METHODS: Merged live births and fetal deaths of 2010-2013 were geocoded and spatially joined to the Census five-year (2009-2013) American Community Survey data to incorporate sociodemographic factors at the Census tract level. VLBW and maternal risk factors were assessed for each Census tract. The K-means approach was used to form spatial peer groups based on the EM algorithm which maximizes variability of studied covariates between groups and minimizes within-group variability.
RESULTS: The social-spatial partition concluded with 4 peer groups, encompassing from 66 to 196 Census tracts. The variables that were most critical in the assignment of Census tracts to peer groups were maternal education, maternal race/ethnicity, median household income, prenatal care adequacy, linguistic isolation, teen motherhood, older maternal age, and maternal obesity pre-pregnancy. Peer groups 1 and 2 stood out as neighborhoods most in need of high risk family support programs, characterized by a high percentage of African American and Hispanic populations, high poverty and teen birth rates, low levels of education, language barriers, low prenatal care uptake, high maternal obesity prevalence, low food access, and above-average VLBW rates. The mapped peer groups were disseminated among stakeholders and used in conjunction with PPOR findings to determine where efforts to improve maternal health and preconception care can be targeted.
CONCLUSIONS: Geospatial analysis was integrated into the PPOR process to investigate and explain spatial patterns of VLBW outcomes. Stakeholders were engaged and motivated to investigate the local rise in excess FIDs and to focus prevention efforts on the most affected peer groups in the community.