Orofacial clefts (OFC) are a commonly diagnosed birth defect in the United States. For every 10 000 livebirths, approximately 10 infants are diagnosed with cleft lip with or without palate (CL/P) and 6 are diagnosed with cleft palate alone (CP), though the prevalence varies by race/ethnicity.^1,2 The critical development window for OFC occurs within the first 9 weeks of pregnancy, often before pregnancy is recognized.^3,4 These anomalies can cause complications in feeding, speech, hearing, and psychological well-being, which are associated with increased morbidity across the lifes-pan.^5-7 Optimal OFC care consists of extensive medical interventions and includes a coordinated interdisciplinary care consisting of medical, surgical, dental, and psychological professionals.⁸

The overall etiology of most OFC is largely unknown, and likely differs to some degree between CL/P and CP given differences in prevalence by race/ethnicity, risk factors, and sensitive periods of development for the lip and palate.^3,9 An estimated 8% of cases are chromosomal.^10,11 Current research suggests that occurrence is likely due to an interaction between genetic, environmental, and lifestyle factors. Prenatal smoking, often evaluated in the first trimester, is the most widely recognized risk factor for both CL/P and CP.^12-18 Thus, it is important to continue assessing the effect of other modifiable exposures on OFC risk, such as maternal diet.

Maternal diet during the preconception and prenatal period has a substantial influence on fetal growth and development. Periconceptional nutrition, in the year prior to pregnancy, aids in the creation of a favorable preimplantation environment for the placenta and fetus^19-21 and is often correlated with nutrition during early pregnancy, the critical development window for OFC.^3,22 Optimal periconceptional diet quality has been associated with a decreased risk of multiple types of birth defects.^23-26 The majority of the published scientific literature has assessed the effect of individual nutrients or single food items on the risk of OFC.^27-33 Current findings suggest that adequate consumption of nutrients such as zinc, magnesium, calcium, and vitamins B, C, and E may decrease the risk of OFC. It has also been suggested that, like neural tube defects (NTDs), OFC risk may decrease with maternal folic acid and multivitamin supplementation.^34,35 However, few studies have holistically evaluated the association between dietary patterns and diet quality on OFC occurrence.^36-38

A focus on overall dietary intake explores the realistic, complex, and synergistic effect of nutrient interactions, since nutrients are rarely consumed in isolation.^39,40 To further explore the relationship between periconceptional diet and OFC, we measured the association between maternal dietary patterns, defined a posteriori with a latent class analysis (LCA), and the occurrence of CL/P and CP using data from a large multisite case-control study of birth defects.

Participants with a control infant were primarily non-Hispanic White (58.3%) with more than 12 years of education (59.4%) (Table 1). Most of the control group took a supplement containing folic acid (75.3%) and did not smoke (84.4%) or report exposure to secondhand smoke (78.5%) or binge drinking (62.8%) at the start of pregnancy. Compared to controls, a higher percentage of CL/P and CP cases reported prenatal smoking (controls: 15.6%, CL/P: 21.4%, CP: 19.9%), exposure to secondhand smoke (controls: 21.5%, CL/P: 27.1%, CP: 25.5%), and family history of OFC (controls: 0.3%, CL/P: 6.1%, CP: 5.7%).

LCA models that specified 4, 5, and 6 classes had similar AIC, BIC, and entropy values (Table S2). We selected the 4-class solution to improve interpretability. Dietary patterns were labeled as Prudent (31.4% of controls), Western (27.0% of controls), Low-Calorie Western (26.3% of controls), and Mexican (15.4% of controls). Pattern labels were based on classes with similar consumption probabilities from previous studies^23,24 (Figure 1). Of note, the Western diet class was more reflective of a diet found in North America, rather than other Western countries. Participants in the Prudent class, our referent group, had a relatively high probability of consuming higher amounts of dairy, nuts, and fish along with a low probability of not consuming vegetables, fruits, and dark breads. Individuals in the prudent diet also had a relatively low probability of “no consumption” for most foods, which may suggest a more diverse diet compared to other patterns. Aside from a higher probability of consuming salsa, chili peppers, and tortillas, the Mexican Diet class (primarily Hispanic participants) closely followed the Prudent diet class (primarily non-Hispanic White participants). This pattern was labeled “Mexican” as it closely follows consumption patterns of individuals living in Mexico. Although similar in food intake, there were notable differences in education between participants in these 2 dietary classes. Among controls, 86.4% of participants in the Prudent diet class, compared to 21.7% of participants in the Mexican diet class, reported more than 12 years of education (Table S3).

