Predicting newborn birth outcomes with prenatal maternal health features and correlates in the United States: a machine learning approach using archival data

Newborns are shaped by prenatal maternal experiences. These include a pregnant person’s physical health, prior pregnancy experiences, emotion regulation, and socially determined health markers. We used a series of machine learning models to predict markers of fetal growth and development—specifically, newborn birthweight and head circumference (HC).

We used a pre-registered archival data analytic approach. These data consisted of maternal and newborn characteristics of 594 maternal-infant dyads in the western U.S. Participants also completed a measure of emotion dysregulation. In total, there were 22 predictors of newborn HC and birthweight. We used regularized regression for predictor selection and linear prediction, followed by nonlinear models if linear models were overfit.

HC was predicted best with a linear model (ridge regression). Newborn sex (male), number of living children, and maternal BMI predicted a larger HC, whereas maternal preeclampsia, number of prior preterm births, and race/ethnicity (Latina) predicted a smaller HC. Birthweight was predicted best with a nonlinear model (support vector machine). Occupational prestige (a marker similar to socioeconomic status) predicted higher birthweight, maternal race/ethnicity (non-White and non-Latina) predicted lower birthweight, and the number of living children, prior preterm births, and difficulty with emotional clarity had nonlinear effects.

HC and birthweight were predicted by a variety of variables associated with prenatal stressful experiences, spanning medical, psychological, and social markers of health and stress. These findings may highlight the importance of viewing prenatal maternal health across multiple dimensions. Findings also suggest that assessing difficulties with emotional clarity during standard obstetric care (in the U.S.) may help identify risk for adverse newborn outcomes.