Disparities Associated with Sepsis Mortality in Critically Ill Children

Abstract

Disparities in health care related to socioeconomic status and race/ethnicity are well documented in adult and neonatal sepsis, but they are less characterized in the critically ill pediatric population. This study investigated whether socioeconomic status and/or race/ethnicity is associated with mortality among children treated for sepsis in the pediatric intensive care unit (PICU). A retrospective cohort study was conducted using information from 48 children's hospitals included in the Pediatric Health Information System database. We included visits by children ≤ 21 years with All Patients Refined Diagnosis-Related Groups (APR-DRG) diagnosis codes of septicemia and disseminated infections that resulted in PICU admission from 2010 to 2016. Multivariable logistic regression was used to measure the effect of race/ethnicity and socioeconomic status (insurance status and median household income for zip code) on mortality after adjustment for age, gender, illness severity, and presence of complex chronic condition. Among the 14,276 patients with sepsis, the mortality rate was 6.8%. In multivariable analysis, socioeconomic status, but not race/ethnicity, was associated with mortality. In comparison to privately insured children, nonprivately insured children had increased odds of mortality (public: adjusted odds ratio [aOR]: 1.2 [1.0, 1.5]; uninsured: aOR: 2.1 [1.2, 3.7]). Similarly, children living in zip codes with the lowest quartile of annual household income had higher odds of mortality than those in the highest quartile (aOR: 1.5 [1.0, 2.2]). These data suggest the presence of socioeconomic, but not racial/ethnic, disparities in mortality among children treated for sepsis. Further research is warranted to understand why such differences exist and how they may be addressed.

Introduction

Pediatric sepsis is a leading cause of pediatric morbidity, mortality, and health care costs. ^{1 2 3} Each year, more than 75,000 children are treated for severe sepsis in the United States, with up to 20% resulting in death. ¹ While there are clear disparities in sepsis mortality in adult and neonatal intensive care unit (ICU) data, much less is known about patients admitted to pediatric ICU (PICU). Disparities are defined as “differences in the quality of health care that are not due to access-related factors or clinical needs, preferences, and appropriateness of intervention.” ⁴ Understanding where disparities exist is imperative to decreasing mortality and providing equitable care.

Recent studies have demonstrated racial/ethnic and socioeconomic (SES) related disparities in sepsis-related mortality among adults. For example, nonwhite adults have almost double the incidence of sepsis ¹ and have higher rates of sepsis-related mortality when compared with white patients. ^{5 6 7} For adults, the loss of life years due to racial disparities in mortality due to infection comes second only to cardiovascular disease. ⁸ With regard to SES-related disparities in sepsis, lack of insurance is associated with increased risk of organ dysfunction at sepsis presentation ⁹ and predicted in-hospital mortality when compared with privately insured patients even after severity adjustment. ¹⁰

In the pediatric population, there are no data showing racial/ethnic disparities in all-cause PICU mortality, ¹¹ however there are clear disparities within certain subgroups such as traumatic brain injury, ¹² kidney transplantation, ^{13 14} diabetes, ^{15 16} and asthma. ^{17 18} Much less is known about disparities in sepsis, apart from emerging data in the neonatal population. Thus far, neonatal data reveal black race, lower household income, and uninsured status are associated with increased sepsis mortality. ^{19 20} To our knowledge, no prior study has evaluated the association of race/ethnicity or SES status on sepsis-related mortality among children admitted to the PICU. We hypothesized that nonwhite race and lower SES status were associated with increased risk of mortality among children treated for sepsis in PICUs.

Materials and Methods

Study Design

We conducted a retrospective study of PICU admissions for sepsis among children's hospitals from 2010 through 2016 using the Pediatric Health Information System (PHIS) database. This study used deidentified data and was determined to be nonhuman subject research by our Institutional Review Board.

Data Source

PHIS is an administrative database that contains inpatient, emergency department, ambulatory surgery, and observation encounter level data from 48 children's hospitals across the United States. Contributing hospitals include pediatric tertiary care centers located in 27 states, including the District of Columbia, affiliated with the Children's Hospital Association (Overland Park, Kansas, United States). PHIS contains deidentified administrative data (demographics, diagnosis codes, procedure codes) and billing data (clinical charges) on all hospital discharges.

