OBJECTIVE: Socioeconomic status (SES) is inversely related to pediatric mortality in the community. However, it is unknown if this association exists for in-hospital pediatric mortality. Our objective was to determine the association of SES with in-hospital pediatric mortality among children’s hospitals and to compare observed mortality with expected mortality generated from national all-hospital inpatient data.
METHODS: This is a retrospective cohort study from 2009 to 2010 of all 1 053 101 hospitalizations at 42 tertiary care, freestanding children’s hospitals. The main exposure was SES, determined by the median annual household income for the patient’s ZIP code. The main outcome measure was death during the admission. Primary outcomes of interest were stratified by income and diagnosis-based service lines. Observed-to-expected mortality ratios were created, and trends across quartiles of SES were examined.
RESULTS: Death occurred in 8950 (0.84%) of the hospitalizations. Overall, mortality rates were associated with SES (P < .0001) and followed an inverse linear association (P < .0001). Similarly, observed-to-expected mortality was associated with SES in an inverse association (P = .014). However, mortality overall was less than expected for all income quartiles (P < .05). The association of SES and mortality varied by service line; only 3 service lines (cardiac, gastrointestinal, and neonatal) demonstrated an inverse association between SES and observed-to-expected mortality.
CONCLUSIONS: Within children’s hospitals, SES is inversely associated with in-hospital mortality, but is lower than expected for even the lowest SES quartile. The association between SES and mortality varies by service line. Multifaceted interventions initiated in the inpatient setting could potentially ameliorate SES disparities in in-hospital pediatric mortality.
- APR-DRG —
- All Patient-Refined, Diagnosis-Related Groups v.24
- CCC —
- complex chronic condition
- ICD-9-CM —
- International Classification of Diseases, Ninth Revision, Clinical Modification
- PHIS —
- Pediatric Health Information System
- Q-AHI —
- quartile of annual median household income
- SES —
- socioeconomic status
What’s Known on This Subject:
Socioeconomic status (SES) is inversely related to mortality and health in children; the higher an individual’s SES, the less likely illness and death. It is unknown whether the association of SES and pediatric mortality exists in the inpatient setting.
What This Study Adds:
Within children’s hospitals, in-hospital mortality is inversely associated with SES, but is lower than expected for even the lowest SES quartile. The association between SES and mortality varies by clinical service line.
Socioeconomic status (SES) is inversely related to mortality and health status in both children and adults; the higher an individual’s SES, the less likely illness and death.1–15 This association is based in part on direct income-related reasons, such as health insurance status,10,16 access to health care,4,16–18 material deprivation,19–23 and education.24 It is also based on differences in psychological factors (eg, sense of control),25–29 social factors (eg, social capital),30–32 and the physical environment (eg, housing and neighborhood conditions).33–39 These factors vary by SES and cause differential effects on health care access, health behaviors, stress, and exposures to pathogens.8,40
Death is a rare outcome in pediatrics, and only half of pediatric deaths occur in hospitals.41 However, because prehospital health is associated with SES, including the prevalence of chronic conditions (more prevalent among those with lower SES), health-related quality of life, and access to care and specialists,10 it would be expected that in-hospital mortality might reflect the broader epidemiological association of SES and pediatric mortality. Previous studies of SES and in-hospital death have focused on a limited number of diagnoses or institutions, making it difficult to determine whether SES–mortality associations were specific only to the examined diagnosis or institution/region.42–45 It remains unknown whether the association of SES and mortality exists generally in the inpatient pediatric setting and how that association might differ across diagnoses.
The objective of this study was to determine the association of SES with in-hospital pediatric mortality. We hypothesized that inpatient mortality, being dependent on prehospital health, would be associated with SES in an inverse gradient relationship for all diagnoses. Similarly, we also hypothesized that patients with lower SES would experience higher mortality rates than expected mortality rates generated from national all-hospital inpatient data.
Data for this retrospective cohort study were obtained from the Pediatric Health Information System (PHIS), which contains administrative data from 43 freestanding tertiary care children’s hospitals across the United States. For each hospital discharge, the PHIS database includes disposition (eg, death, home), patient demographics, up to 41 International Classification of Diseases, Ninth Revision, Clinical Modification (ICD-9-CM) diagnoses, and up to 41 ICD-9-CM procedures. The Children’s Hospital Association (Overland Park, KS), participating hospitals, and Thomson Reuters Healthcare (New York, NY) jointly ensure data quality and reliability as described elsewhere.46–48 This study was approved with informed consent waiver by the Institutional Review Board at Children’s Mercy Hospital.
