Understanding Variation In Nonurgent Pediatric Emergency Department Use In Communities With Concentrated Disadvantage

In 2015, 17 percent of US children had at least one emergency department (ED) encounter, with this percentage being higher for children who were younger, children identified as Black or as American Indian or Alaska Native, children insured by Medicaid, and children living in poverty.¹ Among these pediatric ED visits, 40 percent are initially triaged as nonurgent.² Patient-level factors associated with nonurgent ED visits include factors associated with differential access to care and allocation of resources, including lower income, Medicaid coverage, racial or ethnic minority status, and having a caregiver with lower educational attainment or lower health literacy.^3–7

Multiple interventions have sought to reduce nonurgent pediatric ED visits by redirecting patients to primary care with the goal of improving continuity of care and reducing health care costs.⁸ Reliable access to primary care, in turn, reduces nonurgent ED visits and enhances the receipt of preventive services.^9–11 Interventions focused on reducing repeat nonurgent ED visits through education or follow-up support have had inconsistent results.⁸ Interventions focused on broadly improving access to primary care, including implementing walk-in visit hours, integrating community health workers, or offering educational text messages, have been more promising.^12–14 Qualitative studies similarly emphasize the importance of perceived access, convenience, and cost of different sites of care in care-seeking decisions.^3,15–18

Within the US, access to resources and health care varies across small geographic areas. Health outcomes also vary by neighborhood characteristics. Neighborhood poverty is associated with elevated rates of pediatric hospitalizations,^19,20 asthma hospitalizations,²¹ intensive care hospitalizations,²² injury-related ED visits,⁶ and acute visits.²³ Research has demonstrated an association between community-level structural disadvantage and increased pediatric acute care use. Building on this work, we sought to examine exceptions to these associations by focusing on disadvantaged communities in which health outcomes differed from expected outcomes. Within an ongoing academic-community partnership (the Pittsburgh Study), we identified geographic areas—census block groups in Pittsburgh, Pennsylvania—with better-than-expected rates of pediatric nonurgent ED visits (“low utilization”), drawing on prior work in “positive deviance”^24,25 and “bright spots.”²⁶ We focused our attention on communities with concentrated disadvantage to highlight area-level strengths and assets that support child health within the context of concentrated disadvantage. Informed by Jean-Frederic Levesque and colleagues’ model of patient-centered access to health care,²⁷ we compared these low-utilization block groups with high-utilization block groups (block groups with concentrated disadvantage that had higher-than-expected nonurgent ED visit rates) across area-level factors with the potential to alter access to primary or emergency care. Through this study, we aimed to identify sociodemographic and structural factors associated with lower-than-expected nonurgent pediatric ED use in small geographic areas with concentrated disadvantage. We approached this work as a community-engaged, hypothesis-generating process to guide future interventions.

Study Data And Methods

Study Design, Setting, And Context

We retrospectively reviewed ED visits from January 1, 2017, to December 31, 2017, at an academic pediatric medical center, UPMC Children’s Hospital of Pittsburgh. Children’s is the only hospital dedicated to children in western Pennsylvania and has an annual ED volume of more than 80,000 visits. The city of Pittsburgh has a population of 301,000, of whom 23 percent identify as Black and 3 percent as Hispanic,²⁸ with children who are identified as Black and Hispanic being substantially more likely than White children to live below the federal poverty level.²⁹ Pittsburgh has ninety recognized neighborhoods, each with distinct assets, resources, and histories, with sequelae of historical residential segregation by race, ethnicity, and income still evident today.

This study was part of a larger academic-community research collaborative focused on child health (the Pittsburgh Study). From 2018 to 2020 community members and partners provided input into design, analyses, and interpretation through small- and large-group meetings guided by strategies to share power in research processes.³⁰ Together we developed our research goal of learning from communities with concentrated disadvantage that had better-than-expected child health outcomes to uncover strategies to improve child health that may be transferrable to communities with concentrated disadvantage that have worse-than-expected health outcomes.

