A priori and a posteriori dietary patterns among pregnant women in johannesburg, south africa: The nuped study

Dietary pattern analyses allow assessment of the diet as a whole. Limited studies include both a priori and a posteriori dietary pattern analyses. This study aimed to explore the diet of pregnant women in urban South Africa through both a priori and a posteriori dietary pattern analyses and associated maternal and household factors. Dietary data were collected during early pregnancy using a quantified food frequency questionnaire from 250 pregnant women enrolled in the Nutrition During Pregnancy and Early Development (NuPED) cohort. A priori dietary patterns were determined using the Diet Quality Index-International (DQI-I), and a posteriori nutrient patterns using exploratory factor analysis. Based on the DQI-I, the study population followed a borderline low-quality diet. Three a posteriori nutrient patterns were identified: Pattern 1 “plant protein, iron, thiamine, and folic acid”; pattern 2 “animal protein, copper, vitamin A, and vitamin B12”; pattern 3 “fatty acids and sodium”. Pattern 1 was associated with higher dietary quality (p < 0.001), lower maternal educational level (p = 0.03) and socioeconomic status (p < 0.001). Pattern 3 was significantly associated with lower dietary quality. The low dietary quality among pregnant women residing in urban South Africa should be addressed to ensure optimal maternal and offspring health outcomes. This study was embedded within the Nutrition During Pregnancy and Early Development (NuPED) cohort, a prospective study conducted in Johannesburg, South Africa. The details of the NuPED study have previously been described [25]. Data collection took place at early pregnancy (<18 weeks’ gestation), mid-pregnancy (±22 weeks), late pregnancy (±36 weeks), birth, as well as postnatally (6 weeks, 6 months, 7.5 months, and 12 months’ postnatal age). The current analysis reports on data collected at baseline (early pregnancy). In brief, the recruitment of generally healthy urban pregnant women from primary health care clinics in Johannesburg took place between March 2016 and November 2017. Based on the South African socioeconomic profile, it is mainly the poorer proportion of the population that makes use of medical care provided by primary health care facilities [26]. Women were regarded as eligible for inclusion if they were 18 to 39 years old, less than 18 weeks of gestation with a singleton pregnancy, born within South Africa or in a neighboring country (Lesotho, Swaziland, Zimbabwe, Botswana, or Namibia), and able to effectively communicate in the local languages (English, Afrikaans, Sotho, Zulu or Xhosa). Women born in a neighboring country had to have been living in South Africa for at least 12 months. Pregnant women diagnosed with a non-communicable disease (namely, hypertension, hypercholesterolemia, diabetes, and renal disease), infectious disease (namely, tuberculosis and hepatitis), or serious illness (namely, lupus, cancer, and psychosis) were excluded. Women were also excluded if they used illicit drugs (self-reported) and/or smoked (currently or in the past year). As South Africa has a high prevalence of human immunodeficiency virus (HIV) infections (22.7% of women of reproductive age [27]), HIV status was not considered exclusion criteria for enrolment. Five-hundred and ninety-five pregnant women were invited to take part in the NuPED study of which 313 (53%) women volunteered. A further 63 women were excluded for not meeting the inclusion and exclusion criteria. In total, 250 pregnant women provided informed consent and were enrolled. Participants were followed-up at a tertiary hospital (Rahima Moosa Mother and Child Hospital) where data collection took place alongside routine maternal care. All 250 pregnant women completed data collection at baseline (<18 weeks’ gestation). All women eligible and willing to partake in the study were provided with an informed consent form during recruitment. Written informed consent was obtained during the first study visit before any data were collected. The study was conducted according to the guidelines set out by the World Medical Association Declaration of Helsinki. Ethical approval was obtained from both the Human Research Ethics Committees of the North-West University, Potchefstroom (NWU-00186-15-A1) and the University of the Witwatersrand, Johannesburg (M 150968). The City of Johannesburg District Research Committee (Gauteng Department of Health) and the clinical manager at Rahima Moosa Mother and