The Child Opportunity Index 2.0 and Hospitalizations for Ambulatory Care Sensitive Conditions

In this retrospective cross-sectional analysis, we examined hospitalizations for ACSCs among children <18 years of age within the primary service areas of Cincinnati Children’s Hospital Medical Center (CCHMC) and Children’s Mercy Kansas City (CMKC) occurring from January 1, 2013, to December 31, 2018. Both CCHMC and CMKC are quaternary care, freestanding children’s hospitals that dominate their local market share. That said, to ensure accuracy of pediatric hospitalization rates and approximate population-level samples, we only included hospitalization events originating from census tracts within counties where CCHMC or CMKC had >80% market share for pediatric hospitalizations (Hamilton County for CCHMC and Cass, Clay, Jackson, Johnson, and Platte Counties for CMKC). Hospitalizations were included if the patient had a valid home address that allowed for geocoding to a census tract where COI 2.0 measures could be assigned.

The patient’s home address attached to the hospitalization event allowed for geocoding of records to the census tract geography. This enabled linkage to COI 2.0 data as well as additional key census-available measures, including child population and demographics of children within the identified census tract.²⁸  Census tracts are small, relatively permanent statistical subdivisions of a county, each averaging ∼4000 residents.²⁹  They tend to align closely with local conceptions of neighborhoods or suburbs; this is particularly true in both Greater Cincinnati and Greater Kansas City.

Our primary independent variable was child opportunity, as defined by the 2015 overall and domain-specific COI 2.0 child opportunity levels, for each included census tract in the Cincinnati and Kansas City metropolitan areas.¹⁷  The COI 2.0 includes 29 indicators spanning 3 domains (education, health and environment, and social and economic). Contextual variables within the COI include access to high-quality early childhood education and schools, green space, healthy food, toxin-free environments, and socioeconomic resources (such as employment opportunities). To combine indicators into a composite index, COI developers first standardized them using the z score transformation and then took a weighted average across indicator z scores within each domain to obtain average domain z scores. The indicator-specific weights reflect how strongly each indicator predicts children’s long-term health and economic outcomes. The COI developers then combined average domain z scores and combined them into an overall score using a similar weighting approach. They defined child opportunity levels by ranking all neighborhoods in the United States along their overall and domain-specific average z scores and dividing them into 5 ordered groups, each containing 20% of the US child population. These groups were labeled very low, low, moderate, high, and very high opportunity.

Our primary outcome was the pediatric hospitalization rate at the census tract level for ACSCs, both overall and for specific diagnoses. The Agency of Healthcare Research and Quality and previous pediatric health services research define ACSCs as conditions that may avoid hospitalization with timely and effective outpatient management.^5,30  Using previous work as a guide,^8,31,32  we identified pediatric ACSCs most commonly resulting in hospitalization using codes for specific All Patient Refined Diagnosis Related Groups. These included asthma, bronchiolitis, bacterial pneumonia, cellulitis, gastroenteritis and dehydration, diabetes mellitus with complications, epilepsy, and urinary tract infection. Hospitalizations classified as inpatient and observation were both included. We further characterized included hospitalizations using clinical and resource use data from the Pediatric Health Information System data set, a deidentified administrative and billing database organized and distributed by the Children’s Hospital Association (Lenexa, KS).³³ 

We collected population-level demographic characteristics of included census tracts, including age, sex, race (white, Black or African American, or other), and ethnicity (Hispanic or non-Hispanic).¹⁶  We collected these same demographic characteristics at the patient level for hospitalization events, as well as type of insurance (public, private, or other). Clinical characteristics included the length of stay (in days), need for intensive care services, and comorbid health conditions. Comorbidities were stratified into 3 categories: no chronic disease, presence of a noncomplex chronic disease, and presence of a complex chronic disease. To define each category, we used the Agency for Healthcare Research and Quality’s Chronic Condition Indicator to categorize International Classification of Diseases, Ninth Revision, Clinical Modification and International Classification of Diseases, 10th Revision, Clinical Modification hospital diagnoses as chronic or not chronic.³⁴  Second, we filtered the list of chronic conditions into complex chronic conditions and noncomplex chronic conditions by applying the Feudtner et al³⁵  International Classification of Diseases, Ninth Revision and International Classification of Diseases, 10th Revision coding scheme.

We compared differences in population-level demographic characteristics of included census tracts across the COI 2.0 levels (very low to very high) using Kruskal-Wallis tests. We used frequencies and percentages to describe demographic and clinical characteristics of the hospitalized patients and determined statistical significance with χ² tests. To compare rates of hospitalizations for ACSCs (all cause and by diagnosis) across the 5 COI 2.0 levels, we used unadjusted Poisson regression models. All models were tested for overdispersion and used an offset equal to the log of the child population size in each census tract. We calculated potentially avoidable hospitalizations for ACSCs by finding the difference in hospitalization rates between each COI 2.0 group (ie, very low, low, moderate, and high) and the very high COI 2.0 group and then multiplying by the population of children in the respective COI 2.0 group.

