Association of Income Inequality With Pediatric Hospitalizations for Ambulatory Care–Sensitive Conditions

Key Points

Question  Does income inequality among residents of a geographic area affect pediatric hospitalization rates, resulting in use of hospital resources for ambulatory care–sensitive conditions, even after adjusting for individual income?

Findings  In this cross-sectional analysis of 79 275 hospitalizations, income inequality was associated with increased rates of pediatric hospitalization for ambulatory care–sensitive conditions after adjusting for income. Differences in length of stay and total hospital charges per hospitalization were not clinically meaningful.

Meaning  Income inequality within a geographic area is associated with a higher rate of hospitalization for ambulatory care–sensitive conditions; efforts to reduce rates of these hospitalizations should focus on areas with high income inequality.

Importance  The level of income inequality (ie, the variation in median household income among households within a geographic area), in addition to family-level income, is associated with worsened health outcomes in children.

Objective  To determine the influence of income inequality on pediatric hospitalization rates for ambulatory care–sensitive conditions (ACSCs) and whether income inequality affects use of resources per hospitalization for ACSCs.

Design, Setting, and Participants  This retrospective, cross-sectional analysis used the 2014 State Inpatient Databases of the Healthcare Cost and Utilization Project of 14 states to evaluate all hospital discharges for patients aged 0 to 17 years (hereafter referred to as children) from January 1 through December 31, 2014.

Exposures  Using the 2014 American Community Survey (US Census), income inequality (Gini index; range, 0 [perfect equality] to 1.00 [perfect inequality]), median household income, and total population of children aged 0 to 17 years for each zip code in the 14 states were measured. The Gini index for zip codes was divided into quartiles for low, low-middle, high-middle, and high income inequality.

Main Outcomes and Measures  Rate, length of stay, and charges for pediatric hospitalizations for ACSCs.

Results  A total of 79 275 hospitalizations for ACSCs occurred among the 21 737 661 children living in the 8375 zip codes in the 14 included states. After adjustment for median household income and state of residence, ACSC hospitalization rates per 10 000 children increased significantly as income inequality increased from low (27.2; 95% CI, 26.5-27.9) to low-middle (27.9; 95% CI, 27.4-28.5), high-middle (29.2; 95% CI, 28.6-29.7), and high (31.8; 95% CI, 31.2-32.3) categories (P < .001). A significant, clinically unimportant longer length of stay was found for high inequality (2.5 days; 95% CI, 2.4-2.5 days) compared with low inequality (2.4 days; 95% CI, 2.4-2.5 days; P < .001) zip codes and between charges ($765 difference among groups; P < .001).

Conclusions and Relevance  Children living in areas of high income inequality have higher rates of hospitalizations for ACSCs. Consideration of income inequality, in addition to income level, may provide a better understanding of the complex relationship between socioeconomic status and pediatric health outcomes for ACSCs. Efforts aimed at reducing rates of hospitalizations for ACSCs should consider focusing on areas with high income inequality.

In the United States, hospitalizations for ambulatory care–sensitive conditions (ACSCs), including asthma and bacterial pneumonia, account for nearly 450 000 pediatric hospitalizations, with resulting charges estimated at $4 billion per year.¹ Rates of ACSCs (including hospitalization) are considered to be sensitive to the quality of ambulatory services received.² Children living in low-income households are more likely to experience gaps in insurance coverage, have inconsistent access to primary care clinicians, and lack reliable transportation.^3,4 As such, low income has been shown to be associated with increased hospitalizations for ACSCs.^1,5,6

Income level is not the only financial attribute of patients that influences their health outcomes.⁷ Large differences in income (ie, income inequality) within a geographic region (eg, zip code) may act alone or in combination with the adverse effects of low income level to potentiate poor health outcomes in children.^8-11 Income inequality (as measured by the Gini index; range, 0 [perfect equality] to 1.00 [perfect inequality]) in the United States is the second highest among developed countries^12,13 and is associated with shortened life span, poor child health and well-being, and increased rates of child maltreatment.^8,14 Even small increases in income inequality (eg, increase of Gini index by 0.05) have been associated with significant increases in poor health at the state level.^15,16 In communities of homogenous poverty (low Gini index), poor health is strongly influenced by poverty. In areas of large income disparity (high Gini index), income inequality functions independently and synergistically with poverty to adversely affect children’s health. For instance, income inequality has been shown to weaken social cohesion and subsequently worsen health beyond that associated with income alone.^10,11

Income inequality is not considered when allocating safety-net social programs (including primary care), resulting in a relative scarcity of resources for persons with lower incomes who reside in areas of high income inequality. Use of health services may be affected by income inequality, in particular for ACSCs, because their treatment depends on access to appropriate health care. As such, the objective of this study was to determine whether income inequality at the zip code level is associated with higher rates of pediatric hospitalization for ACSCs and increased length of stay (LOS) and charges for ACSCs.

