Relationships between consumption of ultra-processed foods, gestational weight gain and neonatal outcomes in a sample of US pregnant women

Introduction

It has been well-documented that nutrition before and during pregnancy can have long lasting effects on maternal and neonatal health outcomes (Imhoff-Kunsch & Martorell, 2012). In particular, consumption of ample fruits, vegetables, whole grains, and lean meats, and limited consumption of caffeine, alcohol, and foods high in saturated fat during pregnancy has been recommended (National Health and Medical Research Council (NHMRC), 2013; National Health Service (NHS), 2017). Evidence has emerged showing that consumption of foods high in sugar (Petherick, Goran & Wright, 2014), saturated fat (Park et al., 2013) and sodium during pregnancy can be particularly harmful to both the pregnant woman and their neonates (Tay et al., 2012). Many of these foods can be categorized as ultra-processed foods (UPF), which are assemblages of food substances designed to create durable, accessible, convenient and palatable ready-to-eat or ready-to-heat food products (Monteiro et al., 2017). These products are often consumed as snacks instead of home-prepared dishes, are low in fiber, whole grains, and vitamins (Monteiro et al., 2017) and include artificial colors, flavors, and preservatives, which can be particularly harmful for pregnant women (Halldorsson et al., 2010).

Ultra-processed foods (UPFs) are merely one group in a four-category classification system (NOVA) that was developed to guide consumers towards a healthy diet using food-based, rather than nutrient-based, dietary guidelines (Monteiro et al., 2017). There is increasing evidence that high consumption of UPFs is indicative of poor diet and is associated with obesity, metabolic syndrome and cardiovascular disease in non-gravid adults (Canella et al., 2014; Costa Louzada et al., 2015; Louzada et al., 2015a; Louzada et al., 2015b; Martinez Steele et al., 2016; Moubarac et al., 2013). However, the relationship between the percent of energy intake from ultra-processed foods (PEI-UPF) during pregnancy and maternal and neonatal health outcomes has not been examined. Therefore, the purpose of this study was to determine the association between UPF consumption in pregnant US women and selected maternal/newborn health outcomes.

To do this, we used data collected by Tinius et al. on the health of 45 pregnant women and their neonates in St. Louis, MO, USA (Tinius et al., 2015; Tinius et al., 2016). In the original study’s design, only women within the normal or obese BMI ranges (18.0–24.9 kg/m² or 30.0–45.0 kg/m²) were included. Overweight women (BMI of 25.0–29.9 kg/m²) were excluded. It was found that the lean and obese groups only differed in gestational weight gain and maternal weight. No significant differences in PEI-UPF or other clinical outcomes were found between the two groups. However, the two groups were modeled as having different slopes (with respect to PEI-UPF) as well as intercepts to allow greater model flexibility.

We hypothesize that the percent of energy intake coming from UPF could serve as a concise measure of the diet quality of this sample of pregnant US women. Further, we hypothesize that PEI-UPF could be an efficient predictor of maternal and neonatal health outcomes. These include maternal gestational weight gain (GWG) and neonatal anthropometrics. The ability of UPF consumption to predict these outcomes is clinically important as high GWG is generally associated with high postpartum weight retention (Gunderson & Abrams, 1999), and with the child having a higher BMI early in life (Lau et al., 2014; Mourtakos et al., 2016). More broadly, research has shown that maternal obesity can negatively influence neonatal outcomes in a variety of ways (Castro & Avina, 2002). These patterns almost certainly do not end at birth: Catalano et al. (2003) found that infant body fat percentage in particular (as opposed to body weight) can be a significant predictor of early childhood, and possibly adult, obesity. Additionally, skinfold thickness measurements can be a predictor of insulin resistance and diabetes later in life (Yajnik et al., 2003). Therefore, the ability to determine the role of UPF consumption in maternal and neonatal health is important.

