Statistical Modeling of Dietary Intake and Weight Gain During Pregnancy

Abstract

BACKGROUND: Healthy dietary intake and appropriate weight gain are two key components of an ideal pregnancy. The objective of this thesis was to investigate the weight gain pattern of a large cohort of pregnant women and its association with dietary intakes, which may provide valuable information for the clinical intervention of inappropriate gestational weight gain. Two instruments were used to capture the dietary intakes, the level of agreement and possibility of pooling the results need to be studied. METHOD: For a validation sample of 58 child-bearing-age women, the total calories intakes captured by two instruments were compared by the Bland-Altman plot. The intakes of key nutrient captured by the interviewer-administered instrument version were predicted by the nutrient intakes from the web-based version with a regression model. Then we estimated the weight growth trajectories of each subject through functional principal component analysis techniques. The total weight gain predicted from the trajectory was then regressed on the prepregnancy body mass index, and dietary intakes and physical activities which were measured through pregnancy. RESULTS: We found that the relative bias between the two instruments were small, yet the variances in individuals could be large. Energy-adjusted intakes of macronutrients showed reasonable correlations between the two instruments (0.56 for fat, 0.73 for protein, and 0.67 for carbohydrate). LASSO regularization based multiple regression greatly improved the cross-validated R2 for folate from 0.0033 to 0.46. Our estimated weight growth trajectories showed good accuracy when compared to classic mixedeffect models with significant smaller root mean squared error. The predicted weight gain from trajectory had a strong correlation with prepregnancy body mass index, but adding the dietary intake and physical activity information did not improve the R2 of the model. CONCLUSIONS: Direct pooling of the results from the two instruments may not be feasible. But when pooling is considered, energy-adjustment for macronutrients and the LASSO-based multiple regression for micronutrients are recommended. Functional principal component analysis has significant advantages of flexibility and robustness for the weight growth trajectory modeling. We found that the weight gain during pregnancy negatively correlated to prepregnancy BMI, but the dietary intake and physical activities measured in our study did not provide useful information in predicting the weight gain.