The other 2 classes, Western and Low-Calorie Western, tended to have a relatively high probability of consuming white bread, meat, and chips, and no consumption of vegetables and yogurt (Figure 1). Unlike the Western class, the Low-Calorie Western class was associated with a decrease in class membership probability as energy intake increased, a relatively low likelihood of caffeine intake, and extreme probabilities of relative intake for most food items (high and low). The consistently high probability of no consumption for most foods may suggest a more restrictive diet among the Low-Calorie Western class (Figure 1). Along with differences in consumption, differences in caloric intake and macronutrients by class were also observed (Table S4).

After adjustment, membership in the Western diet class was associated with higher odds of CL/P (aOR: 1.3 [95% CI, 1.2-1.5]) and CP (aOR: 1.2 [95% CI, 1.1-1.4]), compared to the Prudent diet. Associations between OFCs and the Low-Calorie Western and Mexican diets were weaker with wider confidence intervals. Specifically, the aORs for the Low-Calorie Western diet (vs Prudent) were 1.2 (95% CI, 1.0-1.4) and 1.0 (95% CI, 0.9-1.2) for CL/P and CP, respectively. Similarly, aORs associated with the Mexican diet were 1.1 (95% CI, 0.9-1.3) for CL/P and 0.8 (95% CI, 0.6-1.1) for CP (Table 2). Similar results were observed when we excluded 1294 participants with a low probability of class membership (<75%, since K = 4 classes) or 2087 participants with severe morning sickness (Tables S5 and S6). The inclusion of smoking in our adjusted models did not have a material difference on final estimates (Table S7).

Specific periconceptional dietary patterns found in this study population were associated with a modest, yet significant, increase in risk of CL/P and CP, even after controlling for folic acid supplementation and other potential confounding factors including race/ethnicity. Compared to the Prudent diet (high in dairy, fish, dark breads, vegetables, and fruits), the Western diet (high in white bread, chips, and soda) was associated with a significant increase in the odds of both CP and CL/P. The more restrictive Low-Calorie Western diet (similar to Western with a lower energy intake), compared to the more diverse Prudent diet, was associated with a weaker, nonsignificant, increase in CL/P but had a null effect for CP. The Mexican diet (similar to Prudent but also high in salsa, peppers, and tortillas) had a mostly null effect on both phenotypes, when compared to Prudent.

Only 2 studies have evaluated the effect of comprehensive dietary measures on OFC by considering the intake and interaction of multiple food items.^36,38 These studies showed that, after adjustment for folic acid intake, a healthier overall diet quality decreased OFC risk. Our results, although smaller in magnitude, align with these findings. Vujkovic et al. evaluated this association in the Netherlands by deriving dietary patterns using a factor analysis a posteriori.³⁶ Similarly, the highest tertile of their “Western” pattern, which overlaps with our “Western” definition (high in red or processed meat, potatoes, and legumes and low in fruits) was associated with a significant increase in the odds of any OFC (aOR: 1.9 [95% CI, 1.2-3.1]) compared to the lowest tertile of this pattern. Carmichael et al. assessed the effect of diet on OFC in a slightly smaller subset of our NBDPS population from 1997 to 2005 using 2 diet quality indices.³⁸ These indices positively scored dietary components like grains, fruits, vegetables, and several nutrients and negatively scored items such as meat and sweets.^51,52 Higher (healthier) quartiles of the indices decreased the odds of OFC when compared to the lowest quartiles. Hispanic participants were more likely to be in these higher quartiles, similar to the dietary patterns we derived. Although we observed an increased risk of OFC associated with both Western diets, the Mexican diet had a null effect on OFC occurrence, when compared to the Prudent diet. These results remained consistent after controlling for potential confounders. To our knowledge, this is the only prior study to assess the effect of maternal diet on OFC in the United States, which is important to consider since diet is closely linked to geographic residence.⁵³