A secondary aim of the study was to compare the abilities of PEI-UPF and another dietary quality index, the Healthy Eating Index-2010 (HEI-2010), to predict maternal GWG and neonatal body composition. The HEI-2010 is a number ranging from 0 (worst) to 100 (best) that reflects the consumption of desirable macronutrients and food groups (fruits, vegetables, etc.), and avoidance of unhealthy foods (refined grains, sodium, and empty calories). The HEI-2010 measures diet quality according to the 2010 Dietary Guidelines for Americans (Guenther et al., 2014), and has been shown to have significant associations with biomarkers and clinical outcomes in gravid and non-gravid adults (Reedy et al., 2014; Shapiro et al., 2016). However, HEI-2010 has not been directly compared with PEI-UPF in this regard. The HEI-2010 is often computed using 24-hour food recalls or FFQs such as the US National Institutes of Health’s Diet History Questionnaire II (DHQ II) (National Cancer Institute (NCI), 2010), in which subjects reported their consumption of various unprocessed, prepared, and packaged foods over the past month. Tinius et al. administered the DHQ II to participants, and found that macronutrient intake was largely similar between lean and obese study groups, although active obese women tended to consume more fat than inactive obese women (Tinius et al., 2015; Tinius et al., 2016). We note that the DHQ II can be used in a variety of other ways, such as measuring how many servings of a food were consumed (Yang & Rose, 2014), or calculating consumption of ultra-processed foods.

Methods

Results

The present study is based upon previously published data with a sample size of n = 50. However, records with missing FFQ or clinical outcome data had to be excluded from this study. Of the final sample (n = 45), sixteen women are from the lean study group (n = 16) while the remainder (n = 29) are from the obese group. Detailed subject characteristics are presented in Table 1. The majority of women visited a primarily high-income clinic (57.8%), were nulliparous (55.6%), and obese (64.4%). Equal numbers of women were Caucasian and African American (46.7% each), and the remaining 6.7% were Hispanic or Asian. The average PEI-UPF was 54.4 ± 13.2% and the average percentage of energy intake for both processed and ultra-processed foods together was 63.2% (not shown in table). Among ultra-processed foods, the most consumed subgroup was Cakes, Cookies and Pies (5.8% of total energy). Only two out of all thirty-three subgroups had higher average consumption—fruits (9.1% of total energy intake) and grains (9.8%) (not shown in table).

Further detail showing the quantity of nutrients obtained from each main food group is given in Table 2. As with energy intake, the participants’ total carbohydrate, fat, sugar and sodium intakes were primarily derived from ultra-processed foods (57.0%, 58.8%, 57.9% and 65.7% of total dietary intake, respectively). On the other hand, 39.9% of fiber was obtained from Group 4 foods. Indeed, pregnant women who limited their intake of ultra-processed foods tended to have better health outcomes for themselves and their infants. Tables 3 and 4 present the detailed results of multiple regression analysis on newborn and maternal outcomes, respectively. The association between PEI-UPF and GWG was observed only in the fully adjusted model, after controlling for maternal age, race, socioeconomic status, weight status, average daily energy and fat intake, and time spent in moderate physical activity. Likewise, the association of PEI-UPF with newborn body composition was observed only after controlling for maternal age, race, socioeconomic status, weight, average daily energy and fat intake, time spent by the woman in moderate physical activity, and gestational age at time of measurement. However, in each of the four models, the mother’s weight status (lean or obese) had no significant slope or intercept effect on the relation between PEI-UPF and the clinical outcome. A number of biomarkers including blood levels of triglycerides (data available for mother only), free fatty acids, fasting glucose/insulin, and C-reactive protein had no significant association with PEI-UPF in either mothers or infants.

Various interaction terms with PEI-UPF were tested, and only the interaction with age was found to be significant (p ≤ 0.030 for all four outcome variables). Thus, for older pregnant women, increased PEI-UPF has less of an effect on poor health outcomes than for younger women, as indicated by the negative coefficients for the interaction terms. All other interaction terms with PEI-UPF were not significant.

The predictive ability of PEI-UPF was also compared with several other measures. In adjusted models with only one dietary predictor (PEI-UPF, HEI-2010, total energy intake or total fat intake), PEI-UPF was more strongly associated with all clinical outcomes than either total energy or fat intake (Table 5). Indeed, PEI-UPF retained a significant relationship with GWG (p = 0.016) (Table 3), as well as neonatal thigh skinfold thickness (p = 0.045), subscapular skinfold thickness (p = 0.026) and body fat percentage (p = 0.037) (Table 4), even after controlling for total energy and fat intake. This suggests that PEI-UPF measures an aspect of diet that is independent of total energy and total fat intake.