Study Population

We included all visits by children ≤ 21 years old from 2010 to 2016 with All Patients Refined Diagnosis-Related Groups (APR-DRG) diagnosis codes of septicemia and disseminated infections that resulted in PICU admission. Patients admitted to the neonatal ICU (NICU) were excluded. The APR-DRG is a patient classification scheme used by the Centers for Medicare & Medicaid Services (CMS) for hospital payment, which cohorts patients by resource intensity such that they can be divided into meaningful groups. The various individual International Classification of Disease (ICD) codes for illness such as “sepsis,” “severe sepsis,” “SIRS,” “septic shock,” “bacteremia” and more are grouped into the same APR-DRG of septicemia and disseminated infections. APR-DRG has been used in previous studies ^{21 22} to utilize a broader sepsis cohort than individual ICD diagnosis codes alone.

Outcome Measure

The primary outcome was sepsis-related mortality during pediatric ICU admission.

Exposures

The exposure variables were race/ethnicity and SES status. Race and ethnicity were categorized as non-Hispanic (NH) white, NH black, Hispanic, and NH other. SES status was analyzed with two variables: insurance status and median household income by zip code. For insurance status, patients were characterized as being publicly insured, privately insured, and uninsured. Missing data for race and insurance were classified as “not documented” ( Table 1 ). With respect to median household income by zip code, patients were stratified into category quartiles based on 2010 U.S. Census Data: category 1 defined as US$20,000 or less, category 2 as greater than US$20,000 to US$38,043, category 3 as greater than US$38,435 to US$61,753, and category 4 as more than US$61,753. If income was missing, it was categorized as “unknown”. The use of median household income by zip code as a measure of SES has been previously used in the literature. ^{23 24 25 26 27}

Table 1. Characteristics of study population.

Demographics	All, n  = 14,276	Died, n  = 970 (6.8%)
Age categories
0–5 y	6,270 (43.9%)	433 (44.6%)
6–11 y	2,753 (19.3%)	215 (22.2%)
12–17 y	4,194 (29.4%)	226 (23.3%)
18–21 y	1,059 (7.4%)	96 (9.9%)
Female gender	7,175 (50.3%)	483 (49.8%)
Race/ethnicity
Non-Hispanic white	7,084 (49.6%)	426 (43.9%)
Non-Hispanic black	2,288 (16.0%)	175 (18.0%)
Hispanic	2,916 (20.4%)	216 (22.3%)
Other	1,591 (11.1%)	122 (12.6%)
Not documented	397 (2.8%)	31 (3.2%)
Insurance status
Private	5,668 (39.7%)	303 (31.2%)
Public	8,332 (58.4%)	640 (66.0%)
Uninsured	210 (1.5%)	21 (2.2%)
Not documented	66 (0.5%)	6 (0.6%)
Median household income by zip code (2010 U.S. Census quartiles)
More than US$61,735	2,203 (15.4%)	114 (11.8%)
> US$38,043 to US$61,735	6,113 (42.8%)	401 (41.3%)
> US$20,000 to US$38,043	5,287 (37.0%)	401 (41.3%)
US$20,000 or less	271 (1.9%)	23 (2.4%)
Unknown	402 (2.8%)	31 (3.2%)
Presence of complex condition	10,049 (70.4%)	896 (92.4%)
Severity of illness (extreme or major)	12,716 (89.1%)	962 (99.2%)

We examined both patient- and visit-level variables. Patient-level covariables included age, gender, and presence of complex condition. Complex condition is coded in PHIS using Feudtner et al's complex chronic conditions categorization system. ²⁸ This refers to a comprehensive set of ICD codes that meet the following definition: “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 organ system severely enough to require specialty pediatric care and probably some period of hospitalization in a tertiary care center.” ²⁸ Based on this definition, a comprehensive set of ICD-9 codes were identified as indicative of a complex chronic condition. Severity of illness was included as a visit characteristic. Severity of illness is assigned in PHIS using the APR-DRG grouper and defined as 0 = not applicable, 1 = minor, 2 = moderate, 3 = major, and 4 = extreme. ²⁹ We dichotomized this variable by combining minor/moderate and major/extreme for ease of analyzing the impact of higher severity of illness (major/extreme) on outcomes.

Data Analysis

We used standard descriptive statistics to summarize the patient population with sepsis and disseminated infection APR-DRG codes and measured differences in mortality rates. We performed bivariable logistic regression analyses to measure the association of race/ethnicity and SES status with mortality. We examined the relationship between all covariables and mortality. Missing data were included in the model but categorized in as “not documented” or “unknown” category; because we could not make inferences about these categories, we did not present missing data in the final models. Given that insurance status is often used as a proxy for SES, correlation analysis was conducted between insurance status and median household income by zip code, showing a weak positive relationship ( r  = 0.3). We produced one final model that included all exposures as well as covariables with p -Values < 0.2. The following covariables were included: age, presence of a complex condition, and severity of illness. All models used generalized estimating equations to account for clustering by hospital, and a goodness-of-fit test was used to estimate model performance. An α of 0.05 was used to signify statistical significance, and adjusted odds ratios (aORs) with 95% confidence intervals (CIs) were reported. We used SAS Software version 9.3 (SAS Institute Inc.; Cary, North Carolina, United States) to perform all analyses.

