Date of Award

January 2013

Document Type

Open Access Thesis

Degree Name

Master of Public Health (MPH)

Department

School of Public Health

First Advisor

Shuangge Ma

Abstract

Background:

The objective of this paper is to evaluate the role of maternal nutrition intake before and during pregnancy on infant birth weight. Self-reported data on intake of 94 nutrients during four time periods (one year before pregnancy and during the three trimesters of pregnancy) was collected from over 7,000 women during 2010-2012 in Gansu Provincial Maternity and Child Care Hospital in Lanzhou, China.

Methods:

Hierarchical clustering and k-means clustering were performed to detect any potentially influential nutrients. Principal Component Analysis (PCA) was used to reduce the dimension of predictors and linear regression analysis was subsequently conducted to fit new nutrient components and confounders on birth weight. Since the Globaltest showed significant influence of nutrient intake in participant group with low pre-pregnancy maternal Body Mass Index (BMI), stepwise selection and Least Absolute Shrinkage and Selection Operator (Lasso) methods were then performed directly over the 94 nutrients in low BMI group.

Results:

Hierarchical clustering and k-means clustering both resulted in six clusters, but the averages of birth weight in the six clusters were not significantly different from each other based on ANOVA result (p-value=0.1229 and 0.1032, for hierarchical and k-means clustering, respectively). Principal Components Analysis (PCA) was then performed and selected 10 new nutrient components to represent the original 94 nutrients. The following step was to fit the 10 new components and 9 confounders on birth weight in a linear regression model. This procedure was repeated for the four time periods. For each time period, there were two components showing strong association with birth weight. Comparing to the base model with only confounders, full models with nutrient components had very small R-squared increase (around 0.01). According to the Globaltest result, the overall effect of 94 nutrients showed strong association with birth weight in low pre-pregnancy maternal BMI group in all four time periods (p-values=0.0439, 0.0033, 0.0017, and 0.0024, for four time periods, respectively). In the model selection part, stepwise selection resulted in two significant nutrient variables out of 94 variables for each time period. Specifically, variable `insoluble dietary fiber' showed strong association with birth weight (p-value=0.0022, 0.0051 and 0.0062, for pre-pregnancy, the 2nd and 3rd trimester, respectively). In addition, a few vitamin nutrients showed strong association with birth weight (p-values < 0.0001), but the estimated coefficients were very small. Lasso method showed similar results as stepwise selection. Few nutrients showed significant influence but with very small estimated coefficients, and R-squared of the full models had very small increase compared to the base model with only confounders.

Conclusions:

Several methods have been tried to test the association between maternal nutrition intake and birth weight of newborns, such as clustering on predictors and observations, Globaltest, stepwise selection and Lasso method in linear regression. These methods showed consistent results that overall maternal nutrition intake was significantly associated with infant birth weight in low maternal BMI group, and a few individual nutrients showed significant association. It is suggested that instead of focusing on altered consumption of individual nutrients, overall maternal nutrition intake should be improved to help control birth weight in the normal range.

Comments

This is an Open Access Thesis.

Share

COinS