Neighborhood Poverty and Pediatric Intensive Care Use

We identified all PICU admissions that occurred at Cincinnati Children’s Hospital Medical Center (CCHMC) between January 1, 2011, and December 31, 2016, using the electronic health record (EHR). CCHMC is a large quaternary pediatric care facility in Cincinnati, Ohio, that cares for >90% of children who require hospitalization from within Cincinnati and the surrounding Hamilton County.²⁰  The general PICU is a 35-bed unit that admits children with general and subspecialty medical and surgical pediatric conditions. Critically ill cardiac patients are cared for in a separate cardiac ICU; newborns are cared for in a separate NICU. CCHMC is located in Hamilton County, Ohio, which includes 222 census tracts and ∼187 000 children. The overall poverty rate for the county is 16% with a child poverty rate of 24%. The county is 26% African American, 3% Latino, and 66% white.²¹ 

Residential addresses were extracted from the EHR for all children <18 years of age admitted to the PICU during the study period. Addresses were geocoded to census tracts by using custom, Health Insurance Portability and Accountability Act–compliant software on the basis of the 2015 Topologically Integrated Geographic Encoding and Referencing/Line address range files.²²  We chose to include only in-county admissions given the predominance of CCHMC as the pediatric care provider in Hamilton County, which allowed us to capture and analyze virtually all of the county’s intensive care needs; including all PICU admissions at our quaternary referral center in this analysis would have confounded our results by failing to account for an unknown number of children receiving care at other centers. Therefore, children who resided outside Hamilton County, Ohio, were excluded. CCHMC also cares for a large number of international patients, many of whom relocate to temporary housing near the hospital for the duration of their hospital stay; for this reason, any patient who specified an international primary payer in the EHR was also excluded from the analysis even if they presently listed a local address. Analyses were event based, meaning children admitted multiple times were counted each time they contributed a discrete PICU admission. The study received approval from CCHMC’s institutional review board.

Our outcome measures were PICU admission and PICU bed-day rates, each measured at the level of the census tract. As described above, these measures were calculated by dividing the number of PICU admissions and bed days attributable to a specific census tract by the number of children living within the tract. Thus, we obtained a rate measured per 1000 children for each of Hamilton County’s 222 census tracts. These rate variables were measured both on a continuous scale and after tracts were aggregated into quintiles based on the degree of child poverty.

The child population, used as the denominator for our PICU admission and bed-day rate measures, was estimated by using the US Census Bureau’s 2011–2015 American Community Survey (ACS).²³  We also used the child poverty measure from the ACS to characterize and aggregate Hamilton County census tracts. The child poverty rate is calculated in the ACS by dividing the number of children in a tract who live below the federal poverty level (defined in 2017 as an annual income of $25 094 for a family of 4¹³ ) by the number of children in the tract whose poverty status is known. As in our previous work,¹⁹  children were divided into quintiles based on the child poverty rate of the census tract in which they resided. Quintiles were based on the number of children rather than number of census tracts (ie, ∼20% of the child population fell into each quintile). The quintiles were divided from the lowest to highest rates of child poverty as follows: 0% to 5% (Quintile 1), 5% to 10% (Quintile 2), 10% to 18% (Quintile 3), 18% to 31% (Quintile 4), and 31% to 87% (Quintile 5).

We depicted relationships between each of our PICU use measures and neighborhood child poverty rates using scatter plots. To ease visual interpretation of plots, both outcome and child poverty rates were log₁₀ transformed, and we calculated correlation coefficients between use measures and child poverty rates. We also assessed these relationships after categorization of tracts into quintiles using analysis of variance. Finally, we illustrated the geographic distributions of poverty and PICU use throughout the county with choropleth maps.

During the study period, 4071 PICU admissions met criteria for study inclusion, representing 3129 unique children who spent a total of 12 297 days in a PICU bed. The median age at the time of PICU admission was 5.9 years, and female children accounted for 45.3% of the admissions. The sample was 48.4% African American, 44.1% white, and <0.5% Latino. Public insurance covered just less than two-thirds of the children admitted, and <0.5% of children were uninsured. Median PICU length of stay was 2.4 days across all admissions. Two of Hamilton County’s 222 census tracts did not experience any PICU admissions during the study period. We elected to exclude these tracts from the analysis because they also had a low total child population.