Results

During the study period, there were 14,276 sepsis-related admissions to the PICU with 13,172 unique visits; 6.36% of patients were readmitted. The median (interquartile range) patient age was 7.8 (1.5–14.8) years. Half (49.6%) identified as NH white. The majority (58.4%) were publicly insured and 1.5% were uninsured ( Table 1 ). Among the entire population, the mortality rate from sepsis was 6.8%.

Racial/Ethnic Differences

Mortality by race/ethnicity was as follows: 6.0% for NH white, 7.7% for NH black, 7.4% for Hispanic, and 7.7% for NH other. In bivariable analysis, NH black children were more likely to die from sepsis (odds ratio: 1.3; 95% CI: 1.1–1.6) when compared with NH white children. However, after adjusting for age, insurance status, household income, presence of a complex condition, and severity of illness, there were no racial/ethnic differences in sepsis-related mortality ( Table 2 ).

Table 2. Sociodemographic factors associated with sepsis mortality.

	Death crude OR (95% CI), p -Value	Death aOR a (95% CI), p -Value
Race/ethnicity
Non- Hispanic white	Reference	Reference
Non- Hispanic black	1.3 (1.1–1.6) , p = 0.01	1.1 (0.9–1.3), p  = 0.4
Hispanic	1.3 (1.0–1.6), p  = 0.05	1.0 (0.8–1.3), p  = 0.7
Other	1.3 (0.9–1.7), p  = 0.05	1.2 (0.9–1.5), p  = 0.3
Insurance status
Private	Reference	Reference
Public	1.5 (1.2–1.8), p < 0.01	1.2 (1.0–1.5), p = 0.02
Uninsured	2.0 (1.1–3.5), p = 0.02	2.1 (1.2–3.7), p < 0.01
Median household income quartiles
> US$61,735	Reference	Reference
> US$38,043 to US$61,735	1.3 (1.0 – 1.6), p = 0.03	1.2 (1.0 – 1.5), p  = 0.1
> US$20,000 to US$38,043	1.5 (1.2 – 1.9), p < 0.01	1.4 (1.1–1.7) , p = 0.01
≤ US$20,000	1.7 (1.1 – 2.6), p = 0.01	1.5 (1.0–2.2) , p = 0.04

Abbreviations: aOR, adjusted odds ratio; CI, confidence interval; OR, odds ratio.

^aAdjusted for age, race/ethnicity, insurance status, household income, presence of a complex condition, and severity of illness.

Note: model goodness of fit: p  = 0.80. Bold values indicate p  < 0.05.

Socioeconomic Differences

Insurance Status

With respect to insurance status, nonprivately insured patients had higher mortality rates (public: 7.7%, OR: 1.5 [1.2, 1.8]; uninsured: 10%, OR: 2.0 [1.1, 3.5]) compared with privately insured children (5.4%). After adjusting for age, race/ethnicity, household income, presence of a complex condition, and severity of illness, nonprivately insured children had increased odds of mortality when compared with privately insured children (public: aOR: 1.2 [1.0, 1.5]; uninsured: aOR: 2.1 [1.2, 3.7]) ( Table 2 ).

Median Household Income by Zip Code

There were also SES-related differences in mortality with respect to median household income by zip code. Patients in the lowest median household income quartile had higher mortality rates when compared with patients in the highest quartile (lowest quartile: 8.5%; highest quartile: 5.7%). In bivariable analysis, children living in households in the lowest income quartile had higher odds of mortality compared with children living in households with the highest quartile (OR: 1.7 [1.1, 2.6]). In multivariable analyses, children living in the lowest quartile of annual household income had significantly higher mortality rates than children living in zip codes with the highest quartile of annual household income (aOR: 1.5 [1.0, 2.2]) ( Table 2 ).

Discussion

In this study, using data from more than 40 pediatric hospitals, children diagnosed with sepsis who were of lower SES status had a greater risk of mortality. Publicly insured children were more likely to die than privately insured children, and uninsured children had twice the odds of mortality. Similarly, there is an inverse relationship between median household income by zip code and mortality; with each decline in median household income quartile, the likelihood of mortality increased, resulting in a statistically significant difference in mortality between the lowest and highest quartile.

