Is the Association Between Fruits and Vegetables and Preeclampsia Due to Higher Dietary Vitamin C and Carotenoid Intakes?

Abstract

Background

Diets dense in fruits and vegetables are associated with a reduced risk of preeclampsia, but pathways underlying this relationship are unclear. Dietary antioxidants may contribute to the protective effect.

Objective

We determined the extent to which the effect of dietary fruit and vegetable density on preeclampsia is because of high intakes of dietary vitamin C and carotenoids.

Methods

: We used data from 7572 participants in the Nulliparous Pregnancy Outcomes Study: monitoring mothers-to-be (8 United States medical centers, 2010‒2013). Usual daily periconceptional intake of total fruits and total vegetables was estimated from a food frequency questionnaire. We estimated the indirect effect of ≥2.5 cups/1000 kcal of fruits and vegetables through vitamin C and carotenoid on the risk of preeclampsia. We estimated these effects using targeted maximum likelihood estimation and an ensemble of machine learning algorithms, adjusting for confounders, including other dietary components, health behaviors, and psychological, neighborhood, and sociodemographic factors.

Results

Participants who consumed ≥2.5 cups of fruits and vegetables per 1000 kcal were less likely than those who consumed <2.5 cups/1000 kcal to develop preeclampsia (6.4% compared with 8.6%). After confounder adjustment, we observed that higher fruit and vegetable density was associated with 2 fewer cases of preeclampsia (risk difference: –2.0; 95% CI: –3.9, –0.1)/100 pregnancies compared with lower density diets. High dietary vitamin C and carotenoid intake was not associated with preeclampsia. The protective effect of high fruit and vegetable density on the risk of preeclampsia and late-onset preeclampsia was not mediated through dietary vitamin C and carotenoids.

Conclusions

Evaluating other nutrients and bioactives in fruits and vegetables and their synergy is worthwhile, along with characterizing the effect of individual fruits or vegetables on preeclampsia risk.

Introduction

Preeclampsia is a syndrome affecting 5% of United States pregnancies, and its incidence is rising [1]. Preeclampsia has serious consequences for the pregnant individual and the child. It is believed to cause 16% of United States maternal deaths [2], is the top indication for maternal intensive care unit admission [3], and contributes to later-life chronic disease [4]. Prematurity, fetal growth restriction, stillbirth, neonatal death, and long-term child disability are more common in preeclamptic than non–preeclamptic pregnancies [4]. Although preeclampsia typically presents as new-onset hypertension and proteinuria in the third trimester that resolves at delivery, the disorder is multisystemic, involving oxidative stress, inflammation, and endothelial dysfunction in its pathology [5]. The complexity of preeclampsia has resulted in a few advancements in its prevention, identification, or treatment, despite decades of research [6].

Nutritional status around conception is a potentially modifiable factor that has long been of interest in preeclampsia prevention [6,7]. Recently, a systematic review summarized research on dietary patterns relative to preeclampsia and reported that diets high in fruits and vegetables are associated with a risk reduction [8]. There are a tremendous number of bioactive compounds in fruits and vegetables [9], and determining the components that underlie disease etiology would aid in understanding the mechanisms of effect [10]. As substantial evidence implicates oxidative stress very early in pregnancy in the pathophysiology of preeclampsia [11], part of the pathway by which periconceptional diets rich in fruits and vegetables may prevent preeclampsia is through dietary antioxidants, including vitamin C and carotenoids.

Our objective was to determine the extent to which the effect of dietary fruit and vegetable intake on preeclampsia is due to high intakes of dietary vitamin C and carotenoids. Using machine learning-based causal mediation analyses, we estimated the indirect effect of fruit and vegetable intake on preeclampsia risk that occurs through both dietary vitamin C and dietary carotenoid intake while adjusting for dietary patterns and other confounders.

Methods

We performed a secondary analysis of data from the nuMoM2b (Nulliparous Pregnancy Outcomes Study: monitoring mothers-to-be), a large pregnancy cohort study conducted in 8 United States medical centers from 2010‒2013 (n = 10,038) [12]. Individuals with singleton pregnancies at 6‒13 weeks of gestation with no previous pregnancy lasting ≥20 weeks of gestation were eligible. All study sites used trained and credentialed study personnel and a common protocol and operations manual. The Office of human protection at each site approved the study protocol and procedures. All participants provided informed, written consent in their first language.

Participants completed study visits at 6‒13 wk (enrollment, visit 1), 16‒21 wk (visit 2), and 22‒29 wk (visit 3), when research personnel ascertained data on demographics, medical history, behaviors, social factors, psychosocial assessments, and events and complications of pregnancy. Pregnancy and birth outcomes and delivery diagnoses were recorded by study personnel from medical records at ≥30 d after delivery.