This retrospective, cross-sectional analysis primarily used the 2014 State Inpatient Databases (SID),¹⁷ a division of the Healthcare Cost and Utilization Project (January 1 through December 31, 2014). The 2014 SID includes state-specific inpatient care records from 48 states and International Classification of Disease, Ninth Revision, Clinical Modification (ICD-9-CM) diagnosis and procedure codes, patient demographic data, LOS, payment source, and total charges for each hospitalization. Hospital records suppressed at the state level resulted in the following rates of missing data: 0.02% for sex, 8.3% for race (categorized as other), 0.2% for payer, and 0.1% for charges. The SID also assigns each patient encounter to an all patient–refined diagnosis related group (APR-DRG) (3M Health Information Systems) and severity level. The APR-DRGs are based on groupings of similar diagnoses and procedure codes. An APR-DRG severity level is assigned for each hospitalization based on age, comorbidities, and combinations of diagnoses and procedures.¹⁸ We incorporated similar measures from the 2011 SID to compare changes in associations between income inequality and ACSCs over time. The Office of Research Integrity at Children’s Mercy Hospital, Kansas City, Missouri, deemed this study to be exempt from institutional board review.

Children and adolescents aged 0 to 17 years (hereafter referred to as children) living in 14 states (Arizona, Colorado, Florida, Iowa, Kentucky, Nebraska, New Jersey, New York, North Carolina, Oregon, Rhode Island, Vermont, Washington, and Wisconsin) that provided patient-level home zip codes were included. The patient home zip code allowed geocoding of records to census-available measures of child population, including the Gini index and median household income at the zip code tabulation area (ZCTA) level.¹⁹ zip code tabulation areas are generalized representations of zip codes but are organized by census block group rather than US Postal Service boundaries.²⁰ In most cases, zip codes and ZCTAs are similar; however, 76 of the 8451 zip codes (0.9%) in the included states did not map to a ZCTA and were excluded from the analysis.

The Agency of Healthcare Research and Quality defines ACSCs as conditions that may avoid hospitalization with timely and effective outpatient management.^21,22 Pediatric ACSCs are identified through specific ICD-9-CM principal diagnoses codes and include asthma, bacterial pneumonia, gastroenteritis, dehydration, urinary tract infection, uncontrolled diabetes, diabetic short-term complications, and perforated appendicitis.² Hospitalization rates for each ACSC diagnosis were calculated based on populations of children within each zip code. Because the hospitalization rates of uncontrolled diabetes, diabetic short-term complications, and perforated appendicitis among children occurred in less than 1 hospitalization per 10 000 children, these conditions were excluded from individual ACSC analyses.

Our primary independent variable was income inequality (described by Gini index) for each ZCTA. The Gini index, a measure of income inequality in a population, ranges from 0 to 1.00 based on the differences in income of each person in a population; a Gini index of 0 corresponds to perfect equality in income (ie, all persons in a geographic area have the same income), whereas a Gini index of 1.00 corresponds to perfect inequality (ie, 1 person has all the income in a geographic area, and everyone else has none).²³ The ZCTA-level Gini index was extracted from the 2014 American Community Survey²⁴ for the included SID states. The median zip code–level Gini index was the same for included and excluded states (0.46; interquartile range [IQR], 0.45-0.47). We then categorized each zip code into the following 4 income inequality groups based on quartiles of included zip codes: low (0.04-0.38), low-middle (0.39-0.41), high-middle (0.42-0.45), and high (0.46-0.73). For comparison, we included the Gini index for 2011 and categorized zip codes into the following similar income inequality groups: low (0.02-0.39), low-middle (0.40-0.43), high-middle (0.44-0.46), and high (0.47-0.70).

