Statistical Analyses

As total energy intake is correlated with most nutrients, macronutrient intake was expressed as a percentage of total energy intake using the nutrient-density method, and other nutrients were energy adjusted using the residual method (33). Dietary protein intake was categorized into quartiles to account for potential nonlinear associations, with lowest category used as the reference category in all models. For protein intake from red meat, poultry, seafood, egg, and from beans, participants were divided into 4 categories, where ‘no dietary intake’ was the lowest category, with remaining participants divided into tertiles of protein intakes. Protein intakes were also computed as 5% energy units and 2% energy units to compare the association with GDM for protein from different sources.

Characteristics and dietary intakes of the study population according to protein intake are described using proportions or means and standard deviations, with differences tested using χ² test and linear regression. Multivariable logistic regression models were used to estimate odds ratios (ORs) for risk of developing GDM associated with maternal protein intake. Based on previous studies, models were adjusted for maternal age, education attainment, parity, ethnicity, pre-pregnancy BMI, family history of diabetes mellitus, previous history of GDM, smoking status during pregnancy, alcohol consumption during pregnancy and physical activity during pregnancy (15, 34).

The association of dietary protein intake with risk of GDM was evaluated using the substitution model (33). For this, carbohydrate was substituted with protein in an isocaloric model by simultaneously including the percentages of energy from protein and fat, total energy intake, and all other potential confounders in the model. The effect estimate from this model can be interpreted as the effect of increasing intake of protein at the expense of carbohydrate while keeping calories constant. Similarly, when estimating the effects of a major protein source (e.g., animal protein), carbohydrate was substituted with the protein source by simultaneously modeling all other protein sources (% total energy) with total energy and other confounding factors. The associations with increasing protein intake were examined at the expense of carbohydrate, instead of fat, as this appeared the most suitable macronutrient to replace protein as reflected by the lower intake of carbohydrate among participants with higher protein intakes.

To account for covariates with missing data (Supplemental Figure 1), where the proportion missing ranged from 0.1%-6.6%, missing values were imputed with the mode value for categorical variables, or median values for continuous variables. For pre-pregnancy BMI, BMI was imputed using BMI derived from first pregnancy study visit (<14 weeks gestation) which showed a high correlation with pre-pregnancy BMI (r=0.965).

Several sensitivity analyses were performed. Firstly, the associations between dietary protein intake and the risk of GDM were assessed when models were further adjusted for (a) maternal dietary fiber intake, saturated fat intake, heme iron intake and SSB consumption, and (b) gestational weight gain. Secondly, the effects of substituting carbohydrate with protein intake ascertained from 3 day food records were estimated in a smaller subset of participants (n=607). Thirdly, statistical analyses were performed using (a) non-imputed data by excluding all participants with missing data, and (b) data imputed by multiple imputation. Fourth, the effect of substituting fat, instead of carbohydrate, with protein in an isocaloric model was estimated by simultaneously including the percentages of energy from protein and carbohydrate, total energy intake, and all other potential confounders in the model. As some women had blood glucose testing outside of, and prior to, the 26-28wk study visit, sensitivity analyses excluding these women were also performed. Lastly, the associations between dietary protein intake and the risk of GDM were assessed when models were adjusted for covariates which were significantly associated with total protein intake in univariate analyses. Potential effect modification by maternal age and ethnicity on the association of maternal protein intake and the risk for GDM was evaluated by inclusion of multiplicative interaction terms into the multivariable models.

All statistical analyses were performed using SPSS version 20.0 (IBM). P < 0.05 was considered statistically significant.