Association Between Maternal Dietary Pattern and Infant Birth Weight: a Structural Equation Modelling Analysis

Background: The mother’s diet during pregnancy is associated with maternal and child health. However, there are few studies with moderation analysis on maternal dietary patterns and infant birth weight. We aim to analyse the association between dietary patterns during pregnancy and birth weight. Methods: A cohort study was performed with pregnant women in Bahia, Brazil. A food frequency questionnaire was used to evaluate dietary intake. Birth weight was measured by a prenatal service team. Statistical analyses were performed using factor analysis with a principal component extraction technique and structural equation modelling. Results: Four patterns of dietary consumption were identied for each trimester of the pregnancy evaluated. Adherence to the "Meat, Eggs, Fried Snacks and Processed foods” dietary pattern (pattern 1) and the "Sugars and Sweets" dietary pattern (pattern 4) in the third trimester directly reduced birth weight, by 98.42 g (Condence interval (CI) 95%: 24.26, 172.59) and 92,03g (CI 95%: 39.88, 165.30), respectively. Insucient dietary consumption in the third trimester increases maternal complications during pregnancy, indirectly reducing birth weight by 145 g (CI 95%: -21.39, -211.45). Conclusion: Inadequate dietary intake in the third trimester appears to have negative results on birth weight, directly and indirectly.

Thus, dietary pattern can be one of the factors in uencing the health and nutrition of pregnant women and, therefore, is related to birth weight. Inadequate dietary intake during pregnancy increases the risk of maternal morbidity and mortality and affects the health and nutrition of the foetus and child at birth, increasing the prevalence of excess weight and obesity in infancy and adulthood [1,12,13] .
Although several cohort studies have examined the determinants of bight weight, factors associated with moderation analysis on maternal dietary patterns and infant birth weight remain unelucidated among socioeconomically vulnerable populations living in low-and middle-income countries, such as Brazil. In this regard, adequate food consumption can reduce unfavourable results and risks during pregnancy and childbirth for both the mother and foetus, thus offering protection for the nutritional and health status of children, particularly with respect to birth weight. Thus, this study aims to identify the association between dietary patterns during pregnancy and birth weight.

Study design
This is a prospective and dynamic cohort study, involving pregnant women residing in Santo Antonio de Jesus, Bahia, Brazil. The study sample included 185 pregnant women over the age of 18 years, resident and domiciled in the urban area of this municipality, in their rst trimester of pregnancy, registered at the Family Health Units prenatal service, between April 2012 and June 2013.
Sample size calculation was based on average weight at birth of 3196 g and a standard deviation of 456 g [14] , assuming a difference of 100 g in birth weight when coupled with adequate dietary intake, a sample error of 2%, a power of 80%, and acceptable loss of 15%. Thus, a sample size of 175 pregnant women would be needed. However, we decided to include in the sample all 185 pregnant women who were contacted in the cohort.

Exclusion and inclusion criteria
The following women were deemed ineligible for the study: pregnant women residing outside the municipality; adolescents; and those with multiple pregnancies, stillbirths, and anembryonic pregnancies. In the study, a loss was considered an absence of monitoring in any of the study rounds or pregnant women who were not found at their home ( Fig. 1).

Data collection
Dietary intake assessments During the rst, second, and third trimesters of pregnancy, weight measurements were recorded. The dietary consumption in the pregnant women's households was investigated in the rst and third trimesters. We decided not to collect dietary intake data in the second trimester of pregnancy in order to prevent loss to follow-up.
In the remaining rounds, this information was collected using a validated semi-quantitative food frequency questionnaire (FFQ) [15] . It consists of 74 food items that represent the local dietary culture and includes over 11 food groups (Table S1).
The frequency of consumption was represented by the following categories: daily, weekly, monthly, never, or almost never. To estimate the size of the food portions consumed, a photographic record album of food rations and utensils was used [16] . Data on food consumption was elaborated by the researcher [17] .
Initially, the frequency of consumption of each food was converted to a daily basis [18] . Therefore, food was attributed a value of 1 when it was consumed once a day. If the food was consumed more than once a day, the value of one was multiplied by the interval of the daily frequency reported. For the options of weekly and monthly intervals, we used the average of the range of frequencies divided by the period, whether weekly (= 7) or monthly (= 30), thus obtaining the average daily consumption. In all, 11 food groups were constructed (Frame S1).