Child hospital provided permission for the study to take place in the relevant clinical setting. Dietary intake was assessed during early pregnancy using an interviewer-administered quantified food-frequency questionnaire (QFFQ). The 140-item QFFQ was validated previously for energy and nutrient intake in an African population in South Africa [28], and its reproducibility was established [29,30]. Trained fieldworkers used standard probing questions [31] to inquire in as much detail as possible about the type/brand, cooking method, frequency, and amount of all foods and drinks consumed over the past four weeks. Household utensils, together with “dish-up and measure”, food packaging material, and two- and three-dimensional food models, were used to quantify the portion sizes reported. Portion sizes were converted to grams by three registered dietitians/nutritionists using the South African Medical Research Council (SAMRC) Food Quantities Manual [32]. All foods and drinks were coded according to the South African Food Composition Database [33]. Daily energy, macro-, and micronutrient intake were calculated by the SAMRC by linking the coded and quantified dietary intake data to the most recent food composition database available at the time. The daily dietary intake data are reflective of nutrient intake from food, excluding any micronutrient supplementation used. Maize meal and wheat flour are subject to mandatory micronutrient fortification as per South African legislation [34], and these fortified micronutrients (vitamin A, thiamine, riboflavin, niacin, pyridoxine, folic acid, iron, and zinc) were considered in the nutrient analysis. All participants provided complete dietary intake information. The a priori dietary patterns were defined using the Diet Quality Index-International (DQI-I) [35]. Validated for international use and cross-national comparison of dietary quality [36,37], the DQI-I is regarded as a valuable tool for exploring diet quality related to nutrition transition as it accounts for both dietary concerns contributing to chronic disease and undernutrition. South Africa is known to be undergoing a rapid nutrition transition alongside urbanization [38], contributing to a growing burden of malnutrition within the country [39]. In addition, being sensitive towards various dietary practices and allowing for the use of country-specific recommendations in the evaluation of the diet, the DQI-I serves as a good set of guidelines according to which dietary quality of populations can be evaluated [35]. The DQI-I describes all dimensions of a high-quality diet, namely variety, adequacy, moderation, and overall balance. Each of these categories contains components (including both nutrient and food-group indicators) according to which the diet is assessed, allowing for users to determine the aspect(s) of the diet which most likely needs improvement [40]. The four categories of the DQI-I amount to an overall score of 100, with a higher score indicating greater adherence to a higher quality diet. According to the criteria given by Kim et al. [35], a score of less than 60 out of a possible 100 is indicative of a poor quality diet. The daily dietary intake for the study participants, based on the QFFQ, was scored as specified by Kim et al. [35]. Adjustments to the DQI-I criteria were made as follows: (1) serving sizes were determined both according to the guidelines provided by the South African Food-Based Dietary Guidelines [41] and the United States’ Food Guide Pyramid serving size definitions [42]; (2) adequacy of micronutrient intake (iron, calcium, and vitamin C) was based on the Dietary Reference Intakes for pregnant women [43]; (3) the cut-off values for sodium were adjusted as the QFFQ used does not report on discretionary salt intake. According to a paper by Wentzel-Viljoen et al. [44], it is estimated that about 60% of the total salt intake among the South African population is from processed foods, with the remainder (40%) being the discretionary use of salt during food preparation and meals. Based on this observation, the sodium cut-off values were adjusted as follows: ≤1440 mg (6 points), >1440–2040 mg (3 points) and >2040 mg (0 points). The DQI-I scoring criteria, together with how the study participants scored, are indicated in Table 1. Participant scores according to the Diet Quality Index-International (DQI-I) scoring criteria (n = 250) [35]. RDA: recommended dietary allowance; AI: adequate intakes; MUFA: monounsaturated fatty acids; SFA, saturated fatty acids; PUFA: polyunsaturated fatty acids; P/S, ratio of PUFA to SFA intake; M/S, ratio of MUFA to SFA intake. 