At enrollment, usual dietary intake in the 3 mo around conception was assessed using a self-administered modified Block 2005 FFQ (available in English and Spanish). Analysis of the NHANES 1999–2002 24-h dietary recall data formed the basis of the FFQ’s list of ∼120 food and beverage items. A series of “adjustment” questions in the FFQ are used to improve the estimation of fat and carbohydrate intake. Portion size was asked for each food, and pictures of portion sizes were given to participants to enhance accuracy. The FFQ has acceptable validity (most correlations 0.5‒0.6 for nutrients in comparison to 4-d food records) [13]. Study personnel checked all pages of the FFQ for completeness.

Block Dietary Data Systems performed scanning, nutrient and food group mapping, and summary analysis of the FFQ data [14]. The food and beverage items were linked to the nutrient database, based on the USDA Food and Nutrient Database for Dietary Studies [15] and the Food Patterns Equivalents Database [16], to generate nutrient and food group variables.

Our exposure was total fruit and vegetable intake. Grams of fruits and vegetables were summed and then defined as a density (cups/1000 kcal) as per the construction of the healthy eating index (HEI)-2015 [17]. Per the Food Pattern Equivalents Database 2011–2012 [16] used at the time of the FFQ processing, potatoes, tomatoes, and avocados were included in Total Vegetables and fruit juices were included in Total Fruits. Legumes were included in Total Protein Foods and Seafood and Plant Proteins and excluded from Total Vegetables. A list of all FFQ line items classified as fruit or vegetable is provided in the Supplementary Table.

We dichotomized the density of fruit and vegetable intake at 2.5 cups/1000 kcal of fruit/d, reflecting the 80th percentile of the distribution and approximating the recommended intake as defined by USDA Healthy United States-Style Eating Pattern [18]. To account for aspects of dietary patterns independent of fruit and vegetable intake in modeling, we calculated the total HEI-2015 score excluding the fruit and vegetable components [17]. This score, therefore, had a maximum value of 80.

The 2 mediators of interest were dietary vitamin C intake and dietary carotenoid intake. Dietary supplements of these nutrients were not included because our goal was to evaluate whether intake of these nutrients from fruits and vegetables contributed to the association between fruits and vegetables and preeclampsia. We dichotomized dietary vitamin C intake at 70 mg/d [19] [the Estimated Average Requirement (EAR) for pregnant individuals aged 19 or older], which is the intake amount at which the needs of 50% of the pregnant population will be met [20]. This EAR cut-point approach has been used in prior studies with FFQ data [21,22]. The 70 mg/d cutoff represented the 27th percentile of the distribution of vitamin C intake in our cohort. We calculated dietary carotenoid intake by summing the individual carotenoids estimated from the FFQ: α-carotene, β-carotene, β-cryptoxanthin, lutein-zeaxanthin, and lycopene. As recommended amounts of dietary carotenoids do not currently exist [19], we used the same 27th percentile cut-point for carotenoids, which corresponded to 74 mg/d. We did not adjust these nutrient intakes for energy because the EAR is an absolute amount.

We defined preeclampsia based on the 2013 American College of Obstetricians and Gynecologists’ diagnostic criteria [23], which was adapted by the nuMoM2b investigators to the data collected [24]. Cases were identified as having 1) new-onset gestational hypertension (systolic blood pressure ≥140 or diastolic blood pressure ≥90 on 2 occasions ≥6 h apart or 1 occasion with subsequent antihypertensive therapy, excluding blood pressures recorded during the second stage of labor), and 2) 1 or more of the following and/or HELLP syndrome: proteinuria ≥300 mg/24 h; protein/creatinine ratio ≥0.3; or if other measures are not available, dipstick ≥1+, thrombocytopenia (platelet count <100,000/mm³), pulmonary edema, serum creatinine >1.1 mg/dL, severe headache, scotoma, serum AST ≥100 IU/L, and epigastric pain [23].

We defined early-onset preeclampsia as preeclampsia that was diagnosed before 34 weeks of gestation, and late-onset were those diagnosed at ≥34 wk [5]. The date of diagnosis of the hypertensive disorder was ascertained from medical records.

At enrollment, participants self-reported their age, education, race/ethnicity, marital status, prepregnancy smoking, medical insurance, gravidity, and the country of birth for the participant and their parents, and the participant’s age of United States immigration, if applicable. Data were collected in the first trimester using standardized assessment tools to assess nausea and vomiting [25], health literacy [26], depressive symptoms [27], perceived stress [28], and anxiety [29]. United States-Census-based neighborhood and built environment variables, which were based on the participant’s census tract and block group at enrollment, were neighborhood walkability [30], neighborhood deprivation [31], and percentage of neighborhoods with income below the poverty line, and medical record abstraction ascertained data on assisted reproductive technologies, chronic hypertension, and pre-existing diabetes.

