Back to Annual Meeting

Back to Annual Meeting

APHA Scientific Session and Event Listing

Recorded presentation

Sara L. Stewart, MS¹, Karen L. Courtney, RN, MSN², Mihail Popescu, PhD², and Linda K. Goodwin, PhD, RN³. (1) University of Missouri NLM Fellowship, 8247 Starr Grass Dr, Madison, WI 53719, 608-239-6656, sara.seonaid@gmail.com, (2) Department of Health Managemet and Informatics, University of Missouri- Columbia, 324 Clark, Columbia, MO 65211, (3) School of Nursing, Duke University, Trent Dr, Durham, NC 27707

Preterm birth is one of the leading causes of infant mortality in industrialized countries. There is a growing body of literature that suggests socio-economic and socio-demographic factors may be causal factors that either explain or confound disparities. Understanding differential patterns in birth outcomes is necessary to develop effective interventions designed to decrease birth outcome disparities amongst pregnant women.

The purpose of this study is to use statistical and computational modeling to explore predictive factors in existing datasets. This study is a retrospective, secondary analysis of de-identified, North Carolina birth record data (http://www.schs.state.nc.us/SCHS). The following methods are being used to explore predictive models: Logistic regression, CART, Support Vector Machines, and Neural Networks. The parsimonious statistical model for preterm birth prediction includes the following variables: marital status, mother's race, mother's age, father's age, mother's education, number of cigarettes smoked and adequacy of prenatal care (Kotelchuck index). An interaction between mother's age and race is also explored as well as interactions between mother's age and various medical history issues that have been associated with preterm birth. Using logistic regression, this model has a predictive accuracy of .701 (ROC).

This presentation will show the comparison of the statistical modeling with the computational models. Our model results are consistent with other studies found within the literature; however our models were generated from data already being collected routinely during each pregnancy. Public health implications for both the methods used and the factors exposed will be reviewed.

Learning Objectives: Following this session, the participant will be able to

• Describe how existing publicly available databases, such as birth certificates can be used for knowledge discovery for health conditions
• Discuss triangulation of prediction models using multiple statistical and computation modeling approaches
• Identify important prediction factors for preterm birth from routinely collected data
• Discuss the implications of using population databases to identify potential intervention areas

Keywords: Pregnancy Outcomes, Information Databases

Presenting author's disclosure statement:

Any relevant financial relationships? No

Recorded presentation