Identifying Block-Group Visit Rates

ED visits to UPMC Children’s Hospital of Pittsburgh from 2017 were geocoded according to child residential address at the time of the ED visit. Visits by children residing outside the city were excluded from analysis. We identified nonurgent ED visits based on initial ED triage acuity score, limiting our sample to ED visits triaged as low acuity (level 4 or 5),³¹ yielding more than 11,000 nonurgent ED visits. Nonurgent ED visits were then counted by block group, which is a geographic unit usually comprising 600–3,000 people. Use of block groups allowed us to focus on small geographies with consistent boundaries for which census data are available. Using 2012–16 American Community Survey five-year estimates of the pediatric population within each block group, we estimated nonurgent ED visit rates for each of the 355 block groups in Pittsburgh, reported as nonurgent ED visits per 100 children during the 2017 calendar year. We excluded block groups with fewer than fifty children (n = 67) because of concerns that the small denominator made estimates less reliable.

Selecting Block Groups

As a conglomerate measure of concentrated disadvantage, we used the 2015 Area Deprivation Index (ADI).³² The ADI is generated from multiple socioeconomic measures within the US, including income, education, employment, and housing. It is standardized nationally, so a score of 1 represents the 1 percent of block groups with the least structural disadvantage nationally, whereas a score of 100 represents the 1 percent of block groups with the highest structural disadvantage nationally. The ADI scores of block groups within Pittsburgh range from 2 to 100. ADI scores were not available for twenty-four block groups, so these were excluded, leaving 264 block groups for analysis.

We used a spatial regression approach to identify block groups with better- and worse-than-expected ED visit rates given block-group ADI. First, we used the local indicators of spatial autocorrelation test to confirm the presence of statistically significant spatial variation in nonurgent ED visits at the block-group level. We then fit an ordinary least squares spatial regression model with the block-group nonurgent ED visit rate as the dependent variable and ADI score as the independent variable. Through model diagnostics investigating spatial dependence, we identified a significant spatial lag term, which significantly improved model fit (likelihood ratio test, $p<0.001$). Thus, our final model included block-group nonurgent ED visit rate as the dependent variable and ADI score and spatial lag term as independent variables for the 264 included block groups. For each block group, we then determined the difference between predicted and observed nonurgent ED visit rates.

Focusing on block groups with the most concentrated disadvantage (ADI≥90), we identified block groups for which the difference between predicted and observed nonurgent ED visit rates fell outside of the interquartile range. Block groups with ADI≥90 and lower-than-expected nonurgent ED visit rates were considered low utilization. Block groups with ADI≥90 and higher-than-expected nonurgent ED visit rates were considered high utilization.

Block-Group Characteristics

Using 2016 American Community Survey five-year estimates, we determined socioeconomic characteristics of the population living within each low- and high-utilization block group, including household composition, employment, income, race, and access to vehicles. We used Allegheny County property assessment data to assess housing quality and geolocate residential housing parcels within each low- and high-utilization block group,³³ and we used Port Authority of Allegheny County data to geolocate public transit stop locations. We compiled multiple searches and listings to identify health care facilities providing primary care to children. Using service area analysis, we then determined the percentage of housing parcels in each block group within a quarter-mile of public transit, a quarter-mile to one mile of primary care practices, and one to ten miles of the pediatric ED.

Comparing Low- And High-Utilization Block Groups

All analyses were conducted at the block-group level. For the identified low- and high-utilization block groups, we compared sociodemographic characteristics and geographic access to care using Kruskal-Wallis tests. The selection of variables for comparison was informed by the patient-centered access to health care model.²⁷ This model describes system and patient factors influencing the care-seeking process. We included factors potentially affecting the need for health care (child age, housing quality), perceptions of need for care (educational attainment), ability to reach care (adult caregiver demographics; transportation; and proximity to public transit, primary care practices, and the pediatric ED), and ability to afford care (insurance, income, workforce participation, home ownership, and percentage of income spent on housing). We also investigated block-group racial composition, a factor that is not directly causal but is potentially associated with causal factors through historical unequal allocation of resources and current experiences of racism.

We conducted two sensitivity analyses. First, to test the robustness of results to alternative regression approaches, we fit a linear regression instead of spatial regression and followed the same process to identify an alternative set of low- and high-utilization block groups and compare block group–level variables.