1 n = 250. 2 Used as a continuous variable. 3 Based on 1700 kcal (7118 kJ)/2200 kcal (9211 kJ)/2700 kcal (11,304 kJ) diet; 1 kcal = 4.1868 kJ. 4 Based on sodium from food sources, excluding the discretionary use of table salt. 5 Ratio of energy from carbohydrate to protein to fat. Nutrient patterns were determined from 22 nutrients by exploratory factor analysis [45]. Using the dietary intake data obtained from the QFFQ, total protein was divided into animal- and plant protein, and total fat into saturated-, monounsaturated- and polyunsaturated fatty acids. Carbohydrate intake was reported as total carbohydrate- and total sugar intake. The correlation structure of the nutrient intake data was explored for stability using the Spearman and Pearson correlation coefficients. The difference between the coefficients was minimal (less than 0.1) and, therefore, raw nutrient data were used for the analysis. Nutrient data were energy-adjusted according to the nutrient density model proposed by Willet et al. [46]. The principal factor method was applied with the correlation matrix, and the reliability of the factor analysis verified using the Kaiser–Meyer–Olkin measure of sampling adequacy and Bartlett’s test of sphericity. A Kaiser–Meyer–Olkin value of more than 0.5 and significance (p < 0.05) on Bartlett’s test were regarded as adequate. Factors were retained and interpreted for further analysis according to the eigenvalues (more than 1.00), scree plot inspection, the proportion of the total variance explained, and the natural interpretability of the factors. Retained factors (patterns) were rotated by Varimax (orthogonal) rotation for a simpler structure and to improve interpretation. Patterns were named and interpreted according to nutrients with an absolute factor loading of equal or greater than 0.40. A positive factor loading indicates a nutrient to be positively associated with the nutrient pattern and a negative loading that a nutrient is negatively associated with the pattern. Factor scores were calculated for each participant to indicate the degree towards which a participant’s nutrient intake conforms to the identified patterns. Factor scores were computed by weighting the standardized intakes of each nutrient with their factor loadings and then summing the nutrients within the respective patterns [47]. Several maternal and household factors were considered as possible determinants for dietary patterns during pregnancy. Interviewer-administered questionnaires were used to obtain sociodemographic and socioeconomic information, information on household-level food insecurity, medical history, and risk for prenatal depression and fatigue. Sociodemographic and socioeconomic information obtained included maternal age, ethnicity, country of birth, marital status, educational level, employment status, and household beneficiaries of social support grants. Living standard data were also obtained to allow for classification of the socioeconomic status according to the Living Standard Measure in South Africa [48]. Food insecurity was assessed using the Community Childhood Hunger Identification Project (CCHIP) index [49,50]. This index investigates food insecurity as reflected by access to food on three levels, namely, the household-, adult/individual- and the child level. The extent of household food insecurity (CCHIP index total score) could not be determined as a great proportion of participants (n = 113; missing data n = 9) were not the parent or caretaker of a child in that household. We therefore only report on the three questions assessing household-level food insecurity, namely: (1) “does your household ever run out of money to buy food”; (2) “do you ever rely on a limited number of foods to feed your family because you are running out of money to buy food for a meal”; (3) “do you ever cut the size of meals or skip meals because there is not enough money for food” [49]. Households were regarded as “at risk for food insecurity” if the participant answered “yes” to one or more of the three questions; “no risk” was considered when the participant answered “no” to all three questions. Medical history considered in this study included parity, pregnancy symptoms (nausea and/or vomiting experienced), and HIV status (routinely tested as part of antenatal care). Participants recorded pregnancy symptoms (nausea and/or vomiting) during the seven days after enrolment using a pictorial morbidity calendar. Gestational age was confirmed