Adverse pregnancy outcomes are a major contributor to morbidity and mortality and constitute a significant burden on individuals, families, and the healthcare system. Understanding the risk factors for adverse birth outcomes is crucial for prevention and intervention. The aim of this dissertation was to conduct three studies to assess the risk of common and rare medical, behavioral, and pharmacologic exposures on birth defects, low birth weight, preterm birth, and small-for-gestational age, and to compare logistic regression and machine learning methods to predict small-for-gestational age (SGA) using data from the National Birth Defects Prevention Study (NBDPS; 1997–2011).
In Study 1, we assessed the extent to which first trimester nausea and vomiting of pregnancy (NVP) and its treatments were associated with 37 specific cardiac and non-cardiac birth defects. We hypothesized that gestational parents with NVP would have decreased risks of oral clefts and hypospadias and among parents who treat their NVP, common medications and vitamin remedies would not increase risks of specific birth defects. We found compared to no NVP, gestational parents with NVP had ≥10% reduction in risk of cardiac and non-cardiac defects overall, and of 18 specific defects. Over-the-counter antiemetic use, compared to untreated NVP, was associated with ≥10% increase in risk for 9 defects, whereas use of prescription antiemetics increased risk ≥10% for 7 defects. We observed increased risks for promethazine and tetralogy of Fallot (aOR:1.49, 95% CI: 1.05–2.10), promethazine and craniosynostosis (1.44, 1.08–1.92), ondansetron and cleft palate (1.66, 1.18–2.31), pyridoxine and heterotaxy (3.91, 1.49–10.27), and pyridoxine and cataracts (2.57, 1.12–5.88).
In Study 2, we assessed the extent to which nausea and vomiting of pregnancy, according to timing, duration, and severity, and its treatments were associated with adverse pregnancy outcomes, including low birth weight, small-for-gestational age, preterm delivery and gestational length. Additionally, we examined if these effects are modified by periconceptional multivitamin use. We hypothesized that gestational parents with NVP would have a decreased risk of low birth weight, small-for-gestational age, preterm birth, and gestational length compared to those without NVP. Among those who treated their NVP, we hypothesized specific treatments would not be associated with an elevated risk of adverse birth outcomes compared to parents with NVP who do not use any treatments. Additionally, we hypothesized gestational parents with NVP who used multivitamins periconceptionally would have a stronger protective effect for the adverse birth outcomes. We found any versus no NVP was not associated with any of the adverse pregnancy outcomes of interest. NVP in months 4–6 (aOR 1.21, 95% CI: 1.00, 1.47) and 7–9 (aOR 1.57, 95% CI: 1.22, 2.01) of pregnancy were associated with an increase in the risk of preterm birth. NVP lasting one trimester in duration was associated with decrease in risk of SGA (aOR: 0.74, 95% CI: 0.58, 0.95), and NVP present in every trimester of pregnancy had a 50% increase in risk of preterm birth (aOR: 1.50, 95% CI: 1.11, 2.05). For NVP in months 7–9 and preterm birth, ORs were elevated for moderate (aOR: 1.82, 95% CI: 1.26, 2.63), and severe (aOR: 1.53, 95% CI: 1.06, 2.19) symptoms. NVP was not significantly associated with low birth weight. We observed no increased risk of adverse outcomes associated with use of the medication. We could not draw conclusions of interactive effects of multivitamins and NVP because of small sample sizes.
In study 3, we explored using machine learning methods compared with regression methods to predict small-for-gestational age from a variety of gestational parental medical, behavioral, and demographic factors. We hypothesized several important predictive factors for SGA, representing diverse mechanistic pathways, would emerge, and machine learning methods would yield greater predictive value than logistic regression. We created four main models to predict SGA: two using logistic regression methods, and two using random forest methods, using all literature-informed or subsets of variables. With variable selection and other tuning methods, we attempted to maximize predictive ability. However, the models demonstrated low accuracy in predicting SGA, with, contrary to our hypothesis, a slight advantage of the logistic regression models. / 2026-05-10T00:00:00Z
| Identifer | oai:union.ndltd.org:bu.edu/oai:open.bu.edu:2144/48742 |
| Date | 11 May 2024 |
| Creators | Schrager, Nina Lauren |
| Contributors | Werler, Martha M |
| Source Sets | Boston University |
| Language | en_US |
| Detected Language | English |
| Type | Thesis/Dissertation |
| Rights | Attribution-NonCommercial-NoDerivatives 4.0 International, http://creativecommons.org/licenses/by-nc-nd/4.0/ |
Page generated in 0.0085 seconds