Second, to understand how low- and high-utilization block-group selection would differ without the focus specifically being on block groups with ADI≥90, we determined agreement between identified low- and high-utilization block groups compared with block groups that would have been selected without the prespecified focus on block groups with ADI≥90.

Analyses were performed in GeoDa³⁴ version 1.14, ArcGIS version 10.6.1, and SPSS version 25, with statistical significance set at $p<0.05$. This analysis was approved by the UPMC Quality Improvement Review Committee as part of an ongoing quality improvement project within our community. Projects approved by this committee do not meet the formal definition of human subjects research, so approval by an Institutional Review Board is not required.

Limitations

Several limitations to our study warrant discussion. First, we lack data on ED visits to hospitals other than the regional academic children’s hospital. Because the likelihood of children seeking care at another ED increases for outlying communities, we focused this analysis only on the city of Pittsburgh, rather than the surrounding county. Second, we focused our analysis on one year of data. Examining changes in use over time might point to additional block group–level strategies supporting low rates of nonurgent ED visits. Third, this approach relied on an initial regression model and used an ADI threshold for block-group selection. For this reason, we conducted sensitivity analyses to determine the impact of these block-group selection specifications. Fourth, we counted every ED visit within each block group, such that multiple ED visits by the same child contributed multiple times to overall block-group rates, consistent with other analyses.¹ Fifth, we chose intentionally to include only ADI in our model (along with the spatial lag term) because our goal was hypothesis generation to uncover factors that varied between low- and high-utilization block groups with concentrated disadvantage. Although other variables such as insurance type or proximity to primary care are associated in general with nonurgent ED visits, we did not want to assume that these factors varied between block groups with high versus low utilization when focusing specifically on communities with concentrated disadvantage. Finally, we recognize that additional unmeasured confounders may contribute to differences between block groups.

Study Results

Context

A total of 46,102 children younger than age eighteen lived in the 264 included block groups, which had a median ADI score of 76 (range, 5–100; exhibit 1). More than one-quarter of included block groups were within the highest ADI decile nationally (most concentrated disadvantage), and these block groups were home to 27 percent of the children living in the included block groups (exhibit 1).

Block group–level nonurgent ED visit rates ranged from 1 to 98 per 100 children during 2017 (median, 17; exhibit 1). Block groups with the highest nonurgent ED visit rates were clustered in areas of the city with concentrated disadvantage (online appendix exhibit 1).³⁵ Block-group nonurgent ED visits and ADI scores were moderately correlated (ρ = 0.48; $p<0.001$).

Identifying Low- And High-Utilization Block Groups

In the spatial regression model, ADI explained 29 percent of the variation in nonurgent ED use among these 264 block groups. Focusing on block groups with ADI≥90 (n = 70) and with the greatest deviations from predicted nonurgent ED visit rate, we identified low-utilization block groups (n = 17; median, 13 nonurgent ED visits per 100 children) and high-use block groups (n = 17; median, 56 nonurgent ED visits per 100 children; exhibit 2). Block groups identified as low utilization included majority-Black block groups within the East End and Hazelwood and majority-White block groups within western portions of the city (appendix exhibit 2).³⁵ High-utilization block groups included a cluster in the majority-Black East End, the majority-Black Hill District, and a majority-White neighborhood within Hazelwood.

SOURCE Authors’ analysis of 2017 data from UPMC Children’s Hospital of Pittsburgh, 2012–16 American Community Survey five-year estimates, and the 2015 Area Deprivation Index (ADI). NOTES The exhibit illustrates predicted and observed nonurgent ED visit rates by block-group ADI. Each dot represents a block group. Among block groups with the highest concentrated disadvantage (ADI≥90), those with lower-than-expected (low utilization) and higher-than-expected (high utilization) nonurgent ED visits are indicated. An ADI score of 1 represents the 1 percent of block groups with the least structural disadvantage nationally; an ADI score of 100 represents the 1 percent of block groups with the highest structural disadvantage nationally. Block groups with the highest decile of disadvantage nationally (ADI≥90) make up 27 percent of the block groups included in this analysis.