Data on covariates
The details of the cohort procedures are described in previous studies [19,20] . In the rst stage of the cohort (baseline), the information on socio-economic and demographic conditions, reproductive and/or obstetric history, and lifestyle were collected during the enrolment of pregnant women at the prenatal services, and responses were recorded using a questionnaire.
We used the pre-pregnancy body mass index (PP-BMI) and classi ed it according to Institute of Medicine parameters [21] . Maternal weight and height measurements were taken during the interview by trained researchers.
We used gestational BMI as a proxy for gestational anthropometric status, and this was classi ed according to the Atalah curve [22] . Weight gain during pregnancy was assessed according to the categories of PP BMI, following Institute of Medicine (IOM) recommendations [21] . A Marte® portable digital scale with a capacity of 150 kg was used to record weight. Height was measured using a Welmy® brand stadiometer with a 2000 mm maximum height. Anthropometric measurements were taken in duplicate. Maximum variations of 0.5 cm, for measuring height, and 100 g, for measuring weight, were accepted [23] .

Outcomes
In the follow-up of the study, the pregnant women were monitored until the birth of their child [19,20] . The birth weight was measured with the child naked using a digital paediatric scale (Welmy®), with a capacity of 15 kg and a 10 g interval. The anthropometric measurement was performed in duplicate, and a maximum variation of 10 g was acceptable. When different values were obtained, a third measurement was performed. The nal measurement was calculated as the mean of the values of the closest measurements [24] . Information on birth weight was collected from the municipality's Epidemiological Survey (VIEP) registration system. When necessary, home visits were performed at the end of the study for those pregnant women whose data on gestational outcomes were not found in the VIEP [19,20] ..

Theoretical model
The theoretical model was based on knowledge of the relationship between the variables associated with birth weight [23] (Fig. 2). Initially, the socioeconomic, biological, and obstetric environmental variables of the environment that the pregnant women live in were tested.
The observed variables that were associated with the event in this analysis were included in the modelling, under the following headings: PP BMI, weight gain, and complications during pregnancy (anaemia, urinary tract infection, gestational diabetes, hypertensive disorders).

Data management and statistical analyses
Prevalence was determined to describe the categorical variables. Mean and standard deviation (SD) were used for the continuous variables. The mean of birth weight according to the exposure variables was compared using student's t test for equal variances.
To identify the dietary patterns in the sample, factor analysis by principal component was used. Varimax rotation of the factors was employed, with graphic Scree plot use to de ne the extracted (retained) factors. A factorial charge ≥ 0.4 was adopted as the selection criterion for the dietary groups to be included in the pattern. Internal consistency of each factor was evaluated by Cronbach's alpha. Detailed information on the identi cation of dietary patterns during pregnancy is reported elsewhere [17] .
To evaluate the association between dietary patterns during pregnancy and birth weight, structural equation modelling (SEM) was used. In this sense, birth weight represented the response variable (endogenous variable); pattern of dietary consumption (constructed) represented the Exposure variables (exogenous variable); and complications during pregnancy, PP BMI, and weight gain (observable variables), intermediate variables. Associations were accepted with a p < 0.05. The variables (BMI, gestational BMI, weight gain) were included continuously into the SEM. Patterns of dietary consumption (constructed) were entered into the model as a factorial score in a continuous way.
We used the likelihood-ratio Chi-square statistical test to analyse the appropriateness of the model. The results of the analysis for the group of mediating variables were evaluated by means of standardised coe cients (SC) and interpreted according to the results of Kline [25] .
Excel and Virtual Nutri software programs used for processing the dietary consumption data. The Statistical Package for Social Sciences program, version 17, was used for data entry and for descriptive and factor analyses. Stata software, version 12.0 was used for SEM analysis.

Description of participants
The characteristics of the mothers and their new-borns are presented in Table 1; there was a predominance of maternal age lower than 25 (41.1%) and a low level of education (85.9%). The prevalence of anthropometric status characterised by excess weight (overweight/obesity) was 44.0% in the pre-pregnancy period and 48.1% during pregnancy.
Loss over the 12 months of follow-up was 12.1% (n = 24). The most frequent reasons were those related to not nding the address provided by the pregnant woman to the family clinic (6; 25%), absence in one of the rounds of the study (8,33.4%), and refusal to participate in the study (10; 41.6%) (Fig. 1).
The prevalence of adequate weight at birth was 96.2% (average, 3341.18 g; SD: 522.832) (Data not presented in the tables). The highest mean birth weight was found for infants born to mothers with family income of more than one minimum salary (3371.67 SD: 508.31) and for those whose mothers had more than seven medical appointments in the prenatal period (3376.87 g; SD: 440.364), had no complications during pregnancy (337 6.87; SD:528.30), and with more than 10 kg weight gain during pregnancy (3424.36; SD: 490.30).