P values of <.05 were considered statistically significant. All analyses were performed by using SAS software (version 9.4; SAS Institute, Inc, Cary, NC).

The CCHMC Institutional Review Board and the Office of Research Integrity at CMKC both evaluated this study and independently deemed it to be exempt from further review.

We included 604 943 children within 628 census tracts from the Cincinnati and Kansas City metropolitan areas (Table 1). This included 222 (35.4%) census tracts and 187 626 (31.0%) children from Greater Cincinnati and 406 (64.6%) census tracts and 417 317 (69.0%) children from Greater Kansas City. Children living in included census tracts were more commonly white (74.4%) and non-Hispanic (93.6%). Overall, the distribution of census tracts across opportunity levels followed a U-shaped distribution, in which tracts most often had either a very low (21.7%) or a very high (30.9%) opportunity level.

We observed significant racial differences between children residing in very low and very high opportunity census tracts. The majority of children in very low opportunity census tracts were Black or African American (56.1%), whereas the majority of children in very high opportunity census tracts were white (88.8%). In terms of ethnicity, there was a higher percentage of Hispanic children in very low compared with very high opportunity census tracts (10.5% vs 4%). The age distribution skewed slightly younger in very low opportunity census tracts.

There were 20 331 unique children hospitalized for an ACSC during the study period, contributing a total of 26 977 ACSC hospitalizations (Table 2). These children more commonly were <5 years of age (62.3%), were non-Hispanic white (52%), had public insurance (53.5%), and did not have any chronic comorbidities (56.3%). The median length of stay was 2 days (interquartile range 1–3 days), and 7.4% required admission to an ICU. The frequencies of each ACSC diagnostic group were as follows: bronchiolitis, 27.3%; asthma, 25.8%; seizure, 12.1%; bacterial pneumonia, 9.7%; gastroenteritis and dehydration, 8.8%; cellulitis, 7.5%; urinary tract infection, 4.7%; and diabetes mellitus with complications, 4.3%.

Overall, the hospitalization rate for all ACSCs was 46.0 per 1000 children (95% confidence interval [CI]: 45.4–46.5); 50.7 per 1000 children (95% CI: 49.7–51.7) in Cincinnati and 42.6 per 1000 children (95% CI: 42.0–43.2) in Kansas City. In aggregate, hospitalization rates for all ACSCs revealed a stepwise reduction from 79.9 per 1000 children (95% CI: 78.1–81.7) in census tracts with very low opportunity to 31.2 per 1000 children (95% CI: 30.5–32.0) in census tracts with very high opportunity (Table 3). Nearly identical parallel stepwise reductions in hospitalization rates for all ACSCs were observed when disaggregated by city (Fig 1) or by COI domain (Fig 2). Statistically significant decreases in hospitalization rates were observed for all categories of ACSCs when comparing very low with very high opportunity census tracts (all P values < .001; Table 3). This decrease was largest for asthma, with a 5.5-fold reduction in hospitalization rates from very low to very high opportunity census tracts. Pneumonia and gastroenteritis and dehydration revealed the smallest decrease, with a 1.6-fold reduction.

If hospitalization rates among census tracts from all opportunity levels were reduced to the hospitalization rate of census tracts from the very high opportunity level, ∼778 (44.3%) fewer ACSC-related hospitalizations and 2416 (56%) fewer ACSC-related hospital days may have occurred in Cincinnati and Kansas City annually.

In this retrospective cross-sectional analysis, we examined the association between a multidimensional measure of neighborhood context, the COI 2.0, and pediatric hospitalizations for ACSCs in both Greater Cincinnati and Greater Kansas City. We found that lower measured neighborhood opportunity was significantly associated with higher rates of pediatric hospitalizations for ACSCs. This association was present for all 8 of the included pediatric ACSCs, with asthma having the strongest association with opportunity level. When the 3 COI 2.0 domains and the 2 included metropolitan areas were examined separately, each trend was similar to the overall results. Our findings support the hypothesis that neighborhoods with lower opportunity, as defined by measures embedded within the COI 2.0, are at higher risk for adverse population health outcomes. Targeting interventions to lower-opportunity neighborhoods and advocating for policies that equitably bolster opportunity may improve child health outcomes, reduce health-related socioeconomic inequities, and decrease health care costs.^36–39 

