Child Opportunity Index and Rehospitalization for Ambulatory Care Sensitive Conditions at US Children’s Hospitals Free

This retrospective cohort study used the Pediatric Health Information System (PHIS), an administrative database with clinical and resource utilization data for inpatient, ambulatory surgery, ED, and observation unit encounters for >49 children’s hospitals.¹⁴  It is maintained by the Children’s Hospital Association (Lenexa, Kansas). Hospitals submit encounter-level data, including patient demographics, zip code, medications, procedures, billed transactions, utilization, and diagnoses on the basis of International Classification of Disease, Tenth Revision, Clinical Modification codes. Data are deidentified at the time of submission, and data quality and reliability are ensured by the Children’s Hospital Association and participating hospitals. This study was deemed exempt by the Institutional Review Board of the University of California, San Francisco.

Children and adolescents 0 to 17 years of age were eligible for inclusion if they were admitted to a participating hospital (defined by either an inpatient or observation encounter) in calendar year 2017 or 2018 for ACSC, and we only used the index encounter for each patient. Each child was followed for a 1-year period to identify ACSC rehospitalizations to the same hospital. Children were excluded if the hospitalizations were for birth, direct admissions, transfers in, or if the patient encounters had incomplete data. Hospitals with incomplete data during the study period were excluded (N = 2).

The main exposure was the COI using the COI 2.0 approach and measured on the zip code level.⁸  COI 2.0 includes 29 indicators across 3 domains: (1) education, (2) health and environment, and (3) social and economic factors.⁸  These indicators capture important mechanisms through which neighborhoods influence children; for example, availability and quality of neighborhood education centers and schools, neighborhood social structure and economic resources like neighborhood poverty and employment, and resources for healthy living like health food outlets and green space.⁸  Each indicator was transformed to a standardized z-score, which was then used to obtain a weighted mean z-score for each of the 3 domains. An overall z-score was calculated by taking a weighted average across the 3-domain z-scores. The weights applied to the indicator and domain z-scores reflect how strongly the respective variable impacts children’s long-term health and economic outcomes. COI z-scores were used to define COI quintiles, labeled as very low, low, moderate, high, and very high opportunity neighborhoods.⁸  Higher COI reflects better access to beneficial resources.

The primary outcome was rehospitalizations for ACSCs within 1 year of index admission, analyzed as any rehospitalization (yes or no) and ≥2 rehospitalizations (yes or no). Secondary outcomes included cost and length of stay (LOS) for the index hospitalization. The index hospitalization was defined as a hospitalization meeting inclusion criterion between January 1, 2017 and December 31, 2018. Rehospitalization was defined as any repeat hospitalization in the 1-year period after an index hospitalization through December 31, 2019. In addition to the primary and secondary outcomes, we examined the primary diagnosis distribution by COI using an International Classification of Disease, Tenth Revision, Clinical Modification classification system.¹⁵ 

We used descriptive statistics to summarize patient characteristics and to describe COI distribution across index ACSC hospitalization and ACSC rehospitalization/s within 1 year. Race and ethnicity were collected as defined by hospital-specific practices. Race and ethnicity were listed as 2 separate variables in PHIS, and they were collapsed into a single variable for analysis. To evaluate associations between COI and study outcomes, we used generalized estimating equations that clustered on hospital and adjusted for the following covariables: patient age, sex, total number of complex chronic medical conditions,¹⁶  presence of any mental health diagnoses,¹⁵  and Hospitalization Resource Intensity Scores for Kids (H-RISK) score. H-RISK is a measure of clinical severity based on resource utilization, with a higher score indicating greater severity.¹⁷  Race, ethnicity, and individual economic resources (eg, household income) were correlated or collinear with COI; thus, they were not controlled for in regression models. Models used very high COI as the reference group and produced odds ratios with 95% confidence intervals for each of the other 4 COI levels: very low, low, moderate, and high. All analyses were performed by using SAS 9.4 (SAS Institute; Cary, North Carolina), and P values < .05 were considered statistically significant.

A total of 184 478 children were included in the study (Table 1). Compared to the overall cohort, children from very low COI neighborhoods were primarily non-Hispanic Black or Hispanic and publicly insured. Among ACSC hospitalizations, the distribution of COI by individual hospital was highly variable, ranging from 0.4% to 79.9 of children from very low COI neighborhoods and 1.9% to 57.4% from very high COI neighborhoods (Fig 1).

Children from low and very low COI neighborhoods comprised a greater proportion of ACSC hospitalizations compared to children from very high COI neighborhoods; 28.3% of hospitalizations were by children from very low COI neighborhoods with a stepwise decrease to 16.8% by children from very high COI neighborhoods (Table 1). In additional unadjusted analysis, rehospitalization within 1 year and ≥2 rehospitalization within 1 year were significantly different across COI categories, with children from very low COI neighborhoods having higher ACSC rehospitalizations (Table 2).

