Uncovering Patterns for Adverse Pregnancy Outcomes with Spatial Analysis: Evidence from Philadelphia

In this case study, we analyze the risk of stillbirth and preterm birth in Philadelphia from 2010 to 2017. We exploit a rich electronic health records dataset (45,919 deliveries at hospitals within Penn Medicine), and augmented with neighborhood data from 363 census tracts of Philadelphia. We conduct a two-stage statistical analysis. In the first stage, we introduce a Bayesian spatial logistic regression model to study the patient-specific risk of stillbirth and preterm birth. Our model accounts for heterogeneity and spatial autocorrelation between neighborhoods. We find both patient-level characteristics (e.g. self-identified racial/ethnic group) and neighborhood-level characteristics (e.g. violent crime) are highly associated with patient-specific risk of both outcomes. In the second stage, we aggregate the estimates from our spatial model to quantify neighborhood risks of stillbirth and preterm birth. We find that neighborhoods in West Philadelphia and North Philadelphia are at highest risk of these outcomes. Specifically, neighborhoods with higher rates of women living in poverty or on public assistance have higher risk, while neighborhoods with higher rates of women who are college-educated or in the labor force have lower risk. Our Bayesian approach provides meaningful uncertainty measures for these neighborhood risk probabilities and would be useful for public health interventions.