Our study appears to confirm previous work using the COI 1.0 and provides further insight into the impact that neighborhood conditions have on child health and health care use across the care continuum. We were able to leverage the capabilities of the revised COI 2.0 to examine 2 metropolitan areas concurrently. This novel approach revealed that the association between a comprehensive measure of neighborhood social context and a health outcome, such as pediatric hospitalizations for ACSCs, may be generalizable between cities. Our results also revealed lower opportunity levels and higher ACSC hospitalization rates among Black or African American and Hispanic children. This study was not specifically designed to further quantify previously described associations between structural racism, including residential segregation, and health inequities.^40,41  That said, our findings suggest that ramifications of such racism, as highlighted by the clear racial and ethnic disparities present across COI 2.0 child opportunity levels, continues to negatively affect children’s surroundings. This exposure appears to extend from their educational, environmental, and socioeconomic opportunity to their opportunity to avoid potentially preventable hospitalizations.

There has been a growing movement in the field of pediatrics to screen for and intervene to address patients’ social needs, such as housing, food, and transportation. These efforts are increasingly regarded as a central element to promote the well-being of children and families.^20–22  However, focusing only on individual needs falls short of impacting community-level social determinants of health, including the neighborhoods into which children are born and where they grow up.⁴²  A long line of observational and, more recently, experimental studies reveal that neighborhoods have sizeable long-term effects on socioeconomic attainment, physical health, and mental well-being.^43–46  Our results reveal that a multidimensional measure of neighborhood conditions like the COI 2.0 has important prognostic value that could inform patient- and community-level interventions (eg, medical-legal partnerships or home assessment and repair programs), public health planning, and policy.^11,14,15 

One potential critique of a multidimensional measure like the COI 2.0 is that it does not necessarily point to a single factor with the greatest impact on health outcomes and therefore does not suggest a single solution. In reality, the external forces affecting health throughout the life span do not act in isolation. Although the component indicators of a composite index could be used to identify specific contextual factors with greater influence, meaningful improvements in neighborhood opportunity and related health outcomes will most likely require community-driven, multifaceted solutions to stimulate opportunity in neighborhoods that have been neglected by generational disinvestment in infrastructure, business, education, and access to services. By including neighborhood features from different domains, a composite index like the COI 2.0 provides a common metric for cross-sector collaboration in support of comprehensive, community-focused interventions and system-level policy change.^36,47  The COI 2.0 might also serve as a benchmark that could be used to monitor the local impact of these investments, both against other metropolitan areas and over time. The health care sector can play a vital role in this work by supporting community-wide learning networks in research or quality improvement efforts, illustrating how interventions and policies affecting opportunity can improve child health, and advocating that health equity be prioritized when key decisions about resource allocation are made.

Our study did have limitations. First, we recognize that ACSCs have not been as consistently defined in pediatric compared with adult literature and are an imperfect measure of health care access in this population. Although the COI 2.0 accounts for rates of health insurance coverage, other dimensions of health care accessibility and quality went unmeasured and remain areas for future study. Second, the COI 2.0 is measured at the census tract level, which does not necessarily align with how inhabitants of a city define their neighborhoods. To use the COI 2.0 in planning or monitoring the effect of neighborhood-level interventions, it would be necessary to recognize when these different geographic boundaries overlap and when they do not. Third, although we only included counties where the children’s hospital had a high market share for pediatric hospitalizations, it is possible that a small number of relevant hospitalizations could have been excluded if a child were admitted at another hospital. Fourth, the 2 metropolitan areas included in our study share certain geographical and demographic characteristics, including a similar and significant degree of racial residential segregation, which may have contributed to the similar trends we saw in our results. In other words, results may differ in metropolitan areas with different demographic characteristics. Finally, it is possible that individual patient characteristics, which our population-based study design could not account for, may have contributed to hospitalization rates in addition to neighborhood context. The degree to which patient- and population-level factors differentially contribute to ACSCs warrants further inquiry, such as the use of multilevel modeling to compare the performance of patient- and neighborhood-level measures of opportunity. Additional research could also model the trajectory of ACSC hospitalizations or other health care quality measures over time.

ACSC morbidity is disproportionately felt within neighborhoods characterized by lower educational, environmental, and socioeconomic opportunity, as measured by the COI 2.0. This relationship persists across metropolitan areas and for a diverse group of pediatric conditions. Our results suggest that health care systems and policy makers interested in achieving better, more equitable outcomes should target interventions to geographic areas with lower opportunity. The COI 2.0, a multidimensional measure of neighborhood context, appears to be one option for benchmarking the interaction of neighborhood context and associated health outcomes, both over time and within and between cities.

ACSC

ambulatory care sensitive condition

CCHMC

Cincinnati Children’s Hospital Medical Center

CI

confidence interval

CMKC

Children’s Mercy Kansas City

COI

Child Opportunity Index