Statistical analysis

We estimated the relation between dietary fruit and vegetable intake relative to energy, dietary vitamin C and carotenoid intakes, and the risk of preeclampsia using several marginally adjusted risk differences. First, we estimated separate average treatment effects of fruit and vegetable density [≥2.5 cups (exposed) compared with <2.5 cups (referent)/1000 kcal], dietary vitamin C intake [≥70 mg/d (exposed) compared with <70 mg/d (referent)], and dietary carotenoid intake [≥74 mg/d (exposed) compared with <74 mg/d (referent)] on the risk of preeclampsia. These average treatment effects compare the preeclampsia risk that would be observed if fruit and vegetable density, vitamin C intake, or dietary carotenoid intake were set to their exposure compared with referent values. Next, we estimated the average treatment effects of fruit and vegetable density (≥2.5 cups compared with <2.5 cups/1000 kcal) on continuous measures of dietary vitamin C intake (mg/d) and dietary carotenoid intake (mg/d). Finally, we estimated the natural indirect effect of fruit and vegetable density on preeclampsia risk through dietary vitamin C and carotenoids. This approach captured the risk of preeclampsia that would be observed if the sample distribution of dietary vitamin C and carotenoid intakes were set to the values they would be if all participants consumed ≥2.5 cups of fruit and vegetables/1000 kcal relative to what they would be if all participants consumed <2.5 cups/1000 kcal. We also estimated the aforementioned relations with early and late preeclampsia.

Formally, natural indirect effects can be defined using nested counterfactual outcomes [32] such as $Y^{{x^{\prime }M}^x}$. This variable denotes the preeclampsia outcome $Y$ that would be observed if specific fruit and vegetable consumption were set to value x, whereas the mediator $M$ (vitamin C or carotenoids) was set to a value under another specific fruit and vegetable consumption ($x$). Specifically, we represent vitamin C and carotenoid values using $M_1$ and $M_2$, and use $X=1$ to represent fruit and vegetable intake ≥2.5 cups/1000 kcal ($X=0$ otherwise), respectively. Vitamin C and carotenoid values that would be observed if all participants consumed ≥2.5 cups/1000 kcal are then $M_1^{x=1},M_2^{x=1}$, respectively. Similarly, we can define the vitamin C and carotenoid values that would be observed if all participants consumed <2.5 cups/1000 kcal as $M_1^{x=0},M_2^{x=0}$. With these, the natural indirect effect can be defined as:

$$NI{E}_{M_1}=E\left(Y^{x=1,M_1^{x=1}}-Y^{{x=1,M}_1^{x=0}}\right)$$

$$NI{E}_{M_2}=E\left(Y^{{x=1,M}_2^{x=1}}-Y^{x=1,M_2^{x=0}}\right)$$

where $Y^{x,M^x}$ is the preeclampsia outcomes that would be observed under the vitamin C and carotenoid values that would occur if fruit and vegetable consumption was set to some specific value $x$ (similar for $Y^{x^{\prime },M^x}$). By contrasting these nested counterfactual quantities, we can estimate the extent to which the effect of fruit and vegetable consumption on preeclampsia risk occurs exclusively through vitamin C and carotenoids [33].

All effects were adjusted for potential confounding factors, which we identified via directed acyclic graphs [34,35]: participant’s age, race/ethnicity, marital status, insurance status, education, prepregnancy BMI (in kg/m²), prepregnancy smoking, gravidity, HEI-2015 score excluding fruit and vegetable components, prepregnancy physical activity, early-pregnancy nausea and vomiting, use of assisted reproductive technologies, pre-existing diabetes, chronic hypertension, prepregnancy binge drinking, pregnancy planning, acculturation, depressive symptoms, anxiety symptoms, stress, health literacy level, sleep satisfaction, percent of the neighborhood below the poverty index, neighborhood walkability, and neighborhood area deprivation. Some individuals were missing data on select variables, with missingness rates ranging from 0.3%–18%. In particular, select demographic and anthropometric variables (BMI, maternal education, marital status, insurance status) were missing <1% of their information, perceived stress and anxiety were missing 13% of their information, whereas dietary variables were missing 18%. To address missing data, we used mean or mode imputation and adjusted for missingness indicators in the analysis models for all estimators [36]. After mean or mode imputation, our final analytic dataset was N = 10,038.

