(374) Community features incorporating multiple domains outperforms composite indices in predicting NC county-level perinatal outcomes

Community features influence health outcomes; these data are increasingly available and commonly incorporated into research and clinical care. However, it is unknown if multiple individual community domains or composite community indices better predict perinatal outcomes.

Study Design:

NC county-level data were obtained online. Primary perinatal outcomes = county-level PTB < 37 wks, LBW, infant mortality, breastfeeding (BF) initiation, and short IPI (< 6 mo). We adapted an established community framework to create models with 6 community domains (health behaviors, clinical care & quality, SES factors, physical environment, length of life, quality of life), Figure 1A. For each outcome, we selected the most predictive factor from each domain using linear regression; the relative contribution of each domain to each outcome was calculated using general dominance analysis. Next, we selected the most predictive composite index (evaluating the social vulnerability index, childhood opportunity index, and reproductive maternal vulnerability index) for each outcome. Logistic regression was used to evaluate which model best predicted the highest (e.g., worst) quartile for each outcome.

 

Results:

The median rates of PTB (11.5%, IQR 10.1-12.8%), LBW (9.2%, IQR 8.4-10.2%), infant mortality (6.0%, IQR 0.0-8.0%), BF initiation (78.5%, IQR 68.7, 83.5), and short IPI (13.7%, IQR 11.8-15.2%) varied widely across the 100 NC counties. The optimal county-level domain predictor and the relative contribution of each domain to each outcome differed by perinatal outcome (Figure 1B). For all outcomes, the multiple domain models outperformed composite index models (higher AUCs), and the AUC was statistically significantly higher for infant mortality (p< 0.001) and BF-initiation models (p< 0.001), Table 2.

 

Conclusion:

A range of non-conventional community features in multiple domains are associated with perinatal outcomes; these vary by perinatal outcome. These data underscore the importance of considering individual community factors in multiple domains (vs. composite community indices) when quantifying neighborhood exposures.