Our main outcome was the pediatric ACSC hospitalization rate per 10 000 children measured at the zip code level. Secondary outcomes included LOS and total hospital charges per ACSC hospitalization.

From the SID hospitalization data, we identified patient-level demographic variables, including age, sex, race/ethnicity (non-Hispanic white, non-Hispanic black, Hispanic, or other), and insurance type (public, private, or other). Federal poverty level threshold limits for public insurance and rates of uninsured patients varied across states but were not associated with the Gini index (ρ = −0.06 [P = .80] and ρ = 0.33 [P = .25], respectively). We also classified patients by their number and type of complex chronic conditions (CCCs). Complex chronic conditions are medical conditions expected to last for at least 12 months and are severe enough to warrant involvement of multiple specialists and/or a high probability of hospitalization.²⁵

We further classified zip codes by rural-urban commuting area classification code²⁶ and by designation as a health professional shortage area (HPSA). An HPSA is designated by the US Department of Health and Human Services as having shortages of primary care, dental care, or mental health clinicians secondary to geography (eg, rurality), population (eg, large low-income population), or health care facilities.²⁷

We compared differences across the income inequality categories using Kruskal-Wallis tests. We used frequencies and percentages to describe demographic and clinical characteristics of the study population and determined statistical significance with χ² tests.

To compare rates of hospitalizations for ACSCs (all-cause and by type) across the 4 income inequality categories, we used Poisson regression models adjusted for median household income and clustered on state of residence. All models were tested for overdispersion and used an offset equal to the log of the child population size in each zip code. We also calculated the number of potentially avoidable hospitalizations for each income inequality category by determining the difference between the hospitalization rates within the category and the hospitalization rate in the low income inequality category, multiplied by the number of children residing in the income inequality category. For secondary outcomes (LOS and hospital charges), we used generalized linear mixed-effects models with log-transformed outcomes, adjusted for age, payer, race/ethnicity, and CCC count and clustered on state of residence through a random intercept for each state. Estimates were back transformed to their original scale for interpretability. Owing to the low rate of missingness, records with missing outcome data were excluded from analyses (except race/ethnicity, where missing data were categorized as other). All analyses were performed using SAS software (version 9.4; SAS Institute Inc), and P < .05 was considered to be statistically significant.

A total of 8375 zip codes from 14 states met the inclusion criteria (Table 1). These zip codes included 21 737 661 children (approximately 30% of the US population of children). The median household income in the study population was $38 264 (IQR, $32 313-$48 125) and ranged from $5014 to $182 911. The median household income varied by approximately $15 000 between low and high income inequality categories (Table 2). The median Gini index was 0.41 (IQR, 0.38-0.45) and ranged across zip codes from 0.04 to 0.73. The percentage of zip codes classified into each income inequality quartile varied by state. For instance, in the low inequality category, the statewide percentage of low inequality zip codes ranged from 38.8% in Iowa to 9.7% in Florida. In the high inequality categories, the statewide percentage of zip codes ranged from 39.7% in Florida to 9.7% in Wisconsin. From 2011 to 2014, a minimal change was found in the Gini index (median, 0.002; IQR, −0.009 to 0.014) for the zip codes in the 14 states participating in both data sets.

The median number of children per zip code varied across income inequality category from 368 (IQR, 138-1183) for low inequality to 1740 (IQR, 428-4911) for high-middle inequality and 1668 (IQR, 382-4499) for high inequality (P < .001 compared with low inequality). We found small differences between income inequality quartiles in the percentage of zip codes categorized as urban (1012 [48.4%] for low income inequality and 1220 [58.2%] for high income inequality zip codes; P < .001) and those designated as an HPSA (358 [17.1%] for low inequality and 287 [13.7%] for high inequality zip codes; P < .001).

A total of 1 603 108 pediatric hospitalizations occurred, of which 79 275 (4.9%) were for ACSCs. More children hospitalized for ACSCs were aged 1 to 4 years (40.5%), were non-Hispanic white (38.2%), had public insurance (60.5%), and had no underlying CCCSs (85.9%) (Table 3). Patients hospitalized with ACSCs residing in a high inequality zip code were more likely to be covered by public insurance (66.5% vs 45.1% in a low inequality zip code; P < .001) and were less likely to be non-Hispanic white (28.0% vs 55.8% in a low inequality zip code; P < .001). The percentage of hospitalized children with ACSCs complicated by CCCs was low (14.1%) overall. However, lower percentages of CCCs were found among ACSC hospitalizations for high inequality zip codes compared with low inequality zip codes (13.0% vs 16.8%; P < .001).