Comparison Of Area-Level Characteristics Of Block Groups

Among factors potentially affecting need for care and perceived need for care, low-utilization block groups had fewer housing parcels determined to be “unsound” and more adults older than age twenty-five who had completed high school when compared with high-utilization block groups (exhibit 3). Among factors potentially affecting the ability to reach care, low-utilization block groups had higher shares of households headed by two adults (median, 36.7 percent versus 16.9 percent; $p=0.01$) and with access to a vehicle (median, 71.8 percent versus 46.5 percent; $p=0.001$) compared with high-utilization block groups. Related to family economic circumstances, low- and high-utilization block groups differed significantly in home ownership rates.

Low- and high-utilization block groups did not differ across other examined factors, such as percentage of population younger than age five, percentage of families living below the poverty level, adult employment, or children’s insurance status.

In low-utilization block groups, the median percentage of the population identifying as Black was 52.4 percent (interquartile range, 8.7–70.4), whereas the median percentage of the population identifying as Black in high-utilization block groups was 81.0 percent (IQR, 65.8–95.6; $p=0.002$) (data not shown).

The percentage of housing units within a quarter-mile of public transit stops and within a quarter- or half-mile of primary care did not differ significantly between low- and high-utilization block groups (exhibit 3). Compared with high-utilization block groups, low-utilization block groups had a slightly lower percentage of housing units within a mile of primary care (median, 99.0 percent versus 100 percent; $p=0.02$). Low-utilization block groups had significantly fewer housing units within three miles of the pediatric ED (median, 0 percent versus 22.6 percent; $p=0.01$), but no significant differences in the percentage of housing units within one mile and ten miles of the pediatric ED. Four city regions with adjacent low- and high-utilization block groups are highlighted in appendix exhibit 3.³⁵ Map details of housing parcel location and proximity to primary care within these four sets of adjacent low- and high-utilization block groups are presented in appendix exhibit 4,³⁵ and a scatter plot of proximity to primary care across all 264 block groups is presented in appendix exhibit 5.³⁵

Sensitivity Analyses

In a sensitivity analysis using linear regression instead of spatial regression, there was substantial agreement in the identification of low- and high-utilization block groups (kappa coefficient = 0.93). Area-level comparisons using low- and high-utilization block groups identified through this linear regression were similar to the main results, with statistically significant differences persisting for identified variables and with these high-utilization block groups also more likely to have housing costs exceeding 30 percent of household income ($p<0.05$) (data not shown).

As a second sensitivity analysis, we used the spatial regression model to identify potential low-utilization (n = 19) and high-utilization (n = 19) block groups across the full range of ADI scores (appendix exhibit 6).³⁵ This process yielded only moderate agreement with our primary approach in the identification of low- and high-utilization block groups (kappa coefficient = 0.47). Fewer than half of these alternative low-utilization block groups (42 percent) had ADI≥90, whereas a majority (58 percent) of these alternative high-utilization block groups had ADI≥90.

Discussion

Among communities with concentrated disadvantage in Pittsburgh, Pennsylvania, we compared low-utilization block groups, which had lower-than-expected pediatric nonurgent ED visit rates, with high-utilization block groups, which had higher-than-expected pediatric nonurgent ED visit rates. It is notable that low- and high-utilization block groups within communities with concentrated disadvantage did not differ significantly in the percentage of households in very close proximity to public transit or primary care or in children’s health insurance coverage rates, and instead differed in block-group housing quality, high school completion, presence of multiple adult caregivers in a household, access to a vehicle, and home ownership.

Each of the examined block-group characteristics was included because of plausible relationships to care seeking,²⁷ with results informing further hypothesis generation. For example, the presence of fewer unsound houses in low-utilization block groups could reduce the risk for environmental exposures that can contribute to nonurgent ED visits (as well as urgent ED visits) for asthma exacerbations or childhood injuries.³⁶ Greater educational attainment is associated with higher health literacy, which could help families assess the need for care, navigate the health care system, and determine when to use primary care in lieu of nonurgent ED visits.⁷ Higher rates of households headed by two adults in low-utilization block groups could allow greater flexibility in coordinating missed work or care for other household children when seeking care for a sick child, thus potentially allowing nonurgent needs to be met at primary care sites in lieu of the ED. Increased access to vehicles could translate into greater ease, efficiency, and agency in traveling to primary care; in contrast, when families turn to ambulances for transportation, they are directed specifically to the ED.