All inpatient hospitalizations in calendar years 2009 and 2010 at 42 hospitals were included in this analysis; 1 hospital contributing data to PHIS was excluded because patient ZIP code–linked census data were not available. No age range limitations were made because patients >18 years have been shown to be at high risk of mortality at children’s hospitals.49
Patient demographic variables included age, gender, race/ethnicity, and primary payer. Race/ethnicity categories included white, black or African American, Hispanic or Latino, American Indian or Alaska Native, Asian, Native Hawaiian or other Pacific Islander, and other. The “other” category included unreported or missing data or any category not previously mentioned. The primary payer variable of “public” included Medicaid (including Medicaid managed care) and Title V. “Commercial” payer included privately purchased health insurance and TRICARE. “Uninsured” included “self-pay” and “charity.” “Other” indicated Medicare, worker’s compensation, other governmental insurance, missing payer information, and those patients who were not charged for the services provided.
PHIS uses the All Patient-Refined, Diagnosis-Related Groups v.24 (APR-DRGs) (3M Health Information Systems, St. Paul, MN) that are based on ICD-9-CM diagnosis and procedure codes assigned during each patient’s episode of care. These APR-DRGs are further categorized into 11 clinical service lines based on the primary organ system or procedure of the APR-DRG. The service lines were developed by PHIS participating hospitals and the Children’s Hospital Association. The defined service lines are neonatal, cancer/hematology, cardiac, respiratory, orthopedics/joint, transplantation, gastrointestinal, neurologic, and infectious disease; the service lines “other medical condition” and “other surgical condition” are those APR-DRGs not principally described as one of the previously mentioned service lines. Other clinical variables include use of intensive care and mechanical ventilation. ICD-9-CM codes were used to detect the presence of complex chronic conditions (CCCs) by using a previously reported classification scheme.50,51 A CCC is defined as “any medical condition that can be reasonably expected to last at least 12 months (unless death intervenes) and to involve either several different organ systems or one system severely enough to require specialty pediatric care and probably some period of hospitalization in a tertiary care center.”50,51 ICD-9-CM codes were also used to detect the presence of a nosocomial infection by using a previously reported classification scheme.49,52,53
The APR-DRG system also assigns each hospitalization a severity-of-illness level based on the degree of organ dysfunction and a risk-of-mortality level based on the likelihood of death.54,55 The severity and risk-of-mortality scores account for specific patient factors, including primary and secondary diagnoses, the combination of diagnoses, procedures performed, and patient age.49,56,57 Thomson Reuters then assigns an expected mortality rate to each discharge based on the assigned APR-DRG and risk-of-mortality.57 The expected mortality rates are based on Thomson Reuters’ database of >20 million annual pediatric discharges from >2700 US acute, nonfederal, general hospitals.56 Therefore, the expected mortality rates reflect national all-hospital data and not data solely from PHIS hospitals or other children’s hospitals.
The main exposure of interest was SES, represented in this study by the median annual household income for the ZIP code of the patient’s residence. Quartiles of annual median household income (Q-AHI) were generated from 2010 US Census data as quartile 1, $33 311 or less; quartile 2, $33 332 to $41 386; quartile 3, $41 387 to $54 013; and quartile 4, $54 014 or more. Patients were assigned to a quartile based on the median annual household income of their home ZIP code. ZIP code–based median household income has been previously demonstrated to be a useful proxy for patient SES when individual-level data are unavailable.58–60
Main Outcome Measure
The main outcome measure of interest was death occurring during hospitalization.
All statistical analyses were performed by using SAS v.9.3 (SAS Institute, Cary, NC), and P values <.05 were considered statistically significant. First, the percentage of mortality in each Q-AHI was calculated. Bivariate analyses with the use of the χ2 test were performed to compare survivors with nonsurvivors for demographic and clinical characteristics. The χ2 test was also used to determine the association between the percentage of mortality by Q-AHI and each service line. The Cochran-Mantel-Haenszel test was used to assess linear trends across Q-AHI.
Observed-to-expected mortality ratio was also considered. An observed-to-expected mortality <1 indicated that in-hospital mortality was less than expected. The observed-to-expected mortality ratio was analyzed by using the Flora Z score to identify service lines and quartiles of income in which the observed mortality was significantly lower (or higher) than the expected mortality.61 A Poisson regression model for each service line was used to verify whether there was a significant trend in observed-to-expected mortality across Q-AHI.