Our results demonstrate the use of holistic measures of dietary patterns and diet quality, as a complement to the investigation of individual nutrients. Like dietary indices used previously in the NBDPS data,³⁸ our newly derived patterns were more strongly associated with OFC risk compared to single-nutrient estimates previously reported in the NBDPS.³¹ To date, this is the largest single study to assess the impact of maternal diet on OFC occurrence. The NBDPS provides a diverse, population-based study sample with clinically verified cases of OFC and robust dietary information. Given the known etiologic heterogeneity between CL/P and CP, our ability to analyze each phenotype separately is a strength of this study. Our ability to derive dietary patterns, compared to individual nutrients, allowed for a more comprehensive assessment of the ways in which food items are consumed and their interaction with one another. The LCA is optimal for deriving maternal dietary patterns due to its ability to identify complex patterns while adjusting for covariates, such as energy intake.⁴⁷

In terms of possible limitations, our results are potentially susceptible to selection bias due to nonparticipation. Though over 65% of eligible subjects agreed to participate in the NBDPS, it is possible that these individuals were inherently different than those who opted out of the study; yet, it has been noted that selection bias from differential participation in NBDPS may be minimal.⁵⁴ Another potential source of selection bias is missing data; however, this is likely minimal in our study given that 1) our analytic sample included 99% of eligible cases and controls within NBDPS; 2) a comparison of the analytic sample and eligible population showed no differences in maternal and infant characteristics (data not shown); and the frequency of missing data for covariates included in our analysis was minimal (0 to <1%).

Retrospective data collection inherent in a case-control study, such as NBDPS, introduces the potential for exposure misclassification. Self-reported exposures may have been influenced by case/control status, which could lead to differential misclassification. The NBDPS interview ascertained exposures, including dietary intake via the FFQ, retrospectively up to 2 years after the EDD. This extended time between delivery and interview could lead to recall error, which could result in differential or nondifferential misclassification.⁵⁵ However, we see no strong rationale for differential recall between cases and controls in terms of remembering general patterns of food consumption in the year prior to pregnancy, which has been previously correlated with early pregnancy diet.²² Thus, the impact of the longer recall period is likely nondifferential exposure misclassification due to recall error rather than differential recall bias. The semiquantitative nature of our FFQ warrants cautious interpretation of actual nutrient intake values but is a good tool for our purpose of assessing relative dietary patterns.^56,57 Self-reported dietary intake reported through the FFQ is vulnerable to intake-related bias, in which individuals with high consumption report lower than actual consumption and vice versa, and social desirability bias, in which individuals attempt to align their responses to broader social norms.^58-60 Additionally, the LCA assumes that dietary intake is similar within each class, which may be a strong assumption. However, results were robust to the exclusion of participants that had a lower likelihood of class membership, due to their individual intake differing from that of their assigned class. It is also important to note that the labels used for these dietary patterns were selected for ease of comparison with existing literature and do not perfectly represent the frequency of food items consumed by participants in each class. We encourage readers to not solely rely on these labels but rather look closely at the foods consumed within each pattern, provided in Figure 1. Finally, residual confounding is likely as several factors that influence both OFC risk and maternal diet were unmeasured in the NBDPS database, such as household income and other socioeconomic measures that impact the risk of OFC and access to healthy foods.

Dietary patterns that are more likely to include white bread, chips, and soda, compared to diets comprised of dairy, fish, dark breads, vegetables, and fruits, were associated with increased odds of CL/P and CP in our study population. These results remained robust after controlling for factors such as folic acid supplementation and add to the body of literature that suggests that a varied and healthy diet may improve pregnancy and birth outcomes, including a possible reduction in the risk of birth defects.^20,22 Confirmation of these findings is important as periconceptional and prenatal diet is modifiable. Further research could also influence dietary recommendations and food assistance program practices for pregnant women, or those considering pregnancy.

Supplementary material is available at American Journal of Epidemiology online.