Main analysis
In the rst and third trimesters, four dietary consumption patterns were identi ed ( Table 2). The internal consistency of each pattern was assessed using Cronbach's Alpha. All extracted patterns presented an acceptable Cronbach index (> 0.5), indicating homogeneity among the constructs ( Table 2).
In the third trimester, we identi ed four patterns. The rst (meats, eggs, fried snacks and processed and processed products) represented 14.08% of the consumption data, the second (cereals, legumes, vegetables/salads, fruit, and milk/dairy), 13.28%, the third (coffee and butter/margarine), 11.03%, and the fourth (sugars and sweets), 10.62% of the consumption data ( Table 2).
The SEM presented satisfactory t (p > 0.05). The rst (meats, eggs, fried snacks, and processed food products) and fourth patterns (sugars and sweets) in the third trimester of gestation had a direct and signi cant negative impact, where the birth weight was reduced by 98.42 g (p = 0.009) and 92.03 g (p = 0.03), respectively (Table 3; Fig. 3). There was no statistically signi cant difference between the pattern of dietary consumption in the rst trimester and birth weight. Weight gain in pregnancy (β = 160; p = 0.04) and morbid complications during pregnancy (β = -145; p = 0.008) (anaemia, hypertension, diabetes mellitus) had a direct and signi cant impact on birth weight. Gestational weight gain had a direct and positive impact on birth weight (β = 160; p = 0.04). It was observed that for every 1 kg gained during pregnancy, there was a reported increase of 160 g in newborn birth weight. With regard to morbid complications in pregnancy, these directly and negatively impacted the weight of the child, reducing total weight at birth by 145 g (Table 3; Fig. 3).
In addition, it was observed that other variables of the maternal environment mediated the relationship between dietary intake during pregnancy and birth weight. In this sense, the third dietary patterns in the third trimester (coffee and butter/margarine) had an indirect and negative but signi cant impact on birth weight, mediated by the increased occurrence of morbid complications during pregnancy (SC = -0, 11; p = 0.01). This very same pattern of consumption had a strongly negative association with birth weight, reducing it (SC = -0.73 p = 0.04) via the elevation of gestational BMI (Table 3 and Fig. 3).