We estimated the aforementioned effects using targeted maximum likelihood estimation (TMLE) and a library of machine learning algorithms combined into a single ensemble learner (Super Learner). TMLE is a method for estimating targeted effects that relies on fitting models for the outcome, the exposure, and the mediator of interest. These models are then combined into a least-favorable submodel targeting the effect of interest (e.g., natural indirect effect) [37]. Because of how these models are combined into the least-favorable submodel, one can use complex machine learning libraries to fit the outcome, exposure, and mediator models and still obtain valid estimates of the targeted effect of interest [38,39].

Our machine learning library included extreme gradient boosting (xgboost) with 200, 500, and 1000 trees; least absolute shrinkage and selection operator and elastic-net regularized generalized linear models (glmnet); random forests with minimum node size 10, 500, and 2500 trees, and 2, 3, and 4 predictor variables selected at random for each split (ranger); classification and regression trees with default tuning parameters (rpart); generalized linear models (glm); and simple mean.

We also included screening algorithms in the ensemble that employed impurity and permutation-based variable importance measures to select the top 10–15 covariates. To reduce the potential for overfitting, each ensemble learner was fit using 10-fold cross-validation, and we used 10-fold cross-fitting to avoid empirical process conditions and improve SEE. TMLE is doubly robust, which ensures that parameter estimates generated by TMLE will be asymptotically unbiased if ≥1 of the exposure or outcome mechanisms is consistently estimated [40]. We and others have previously shown that doubly robust estimators like TMLE are less susceptible to the problems that result from the curse of dimensionality [[41], [42], [43]].

Results

Participants were primarily aged 25‒34 y, non-Hispanic White, college-educated, normal weight, married, had private insurance, and planned their pregnancies (Table 1). HEI-2015 scores indicated low adherence to the Dietary Guidelines for Americans [17]. Approximately 16% of the cohort consumed a diet with ≥2.5 cups of fruits and vegetables/1000 kcal. Compared with people who had fruit and vegetable intake densities <2.5 cups/1000 kcal, participants with higher intakes were more likely to be non-Hispanic White, older, married, nonsmokers, more physically active, and have more formal education and lower BMI. They also had higher overall HEI-2015 scores and fewer symptoms of depression, stress, and anxiety.

TABLE 1.

Characteristics of deliveries in the Nulliparous Pregnancy Outcomes Study: monitoring mothers-to-be overall and according to usual daily fruit and vegetable density relative to energy intake in the periconceptional period

	Overall	Total fruit and vegetable density
	n = 10,038	≥2.5 cups/1000 kcal n = 8659	<2.5 cups/1000 kcal n = 1379
	% or mean (SD)	% or mean (SD)	% or mean (SD)
Maternal age, y
<25	36	15	39
25‒34	55	70	55
≥35	9	15	8
Maternal race/ethnicity
Non-Hispanic White	59	72	57
Non-Hispanic Black	14	4	16
Hispanic	17	14	18
Other	9	10	9
Maternal education
High school or less	20	6	22
Some college	29	18	31
College graduate	28	37	26
Graduate degree	23	39	20
Prepregnancy BMI
Underweight	4	3	4
Normal weight	55	63	54
Overweight	22	20	22
Class 1 obese	10	8	10
Class 2 obese	5	4	6
Class 3 obese	4	2	4
Gravidity
1	74	74	74
2 or more	19	19	19
Marital status
Not married	39	17	43
Married	61	83	57
Insurance at delivery
Private	68	86	65
Public or self-pay	32	14	35
Pregnancy was planned
No	41	23	44
Yes	59	77	56
Assisted reproductive technologies
No	96	93	97
Yes	4	7	3
pre-existing diabetes
No	99	98	99
Yes	1	0	1
Chronic hypertension
No	97	98	97
Yes	3	2	3
Preconception smoking status
Nonsmoker	82	93	81
Smoker	18	7	19
Healthy Eating Index-2015 score1	65 (10)	63 (10)	75 (7)
Preconception physical activity
<450 MET h/wk	54	36	57
450‒899 MET h/wk	20	24	19
≥900 MET h/wk	26	40	24
Any preconception binge drinking
No	66	66	66
Yes	34	34	34
Severity of nausea and vomiting1
Mild (0–6)	84	85	84
Moderate or severe (7 or more)	16	15	16
First-trimester sleep satisfaction1
Restless or average	66	60	67
Restful or very restful	34	40	33
First-trimester depressive symptoms1
Low	86	93	85
High	14	7	15
First-trimester perceived stress scale1
Low	58	73	56
Moderate	38	26	40
High	4	1	4
First-trimester anxiety1
Low	76	85	75
Moderate or high	24	15	25
Health literacy1
Sixth grade or lower reading level	3	2	4
Seventh- to eighth-grade reading level	14	6	16
Ninth grade or higher reading level	82	92	81
Acculturation
Born in the United States/parent(s) born in the United States	72	69	72
Born in the United States to ≥1 immigrant parent	13	13	13
Born outside the United States, immigrated to the United States at <18	8	9	8
Born outside the United States, immigrated to the United States at ≥18	7	9	6
Percent of neighborhoods in poverty	18 (13)	18 (13)	15 (12)
Area Deprivation Index1	45 (30)	47 (30)	37 (28)

Abbreviations: MET, metabolic equivalent of task.

