Dietary patterns during pregnancy and risk of gestational diabetes: a prospective cohort study in Western China

A prospective cohort study was conducted among pregnant women in Western China to identify modifiable risk factors (maternal lifestyle and nutritional status) of adverse pregnancy and infant health outcomes. Full details of the study protocol and recruitment procedures were described elsewhere [29]. Briefly, participants were recruited from four maternity hospitals in Chengdu, Sichuan Province. From May 2015 to August 2015, all pregnant women at 15 to 20 weeks of gestation who attended antenatal care at these hospitals were approached and assessed for eligibility. Eligible women were those aged 18–40 years, with singleton pregnancy, without severe chronic or infectious disease (e.g. HIV), and without any infertility treatment such as in vitro fertilization and intrauterine insemination. Only women who returned their signed consent form were included. In total, 1901 pregnant women were followed up from 15 to 20 weeks of gestation until 12 months postpartum. The study protocol was approved by the Ethics Committee of West China School of Public Health, Sichuan University, and the Human Research Ethics Committee of Curtin University.

Dietary intake was assessed at baseline during 15–20 weeks of gestation via face-to-face interviews conducted by trained nurses, using a semi-quantitative Food Frequency Questionnaire (FFQ) originally developed for residents in Chongqing city (a municipality geographically adjacent to Sichuan province), Western China, with the exception of 10 items (four for oil and six for condiments) which are uncommon or deemed difficult to estimate [30]. The FFQ is based on a previous semi-quantitative FFQ validated among women in Chengdu, Sichuan province [31], with reported mean correlation coefficient of 22 nutrients and intra-class correlation coefficients being 0.66 and 0.65, respectively. It contained 109 food and beverage items in 17 predefined groups, with quantities and frequencies recorded in detail. A standard portion (in grams) was defined for each food item listed. A photo booklet was available to show participants about the standard portion sizes, and they were asked to estimate their intake portion for each food, with the options of 0, 0.5, 0.8, 1.0, 1.5, 2.0 and 3.0 or above. They also chose their intake frequencies: 1–3 times per month, 1–2 times per week, 3–4 times per week, 5–6 times per week, once per day, twice per day, three times per day, and four times or more per day. A daily food frequency intake was then computed. Total food intakes in grams per day were calculated using the product of daily frequency intake and amount of food intake per meal in standard portions. In addition, we adjusted food intakes for energy consumption using a density method by calculating the amount of food intake per 1000 kcal of energy [32].

For the analysis of dietary patterns, 109 food and beverage items originally included in the FFQ were combined into 32 food and beverage groups based on their similarities in nutrient profile (see Appendix 1). Exploratory factor analysis with the principle component method [33] was applied to derive dietary patterns, based on the energy-adjusted intakes of food. Number of factors retained was determined by identifying a break point in the scree plot [34] and interpretability of patterns. Factors were rotated by orthogonal/varimax transformation to improve interpretability and utility. Dietary patterns were named according to the food items that contributed most to the pattern. We chose factor loadings ≥0.3 to represent a high correlation between the food group and the dietary pattern. Accordingly, factor scores for each dietary pattern were calculated by the sum of food intake weighted by the corresponding factor loadings.

GDM status was determined using oral glucose tolerance tests (OGTT). Between 24 and 28 weeks of gestation, participants who were not previously diagnosed as diabetic were routinely scheduled for a 75 g, 2-h OGTT. Blood samples were collected at fasting, at 1 and 2 h after they received the 75 g oral glucose. GDM cases were confirmed using the IADPSG diagnostic criteria [35], including at least one of the following values being met: fasting serum glucose ≥5.1 mmol/L, 1-h serum glucose ≥10.0 mmol/L, or 2-h post prandial serum glucose ≥8.5 mmol/L.

Information on socio-demographic and health characteristics, including maternal age, education, parity, family history of diabetes and cigarette smoking, was collected during the baseline survey. Maternal age was categorized into four groups (< 25, 25–29, 30–34, ≥35 years) for statistical analysis. Educational status was classified by three levels (junior secondary school or below, senior/technical secondary school, university or above). Physical activity during pregnancy was assessed using a validated Pregnancy Physical Activity Questionnaire [36] and expressed in terms of metabolic equivalent tasks (MET) [37]. Height was measured by a stadiometer, and self-reported pre-pregnancy weight was obtained from medical records. Pre-pregnancy body mass index (BMI) was then calculated as pre-pregnancy weight (kg) divided by height squared (m²). Participants with BMI < 18.5, 18.5–23.9 and ≥ 24 kg/m² were classified as underweight, normal and overweight, respectively [38].