Discussion
This is one of the few studies conducted in Brazil that aims to identify patterns of dietary consumption during pregnancy and their relation to birth weight. It uses robust methodological modelling of follow-up and a convenient framework of multivariate data analysis, permitting the estimation of a series of relationships between variables. This study provides a convenient framework of multivariate data analysis and allows the estimation of a series of relationships between variables.
In this sense, it was found that consumption of meat, eggs, fried/savoury foods, and processed food products (pattern 1) and consumption of sugar and sweets (pattern 4) in the third trimester had a direct and negative impact on birth weight, reducing it by 98.42 g and 92.03 g, respectively.
Although some of these foods, such as meat and eggs, included in pattern 1, contain protein of high biological value that has a recognised physiological function in pregnancy, these foods are mainly fried because of the cultural in uence in the population of the local municipality. Savoury, fried, and processed food products, contain high levels of saturated fat, sodium and cholesterol in their composition, which negatively affect the health of the individual and can also adversely affect new born birth weight [26] .
Further, dietary pattern 3 (coffee and butter/margarine) during the third trimester signi cantly impacts birth weight in an indirect fashion, causing complications during pregnancy. Although this effect was small (β = -0.11) at the sample level, the overall risk could be signi cant.
The third dietary pattern had an indirect association with anaemia. Observational studies show that coffee consumption is associated with anaemia [26] . However, the inhibitory effect of coffee on iron absorption can be partially reduced by the concurrent intake of vitamin C-rich foods and foods of animal origin [26] . Therefore, pregnant women should limit coffee consumption and avoid drinking coffee with meals [26] .
Thus, the third dietary pattern may favour weight gain during pregnancy, increasing the likelihood of complications during pregnancy (hypertension, diabetes, anaemia) as well as reducing the availability of essential nutrients for the proper development of the foetus.
The patterns of dietary consumption identi ed among the women in this study are similar to those found in the Brazilian population as a whole, which is explained by changes taking place in this population's diet, re ected in its epidemiological and nutritional pro le at the different stages of life.
In Brazil, as in other countries, these changes are related to changes in lifestyle, which are expressed in the reduction in daily caloric expenditure; increased consumption of saturated fats, simple sugars, and processed foods; and reduced consumption of fruits and vegetables. This dietary pattern is a matter of concern at all stages of life, and during pregnancy, it can be associated with poor maternal health conditions and compromise foetal growth and development. The lack or excess of nutrients during pregnancy can lead to morbid complications for both the mother and foetus, impacting the child's health at later stages of life [1,8,26,27] .
The four different dietary patterns identi ed in the rst and third trimesters of pregnancy represented 50.45% and 49.03%, respectively, of the total consumption variance, indicating that the patterns extracted actually represent the eating habits of the population being studied.
The dietary intake pro le of the pregnant women in this study predominantly consisted of foods rich in saturated fats, trans fats, cholesterol, sugars, and sodium and poor in vegetables and fruit, which are good sources of bre and micronutrients. This dietary pro le, in a critical period of human development such as pregnancy, has consequences for the mother and foetus [28,29] .
In the trimesters evaluated, we reported the consumption of fruits, but at an extremely low frequency on a daily basis, accounting for only 9.11% in the rst and 13.28% in the third trimester. This low consumption particularly restricts the availability of vitamins, minerals, nutrients that directly impact appropriate weight and size at birth. Observational studies found that fruit consumption during pregnancy is conducive to adequate weight and size at birth and associated with a low incidence of preterm birth [29,30] .
One similar study involving pregnant women reported two patterns of food consumption: healthy (local preparations containing maize our and maize meal, yams, fruits, vegetables, meat, and eggs) and unhealthy (sweetened drinks, ice cream, chocolate drinks, and soda) [31] . The healthy pattern was found to offer protection against low birth weight. In another study, maternal food pattern consisting of fruits, nuts, Cantonese desserts, and a varied diet (pasta, bread, vegetables, poultry, meats, sh, seafood, yogurt, and beans) was associated with high birth weight [32] .
This study highlights the epidemiological importance of assessing the dietary intake of pregnant women during pregnancy; we identi ed the different consumption patterns during each trimester and associated these patterns to the different nutritional needs in these periods.
The need to avoid imbalanced food consumption must be emphasised, both in relation to excess and de cit. The adoption of healthy living habits, with a varied dietary intake, qualitatively balanced in nutrients and energy so as to meet each trimester's nutritional needs, allows satisfactory physiological adaptations of the maternal organism and adequate foetal growth and development. Women who intend to become pregnant should maintain a healthy lifestyle, which includes maintaining a healthy weight, practicing physical activity, stopping tobacco and alcohol use, and adopting a balanced diet [33] . In addition to all of these measures, pregnant woman must maintain a balanced weight and adequate vitamin and mineral supplementation. All of these factors contribute to maternal health and reduce the risk of negative outcomes for newborns [31] . In this sense, maintaining healthy eating habits is a requirement both before and during gestation. In Brazil, the Ministry of Health recommends following the guidelines of the Food Guide for the Brazilian Population as a practice of healthy eating [34] .
Despite the need to follow the nutritional recommendations in this stage of life, evidence from a meta-analysis on micronutrient intake during pregnancy shows that women in this period ingest an insu cient quantity of micronutrients as opposed to the recommended values [35] .
It is known that when pregnant women begin their pregnancy with an age-appropriate weight, their nutritional reserves (energy and nutrients) supply the foetus with its nutritional needs in the rst trimester [36] , which may be associated with the absence of a correlation between maternal dietary consumption in the rst trimester and foetal growth and development in this period, as seen in this study.
However, some methodological issues of this study should be considered, especially regarding techniques of collection and analysis of dietary intake; su cient measures were adopted to minimise the measurement biases that could be caused by recall, such as the use of a photo album containing pictures of food portions during the investigation of dietary consumption.
In this study, factor analysis with the principal component technique was adopted to identify dietary consumption patterns. Still, this method is not without criticism, mainly due to the subjectivity in determining the number of factors to be extracted. This restriction has been minimised with the adoption of methodological assumptions in the statistical model [37] .
It is understood that losses in monitoring may contribute to selection bias in the investigated sample. However, on statistical analysis of the losses, it was noted that these did not differ in the sample throughout the monitoring period for some exposure variables, for example, pre-gestational anthropometric status, maternal age, income, education, smoking and alcohol consumption (data not shown in table), which indicates that the losses occurred at random. Of note, the losses were negligible (12%), so it was unlikely that they resulted in biased results.
Thus, longitudinal studies with a broad methodology for assessing consumption, suitable to the complexity of the Brazilian population's diet, generate consistent information that can support the de nition and implementation of health measures for ensuring adequate nutrition during pregnancy. The objective is to reduce maternal morbidity and mortality rates and guarantee ideal health and nutritional conditions for foetal growth and development.
The results of this study indicate that unhealthy dietary consumption in the third trimester has negative effects on birth weight in two ways. Patterns one (meat, eggs, fried snacks, processed foods) and four (sugars and sweets) directly impact birth weight, by reducing it. The third pattern (coffee and butter / margarine), which also indicates inadequate intake, has the same impact on birth weight, however it acts indirectly, mediating the occurrence of increased morbidity.
These ndings are relevant to the eld of nutritional health for mothers and their infants, since food consumption is considered a modi able factor, and dietary and nutritional interventions during gestation, which are important determinants of birth weight, could contribute positively to the health of the mother and foetus. Pearson correlation between the dietary patterns of the rst and third trimester: Pattern 1: 0,035 (p=0,6); Pattern 2: 0,023 (p=0,7); Pattern 3: 0,085 (p=0,2); Pattern 4: 0,051 (p=0,4).