RESULTS
The analytical sample included 709 patients. The geographic distribution of the population is visualized in Figure 2 with wide spread of the subjects and increased concentration around the major cities of north and central Alabama. Characteristics of the sample, overall and by ACS status, are presented in Table 1 . Median age was 13.6 years, and the majority (72%) had public health insurance. Those who resided in highly deprived areas made up 45% of the sample, while 21% resided in neighborhoods that were ≥90% African American in composition, and 17% resided in rural areas. The most prevalent phenotype was the SS phenotype (54%). When compared to children with non-recurrent ACS (no ACS and single ACS), those with recurrent ACS were older (IRR 1.05, CI 1.02–1.07, p<0.001), more likely to have asthma (IRR 3.93, CI 2.84–5.44, p<0.001), to be a male (IRR 1.91, CI 1.36–2.68, p<0.001), and to be on chronic blood transfusion (IRR 1.78, CI 1.26–2.52, p=0.001) and hydroxyurea (IRR 6.3, CI 4.01–9.91, p<0.001). Recurrent ACS was lower among those living in a highly deprived area (IRR 0.72, CI 0.52–0.99, p=0.044) and in a predominantly African American community (IRR 0.56, CI 0.35–0.90, p=0.017).
To assess the adjusted association between residential characteristics and ACS categories (no ACS, single ACS, any ACS, non-recurrent ACS, and recurrent ACS), we estimated a separate model for each neighborhood characteristic (Model 1: deprivation; Model 2: racial composition; Model 3: rurality) and a full model with all three neighborhood characteristics together (Model 4) Table 2 . All four models controlled for the same covariates: age, sex, insurance type, BMI, chronic transfusions, hydroxyurea, asthma, and phenotype. Residence in high-deprivation neighborhood was associated with 27% lower risk of recurrent ACS vs non-recurrent ACS, 43% lower risk of recurrent ACS vs single ACS, and 42% lower risk of recurrent ACS vs no ACS (Model 1,Table 2 ). Residence in predominantly African American neighborhood was associated with 43% lower risk of recurrent ACS vs non-recurrent ACS, 57% lower risk of recurrent ACS vs single ACS, and 59% lower risk of recurrent ACS vs no ACS (Model 2, Table 2 ). However, when all residential characteristics were assessed together, area deprivation was no longer significant while neighborhood racial composition remained significant, indicating that the neighborhood’s racial composition accounts for the association between socioeconomic deprivation and decreased ACS recurrence (Model 4, Table 2 ). A separate analysis showed that those who reside in a predominantly African American neighborhood are 1.24 times more likely to reside in a socioeconomically deprived neighborhood (CI 1.04–1.48, p=0.016). Rurality was not associated with ACS, neither by itself (Model 3,Table 2 ) nor together with the other neighborhood characteristics (Model 4, Table 2 ). There was no difference in residential characteristics when comparing single ACS to no ACS (Table 2 ).
Multiple regression of recurrent ACS adjusted for all covariates showed that males have nearly two-fold risk of recurrent ACS compared to females Table 3 . Clinical covariates associated with increased risk of recurrent ACS included asthma (IRR 2.62, CI 1.89–3.63, p<0.001) and being on hydroxyurea (IRR 2.89, CI 1.74-4.8, p<0.001). In contrast, patients with the SB+ and SC phenotypes had 82% (CI 0.04–0.72, p=0.015) and 66% (CI 0.19–0.61, p<0.001) less risk of recurrent ACS compared to those with the SS phenotype. Because asthma was a significant risk factor for recurrent ACS, in a separate analysis we assessed the relationship between living in a highly deprived area and having asthma, but found no significant association (IRR 1.10, CI 0.93–1.30, p=0.28).
Next, we sought to determine if the adverse relationship of neighborhood socioeconomic deprivation with recurrent ACS is explained by access to care. While mild ACS may be managed at home or local hospitals, severe ACS has to be managed in a tertiary pediatric hospital with intensive care unit abilities. We therefore used the medical records to determine the severity of ACS (mild vs severe) and replicated the model build-ups to determine the odds of experiencing severe ACS. These additional analyses found no association between neighborhood characteristics and ACS severity Supporting Information Table S1 , indicating that the association of neighborhood characteristics with recurrent ACS is not explained by access to care. We also estimated a model of ACS recurrence with distance to the hospital in miles as another marker of access to care, and again found no significant associationSupporting Information Table S2 .
Finally, we performed sensitivity analyses where all area-level measures were treated as continuous variables Supporting Information Table S2 . These analyses confirmed a dose-response relationship between neighborhood socioeconomic deprivation, neighborhood racial composition, and recurrent ACS. We also tested for interactions between neighborhood socioeconomic deprivation and each of rurality and racial compositionSupporting Information Table S3 . None were found to be significant, suggesting that the likelihood of recurrent ACS does not vary by the relationship between deprivation and rurality or racial composition.