Risk-adjusted outcomes by COI level are presented in Tables 2 and 3, with very high COI used as the reference. ACSC rehospitalization within 1 year was significantly higher for children from lower COI neighborhoods, with rehospitalization 18.7% for very low COI versus 13.5% for very high COI (adjusted odds ratio 1.14, 1.05–1.23) and ≥2 rehospitalization 4.8% for very low COI versus 3.2% for very high COI (odds ratio 1.51, 1.29–1.75).

Discharge diagnoses varied by COI category (Fig 2). Bronchiolitis and seizure were among the most frequent diagnoses for all COI categories. Asthma hospitalization was the most frequent diagnosis for very low COI and trended down with increased COI category (test of trend: P < .001).

We found that children from lower opportunity neighborhoods have greater odds of ACSC rehospitalization within 1 year. These findings reinforce emerging data demonstrating how social factors influence children’s health. Previous single-center studies have described the prevalence of social risks in neighborhoods,¹³  whereas national studies have demonstrated the impact of neighborhood context on emergency department visits,^11,12  health care utilization among children with medical complexity,¹⁸  hospitalizations for common diagnoses at US children’s hospitals,¹⁹  and multisystem inflammatory syndrome in children during the coronavirus disease 2019 pandemic.²⁰  Multiple opportunities are available for hospital systems to improve outcomes for these children at high risk. These include mapping and integrating COI into hospital data systems to target discharge interventions, screening for social risks, placing referrals to community resources, partnering with communities to improve neighborhood context, and influencing policy.

We used the COI because it is a readily available tool for health systems to engage in efforts to address social determinants of health. At the census level, COI captures multiple sources of disadvantage that are known predictors of healthy child development. Specifically, it reflects how structural racism has generated multidimensional, spatial inequities in access to neighborhood opportunity.⁸  Health systems can map COI for geographic areas served to better understand the neighborhood context experienced by children. Then, they can collaborate with community programs in high COI neighborhoods to reduce preventable health care utilization. Additionally, health systems engaged in these “hot-spotting” efforts^21–23  can use the domain information within COI to understand potential drivers of utilization.

While our findings are consistent with previous connections made between ACSC hospitalizations and patient socioeconomic characteristics,^13,24,25  our work uniquely compares neighborhoods across the United States rather than within a single region. We also use COI, as opposed to socioeconomic markers that are not child-specific. ACSCs are defined as conditions or diagnoses “for which timely and effective outpatient care can help to reduce the risk of hospitalizations by either preventing the onset of an illness or condition, controlling an acute episodic illness or condition, or managing chronic disease or condition.”²⁶  Although the original ACSC work focused on access and quality of primary care, it is important to acknowledge that health outcomes are multifactorial and dependent on the interactions between health care provider, health care systems, and the environment in which the child lives.²⁷  For example, external factors such as air quality and other environmental toxins may drive hospitalizations and rehospitalizations for asthma (an ACSC condition), so interventions to prevent these hospitalizations must go beyond simply improving access and quality of primary care. To improve outcomes for ACSC conditions, we must understand which conditions respond to interventions at the health care system level versus the community level, or some combination of the two approaches.

As research continues to highlight the role of neighborhood context and child opportunity on health outcomes,^11,12,18,20  it becomes more imperative for hospital systems to build processes for addressing these factors in addition to the acute illness episode. One opportunity is for hospital systems is to identify patients with social needs affecting their health, and to connect those patients with corresponding resources in a discharge program. A systematic review of Medicaid managed care organizations found that despite growing interest in interventions to address social determinants of health, implementation has been limited and best practices for widespread dissemination have yet to be identified.²⁸  Fortunately, in pediatrics, promising qualitative work shows that most caregivers believe social risk screening is appropriate, so acceptability to families is unlikely to be a major implementation barrier.^29,30  Moreover, promising approaches exist to link publicly available neighborhood-level data, such as the COI, to patients’ electronic medical records systems. Although the technology exists to make neighborhood data available at the point of care, pediatricians often still lack access to this data or have not been educated on how to interpret and use it to inform patient care.^22,23  However, our findings suggest that neighborhood opportunity is an important predictor of health care utilization and risk factor for rehospitalization. Access to this information at the point of care could elicit further screening, for example, for potential barriers to accessing care in patient’s neighborhoods, that results in referrals to nonclinical services. However, despite calls for more comprehensive collection and utilization of social and community-level information, health care systems often lack infrastructure, resources, and support to develop screening and referral protocols to systematically address health-related social needs.³¹  Additionally, nonprofit hospital organizations, which includes many of the hospital in the study cohort, are required by federal tax law to spend some of their surplus on “community benefits.” These are goods and services that address a community need, including housing, environment, and community health delivery. Partnering local COI research with hospital community benefits needs assessments and funding initiatives has the potential to address neighborhood COI needs.³² 