Figure 1 depicts the scatter plots for our log₁₀-transformed outcome rates and child poverty rates. Figure 1A illustrates the census tract-level PICU admission rate on the y-axis and the child poverty rate on the x-axis. When these data were subject to correlation analysis, we found a significant positive relationship, as indicated by a Spearman correlation coefficient of 0.59 (P < .001). Figure 1B illustrates the similar relationship between census tract-level PICU bed-day rate and child poverty rate (r = 0.47; P < .001).

We next categorized census tracts according to child poverty rates, from lowest poverty (Quintile 1) to highest poverty (Quintile 5), as depicted in Fig 2. Although analysis of continuous data provided us with more power to illustrate the correlation between PICU use and neighborhood child poverty, condensing the analysis from 220 tracts into quintiles allowed us to make discrete comparisons between low- and high-poverty communities across the county. Figure 2A illustrates the distribution of census tract admission rates across each poverty quintile. The county-wide PICU admission rate was 20.6 admissions per 1000 children over the study period. Quintile 1 had a median admission rate of 11.3 admissions per 1000 children, which increased to 32.0 per 1000 children in Quintile 5 (P < .001). Figure 2B illustrates PICU bed-day rates across quintiles. The county-wide PICU bed-day rate was 55.3 days per 1000 children. Quintile 1 had a median PICU bed-day rate of 23.4 days per 1000 children, which increased to 81.9 days per 1000 children in Quintile 5 (P < .001).

Finally, we visualized geospatial patterns of each PICU use outcome and child poverty. The panels in Fig 3 depict the large degree of overlap between these measures. In each map, darker shades indicate higher rates of each variable; gradations are continuous rather than quintile based. The figure in the lower right illustrates how the tracts aggregated into the 5 poverty-based quintiles. This figure also illustrates how areas with higher rates of PICU use and poverty were not uniformly distributed throughout the county; clusters of both were observed in Cincinnati’s urban core.

Our study evaluated pediatric intensive care use patterns in Cincinnati and the surrounding Hamilton County over a multiyear period. We found significant correlations between neighborhood child poverty rates and neighborhood rates of PICU use, providing further evidence of income-based disparities in the need for intensive care services. By our estimates, if all children in Hamilton County required PICU care at rates equivalent to children in the quintile with the lowest rates of child poverty, there would have been nearly 2000 fewer PICU admissions and nearly 8000 fewer PICU bed days during our study period, which is almost 22 child-years of time our region’s children could have avoided spending in the PICU.

The disproportionate burden of critical illness on low-income children affects the physical, mental, emotional, and financial health of children and families in ways that are both obvious and subtle. As stated previously, evidence suggests that lower-income children are more ill when they arrive to the PICU and may be more likely to die before PICU or hospital discharge. Even when children survive a PICU stay, their quality of life may subsequently be diminished compared with their experience before admission; they may also experience lower functioning and quality of life compared with same-age controls who did not experience a PICU stay.^{15–17,24–26}  Although there is not currently evidence to suggest that this “post-intensive care syndrome” is more likely to develop in lower-income children after a PICU admission, the disproportionate burden of critical illness in lower-income children also results in increased exposure to the risks of its subsequent effects.

The impact of a PICU stay on a family is also substantial. The medical and nonmedical financial costs of a hospital admission may be especially catastrophic for the already financially distressed family^27,28  and may be worsened by the comparatively higher costs of a PICU admission. Lower-income earners commonly have jobs that tend to have less work flexibility,^28,29  causing these parents the added distress of having to choose between dedicating time to earning income or being with their critically ill child. The inequitable acute and longer-term burdens of critical illness on low-income children, families, and communities merit focused investigation and action.

Although research has begun to explore the relationships between poverty and pediatric intensive care needs, it has largely, although not exclusively, used a patient-level frame. We suggest that taking a neighborhood-level view of poverty-related health disparities, as some authors have done, may support both public health– and health system–based approaches to pattern recognition and preventive strategies aimed at disparity reduction.^1,30–37  We also agree with authors who argue that worse health outcomes seen in higher-poverty neighborhoods represent more than just the aggregate of individual health risks, behaviors, and disparities.^38–40  Delineating the complex, intersectional relationships between individual demographic variables; patient and family health risk factors, assets, and exposures; health infrastructure of communities; and cultural and community health norms and practices offers a broader and more nuanced understanding of health inequity and the roots of disparities. Our data support the potential value of identifying high-risk neighborhoods or “hot spots” to inform future community-hospital partnership interventions. Our findings also provide important context for future studies that will seek to explore disparities in PICU use using a multilevel approach.