Overall, PEI-UPF had a strong negative correlation with HEI-2010, with r =  − 0.74 95% CI [−0.85, −0.56], indicating that both measures of diet quality are fairly consistent. Additionally, in models with one dietary predictor, maternal HEI-2010 scores were strongly associated with HDL cholesterol (p = 0.0020) (not shown in table), GWG (p = 0.0011), and neonatal subscapular skinfold thickness (p = 0.026) (Table 5).

In fully adjusted models including total energy and fat intake, the HEI-2010 was a better predictor of gestational weight gain than PEI-UPF (Adj. R² = 0.26, as opposed to 0.14 for PEI-UPF). However, PEI-UPF was still a better predictor of infant body fat percentage, thigh skinfold thickness, and subscapular skinfold thickness than the HEI-2010 (Adj. R² = 0.01, 0.14, and 0.10, as opposed to −0.09, −0.02, and −0.02, respectively) (not shown in table). Although HEI-2010 has a greater association with subscapular skinfold thickness than PEI-UPF (according to P-values in Table 5), Adjusted R² values indicate that overall the HEI-2010 model is a worse predictor than the PEI-UPF model. Furthermore, adding HEI-2010 as a predictor in our four fully adjusted PEI-UPF models did not significantly improve fit (p ≥ 0.097 from extra sum-of-squares F-test in all cases). The failure of HEI-2010 to improve model fit was likely caused by the strong (negative) correlation between PEI-UPF and HEI-2010.

Discussion

The results show a strong positive association of PEI-UPF with GWG and with neonatal anthropometrics (i.e., subscapularis and thigh skinfold thicknesses and body fat percentage). This study demonstrates that many pregnant women are obtaining the majority of their energy from ultra-processed foods, and these ultra-processed foods may also be worsening health outcomes for themselves and their children. These relationships are essentially the same in both lean and obese mothers. Indeed, the majority of participants’ carbohydrate, fat, sugar, sodium, and energy were obtained from UPF, which is consistent with the refined ingredients and highly palatable nature of such foods. As such, it is not surprising that UPF consumption negatively affects health.

The identification of causes of excessive gestational weight gain is clinically important as excessive gestational weight gain can have serious consequences for the postpartum women and their neonates. It leads to excessive postpartum weight retention (Gunderson & Abrams, 1999), which in turn can contribute to long-term obesity and associated comorbidities including type 2 diabetes, cardiovascular disease, mental health issues, and cancer (Institute of Medicine and National Research Council, 2009). For the neonate, excess adiposity is likely to continue into childhood (Mei, Grummer-Strawn & Scanlon, 2003), and childhood obesity is a strong predictor of adult obesity (Freedman et al., 2005). Thus, higher body fat as an infant may contribute to long-term risk for obesity and its associated comorbidities (Catalano & Ehrenberg, 2006; Tinius, Cahill & Cade, 2016). Because UPF consumption was related not only to excessive gestational weight gain, but also neonatal adiposity, maternal diet quality modification could substantially improve long-term health outcomes for mother and child.

Interestingly, a number of successful interventions to limit excessive GWG emphasize energy or fat restriction, or other macronutrient targets (Gardner et al., 2011; Phelan et al., 2011). Despite the popularity of energy- and fat-restricting diets, GWG was more strongly associated with PEI-UPF than total energy or fat intake (p = 0.017 for PEI-UPF compared to p = 0.73 and p = 0.88 for total energy and fat intake). More generally, although low fat ultra-processed foods are ubiquitous, such results cast doubt on the health benefits of these low fat foods. We believe interventions to limit GWG could be even more successful if they also emphasized a minimally-processed diet, since our results show that PEI-UPF captures information about diet quality, which total fat or energy intake cannot. In general, our results suggest that consumption of UPF may be a key factor contributing to unfavorable maternal and neonatal outcomes. This study showed that poor diet quality during pregnancy increases neonatal adiposity independent of maternal weight and maternal moderate physical activity; thus, maternal diet quality is an important direction of future study. Specifically, diet quality seems to be more important than the amount of energy consumed. Thus, from a clinical standpoint, pregnant women should be educated to focus less on the total energy consumed, and more on the source of that energy.