Factors that were not significantly different between block groups with lower- and higher-than-expected rates of nonurgent pediatric ED use, such as the percentage of households in close proximity to public transit and the percentage with children covered by Medicaid, may nevertheless affect access to care in these communities. Our study did not focus on overall access but, rather, on differences in nonurgent pediatric ED use between low- and high-utilization block groups. Proximity to primary care, which other studies found to be a significant factor, was not generally significant in this urban setting, in which no block groups were far from primary care. Nongeographic factors affecting access to primary care such as evening hours or walk-in availability may have differed between the low- and high-utilization block groups, but these factors were not measured in this analysis.

By examining variations through a geospatial lens, we emphasized area-level opportunity rather than individual-level decisions.

By examining variations in nonurgent pediatric ED use through a geospatial lens, we emphasized area-level opportunity rather than individual-level decisions. Highlighting that families in one block group with certain characteristics seek care in a different way than families in another block group with different characteristics illuminates the contribution of structural inequities in shaping care-seeking decisions. Through this lens, then, differences between low- and high-utilization block groups, when investigated further, could translate into further hypothesis generation about addressing community-level factors to alter nonurgent ED use. For example, improvements in housing quality and support for flexible work arrangements could be opportunities to extend some of the advantageous circumstances seen in low-utilization block groups to high-utilization block groups. Stakeholders that could support potentially relevant interventions include health systems (which could invest in local housing quality³⁷ and provide access to medical-legal partnerships¹⁹) and employers (which could promote schedule flexibility and adequate wage levels to support better-quality housing and transportation).^38,39

Our results also suggest opportunities for primary care practitioners and payers to design care to better meet the needs of families in high-utilization block groups, who may have significant time and transportation constraints. Interventions such as extended primary care hours⁴⁰ and walk-in visits¹² may address some of these concerns. In addition, payer policies supporting primary care in integrating high-quality, accessible telehealth (building on the expansion of primary care telemedicine during the coronavirus disease 2019 pandemic) could also make primary care more tenable for families with constrained time and transportation options.

The effects of historical inequities and structural racism are evident, thus pointing to the need for structural solutions.

Another consequence of focusing on area-level variation is that the effects of historical inequities and structural racism are evident, thus pointing to the need for structural solutions for communities experiencing intergenerational disadvantage. Specifically, in this analysis, high-utilization compared with low-utilization block groups had a higher percentage of block-group residents identifying as Black. Again, as a hypothesis-generating process, this finding points to at least two hypothesized mechanisms. First, past and present structural racism, including housing-specific policies such as redlining,⁴¹ has created unequal distribution of resources and opportunity. Housing quality, educational opportunity, and health care access, for example, often differ in majority-Black communities compared with majority-White communities.^42,43 Thus, structural racism may be an underlying explanation for why some of the identified factors differ between low- and high-utilization block groups that also have different racial compositions. Second, past and present experiences of prejudice and racism within health care encounters may alter care-seeking decisions.⁴³ This individual-level, experiential factor could not be examined in the current analysis.

Finally, we reflect on the value of this “positive deviance” or “bright spot” method of examining geographic variation in child health outcomes. An underlying assumption of this method is that local communities may have solutions and strategies that can be uncovered and replicated in neighboring communities. Our sensitivity analysis selecting low- and high-utilization block groups without the ADI threshold affirmed the importance of intentionally focusing on block groups with concentrated disadvantage if we wish to understand the barriers and facilitators of child health in these specific parts of the city—after all, selection processes across all ADI scores would have compared low-utilization block groups mostly outside of the highest decile of disadvantage with high-utilization block groups mostly within the highest decile of disadvantage. We plan to follow up this first analysis with asset mapping and qualitative data collection to further identify strengths, assets, and resources identified by families in low-utilization block groups. As one component of an ongoing community-partnered longitudinal study, this approach has proved to be a powerful way of engaging partners with varied expertise and experience in reciprocal discovery.