Mechanisms by which a physical and social environment may influence need for pediatric intensive care become apparent when examining features common in high-poverty, racially segregated neighborhoods.^{5,6,16,30–45}  Higher-poverty neighborhoods may lack health care infrastructure equivalent to that found in lower-poverty neighborhoods (eg, fewer or more distant points of care access or a limited number of providers who accept public insurance).^46,47  Physical living conditions and a more distressed social environment likely also contribute.^30,40–44  For example, dilapidated, substandard, or overcrowded housing; environmental pollution; a lack of safe public spaces for play and recreation; and increased exposure to tobacco smoke may increase a child’s risk of injury or exposure to allergens, toxins, or infections.^6,35,40–44  Neighborhoods with higher poverty rates bear a greater burden of community violence, which may directly affect a child’s need for critical care by way of injury^{5,32,42–44,48} ; community violence (along with the aforementioned exposures) also serves as a toxic stressor, which can affect health in ways that are becoming increasingly well characterized.^{11,30,45,49–52}  Even if a neighborhood does not have comparatively higher crime rates, the perception of poor neighborhood safety can affect the physical health of residents,⁵³  as can the chronic sense of social devaluation based on socioeconomic status, race, or place of residence.^38,39  When experienced within the health care system, this sense of social devaluation may create a community culture in which care is not routinely sought or is sought late in the course of illness even if resources are physically and financially available.

The intensive care setting provides a unique platform for evaluating the contribution of poverty to childhood illness. When examining disparities in admissions across the general inpatient ward, it is occasionally argued that identified disparities in use are rooted, at least in part, in providers’ “social concerns” about particular patients or families; for example, a provider may be more likely to admit rather than discharge a low-income patient from the emergency department on the basis of perceptions about access to follow-up care or adherence to a medication regimen even if that patient is less ill compared with a child from a higher-income family. Data do not support this argument,^54–56  and we further argue that even if it is true, this is a symptom of rather than an explanation for income-based disparity. Nonetheless, the more severe physiologic derangements typically observed in children who are ill enough to be admitted to an ICU^57,58  serve to remove this potential social confounder, isolating relationships between childhood illness and poverty.

There are important limitations to our study. First, although we chose to focus on in-county admissions given that nearly all children admitted from within Hamilton County are admitted to CCHMC,²³  this resulted in single-center data from 1 county. There are also still in-county patients we may have missed who used intensive care services at other centers, although we suspect this number is small. The racial and economic makeup of Hamilton County is also not necessarily generalizable, with a racial composition that is different from the United States as a whole and slightly higher poverty levels than the national average.^31,59  Third, we used census tracts as a surrogate marker for neighborhood, and although this provided standardized measures for our study, the accepted geographic and social boundaries of residential neighborhoods may not precisely correlate with census tract divisions. That said, analyses of disparities between neighborhoods as defined by communities has also been described, with analogous relationships between poverty and health care use being observed.⁴⁴ 

At present, our study examines only disparities in CCHMC PICU admissions and bed days; clinical outcomes were not a component of this investigation. Future research will assess for disparities in the severity of illness on presentation, clinical courses, and dispositions of critically ill children by poverty level of their residential neighborhoods, as defined by census tracts and/or community-defined boundaries. Additional directions include exploration of whether particular diagnoses (eg, lower respiratory infection, trauma, and perhaps even sepsis) drive the disparities observed in critical care use between low- and high-poverty communities. These data may be particularly useful to inform targeted intervention efforts (eg, community outreach on asthma recognition and management in certain neighborhoods or bolstering discharge processes for families from other high-risk neighborhoods).

Finally, there were some high-poverty tracts within Hamilton County that did not have a high rate of PICU use and, conversely, had comparatively lower-poverty tracts that demonstrated high PICU use. Evaluating which characteristics of these communities may help to shield children from critical illness or confer elevated risk could also help in focusing intervention efforts.

Children who live in poverty in the United States have inferior health outcomes compared with those who live in more privileged communities. Our study suggests that these inequities include a disproportionate burden of critical illness. Further delineating the extent of that burden, including contributory disease states and outcomes of critical illness, is important future work. Understanding how physical and cultural environments within communities contribute to need for intensive care may reveal opportunities for interventions that promote health equity across health systems, including in the ICU.

ACS

American Community Survey

CCHMC

Cincinnati Children’s Hospital Medical Center

EHR

electronic health record

IPBD

inpatient bed-day