Interestingly, the nutrient profiles indicate that processed foods generally have less fiber and more sodium than UPF. Thus, while small amounts of processed foods are part of any diet and acceptable in moderation, clear preference should be given to unprocessed/minimally processed foods. It is also important to highlight the need for moderation when using salt, sugar and oil in home based meal preparations.

Furthermore, almost all currently used tools for assessing diet are largely quantity based instead of quality based, which is one of the main reasons for measuring UPF consumption. Because many currently used tools assess quantity, we also wanted to assess the ability of the HEI-2010 to predict maternal and neonatal outcomes. As part of our secondary aim, our results did show that a quantity-focused measure such as HEI-2010 can be a useful predictor of gestational weight gain. Despite the high correlation between PEI-UPF and HEI-2010, PEI-UPF is a better predictor of neonatal body fat percentage and skinfold thickness at the thigh and subscapularis. This comparison is based on Adjusted R² values (0.01, 0.14, and 0.10 for PEI-UPF, as opposed to −0.09, −0.02, and −0.02 for HEI-2010). Interestingly, Shapiro et al. (2016) found that a low maternal HEI-2010 score was associated with higher neonatal body fat percentage, in a sample size of >1,000 woman and infant pairs. We were unable to confirm this finding (in our study, p = 0.30 for association with body fat percentage). The differences in predictive ability between HEI-2010 and PEI-UPF indicate that each statistic measures different aspects of the diet, and therefore both are useful. To achieve the optimal diet, one must both limit intake of UPFs as well as eat a variety of different nutrients.

This study has several notable strengths, including being the first effort to measure UPF consumption in pregnant women, and to correlate PEI-UPF with maternal and neonatal clinical outcomes. Additionally, it is only the second study to examine PEI-UPF in the United States, where the percent of the diet coming from UPFs is much higher than in some other countries (Canella et al., 2014; Martinez Steele et al., 2016; Monteiro et al., 2013). However, this study presents some limitations. Due to the design of the original longitudinal study, only women within the normal or obese weight ranges were included (BMI between 18.0 kg/m² and 24.9 kg/m² or between 30.0 kg/m² or 45.0 kg/m²). Thus, women in the overweight range were excluded, and the study results may not be applicable to such women. Additionally, the racial composition, with essentially equal numbers of Caucasians and African Americans, and very few other minorities, is not representative of the entire US population. The design of the survey instrument presents further limitations. Since food frequency and portion sizes were collected in a semi-quantitative/categorical format, often with as few as three options, there was some error simply because respondents had to round off quantities. Additionally, somewhat subjective researcher input was required to categorize each DHQ II food according to the NOVA scheme. For example, homemade bread would be a processed food, but for this study, bread was classified as ultra-processed since most bread consumed in the US meets this definition. A full listing of classifications (along with justification) can be found in the Appendix S1.

A greater error we could not eliminate is the fact that participants may underreport their food intake. Previous research found that postmenopausal women underestimated their energy intake by 21% on a FFQ (Horner et al., 2002). However, another study found that food frequency questionnaires (FFQs) inquiring about consumption over a several-month period provide reproducible and valid measures of relative dietary intakes in pregnant populations (Vioque et al., 2013). However, since we are using percentages of energy intake as the main predictor rather than absolute energy, we feel our data may not be subject to the same degree of error. Finally, another major limitation is that administering the DHQ II once, at Visit 1, effectively only assesses maternal diet at 30–34 weeks gestation. It is unlikely that this assessed diet accurately represents diet across the entire pregnancy, since previous research indicates that intake of certain foods and overall caloric intake vary across the three trimesters (Durnin, 1991; Rifas-Shiman et al., 2006).

Conclusions

This study showed that consumption of ultra-processed foods leads to unfavorable pregnancy outcomes including excessive maternal gestational weight gain and increased neonatal body fatness. For both mother and neonate, excess adiposity is likely to remain, contributing to associated comorbidities such as Type II diabetes, cardiovascular disease, mental health issues, cancer. Reducing dietary consumption of ultra-processed foods may be a potential avenue for improving short and long term maternal and neonatal health, making this an important direction for future research. A natural, minimally-processed diet centered on home cooking should be promoted among pregnant women.

Supplemental Information