¹References for standardized tools are healthy eating index-2015 [17], PUQE (Pregnancy-Unique Quantification of Emesis and Nausea) index [25], REALM-SF (Rapid Estimate of Adult Literacy in Medicine short form)[26], Edinburgh Postnatal Depression Scale [27], Perceived Stress Scale [28], State-Trait Anxiety Inventory [29], Area Deprivation Index [31].

Intakes of dietary vitamin C and dietary carotenoids were substantially higher among participants with ≥2.5 cups/1000 kcal fruits and vegetables than <2.5 cups/1000 kcal (Table 2). Over 90% of individuals with high fruit and vegetable density diets met the EAR (70 mg/d) for vitamin C compared with 69% of people with lower intakes. Results were similar for carotenoids ≥74 mg/d. After adjustment for confounders using TMLE and SuperLearner, individuals with higher fruit and vegetable density diets had dietary vitamin C intakes that were 70 (95% CI: 63, 77) mg/d higher and dietary carotenoid intakes that were 85 (77, 92) mg/d higher compared with participants who had lower fruit and vegetable density.

TABLE 2.

Association between total fruit and vegetable density and dietary vitamin C and dietary carotenoids, Nulliparous Pregnancy Outcomes Study: monitoring mothers-to-be (2010‒2013)

	Total fruit and vegetable density
	≥2.5 cups/1000 kcal n = 1379	<2.5 cups/1000 kcal n = 8659
Total dietary vitamin C
≥70 mg/d, %	94	69
Median [25th, 75th], mg/d	140 [102, 184]	103 [63, 113]
Difference in dietary vitamin C (95% CI),1 mg/d	70 (63, 77)	referent
Total dietary carotenoids
≥74 mg/d, %	91	70
Median [25th, 75th], mg/d	169 [114, 239]	111 [66, 127]
Difference in dietary carotenoids (95% CI),1 mg/d	85 (77, 92)	referent
Individual carotenoids, mg/d
Total dietary α-carotene, median [25th, 75th], mg/d	5.5 [3.0, 10]	2.2 [1.2, 4.2]
Total dietary β-carotene, median [25th, 75th], mg/d	60 [39, 90]	24 [14, 40]
Total dietary β-cryptoxanthin, median [25th, 75th], mg/d	2.0 [1.3, 2.8]	1.1 [0.7, 1.8]
Total dietary lutein/zeaxanthin, median [25th, 75th], mg/d	51 [30, 79]	19 [11, 32]
Total dietary lycopene, median [25th, 75th], mg/d	39 [25, 59]	37 [23, 58]

¹Based on targeted maximum likelihood estimation with SuperLearner with adjustment for participant’s age, education, race/ethnicity, marital status, prepregnancy smoking, medical insurance, gravidity, pre-existing diabetes, chronic hypertension, use of assisted reproductive technologies, periconceptional healthy eating index score, periconceptional physical activity, nausea and vomiting, health literacy, depressive symptoms, perceived stress, anxiety, neighborhood walkability, neighborhood deprivation, and percent of the neighborhood with income below the federal poverty line.

A total of 8.3% of pregnancies developed preeclampsia, with 1.5% developing early-onset and 6.8% developing late-onset preeclampsia. Participants who consumed ≥2.5 cups of fruits and vegetables per 1000 kcal were less likely than those who consumed <2.5 cups to develop preeclampsia (6.4% compared with 8.6%) (Table 3). The difference in incidence was larger for late-onset preeclampsia (4.9% compared with 7.1%). Using TMLE and SuperLearner to adjust for confounders, we observed that higher fruit and vegetable density was associated with 2 fewer cases of preeclampsia (–2.0; 95% CI: –3.9, –0.1) and 2.2 fewer cases of late-onset preeclampsia (–2.2; 95% CI: –3.8, –0.6)/100 pregnancies compared with lower density. Fruit and vegetable density was not associated with early-onset preeclampsia.

TABLE 3.