using fetal ultrasonography during early pregnancy. Prenatal fatigue and depression were assessed using the Multidimensional Assessment of Fatigue (MAF) scale [51] and the Edinburgh Postnatal Depression Score (EPDS) [52,53], respectively. MAF measures the severity of fatigue, distress, and the degree to which it influences daily activities. The global fatigue index (GFI) was calculated with higher scores indicating higher levels of fatigue, and a cut-off of 28 out of a possible 50 was used to define women with prenatal fatigue [54]. A cut-off value of ≥9 out of a possible 30 was used for the EPDS to indicate the risk for depressive symptoms [53]. Maternal anthropometric measures were also assessed although not considered as possible determinants for the dietary patterns in this analysis. The exclusion of maternal weight as a determinant is based on the observation by Doyle et al. [55] that the relationship between maternal weight and dietary patterns is more difficult to interpret as maternal weight can serve both as an outcome and a determinant of dietary patterns. Maternal height, weight, and mid-upper arm circumference (MUAC) were assessed during early pregnancy using the standardized methods indicated by the International Society for the Advancement of Kinanthropometry [56]. All measurements were taken twice and recorded to the nearest 0.05 kg for weight and 0.1 cm for height and MUAC. A calibrated scale was used for weight measurements (Seca 813, Hamburg, Germany), a mobile stadiometer for height (Seca 213, Hamburg, Germany), and a non-stretchable metal tape for the MUAC. As weight gain and edema during pregnancy decrease the reliability of the body mass index (BMI) [57], maternal weight at enrolment was adjusted (assuming that 2.5 kg was gained during the first 14 weeks of gestation) according to the gestational weight gain guidelines provided by the Institute of Medicine [58] to calculate an approximate pre-pregnancy BMI. The MUAC cut-offs for underweight and obesity as indicated by the South African Maternity Care guidelines [59] were also considered to evaluate maternal nutritional status, i.e., ≤23 cm considered as underweight and ≥33 cm as obese. The sample size was set as 250 participants; calculations of the sample size have been previously published [25]. IBM SPSS version 26 (SPSS Inc., Chicago, IL, USA) was used for data processing and statistical analysis. Dietary data were captured in Microsoft Excel, with all other raw data captured in Microsoft Access (Microsoft Corporation, Washington, DC, USA). All dietary data were double-checked to ensure accuracy of food codes used and amounts captured. Twenty percent of all other raw data were randomly checked for correctness. Data were tested for normality using histograms, q-q plots, and the Shapiro–Wilk test. For normally distributed data, continued variables are reported as mean and standard deviation, and for non-normally distributed data as the median and interquartile range (IQR). Categorical data are reported as frequencies and percentages. Both the DQI-I total scores and nutrient pattern factor scores were divided into tertiles for further analysis. To determine food groups associated with each nutrient pattern, analysis of variance (ANOVA) was used to determine the difference between each food group and the nutrient pattern tertiles. The 140-items of the QFFQ were aggregated into 39 food groups expressed as a percentage of the total energy intake (Tables S1–S3). ANOVA was also used to determine differences between DQI-I scores and the nutrient pattern tertiles. In both instances, post-hoc comparisons for unequal variance were carried out using the Games–Howell test. Hierarchical multiple linear regression models were applied to test for associations between the dietary patterns and maternal- and household factors. Hierarchical steps were based on the conceptual framework for multiple determinants of diet during pregnancy as proposed by Doyle et al. [55] and were as follows: step 1 investigates all sociodemographic factors (maternal age, level of education, and marital status), step 2 the sociodemographic and socioeconomic factors (employment, social grants received by the household, household-level food insecurity, and living standard measure), step 3 the sociodemographic-, socioeconomic-, pregnancy-related factors (parity, nausea- and/or vomiting experienced) and HIV status, and step 4 the sociodemographic-, socioeconomic-, pregnancy-related factors, HIV status and psychological factors. Significance was defined as p < 0.05.