부산대학교 도서관

The rate of preterm birth in Puerto Rico is among the highest in the United States, a serious public health issue. Environmental exposures, particularly contaminated drinking water and phthalates, chemicals widely used in industrial and consumer applications, are thought to contribute to preterm birth. We used data from a birth cohort based in the northern region of Puerto Rico, known as PROTECT, to examine phthalate levels in drinking water and corresponding phthalate concentrations in the urine of pregnant women. We identified and evaluated predictors of preterm birth based on these findings. In this project, we performed two analyses: i) compared tap water contaminant concentrations and participant health status, and ii) evaluated associations between water and urinary concentrations of phthalate metabolites. We leveraged Decision Trees to predict preterm birth and learn the important features. We addressed the issue of using highly imbalanced data by utilizing an AUC split criterion. For the tap water model we found the key features included health indicators, phthalate concentrations, and water pipe materials. For the urine phthalate model we found the most important features to be MCNP and MCOP, followed by 12 additional phthalates. The most important features for the tap water model indicated that health, water processing, and phthalate levels were crucial to accurately predicting adverse birth outcomes using our analysis.