Conclusion

Among block groups with high structural disadvantage, those with lower-than-expected nonurgent pediatric ED visit rates had better housing quality, more high school graduates and households headed by two adults, higher access to personal vehicles, and more home ownership. These factors may enable households to seek primary care for nonurgent treatment in lieu of the ED. Our results suggest that governments, payers, health care systems, employers, and clinicians wishing to reduce pediatric nonurgent ED visit rates in neighborhoods with high levels of disadvantage should consider strategies to mitigate financial, time, transportation, and health literacy constraints that may affect families’ access to primary care.

ACKNOWLEDGMENTS

Jamil Bey and Elizabeth Miller were supported through funding to the Pittsburgh Study (Grable Foundation, Shear Family Foundation, University of Pittsburgh Department of Pediatrics, and UPMC Children’s Hospital of Pittsburgh Foundation).

NOTES

• 1 National Center for Health Statistics. Health, United States, 2016, with chartbook on long-term trends in health [Internet]. Hyattsville (MD): NCHS; 2017 [cited 2020 Nov 30]. Available from: https://www.cdc.gov/nchs/data/hus/hus16.pdf Google Scholar
• 2 Rui P, Kang K, Ashman JJ. National Hospital Ambulatory Medical Care Survey: 2016 emergency department summary tables [Internet]. Hyattsville (MD): National Center for Health Statistics; 2016 [cited 2020 Nov 4]. Available from: https://www.cdc.gov/nchs/data/nhamcs/web_tables/2016_ed_web_tables.pdf Google Scholar
• 3 Uscher-Pines L, Pines J, Kellermann A, Gillen E, Mehrotra A. Emergency department visits for nonurgent conditions: systematic literature review. Am J Manag Care. 2013;19(1):47–59. Medline, Google Scholar
• 4 Schlichting LE, Rogers ML, Gjelsvik A, Linakis JG, Vivier PM. Pediatric emergency department utilization and reliance by insurance coverage in the United States. Acad Emerg Med. 2017;24(12):1483–90. Crossref, Medline, Google Scholar
• 5 Samuels-Kalow ME, Bryan MW, Shaw KN. Predicting subsequent high-frequency, low-acuity utilization of the pediatric emergency department. Acad Pediatr. 2017;17(3):256–60. Crossref, Medline, Google Scholar
• 6 Macy ML, Zonfrillo MR, Cook LJ, Funai T, Goldstick J, Stanley RMet al. Patient- and community-level sociodemographic characteristics associated with emergency department visits for childhood injury. J Pediatr. 2015;167(3):711–8. Crossref, Medline, Google Scholar
• 7 May M, Brousseau DC, Nelson DA, Flynn KE, Wolf MS, Lepley Bet al. Why parents seek care for acute illness in the clinic or the ED: the role of health literacy. Acad Pediatr. 2018;18(3):289–96. Crossref, Medline, Google Scholar
• 8 Poku BA, Hemingway P. Reducing repeat paediatric emergency department attendance for non-urgent care: a systematic review of the effectiveness of interventions. Emerg Med J. 2019;36(7):435–42. Medline, Google Scholar
• 9 Strickland BB, Jones JR, Ghandour RM, Kogan MD, Newacheck PW. The medical home: health care access and impact for children and youth in the United States. Pediatrics. 2011;127(4):604–11. Crossref, Medline, Google Scholar
• 10 Garcia-Huidobro D, Shippee N, Joseph-DiCaprio J, O’Brien JM, Svetaz MV. Effect of patient-centered medical home on preventive services for adolescents and young adults. Pediatrics. 2016;137(6):e20153813. Crossref, Medline, Google Scholar
• 11 Enlow E, Passarella M, Lorch SA. Continuity of care in infancy and early childhood health outcomes. Pediatrics. 2017;140(1):e20170339. Crossref, Medline, Google Scholar