Demographic and Clinical Characteristics
Of the 1 053 101 hospitalizations in years 2009 and 2010, death occurred in 8950 (0.85%). Demographic and clinical characteristics are shown in Table 1. Nonsurvivors were younger than survivors with 41.2% of nonsurvivors being neonates, compared with 11.2% of survivors (P < .001). Nonsurvivors had a significantly higher percentage of governmental insurance compared with survivors (51.5% vs 47.5%, P < .001). In comparison with survivors, nonsurvivors had more CCCs (73.6% vs 37.6%, P < .001), and required more ICU services (55.1% vs 14.1%, P < .001), NICU services (35.8% vs 5.9%, P < .001), and mechanical ventilation (87.1% vs 8.2%, P < .001). Nonsurvivors had more admissions associated with the neonatal service line than survivors (36.0% vs 7.9%, P < .001). Nonsurvivors also had more nosocomial infections (5.4% vs 0.3%, P < .001).
Demographic and clinical characteristics are stratified by income quartile and survival in Table 2. Within each income quartile, differences between survivors and nonsurvivors were statistically significant (P < .001) for all demographic and clinical characteristics, with the exception of the presence of a hematologic/immunologic CCC in the lowest income quartile (P = .264). In comparison with survivor patients, the distribution of nonsurvivor patients was shifted toward the neonatal age category and neonatal service line for all income quartiles. Nonsurvivors also had higher percentages of patients in the cardiac, transplantation, and other surgical condition service lines in all income quartiles. With the exception of the first income quartile of the hematology/immunology CCC, nonsurvivors had higher percentages of all CCCs for all income quartiles. Nonsurvivors also had >87% of patients in the major/extreme severity level for all income quartiles. Across all income quartiles, nonsurvivors also had 3 times higher usage of intensive care, neonatal intensive care, or mechanical ventilation in comparison with survivors.
Association of Mortality and Income
The percentage of mortality by income quartile and service line is shown in Table 3. For all hospitalizations, mortality was associated with income (P < .0001) in an inverse linear association (P < .0001), with mortality rates decreasing as income increased. The relationship between income and mortality, however, varied by service line. Five of the 11 service lines demonstrated a statistically significant, inverse linear association between mortality and income. One additional service line (other medical condition) demonstrated an inverse linear association (P = .012) between income and mortality across income quartiles; that service line, however, did not have statistically significant differences (P = .072) in mortality between income quartiles.
To assess for differences in baseline expected mortality between income quartiles, the ratio of observed-to-expected mortality by income quartile is displayed in Fig 1. For all hospitalizations, a linear inverse association between observed-to-expected mortality and income existed (P = .014). The observed mortality, however, was significantly less than expected for all income quartiles. The association of observed-to-expected mortality to income varied by service line. Only 3 of 11 service lines demonstrated an inverse linear association between observed-to-expected mortality and income; the other 8 service lines had no association. Even for the 3 service lines with an inverse linear association between observed-to-expected mortality and income (neonatal, cardiac, gastroenterology), observed mortality was significantly lower than expected for most income quartiles. Observed mortality was significantly lower than expected for all income quartiles in the other 8 service lines, with the exception of orthopedics and transplantation.
This study describes associations between SES and inpatient mortality in children hospitalized at freestanding children’s hospitals in the United States. We identified an inverse relationship between pediatric inpatient mortality and household income. This relationship is similar to that seen epidemiologically,7,9,11,12 with death highest in the lowest income quartile and decreasing as income increases. However, we found variations across clinical service lines. When examined individually, most service lines demonstrated no association between income and mortality. In addition, we find that observed mortality was less than expected. Even for the 3 of 11 service lines with associations between observed-to-expected mortality and income, observed mortality was lower than expected for all income quartiles.
Previous diagnosis- and age-specific studies have found differing relationships between inpatient pediatric mortality and SES. In a Canadian sample, Wang et al42 found that inpatient mortality for infants with CCCs was highest in the bottom income quartile. Similarly, in patients with congenital diaphragmatic hernia, Sola et al43 found inpatient mortality differences between children from the highest and lowest income quartiles. In contrast, Chang et al44 found a nonstatistically significant SES gradient in pediatric inpatient mortality for cardiac surgery. McCavit et al45 found no association between SES and inpatient mortality in patients with sickle cell disease. The current study offers an opportunity to look across multiple diagnostic groups (service lines) and at a larger number of institutions, in part, overcoming the limitations of previous analyses. The strong relationships seen in some, but not all, service lines confirm that, although inconsistent, the relationship between SES and inpatient mortality can be important.
