Statistical methods for modeling the association between repeated measures of hemoglobin during pregnancy and adverse birth outcomes Restricted; Files Only

Abstract

Introduction: 

Maternal nutrition status during pregnancy plays a crucial role in the health, growth, and development of the fetus and the newborn infant. While some studies may investigate one single measurement, more measurements will provide more information about the biomarker trajectory and how this will associate with birth outcomes. 

Methods: 

We applied 7 methods that model the association between a longitudinal biomarker and a binary, time-invariant outcome: 1) Window-specific regression; 2) Min/Max/Mean value; 3) Multivariable logistic regression; 4) Conditional Method; 5) Distributed Non-Linear Model; 6) Two-Stage Mixed Effect Model; 7) Generalized Additive Mixed Model (GAMM). We examined the repeated measures of hemoglobin (Hb) concentration across pregnancy in association with low birth weight (LBW), preterm birth (PTB), and small for gestational age (SGA) using the pregnancy woman data from the Biomarkers Reflecting Inflammation and Nutritional Determinants of Anemia (BRINDA) which contains 8 datasets from different countries.

Results: 1) The parallel models identified the sensitive window of SGA in 3rd trimester and that of preterm in 1st and 2nd trimesters. 2) Model using maximum Hb detected a risk of higher maximum value Hb in preterm. 3) Multivariable logistic regression indicated that a lower value in 2nd trimester tends to be associated with the probability of SGA. 4) Distributed Non-Linear Model indicated a sensitive window of SGA in 2nd trimester. 5) Two-Stage mixed effect model found that a higher initial value of Hb at the start of gestation significantly decreased the risk of LBW and preterm and a rapid increase in Hb or slower decrease in Hb during pregnancy will decrease the risk of SGA. 6) GAMM shown a significantly lower Hb value in women with preterm around 1st and 2nd trimester.

Conclusions: Methods 4, 5, 6 are preferred when investigators are interested in the biomarker pattern. GAMM emphasizes visualization while others could provide effect estimations. Our study provides extremely useful insight into the pros and cons of statistical methods modeling time-varying predictors in association with not-time-varying outcomes.

Keywords: Biomarkers, Birth outcomes, Repeated measures, Statistical methods

Table of Contents

Introduction 1

Methods 2

Data source 2

Notation 6

Statistical Methods 6

Results and Discussion 7

Standard Methods 7

Logistic regression using conditional Hb 10

Distributed lag non-linear model 11

Two-stage mixed effects model 13

Generalized additive mixed model 15

Conclusion: 17

References 18

About this Master's Thesis

Rights statement

• Permission granted by the author to include this thesis or dissertation in this repository. All rights reserved by the author. Please contact the author for information regarding the reproduction and use of this thesis or dissertation.

School	Rollins School of Public Health
Department	Biostatistics
Subfield / Discipline	Biostatistics - MPH & MSPH
Degree	M.P.H.
Submission	Master's Thesis
Language	English
Research Field	Biology, Biostatistics Health Sciences, Nutrition
Keyword	Biomarkers Birth outcomes Repeated measures Statistical methods
Committee Chair / Thesis Advisor	Yi-An, Ko, Emory University
Committee Members	Hanqi, Luo, Emory University Young, Melissa Fox, Emory University

Last modified

Primary PDF

Thumbnail	Title	Date Uploaded	Actions
	File download under embargo until 18 May 2025	2023-04-10 10:47:31 -0400	File download under embargo until 18 May 2025

Supplemental Files

Thumbnail	Title	Date Uploaded	Actions