Baseline characteristics were compared between groups using t-tests or analysis of variance (ANOVA) for continuous variables, and χ² tests for categorical variables. Logistic regression models were used to ascertain the associations between different dietary patterns and risk of GDM, with the lowest tertile taken as the reference level. Crude and adjusted odds ratios (OR) together with 95% confidence intervals (CI) were presented (see Appendix 2). Linear trend analysis was also performed for the logistic regressions. Potential covariates included age, pre-pregnancy BMI, family history of diabetes, parity, education and total physical activity. We examined potential effect modification by age groups, pre-pregnancy BMI (< 24 kg/m², or ≥ 24 kg/m²) and family history of diabetes (yes, no, or don’t know) on the association between dietary patterns and GDM risk. We tested the significance of interaction by adding multiplicative interaction terms in the models. All statistical analyses were performed using Stata (version 14.2, Stata Corp, Texas, USA). A two-tailed p-value less than 0.05 was considered as statistically significant.

After excluding participants with a history of diabetes before pregnancy (n = 2), incomplete information on OGTT (n = 450) and those with missing data on diet (n = 112), 1337 women were available for analysis. No significant difference was observed between the included and excluded women in terms of age, pre-pregnancy BMI, education, parity, smoking habit and family history of diabetes.

Table 1 summarizes participant characteristics according to the GDM status. Overall, the mean age of participants was 25.8 years (SD = 4.1), and women aged below 25 years accounted for 39.4% of the cohort. The mean pre-pregnancy BMI was 20.7 kg/m² (SD = 2.8), while the prevalence of overweight or obese (BMI ≥24 kg/m²) was 12.6%. In total, 199 women (14.9%) were diagnosed with GDM, who tended to be older, had a higher pre-pregnancy BMI and a greater likelihood of family history of diabetes when compared to their non-GDM counterparts (p < 0.05).

We identified three major dietary patterns that accounted for 21.2% of the total variation, with their rotated factor loadings presented in Table 2. The first pattern, named the ‘plant-based pattern’, was characterized by high intakes of green leafy vegetables, cruciferous vegetables, gourd/melon family vegetables, red or orange vegetables, potatoes, root vegetables, bean vegetables, bean products, mushrooms, fruits, and low intake of lean pork meat. It accounted for 9.0% of the total variance. The second pattern, called the ‘meat-based pattern’, was represented by high intakes of pork, pig blood curd, ox tripe, organ meat, processed meat, squid and mushrooms. It explained 7.1% of the total variance. The third pattern was typified by high intakes of foods rich in protein, including eggs, milk, fish and lean pork meat, and low intakes of noodles and bread. We named it the ‘high protein-low starch pattern’, which explained 5.0% of the total variance.

Table 3 describes participants’ characteristics and their dietary intakes according to tertiles of dietary pattern scores. For the plant-based pattern, women with a higher score were older, more highly educated and physically active than those with a lower score. They had higher intakes of total energy and lower intake of total fat. Also, they consumed more vegetables, fruit, nuts/seeds, fish, eggs, protein and carbohydrate, but less red meat. Regarding the meat-based pattern, women with a higher score were older and had more children than those with a lower score. They had higher intakes of total energy, vegetables, red meat, protein and total fat, but had lower intakes of fruit, nuts/seeds, milk, eggs, grains and carbohydrate. For the high protein-low starch pattern, women with a higher score were more educated and had a lower pregravid BMI when compared to women with a lower score. Those with a higher versus lower score had higher total energy intake and higher consumption of fruit, nuts/seeds, red meat, lean pork, fish, milk, eggs, protein and total fat, while their intake of vegetables, grains, and carbohydrate was lower.

There was no significant association between any dietary pattern and the risk of GDM. Compared with the lowest tertile of dietary pattern scores, multivariable-adjusted ORs for the corresponding highest tertile for the plant-based, meat-based and high protein-low starch patterns were 0.97 (95% CI 0.64 to 1.47, p > 0.05), 0.89 (95% CI 0.58 to 1.36, p > 0.05) and 0.73 (95% CI 0.48 to 1.10, p > 0.05), respectively (Appendix 2).

In subgroup analyses, a significant reduction in GDM risk was observed among overweight women, when comparing the highest tertile of high protein-low starch pattern scores to the lowest tertile (OR 0.29; 95% CI 0.09 to 0.94; p for trend = 0.049, Figure 1), despite the lack of significance for the interaction between pre-pregnancy BMI and high protein-low starch pattern score (p for interaction = 0.134). There was no effect modification by pregravid BMI on the association between the plant-based pattern and meat-based pattern and GDM risk. Neither age nor family history of diabetes modified the risk of GDM in relation to the identified dietary patterns (data not shown for brevity).