• 12 Warrick S, Morehous J, Samaan ZM, Mansour M, Huentelman T, Schoettker PJet al. Walk-in model for ill care in an urban academic pediatric clinic. Acad Pediatr. 2018;18(3):281–8. Crossref, Medline, Google Scholar
• 13 Anugu M, Braksmajer A, Huang J, Yang J, Ladowski KL, Pati S. Enriched medical home intervention using community health worker home visitation and ED use. Pediatrics. 2017;139(5):e20161849. Crossref, Medline, Google Scholar
• 14 Ladley A, Hieger AW, Arthur J, Broom M. Educational text messages decreased emergency department utilization among infant caregivers: a randomized trial. Acad Pediatr. 2018;18(6):636–41. Crossref, Medline, Google Scholar
• 15 Kangovi S, Barg FK, Carter T, Long JA, Shannon R, Grande D. Understanding why patients of low socioeconomic status prefer hospitals over ambulatory care. Health Aff (Millwood). 2013;32(7):1196–203. Go to the article, Google Scholar
• 16 Coster JE, Turner JK, Bradbury D, Cantrell A. Why do people choose emergency and urgent care services? A rapid review utilizing a systematic literature search and narrative synthesis. Acad Emerg Med. 2017;24(9):1137–49. Crossref, Medline, Google Scholar
• 17 Butun A, Linden M, Lynn F, McGaughey J. Exploring parents’ reasons for attending the emergency department for children with minor illnesses: a mixed methods systematic review. Emerg Med J. 2019;36(1):39–46. Medline, Google Scholar
• 18 Vogel JA, Rising KL, Jones J, Bowden ML, Ginde AA, Havranek EP. Reasons patients choose the emergency department over primary care: a qualitative metasynthesis. J Gen Intern Med. 2019;34(11):2610–9. Crossref, Medline, Google Scholar
• 19 Beck AF, Anderson KL, Rich K, Taylor SC, Iyer SB, Kotagal URet al. Cooling the hot spots where child hospitalization rates are high: a neighborhood approach to population health. Health Aff (Millwood). 2019;38(9):1433–41. Go to the article, Google Scholar
• 20 Brokamp C, Beck AF, Goyal NK, Ryan P, Greenberg JM, Hall ES. Material community deprivation and hospital utilization during the first year of life: an urban population-based cohort study. Ann Epidemiol. 2019;30:37–43. Crossref, Medline, Google Scholar
• 21 Beck AF, Huang B, Wheeler K, Lawson NR, Kahn RS, Riley CL. The Child Opportunity Index and disparities in pediatric asthma hospitalizations across one Ohio metropolitan area, 2011–2013. J Pediatr. 2017;190:200–206.e1. Crossref, Medline, Google Scholar
• 22 Andrist E, Riley CL, Brokamp C, Taylor S, Beck AF. Neighborhood poverty and pediatric intensive care use. Pediatrics. 2019;144(6):e20190748. Crossref, Medline, Google Scholar
• 23 Kersten EE, Adler NE, Gottlieb L, Jutte DP, Robinson S, Roundfield Ket al. Neighborhood child opportunity and individual-level pediatric acute care use and diagnoses. Pediatrics. 2018;141(5):e20172309. Crossref, Medline, Google Scholar
• 24 Baxter R, Taylor N, Kellar I, Lawton R. What methods are used to apply positive deviance within healthcare organisations? A systematic review. BMJ Qual Saf. 2016;25(3):190–201. Crossref, Medline, Google Scholar
• 25 Rose AJ, McCullough MB. A practical guide to using the positive deviance method in health services research. Health Serv Res. 2017;52(3):1207–22. Crossref, Medline, Google Scholar
• 26 Holmes GM, Lane NM, Holding W, Randolph R, Rodgers J, Silberman Pet al. Identifying bright spots in Appalachian health: statistical analysis [Internet]. Washington (DC): Appalachian Regional Commission; 2018 Jul [cited 2020 Nov 30]. Available from: https://www.arc.gov/wp-content/uploads/2020/06/BrightSpotsStatisticalAnalysisJuly2018.pdf Google Scholar
• 27 Levesque JF, Harris MF, Russell G. Patient-centred access to health care: conceptualising access at the interface of health systems and populations. Int J Equity Health. 2013;12(18):18. Crossref, Medline, Google Scholar