It remains unclear whether this inconsistent relationship between SES and pediatric mortality is due to differences in the prehospitalization SES gradient across services. In other words, it is possible that SES variably affects certain diagnoses (and associated service lines), shaping children’s prehospital risk of mortality. This would produce the observed inconsistent relationships between SES and inpatient mortality across service lines. Alternatively, in-hospital processes may be able to decrease SES disparities in some service lines more than others. According to this explanation, high-quality care, provided equitably to all children within some service lines, would be able to overcome prehospital mortality risk factors. Possible ways that hospital service lines might reduce SES disparities include the structure of providing care (eg, standardization of care) or the provision of additional services designed to assist vulnerable patients (eg, social work, medical–legal partnerships, standardized screening for social problems).62–66 However, the provision of those additional services could only be expected to reduce mortality in future hospitalizations. No data were available in this study to determine how the care structure or processes might affect the hospital care provided to vulnerable populations.
These explanations are not mutually exclusive. Additional studies could explore whether the absence of an SES gradient in inpatient mortality for some service lines is at least in part due to in-hospital processes or to an absence of an SES-based risk gradient at the time of admission. If in-hospital care attenuates prehospital SES differences for certain diagnoses, this should be examined, and ideally replicated so that low-SES patients might benefit. Additional training or inpatient service programs could reduce SES disparities further. For instance, previous studies have indicated that physicians recognize the importance of social factors in determining patient outcomes,67 but they are either unable to identify critical social factors68 or lack the capacity67,69 to connect patients to resources to ameliorate those factors. Further physician training and an increase in available resources (eg, intensive social work interventions, medical–legal partnerships) could lessen those deficiencies. These strategies may not impact the risk of mortality in the immediate hospitalization but would have a global impact on SES’s association with child health, thereby affecting the risk of mortality in subsequent hospitalizations.
This study also finds that overall mortality, regardless of income quartile, was lower in freestanding children’s hospitals than expected mortality rates generated from national all-hospital inpatient data. One plausible explanation for lower-than-expected mortality rates may be that children in this study are essentially all covered by insurance; only a small percentage (2%) of the study population was not covered by some form of either private or public health insurance. This seems to reaffirm the importance of efforts to preserve and expand access to coverage for children.70 Because of colinearity, however, our study was not able to separate the independent effects of SES and insurance status. Although patients may be admitted without insurance, children’s hospitals often assist patients in obtaining coverage. Our administrative data, which reflect insurance status at the time of discharge, do not permit us to know prehospital insurance status, and, therefore, findings may underestimate the number of uninsured patients at the time of admission. It is also possible that the children’s hospitals in this study were able to provide high-quality care that overcame expected survival rates.
Limitations to this study are important to consider. We were unable to account for SES differences in out-of-hospital mortality, such as the use of home palliative care. We also were unable to assess for any SES differences in referral patterns or family preferences for treatment at children’s hospitals. Therefore, differences based on SES may exist because of differences in the types of patients seeking care at children’s hospitals and whether out-of-hospital palliative care is used. Because the studied hospitals represent tertiary care, academic medical centers dedicated to pediatric care, it will be important in future studies to determine if the associations found here exist in other settings, such as nonchildren’s hospitals. In addition, because of colinearity, we were unable to study the independent effects of race/ethnicity. As a result, we were unable to determine if SES was a proxy for race/ethnicity in this study. Furthermore, we used ZIP code–level income data to approximate patient-level income. Although this approximation has been used previously, it may result in biases of unknown direction or significance, and individual experiences within a ZIP code may also differ from the ZIP code–wide experience (eg, an individual household may have an income higher or lower than the median).71 In addition, grouping related diagnoses into service lines may obscure important SES differences between diagnoses within a service line.
Finally, as described above, we were unable to separate prehospital differences owing to SES from in-hospital factors. One possible way to help differentiate prehospital differences from in-hospital factors would be to exclude deaths occurring within the first 24 hours of hospitalization. This would assume that deaths occurring in the first 24 hours of hospitalization are more likely to be unpreventable and more related to prehospital factors, an assumption that has not been directly studied to our knowledge. However, in the current study, we were unable to exclude deaths occurring in the first 24 hours of admission because the expected mortality calculations were based on APR-DRG and Thomson Reuters calculations that use all hospitalizations without excluding deaths based on when they occur in the hospitalization. To exclude deaths occurring in the first 24 hours in the observed data but not the expected data would result in misleading observed-to-expected mortality ratios. Future studies should attempt to separate prehospital and in-hospital factors relating to SES differences in hospital pediatric mortality.