After adjusting for median household income and state of residence, all-cause ACSC hospitalization rates per 10 000 children in 2014 were highest in the high inequality zip codes (31.8; 95% CI, 31.2-32.3) and lowest in low inequality zip codes (27.2; 95% CI, 26.5-27.9; P < .001) (Table 4). Adjusted hospitalization rates increased for bacterial pneumonia, dehydration, and asthma with increasing income inequality. Differences in hospitalization rates for children with gastroenteritis and urinary tract infection among income inequality categories were not statistically significant. If hospitalization rates were reduced for all inequality categories to the rate of the low inequality category, 2989 (3.8%) fewer ACSC-related hospitalizations per year would occur in these 14 states. Data from 2011 demonstrated similar findings (eTable 1 in the Supplement) with the exception that hospitalization rates for all ACSCs decreased by a median of 4.7 (IQR, −25.6 to 1.4) per 10 000 children from 2011 to 2014, which may represent reclassification into inpatient vs observation criteria rather than decreased hospitalization rates overall.

For all-cause ACSC hospitalizations, a significant but clinically unimportant longer LOS was found for high inequality (2.5 days; 95% CI, 2.4-2.5 days) compared with low inequality (2.4 days; 95% CI, 2.4-2.5 days; P < .001) zip codes (Table 5 and eFigure in the Supplement). A $765 difference in charges between quartiles was found between high inequality ($14 922; 95% CI, $14 669-$15 179) and low inequality ($14 239; 95% CI, $13 935-$14 549; P < .001) zip codes. Similar findings were observed for hospitalizations for ACSCs from 2011 (eTable 2 in the Supplement).

In this study of 21 737 661 children from 8375 zip codes in 14 states, we found that high income inequality was independently associated with higher pediatric hospitalization rates for ACSCs but did not meaningfully affect LOS or total hospital charges during the hospital stay. Approximately 3.8% of all hospitalizations for ACSCs could potentially be avoided if hospitalization rates in high inequality zip codes were reduced to those observed in low inequality zip codes. This reduction would represent a savings of nearly $67 million annually for these 14 states alone. These findings underscore the potential downstream effects of social inequality resulting in use of health care resources (eg, potentially preventable hospitalizations).

Our finding of the association of higher hospitalization rates for ACSCs with high income inequality is consistent with prior research describing associations of societies with less income disparity with improved population health for adults and children.^8-11,16,28 We also found that after children were admitted, hospital-level outcomes (LOS and charges) were similar. This finding suggests that the main difference in hospitalization rates is highly related to the circumstances preceding admission (eg, access to appropriate outpatient care) rather than the care received during a hospitalization.

Most pediatric studies use household income alone to approximate the effects of socioeconomic status on health outcomes.⁷ However, several international studies evaluating the relationship of gross domestic product against measures of income inequality at the country level suggest that income inequality is more accurately associated with health outcomes.²⁸ However, ample evidence suggests that child health is greatly affected by poverty at the family level. Our study and previous work in the United States^14,29,30 support that income and income inequality aggregated to the state and county level act individually to negatively affect health outcomes. In this context, income inequality may provide a measure in addition to poverty to refine the estimated effect of socioeconomic status overall but is unlikely to replace poverty.

Although our findings suggest that areas of high inequality represent medically underserved populations, the percentage of zip codes classified as HPSAs was highest for zip codes classified in the low inequality category. Criteria for HPSA designation include distance to health care facilities and populations with unusually high needs for primary care services, including those with greater than 20% poverty.²⁷ Our results and previous work¹⁰ suggest that income inequality may aid in the identification of geographic areas in need of enhanced primary care beyond that revealed through poverty measures alone. The addition of a measure of income inequality to the existing HPSA criteria may help better allocate primary care services to geographic areas in need.