• 28 Census Bureau. Quickfacts: Pittsburgh city, Pennsylvania [Internet]. Washington (DC): Census Bureau; 2019 [cited 2020 Nov 4]. Available from: https://www.census.gov/quickfacts/pittsburghcitypennsylvania Google Scholar
• 29 Allegheny County Health Department. Allegheny County community indicators [Internet]. Pittsburgh (PA): ACHD; 2019 [cited 2020 Nov 4]. Available from: https://alleghenycounty.us/Health-Department/Resources/Data-and-Reporting/Chronic-Disease-Epidemiology/Allegheny-County-Community-Indicators.aspx Google Scholar
• 30 Chicago Beyond. Why am I always being researched? [Internet]. Chicago (IL): Chicago Beyond; 2018 [cited 2020 Nov 4]. Available from: https://chicagobeyond.org/wp-content/uploads/2019/05/ChicagoBeyond_2019Guidebook.pdf Google Scholar
• 31 Agency for Healthcare Research and Quality. Emergency Severity Index (ESI): a triage tool for emergency departments [Internet]. Rockville (MD): AHRQ; 2012 Sep [cited 2020 Nov 4]. Available from: https://www.ahrq.gov/professionals/systems/hospital/esi/index.html Google Scholar
• 32 Kind AJH, Buckingham WR. Making neighborhood-disadvantage metrics accessible—the Neighborhood Atlas. N Engl J Med. 2018;378(26):2456–8. Crossref, Medline, Google Scholar
• 33 Western Pennsylvania Regional Data Center. WPRDC parcels n’at [Internet]. Pittsburgh (PA): University of Pittsburgh, WPRDC; 2017 [cited 2020 Nov 4]. Available from: https://tools.wprdc.org/parcels-n-at/# Google Scholar
• 34 Anselin L, Syabri I, Kho Y. Geoda: an introduction to spatial data analysis. Geogr Anal. 2006;38(1):5–22. Crossref, Google Scholar
• 35 To access the appendix, click on the Details tab of the article online.
• 36 Hughes HK, Matsui EC, Tschudy MM, Pollack CE, Keet CA. Pediatric asthma health disparities: race, hardship, housing, and asthma in a national survey. Acad Pediatr. 2017;17(2):127–34. Crossref, Medline, Google Scholar
• 37 Kelleher K, Reece J, Sandel M. The Healthy Neighborhood, Healthy Families Initiative. Pediatrics. 2018;142(3):e20180261. Crossref, Medline, Google Scholar
• 38 Rogers Z. Cincinnati Children’s is raising its minimum wage to $15 an hour. Cincinnati: WKRC Local 12 News [serial on the Internet]. 2019 Oct 2 [cited 2020 Nov 4]. Available from: https://local12.com/news/local/cincinnati-childrens-is-raising-its-minimum-wage-to-15-an-hour Google Scholar
• 39 Glasmeier AK. Living wage calculation for Pittsburgh, PA [Internet]. Cambridge (MA): Massachusetts Institute of Technology; 2020 [cited 2020 Nov 4]. Available from: https://livingwage.mit.edu/metros/38300 Google Scholar
• 40 Zickafoose JS, DeCamp LR, Prosser LA. Association between enhanced access services in pediatric primary care and utilization of emergency departments: a national parent survey. J Pediatr. 2013;163(5):1389–95. Crossref, Medline, Google Scholar
• 41 Nelson RK, Winling L, Marciano R, Connolly Net al. Mapping inequality: redlining in New Deal America [Internet]. Richmond (VA): University of Richmond; 2020 [cited 2020 Nov 4]. Available from: https://dsl.richmond.edu/panorama/redlining/#loc=11/40.442/-80.134&city=pittsburgh-pa Google Scholar
• 42 Bell CN, Bowie JV, Thorpe RJ Jr, Levine DM. A spatial analysis of race, local health-promoting resources, and preventable hospitalizations. Prev Med. 2017;105:149–55. Crossref, Medline, Google Scholar
• 43 Trent M, Dooley DG, Dougé JSection on Adolescent Health, Council on Community Pediatrics, Committee on Adolescence. The impact of racism on child and adolescent health. Pediatrics. 2019;144(2):e20191765. Crossref, Medline, Google Scholar