision to only consider a dichotomization of exposure limited our ability to explore an exposure-response relationship, which is one criterion of causality. Although information was available on the metrics that could allow such an investigation (intensity and frequency of exposure), our limited sample size of exposed mothers within each CHD subtype restricted our ability to further classify them by these metrics. In addition, there was insufficient variation in these other metrics to adequately conduct a CHD-subtype-specific exposure-response analysis.

Second, we were unable to account for potential exposures to solvents outside of the work environment, such as through painting or cleaning/degreasing activities at home or as part of hobbies. Some solvents, such as trichloroethylene and its metabolites, are known drinking water contaminants in the United States; our study did not assess solvent exposure via drinking water.³⁹ The extent to which these additional sources of solvent exposure might have resulted in appreciable misclassification depends on whether such exposures occurred with comparable frequency and intensity during the period of interest. Unfortunately, we had no data to assess the extent to which non-occupational exposures contributed to the total exposure burden.

Third, biased or inaccurate recall are possibilities for at least two reasons – first, because of the time delay from delivery to interview (in our data, approximately 11 months for mothers of CHD cases and 9 months for mothers of controls), and second, because mothers of CHD cases could be more concerned than mothers of controls about their on-the-job exposures and be more likely to report exposure to substances that they perceived as hazardous. However, in our data, there were no differences in the rating of solvent exposure (using either method) by the time between delivery and interview (not shown). In addition, because the foundation of the occupational exposure assessment was the maternal report of job title and tasks (rather than specifics of her potential on-the-job solvent exposures) we think it is relatively unlikely that these data were affected by biased recall. In theory, however, if mothers of CHD cases reported a more complete occupational history, they would have a greater probability of an accurate assignment of exposure. In our data, mothers of CHD cases and controls had a very similar distribution of the total number of jobs held (not shown), suggesting that a differential recall of occupational history did not occur.

Fourth, selection bias could be a possibility if participation in the study were influenced by both exposure and outcome status. Participation in the NBDPS is, by design, influenced by outcome status; it is unlikely to be influenced by exposure status. As discussed above, clinical reviewers with expertise in pediatric cardiology review every CHD case to determine whether the case meets the criteria for inclusion in the study. This process of clinical review is undertaken without regard for the nature of the interview data collected (e.g. completion of the questionnaire, specific responses to questions of exposure). With respect to NBDPS controls, they have been found to generally represent their base populations with respect to several maternal and paternal demographic factors.⁴⁰ As stated above, the participation rate was 69% among mothers of CHD cases and 67% among mothers of controls, and these analyses were limited to mothers who reported working for at least one month in duration, at some point from three months before pregnancy through the end of pregnancy. Mothers who noted no employment during this time period were not assessed for occupational exposures – and were therefore excluded from this analysis.

Fifth, we conducted a large number of analyses (n=105; 15 CHD categories * 7 exposure variables in the two exposure assessment approaches combined) and, given a 0.05 probability of a false positive result, all of our findings could be due to chance alone. And finally, although approximately 90% of CHDs are detected prenatally or before the infant reaches one year of age⁴¹ and the NBDPS relies on high quality birth defects surveillance systems to ascertain cases, CHDs that manifest later in childhood or in adulthood are not included in the NBDPS. Therefore, this analysis does not reflect potential associations with these late-diagnosed CHDs.

One strength of our study was that both the expert consensus-based approach and the literature-based approach were expected to provide more accurate exposure ratings than would have been available through a strategy that relied solely on maternal self-reports. The NBDPS questions asking about solvent exposure (Online Appendix, Question 12e) were relatively insensitive for ascertaining exposure; if we had relied solely on these maternal reports, our results would have been quite different. For example, as reported in the Results, the literature-based approach identified 61 control mothers as potentially exposed to aromatic solvents. Only 10 out of these 61 mothers affirmatively answered the question about solvent exposure in the NBDPS questionnaire; three of the 10 responses were a specified aromatic solvent (two reported toluene and one reported xylene exposure) (data not shown).

A second strength is the large number of CHD cases available for study, all of which were carefully reviewed by clinical geneticists and clinicians with expertise in pediatric cardiology to ensure that case inclusion criteria were met and that the cases were accurately classified for analysis. The inclusion of only simple, isolated CHDs ensured that we had the most homogeneous case groupings possible and therefore we likely substantially reduced the possibility of outcome misclassification.

Finally this study provides results with a greater level of detail than those currently available in the literature, by reporting associations between specific solvent classes and CHD subtypes.

We observed associations between occupational exposure to solvents and several types of simple, isolated CHDs. Some of these findings were consistent with those previously reported in the literature, and other findings were new, yet all warrant corroboration in other study populations. Despite the strengths of this analysis, the results do not allow for the drawing of definitive conclusions on specific exposure-CHD combinations. These results should be interpreted with caution in light of the potential for misclassification in both exposure assessment methods.