Effect of fruit and vegetable density, dietary vitamin C, and dietary carotenoids on the risk of preeclampsia and early-onset preeclampsia, Nulliparous Pregnancy Outcomes Study: monitoring mothers-to-be (2010‒2013)

	Population at risk	Preeclampsia		Early-onset preeclampsia		Late-onset preeclampsia
	n	n (%)	Adjusted1 number of excess cases/100 pregnancies (95% CI)	n (%)	Adjusted1 number of excess cases/100 pregnancies (95% CI)	n (%)	Adjusted1 number of excess cases/100 pregnancies (95% CI)
Total fruit and vegetable density
<2.5 cups/1000 kcal	8659	746 (8.6)	referent	130 (1.5)	referent	616 (7.1)	referent
≥2.5 cups/1000 kcal	1379	88 (6.4)	–2.0 (–3.9, –0.1)	20 (1.5)	–0.2 (–0.8, 0.3)	68 (4.9)	–2.2 (–3.8, –0.6)
Dietary vitamin C
<70 mg/d	7304	605 (8.3)	referent	115 (1.6)	referent	194 (7.1)	referent
≥70 mg/d	2734	229 (8.4)	0.0 (–2.7, 4.0)	35 (1.3)	0.6 (0.1, 1.2)	490 (6.7)	–0.9 (–2.2, 0.4)
Dietary carotenoids
<74 mg/d	7360	594 (8.1)	referent	108 (1.5)	referent	198 (7.2)	referent
≥74 mg/d	2678	240 (8.8)	0.2 (–7.2, 3.6)	42 (1.5)	–0.3 (–1.0, 0.4)	486 (6.7)	0.2 (–0.9, 1.4)

¹Based on targeted maximum likelihood estimation with SuperLearner for each exposure, with adjustment for participant’s age, education, race/ethnicity, marital status, prepregnancy smoking, medical insurance, gravidity, pre-existing diabetes, chronic hypertension, use of assisted reproductive technologies, periconceptional healthy eating index score, periconceptional physical activity, nausea and vomiting, health literacy, depressive symptoms, perceived stress, anxiety, neighborhood walkability, neighborhood deprivation, and percent of the neighborhood with income below the federal poverty line.

Dietary vitamin C intakes at or above the EAR or dietary carotenoid intakes ≥74 mg/d relative to lower intakes were not related to risks of preeclampsia or early- or late-onset preeclampsia in bivariate analyses (Table 3). There were also no effects after confounder adjustment with TMLE and SuperLearner.

The protective effect of high fruit and vegetable density on the risk of preeclampsia and late-onset preeclampsia was not mediated through dietary vitamin C and carotenoids (Table 4).

TABLE 4.

Effect of fruit and vegetable density on risk of preeclampsia and early-onset mediated through dietary vitamin C and carotenoids, Nulliparous Pregnancy Outcomes Study: monitoring mothers-to-be (2010‒2013)

	Preeclampsia	Early-onset preeclampsia	Late-onset preeclampsia
	Adjusted number of excess cases/100 pregnancies (95% CI)	Adjusted number of excess cases/100 pregnancies (95% CI)	Adjusted number of excess cases/100 pregnancies (95% CI)
≥2.5 cups of fruits and vegetables/1000 kcal mediated through dietary vitamin C	0 (–12, 12)	0 (–1.4, 1.4)	0 (–12, 12)
≥2.5 cups of fruits and vegetables/1000 kcal mediated through dietary carotenoids	0 (–13, 13)	0 (–1.8, 1.8)	0 (–12, 12)

These effects correspond to the natural indirect effect of fruit and vegetable density on preeclampsia occurring jointly through vitamin C and carotenoids. The total effects of fruit and vegetable density on the risk of preeclampsia are shown in Table 3.

Adjusted for participant’s age, education, race/ethnicity, marital status, prepregnancy smoking, medical insurance, gravidity, pre-existing diabetes, chronic hypertension, use of assisted reproductive technologies, periconceptional healthy eating index score, periconceptional physical activity, nausea and vomiting, health literacy, depressive symptoms, perceived stress, anxiety, neighborhood walkability, neighborhood deprivation, and percent of the neighborhood with income below the federal poverty line.

Further adjustment for regular multivitamin use had no impact on the results (data available upon request).

Discussion

We observed that the high vitamin C and carotenoid intakes of diets dense in fruits and vegetables did not contribute to the protective association between high fruit and vegetable intake and preeclampsia or late-onset preeclampsia. Neither fruit and vegetable density nor intake of vitamin C and carotenoids were related to early-onset preeclampsia. These relations did not change after adjustment for a wide range of confounders, including other HEI-2015 components, physical activity and other behaviors, and psychological, neighborhood, and sociodemographic factors.