Lower income is associated with all-cause inpatient mortality at freestanding children’s hospitals, but this association varies by service line, and mortality is lower than expected for all income quartiles. Further research is needed to determine the potential effects of in-hospital processes and interventions on the relationship between SES and pediatric mortality.
- Accepted September 17, 2012.
- Address correspondence to Jeffrey Colvin, MD, JD, Department of Pediatrics, Children’s Mercy Hospitals and Clinics, 2401 Gillham Rd, Kansas City, MO 64108. E-mail:
Dr Colvin has participated in the study design, analysis and interpretation of the manuscript, provided critical intellectual content in the revision of the manuscript, was the primary author of the manuscript, and has approved the final version of the manuscript being submitted; and Drs Zaniletti, Fieldston, Gottlieb, Raphael, Hall, Cowden, and Shah have participated in the study design, analysis and interpretation of the manuscript, provided critical intellectual content in the revision of the manuscript, and have approved the final version of the manuscript being submitted.
FINANCIAL DISCLOSURE: The authors have indicated they have no financial relationships relevant to this article to disclose.
FUNDING: No external funding.
- Singh GK,
- Siahpush M
- van Rossum CT,
- Shipley MJ,
- van de Mheen H,
- Grobbee DE,
- Marmot MG
- ↵Kenney G, McFeeters J, Yee J. Variation in access to care for low-income children with public coverage: a baseline analysis with the 2002/2003 Los Angeles County Health Survey. 2006. Available at: www.urban.org/uploadedpdf/411465_low_income_care.pdf. Accessed February 7, 2012
- Adler N,
- Snibbe AC
- Berkman L,
- Glass T
- Kawachi I,
- Berkman L
- Aysola J,
- Orav EJ,
- Ayanian JZ
- Do DP,
- Finch BK
- Feudtner C,
- Silveira MJ,
- Christakis DA
- Chang RK,
- Chen AY,
- Klitzner TS
- Slonim AD,
- Khandelwal S,
- He J,
- et al
- Feudtner C,
- Christakis DA,
- Connell FA
- ↵Scanlon MC, Harris JM, Levy F, Sedman A. Evaluation of the Agency for Healthcare Research and Quality pediatric quality indicators. Pediatrics. 2008;121(6). Available at: www.pediatrics.org/cgi/content/full/121/6/e1723
- ↵Agency for Healthcare Research and Quality. Neonatal Blood Stream Infection Rate. Neonatal Quality Indicators Technical Specifications. Version 4. June 3, 2011:1–2. Available at: www.qualityindicators.ahrq.gov/Modules/PDI_TechSpec.aspx. Accessed March 5, 2012
- ↵Hughes J. Development of the 3M all patient-refined diagnosis-related groups (APR DRGs). Available at: www.ahrq.gov/qual/mortality/Hughes.htm. Accessed February 7, 2012
- ↵Hughes J. 3M Health Information Systems (HIS) APR-DRG classification software: overview. Available at: www.ahrq.gov/qual/mortality/Hughessumm.htm. Accessed February 7, 2012
- ↵Foster D. IHI Mortality Measurement Meeting. Available at: www.ahrq.gov/qual/mortality/foster.pdf. Accessed February 7, 2012
- ↵Children’s Hospitals and Clinics of Minnesota. Outcomes: Quality of Care. Available at: www.childrensmn.org/Web/outcomes/190737.asp. Accessed March 6, 2012
- Krieger N,
- Chen JT,
- Waterman PD,
- Soobader M-J,
- Subramanian SV,
- Carson R
- Geronimus AT,
- Bound J
- Garg A,
- Butz AM,
- Dworkin PH,
- Lewis RA,
- Thompson RE,
- Serwint JR
- Garg A,
- Butz AM,
- Dworkin PH,
- Lewis RA,
- Serwint JR
- ↵Goldstein D, Holmes J. 2011 Physicians’ Daily Life Report. 2011:1–104. Available at: www.rwjf.org/content/dam/web-assets/2011/11/2011-physicians–daily-life-report. Accessed February 7, 2012
- Abdullah F,
- Zhang Y,
- Lardaro T,
- et al
- Copyright © 2013 by the American Academy of Pediatrics