Consistent identification and attention to social needs during pediatric preventative and acute care visits may moderate the downstream effects of income inequality.^22,31-34 A recent policy statement from the American Academy of Pediatrics³⁵ defines identification and treatment of social needs within the scope of pediatric practice and recommends screening for basic needs (eg, food, housing, and heat) during each patient encounter. In addition, the Accountable Health Communities Model, a novel program developed by the Centers for Medicare & Medicaid Services,³⁶ aims to narrow the gap between clinical care and community services during clinical visits. This model specifically emphasizes screening for unmet needs, referral to community services, support for high-risk patients through applications for community services, and alignment of clinical and community services. Although the most important effects of income inequality occur at the patient level, policy-focused concepts such as these, with increased accounting of the association of income inequality with children’s health, may lead to significant reductions in health care spending for ACSCs over time.

Our findings need to be viewed in light of several limitations. First, although zip code–based median household income has been previously demonstrated to be a useful proxy for patient socioeconomic status when individual-level data are unavailable, estimates of income at more granular levels of geography (eg, Census tract) may result in different findings.^37-40 Second, designation as an HPSA refers to an indirect measure of medical home and health care accessibility. Health professional shortage areas estimate but may not accurately reflect true access to outpatient health care. Third, although we included 14 geographically and politically diverse states from the SID of the Healthcare Cost and Utilization Project, our results may not be nationally generalizable. Fourth, our data set is limited to patient home zip codes. Zip codes relate to ZCTAs but, less accurately, to other geopolitical units where the Gini index is available (eg, county, metropolitan statistical area). The most appropriate geographic unit of measurement is uncertain; the zip code–level Gini index may reflect the inequality within that zip code or larger areas of income level transitions. Future studies defining the appropriate geographical level for measurement of income inequality are needed.^28,30,41 Fifth, variation in ACSC coding across hospitals could have influenced the study findings. The clinical overlap of many ACSCs (eg, ACSC for gastroenteritis vs dehydration) probably limits the magnitude of such influence. Sixth, the SID does not contain information on emergency department, primary care, urgent care, or other preadmission health services that might influence ACSC hospitalizations. Therefore, we were unable to assess use of these services in relation to hospital use. Finally, the associations between pediatric ACSC hospitalizations, income, and income inequality may vary widely based on variation in state-level politics and other social factors—factors accounted for in this study by clustering our statistical models on state of residence.

After adjusting for household income, zip code–level income inequality was associated with increased hospitalizations for pediatric ACSCs. The LOS and total hospital charges for ACSCs demonstrated clinically insignificant differences across inequality categories. Our findings suggest that consideration of income inequality, in addition to income level, better estimates the relationship between socioeconomic status and pediatric health outcomes for ACSCs. Measurements of income inequality should be used in additional health policy research to refine measurement of socioeconomic status and patients’ access to ambulatory care. Efforts aimed at reducing rates of ACSC hospitalizations should consider focusing on areas with high income inequality.

Corresponding Author: Jessica L. Bettenhausen, MD, Department of Pediatrics, Children’s Mercy Hospitals and Clinics, 2401 Gillham Rd, Kansas City, MO 64108 (jlbettenhausen@cmh.edu).

Accepted for Publication: February 1, 2017.

Published Online: April 3, 2017. doi:10.1001/jamapediatrics.2017.0322

Author Contributions: Drs Bettenhausen and Hall had full access to all the data in the study and take responsibility for the integrity of the data and the accuracy of the data analysis.

Study concept and design: Bettenhausen, Colvin, Berry, Puls, Markham, Plencner, Krager, Johnson, Queen, Latta, Riss, Hall.

Acquisition, analysis, or interpretation of data: Bettenhausen, Colvin, Berry, Puls, Markham, Plencner, Krager, Queen, Walker, Riss, Hall.

Drafting of the manuscript: Bettenhausen, Berry, Hall.

Critical revision of the manuscript for important intellectual content: Bettenhausen, Colvin, Berry, Puls, Markham, Plencner, Krager, Johnson, Queen, Walker, Latta, Riss.

Statistical analysis: Bettenhausen, Plencner, Hall.

Study supervision: Berry, Walker.

Conflict of Interest Disclosures: None reported.

Funding/Support: This study was supported by internal funds of Children’s Mercy Hospitals and Clinics.

Role of the Funder/Sponsor: The sponsor had no role in the design and conduct of the study; collection, management, analysis, and interpretation of the data; preparation, review, or approval of the manuscript; and decision to submit the manuscript for publication.