Diets dense in fruits and vegetables in the periconceptional period [44,45] and mid-pregnancy [[46], [47], [48], [49]] have been associated with a reduced risk of preeclampsia in a variety of populations. However, we are unaware of studies that evaluated the extent to which the effect of dietary fruit and vegetable intake on preeclampsia or subtypes of the syndrome is due to high intakes of dietary vitamin C and carotenoids. While dietary recommendations for disease prevention are food-based for dissemination to the public [18,50], understanding the nutrients or other dietary constituents that mediate the effects of foods on health outcomes is critical for uncovering mechanisms of effect [10,51] and refining guidance. As nutrient composition differs among fruits and vegetables, understanding important nutrients underlying health effects can inform more specific dietary guidance.

We pursued vitamin C and carotenoids as mediators of the fruit and vegetable association with preeclampsia because their potent antioxidant effect may target the oxidative stress component of preeclampsia pathology [11]. Trials of vitamin C and α-tocopherol supplementation (typically 1000 mg vitamin C and 400 mg α-tocopherol) during pregnancy have proven ineffective at preventing preeclampsia [52,53], but it is possible that baseline intake, including nutrients in prenatal vitamins, may exceed a threshold needed for vitamin C supplements to have an effect. There is also concern that in a setting of oxidative stress, high-dose vitamin C might act as an oxidant [54]. Further, large doses of α-tocopherol supplements may suppress the concentrations of circulating γ-tocopherol, which is a more potent anti-inflammatory effect than α-tocopherol [55].

There are several explanations for why the association between fruit and vegetable density and preeclampsia risk is not mediated by vitamin C and carotenoid intake. First, the bioavailability of vitamin C and carotenoids in fruits in vegetables is variable. Internal concentrations are affected by the structure of the food, food preparation method, fat intake, medications, interactions with other nutrients, and more [56]. We lacked biomarkers of vitamin C and carotenoids to reflect their internal doses more accurately. Second, although fruits and vegetables are high in vitamin C and carotenoids, they also contain a vast number of nutrients and dietary constituents (e.g., fiber, folate, vitamin A, magnesium, and potassium) and bioactive compounds (e.g., phytochemicals) that are known and some of which may be undiscovered [9]. There is also synergy among these bioactive compounds in fruits and vegetables [9,57] that cannot be captured by examining intake of vitamin C and carotenoids alone. Third, a diverse collection of foods fall within fruits and vegetables as a food group. Certain fruits or vegetables may drive a protective effect on preeclampsia but may be lower in vitamin C or carotenoids or have varying bioavailabilities for these nutrients.

The relationship we observed between fruit and vegetable density and preeclampsia was limited to late-onset preeclampsia; associations were null for early-onset disease. It is possible that our sample of early-onset cases was too small to observe an effect. Additionally, the difference with respect to early- compared with late-onset disease may be explained by variations in duration, timing, or severity of these 2 preeclampsia variants. For instance, early-onset preeclampsia is more severe and more often associated with later-life CVD than late-onset [58]. A current hypothesis about the pathophysiology of these subtypes proposes oxidative stress as an important insult, but through different pathways for each subtype [59]. In early-onset preeclampsia, early-pregnancy oxidative stress is proposed to be due to failure of the uterine arteries to remodel to increase oxygen and nutrient delivery to the placenta, as normally occurs in pregnancy. However, in late-onset preeclampsia, remodeling of these vessels is normal, and it is posited that in late pregnancy, the placenta outgrows its blood supply or becomes senescent, prematurely leading to oxidative stress [59,60].

We used machine learning to estimate average treatment and natural indirect effects of interest. Natural indirect effects capture the effect of the exposure (fruit and vegetable density) on an outcome (preeclampsia) that occurs through a mediator of interest (dietary antioxidants). Although these represent relevant effects that target our study question, the validity of natural effects rests upon several assumptions. These include no confounding assumptions and the absence of mediator-outcome confounders affected by the exposure [61]. This latter assumption may be violated in our setting. However, it is unlikely that the violation of this assumption in our setting will lead to the null indirect effect estimates across all outcomes, mediators, and exposures, as observed in this work.

Our approach addressed many of the common errors in studies of nutrients as mediators of diet-disease associations. Researchers often fail to account for 1) confounders of the nutrient and outcome relationship or 2) the effects of the food group on these confounders [62,63]. These omissions can have an important effect on estimated associations. We accounted for numerous confounders in the recognition that many characteristics may impact dietary quality and adverse birth outcomes. However, we cannot rule out the potential for unmeasured confounding. FFQ data are affected by more systematic measurement errors than those collected using other methods, such as 24-h dietary recalls [64,65]. There are no recovery markers for fruit and vegetable intake, but recovery biomarker-based validation studies have shown that self-reported data may be more accurate for potassium than for sodium or energy [66,67]. Given their energy and nutrient composition, fruits and vegetables, therefore, may be reported with less error than other foods. Our adjustment for EI attenuates the potential effect of bias because densities are more accurately measured in FFQ than in absolute intake of nutrients [68]. Lastly, we have no evidence of differential measurement error related to preeclampsia cases compared with non-cases.

