319 - Effect of Improving Neighborhood Resources on Pediatric Asthma Outcomes: Insights for Targeted Pathways

Publication Number: 2308.319

Jonathan M. Gabbay, The Children's Hospital at Montefiore, White Plains, NY, United States; Melissa Fazzari, Albert Einstein College of Medicine, Bronx, NY, United States; Benjamin Bajaj, Massachusetts General Brigham, Durham, NC, United States; Samantha R. Levano, Albert Einstein College of Medicine, BRONX, NY, United States; Deepa Rastogi, Albert Einstein College of Medicine, Bronx, NY, United States; Florinda Islamovic, The Children's Hospital at Montefiore, Bronx, NY, United States; Kevin Fiori, The Children's Hospital at Montefiore, Bronx, NY, United States

Background: Pediatric asthma is one of the most common chronic childhood conditions and is mediated by neighborhood social risk. Neighborhood risk indices, such as the Child Opportunity Index (COI), comprise multiple domains. Understanding how differences in neighborhood resources impact asthma outcomes can inform targeted interventions and advance precision social medicine.

Objective: To analyze the effect of improving neighborhood resources across all 14 subdomains of the COI on pediatric asthma admissions, emergency department (ED) visits, and hospital days.

Design/Methods: We used the Pediatric Health Information System to identify children aged 2-18 years with asthma living in Bronx County from January 2021 to July 2025. We transformed z-scores of the overall COI and its subdomains into Bronx-specific, metro-normed, tertiles. We excluded children who died during the study period (n = 1; < 0.1%) and those who changed ZIP codes (n = 44; 1.0%). We applied augmented inverse probability weighting across each subdomain of the COI to estimate the average causal effect of improving COI subdomain levels (improving opportunity) on rates of admissions, ED visits, and hospital days for asthma. Estimates are provided as incidence risk ratios (IRR) and reflect the change in outcome rate (monthly) per one-level increase in COI subdomains. Figure 1 illustrates the conceptual framework of this study. Analyses were conducted in R (R Core Team, Version 4.5.1).

Results: There were 4,330 patients in our cohort. Patient measures stratified by the overall COI tertiles are provided in Table 1. Improvements in early childhood education (IRR: 0.93, 95% CI: 0.89-0.97, adjusted p-value = 0.026) and wealth (IRR: 0.95, 95% CI: 0.92-0.98, adjusted p-value = 0.026) levels were significantly associated with fewer ED visit rates. However, across each of the 14 subdomains of the COI, improvements in opportunity were not associated with reductions in admission or hospital day rates.

Conclusion(s): In children with asthma in a low-resource urban county, most neighborhood subdomain improvements were not associated with lower adverse utilization. However, improvements in early childhood education and wealth were associated with reduced ED visits, suggesting that investments in such upstream factors may modestly reduce acute care utilization for patients. Our findings underscore the importance of disaggregating neighborhood indices to identify actionable domains for precision social medicine interventions.

Figure 1. Conceptual framework of analysis
Figure 1 PAS.pdf

Table 1. Patient measures by overall Bronx-specific Child Opportunity Index
Table 1 PAS.pdf1; Median (Q1, Q3); n (%)
Other payer includes self-pay, 'other' specified, and unknown.
Other race/ethnicity includes Pacific Islander, American Indian, 'other' specified, and unknown. Pacific Islander and American Indian are included in other race/ethnicity due to low cell counts and risk of identification.

Table 2. Adjusted incidence rate ratios associated with increasing Child Opportunity Index (COI) subdomain levels
Table 2 PAS.pdfIncidence rate ratios (IRRs) are presented with 95% confidence intervals. Given multiple hypothesis testing, p-value adjustment was performed using the Benjamini-Hochberg procedure/false-discovery-rate (FDR) adjustment. Adjusted p-values are given as a q-value. Models are doubly robust and adjusted for age, sex, payer, number of complex chronic conditions, race/ethnicity, and distance to hospital in kilometers (measured as ZIP centroid to hospital location, log-transformed) with an offset for log-transformed observed months (defined as time in months from first encounter to end of study period). Weights used in regression models were derived from inverse probability weighting using an ATT framework.