Extending Functional Principal Component Analysis to Model Weight Gain During Pregnancy

• Author / Creator

  Shulman, Lisa

• BACKGROUND: Achieving an appropriate gestational weight gain (GWG) is important during pregnancy. Inadequate and excessive GWG have been linked to various negative pregnancy outcomes and future health issues for both the mother and fetus. One useful intervention is to provide individualized assessment and counselling at the beginning of pregnancy and throughout pregnancy as needed, to help women follow dietary and physical activity patterns that support optimal weight gain. Currently, pregnant women receive a personalized weight gain goal, usually a weight gain chart, that is based on their pre-pregnancy body mass index (BMI). However, these weight gain charts cannot be adjusted to account for individual weight measurements during pregnancy. Interventions focused on the progress of weight gain throughout pregnancy could thus benefit from a personalized weight growth trajectory.

  A previous study used the functional principal component analysis (FPCA) approach and successfully estimated individual weight trajectories in pregnant women. However, the FPCA method borrows information from the whole cohort, and does not incorporate information on women's pre-pregnancy BMI. The objective of this thesis was to extend FPCA by incorporating additional BMI category-specific principal components (PCs) in trajectory modelling of gestational weight gain. The new proposed method, called JIVE--FPCA, applies the joint and individual variation explained (JIVE) algorithm to FPCA, and can be used to estimate individual trajectories from any sparse, longitudinal multi-block dataset.

  METHODS: Weight data during pregnancy and early postpartum were collected from a large cohort ($n = 1648$) of pregnant and postpartum women. Longitudinal weight measurements were irregularly spaced and obtained from multiple data sources. Sparse, longitudinal multi-block data were simulated according to the JIVE--FPCA model, and closely mimicked the real, gestational weight data. The FPCA and JIVE--FPCA methods were then applied to both the simulated and real datasets. The performances of the two methods were compared on the basis of the mean squared error and average confidence bandwidth.

  RESULTS: Studies with both simulated and real data showed that the JIVE--FPCA method provides a significantly better fit to individual trajectories than the original FPCA approach. This was evident both visually and numerically. The mean squared error and average confidence bandwidth were appreciably lower for the JIVE--FPCA method than for FPCA. In the application to gestational weight data, the JIVE--FPCA approach successfully captured differences in GWG patterns among different pre-pregnancy BMI categories. In particular, it was found that the weight trajectories for women with pre-pregnancy obesity were more gradually rising than those for women with lower pre-pregnancy BMI. The mean JIVE--FPCA weight trajectories for overweight and obese women were similar, and exhibited a different pattern than the trend displayed by the mean weight curves for underweight and normal weight women.

  CONCLUSIONS: This thesis presents the JIVE--FPCA approach, which is an extension of the existing FPCA method. The advantage offered by JIVE--FPCA is that it can simultaneously capture patterns that are shared across multiple blocks of data and patterns that are specific to a particular block. Trajectory modelling of gestational weight gain indicated that the JIVE--FPCA method leads to a significant improvement in explaining the weight variation. The new method also highlights the differences in GWG patterns between pre-pregnancy BMI classes.

• Subjects / Keywords

  • functional principal component analysis
  • gestational weight gain
  • pregnancy
  • simulation study

• Graduation date

  Fall 2019

• Type of Item

  Thesis

• Degree

  Master of Science

• DOI

  https://doi.org/10.7939/r3-w7pz-c198

• License

  Permission is hereby granted to the University of Alberta Libraries to reproduce single copies of this thesis and to lend or sell such copies for private, scholarly or scientific research purposes only. Where the thesis is converted to, or otherwise made available in digital form, the University of Alberta will advise potential users of the thesis of these terms. The author reserves all other publication and other rights in association with the copyright in the thesis and, except as herein before provided, neither the thesis nor any substantial portion thereof may be printed or otherwise reproduced in any material form whatsoever without the author's prior written permission.