In the absence of a role for vitamin C and carotenoids in explaining the association between fruit and vegetable intake on preeclampsia, other pathways should be explored. Evaluating other nutrients and bioactives in fruits and vegetables and their synergy is worthwhile, along with characterizing the effect of individual fruits or vegetables on preeclampsia risk. Certain subtypes of fruits (e.g., berries) and vegetables (e.g., cruciferous vegetables) are more strongly associated with chronic disease risk than others [69]. Increasing servings of fruits and vegetables in the diets of pregnant individuals through individual-, community-, environment-, and systems-level interventions may be a safe and effective approach to lessen the burden of preeclampsia.

Acknowledgments

We thank the following institutions and researchers who compose the Nulliparous Pregnancy Outcomes Study: Monitoring Mothers-to-be (nuMoM2b) Network: Case Western Reserve University/Ohio State University - Brian M Mercer, MD, Jay Iams, MD, Wendy Dalton, RN, Cheryl Latimer, RN, LuAnn Polito, RN, JD; Columbia University/Christiana Care - Matthew K Hoffman, MD, MPH, Ronald Wapner, MD, Karin Fuchs, MD, Caroline Torres, MD, Stephanie Lynch, RN, BSN, CCRC, Ameneh Onativia, MD, Michelle DiVito, MSN, CCRC; Indiana University - David M Haas, MD, MS, Tatiana Foroud, PhD, Emily Perkins, BS, MA, CCRP, Shannon Barnes, RN, MSN, Alicia Winters, BS, Catherine L McCormick, RN; University of Pittsburgh - Hyagriv N Simhan, MD, MSCR, Steve N Caritis, MD, Melissa Bickus, RN, BS, Paul D Speer, MD, Stephen P Emery, MD, Ashi R Daftary, MD; Northwestern University - William A Grobman, MD, MBA, Alan M Peaceman, MD, Peggy Campbell, RN, BSN, CCRC, Jessica S Shepard, MPH, Crystal N Williams, BA; University of California at Irvine - Deborah A Wing, MD, Pathik D Wadhwa, MD, PhD, Michael P Nageotte, MD, Pamela J Rumney, RNC, CCRC, Manuel Porto, MD, Valerie Pham, RDMS; University of Pennsylvania - Samuel Parry, MD, Jack Ludmir, MD, Michal Elovitz, MD, Mary Peters, BA, MPH, Brittany Araujo, BS; University of Utah - Robert M Silver, MD, M Sean Esplin, MD, Kelly Vorwaller, RN, Julie Postma, RN, Valerie Morby, RN, Melanie Williams, RN, Linda Meadows, RN; RTI International - Corette B Parker, DrPH, Matthew A Koch, MD, PhD, Deborah W McFadden, MBA, Barbara V Alexander, MSPH, Venkat Yetukuri, MS, Shannon Hunter, MS, Tommy E Holder, Jr, BS, Holly L Franklin, MPH, Martha J DeCain, BS, Christopher Griggs, BS; Eunice Kennedy Shriver NICHD- Uma M Reddy, MD, MPH, Marian Willinger, PhD, Maurice Davis, DHA, MPA, MHSA; University of Texas Medical Branch at Galveston - George R Saade, MD.

Author Contributions

The authors’ responsibilities were as follows – LMB, AIN: designed research; LMB, SIK, AIN: conducted research; AIN: analyzed data or performed statistical analysis; LMB, SIK, EHK, JMR, AIN: wrote the paper; LMB: had primary responsibility for final content, and all authors: approved the final manuscript.

Conflict of Interest

The authors report no conflicts of interest.

Funding

This study was supported by grant funding from the Eunice Kennedy Shriver NICHD: R01 HD102313 to LMB and AIN as well as U10 HD063036, RTI International; U10 HD063072, Case Western Reserve University; U10 HD063047, Columbia University; U10 HD063037, Indiana University; U10 HD063041, University of Pittsburgh; U10 HD063020, Northwestern University; U10 HD063046, the University of California Irvine; U10 HD063048, University of Pennsylvania; and U10 HD063053, University of Utah. Support was also provided by respective Clinical and Translational Science Institutes to Indiana University (UL1TR001108) and the University of California Irvine (UL1TR000153). The study sponsor had no role in the study design; collection, analysis, and interpretation of data; writing the report; or the decision to submit the report for publication.

Data Availability

Data described in the manuscript and code book will not be made available until it is released to the public by the Eunice Kennedy Shriver NICHD. All code to reproduce this analysis can be found at https://github.com/ainaimi/tmle3_numom.