Additionally, Centers for Medicare and Medicaid Services (CMS) launched the Accountable Health Communities (AHC) Model in 2016 to systematically identify and address social determinants of health among populations insured by Medicare and Medicaid through screening, referral, and community navigation services. This AHC model is based on evidence that addressing health-related social needs through enhanced clinical-community linkages can improve health outcomes and reduce health care costs.^31,33  As of July 2021, 28 organizations were participating in ACH Model testing for 5 core areas of needs (housing instability, food insecurity, interpersonal violence, transportation problems, and utility difficulties). In an evaluation of performance during the first 3 years, the AHC Model effectively identified beneficiaries with higher cost and utilization, and these individuals’ accepted navigation at higher rates than anticipated. However, effectiveness in resolving health-related social needs was low in early stages of implementation, and further results are in process.³⁴  Because of competing hospital priorities and limited resources, identifying social needs and referring patients may not be feasible without a broader and more integrated approach with community partners. Although this CMS model primarily focuses on adults, best practices learned from this initiative could be used to design future pediatric programs to address social determinants of health, including programs that focus on discharge follow-up. CMS is a natural partner for this work, given the high proportion of publicly insured payers in very low COI. Long-term success will be dependent upon payers incentivizing and supporting these activities through policy change.

The potential of hospital-community partnerships to improve health outcomes in children is drawing interest and investment by health systems, payers, and funders; yet research on community interventions and resulting outcomes is limited.³⁵  In children, neighborhood and housing factors impact a range of child health outcomes, including asthma^36–38  and injuries.³⁹  One example is the Healthy Neighborhoods Health Families community program in Columbus, Ohio, which was established in 2008 to address housing, education, and employment with the goal of improving child health and well-being.⁴⁰  The team has improved >380 residential properties and led >$40 million in direct and indirect investment in the surrounding neighborhoods. Using a quasi-experimental study design, the team used a difference-in-difference approach to compare emergency department visits and inpatient stays between the intervention neighborhood and propensity-matched, pooled comparator neighborhoods in the same city.⁴⁰  They found that the intervention neighborhood trended nonsignificantly toward greater decreases in inpatient stays and ED visits. Their results suggest that community programs may influence health outcomes, but early years of the intervention yield only modest changes that were variable on the basis of the comparator neighborhoods.⁴⁰  One key lesson that the authors highlighted is that expecting large-scale changes in neighborhood health outcomes may not be realistic because proximate person-level health factors serve as intermediaries to neighborhood-level improvements. Rather than being disheartening, the identified early improvements continue to add to the discussion of return on neighborhood-level investments by health systems.

Ultimately, neighborhood and community level changes will also need to be addressed from a health policy perspective. Historical practices in discriminatory federal laws and policies such as redlining, school segregation, and the resulting concentrated neighborhood poverty are structural factors⁴¹  that likely perpetuate health inequities in lower COI neighborhoods. Further research into the specific drivers of poor health outcomes may inform neighborhood level investments and government policies to mitigate these factors.

Our study has the limitations inherent in using administrative databases, as well as limitations in drawing conclusions about causality. The PHIS database includes demographic and clinical information for children’s hospitals and cannot be generalized to all hospital types or capture rehospitalizations at non-PHIS hospitals. However, pediatrics admissions to community hospitals may draw from a small catchment area. In addition, this administrative database does not include detailed clinical variables; although we adjusted our analysis with factors to address clinical severity (eg, H-RISK, complex conditions), there may be residual confounding of our results. Next, we were not able to control for variables that are highly correlated or collinear with COI, including race, ethnicity, and socioeconomic status. These correlations reflect neighborhood context and effects of structural racism that COI is meant to capture. The COI measure is a composite of several indicators of neighborhood resources and conditions, and further research is needed to understand the mechanisms of these individual indicators on hospitalizations. For example, specific indicators and/or combinations may be more strongly associated with health outcomes of specific conditions (eg, environmental quality and asthma outcomes, or availability of healthy food and diabetes outcomes). Lastly, PHIS data only includes information on the 5-digit home ZIP code. Although measuring COI at the 9-digit zip code or census tract level would be more precise, we do not believe that our findings were significantly impacted by this limitation, and our results are similar to that of a local study, which used the census track level analysis. When feasible, however, local COI studies should use the more discrete census tract level when designing interventions.

Our study findings may highlight novel opportunities to improve health outcomes for children via health system–level interventions, such as improving access to health care and quality of outpatient care in low opportunity neighborhoods; however, it is critical to consider that upstream neighborhood-level factors may also contribute to disparities. Additional work is needed to evaluate drivers of health disparities in children at the health system and neighborhood levels.