These results support our hypothesis that lower SES status is associated with greater risk of sepsis mortality and align with adult data ^{8 9} as well as neonatal ICU data. ^{10 19} These data reflect the same inverse relationship found between median income quartile for zip code and pediatric inhospital mortality. ²⁴ However, while prior studies treat insurance status as a proxy for SES, this study suggests that insurance status may be an independent risk factor for death, as significant differences in mortality by insurance status persisted even after adjustment for median household income by zip code.

Our study did not find an association between race/ethnicity and sepsis-related mortality. This differs from adult ^{5 6 7} and neonatal ²⁰ sepsis studies. These studies may not have adequately adjusted for SES status, or, conversely, it may be that differences in race are actually mediated by SES status. The first (unadjusted) column in Table 2 may reflect the mediating effects of SES on race; hence, we see significantly higher odds of mortality in the NH black group and more pronounced increased odds of mortality with respect to insurance status and median household income. The subsequent adjusted values ( Table 2 ) limit confounders and still unmask important disparities with regard to SES status. Our data align with previous studies that did not show racial disparities in all-cause PICU mortality, ^{11 30} though, overall, the data are mixed, and all-cause mortality cannot be adequately compared with sepsis mortality alone.

While it was beyond the scope of this study to identify causes for SES disparities in mortality, we can surmise a few reasons for the results observed. There is evidence in adult data that the prevalence of sepsis is higher in low-income areas ³¹ and that stress and poor nutrition, which often accompanies poverty, can increase susceptibility to infection. ^{32 33 34} Additionally, SES status is thought to be associated with lower health literacy and greater barriers to access care ^{35 36} and may contribute to worse organ dysfunction at presentation. That said, we found SES-related disparities even after adjustment for severity of illness. Additionally, we found differences by median household income for patient zip code that were independent of insurance status; this may be because this was a distinct marker of poverty (and its associated resource deprivation and stress), whereas insurance status represented access to and quality of care. Prior literature has also discussed the role of the health care system in health outcomes for vulnerable populations. ³⁷ For instance, minorities and Medicaid patients are more likely to receive care at hospitals that offer lower quality care. ³⁸ Other studies have shown that uninsured patients have increased adverse events, ³⁹ lower resource utilization, ⁴⁰ and shorter length of stay, ³⁰ whereas public insurance has been associated with higher PICU readmission rates. ² What is less clear is whether the treatment received is influenced by factors such as implicit bias. A recent single-center study utilizing an electronic sepsis alert showed that NH white patients were more likely to be treated for sepsis outside of the alert (through clinician judgment alone) when compared with NH black patients. ⁴¹ Overall, these data raise important questions regarding why SES disparities exist and introduce a potential area for intervention to decrease PICU sepsis mortality.

This study has several important limitations. First, PHIS is an administrative database based on billing codes, thus coding errors would result in misclassification bias; to address this, PHIS undergoes rigorous reliability and validity checks before data can be analyzed. Moreover, severity of illness is limited with PHIS because it is derived from billing codes as opposed to granular data such as vital signs or laboratory values. Another limitation of this study is that it did not investigate specific sepsis syndromes or primary organ dysfunction, for which mortality rates and association with SES status may differ. Along the same lines, our study is unable to delineate if sepsis was the primary cause of mortality or an associated diagnosis. In prior studies, investigating sepsis mortality by different ICD-9 codes yielded different mortality rates; for example, using ICD-9 for “infection plus organ dysfunction” yielded a lower mortality rate when compared with “severe sepsis and septic shock.” ¹ Our study, which encompassed an even broader set of diagnoses (all those in the APR-DRG category of “septicemia and disseminated infections”), had a mortality rate of 6.8% comparable to the former study's “infection plus organ dysfunction” mortality rate of 8.2%. ¹ For the purposes of our study, in which we wanted to reduce provider bias inherent in diagnosis selection, we intentionally selected the broadest classification of sepsis-related illness. Lastly, this study is unable to account for patients who were evaluated prior to admission, for example, at a primary care clinic or from a transferring hospital. In an illness process such as sepsis, for which early recognition, time to antibiotics, and fluid resuscitation are known to improve outcomes, any delay to care may disproportionately contribute to patient admission to the ICU.

Conclusion

This study suggests potential disparities in sepsis mortality among children based on SES differences and insurance status but not race/ethnicity. Identifying this gap is important because it illuminates a potential area of investigation and intervention that may help decrease pediatric sepsis mortality. Further research is warranted to understand why such disparities exist and how they may be addressed.

Funding Statement

Funding None.