The Effect of Community Socioeconomic Status on Sepsis-attributable Mortality

Abstract

Purpose

Community factors may play a role in determining individual risk for sepsis, as well as sepsis-related morbidity and mortality. We sought to define the relationship between community socioeconomic status and mortality due to sepsis in an urban locale.

Methods

Using community statistical areas of Baltimore City, we dichotomized neighborhoods at median household income, and compared distribution of outcomes of interest within the two income categories. We performed multivariable regression analyses to determine the relationship between socioeconomic variables and sepsis-attributable mortality.

Results

The collective median household income was $38,660 (IQR $32,530, 54,480), family poverty rate was 28.4% (IQR 13.5, 38.8%), and rate of death from sepsis was 3.1 per 10,000 persons (IQR 2.60, 4.10). Lower household income communities demonstrated higher rates of death from sepsis (3.65 (IQR 2.78, 4.40)) than higher household income communities (2.80 (IQR 2.05, 3.55)) (p=0.02). In regression models, household income (β= −8.42, p=0.006) and percentage of poverty in communities (β= 2.71, p=0.01) demonstrated associations with sepsis-attributable mortality.

Discussion

Our findings suggest that socioeconomic variables play significant role in sepsis-attributable mortality. Such confirmation of regional disparities in mortality due to sepsis warrants further consideration, as well as integration, for future national sepsis policies.

INTRODUCTION

Sepsis is a clinical syndrome defined by acute organ dysfunction due to dysregulated host response to infection.¹ Early epidemiological studies have highlighted racial disparities existing in the incidence of sepsis, where blacks have higher rates of sepsis than whites, without adjusting for other socioeconomic variables.^2–4 It is unclear if the disproportionate incidence rate is due to greater susceptibility to infection, risk of developing organ dysfunction or a combination thereof.⁵ Racial differences in sepsis incidence could be attributable to socioeconomic variables, as these factors have been found to be associated with infection burden⁶ and differences in health care utilization.⁷ Therefore, further investigation is warranted to better understand socioeconomic variables beyond race (which is often used as a proxy for these other socioeconomic variables)^18,19 in regards to their involvement in sepsis-related outcomes.

Community socioeconomic characteristics have been shown to influence the health of an individual through multiple pathways.^8–10 Socioeconomic characteristics, such as poverty, have also been linked with sepsis disparities, raising the concern for residual confounding among specific populations.⁴ Therefore, when investigating sepsis-related outcomes, evaluation of socioeconomic status and factors that co-localize with community socioeconomic characteristics should be targeted to best understand factors influencing sepsis disparities.

The objective of this study is to define the relationship between community socioeconomic status and sepsis, specifically regarding sepsis-attributable mortality. We hypothesized that communities with higher rates of poverty would experience higher rates of mortality due to sepsis. Further, we investigated the impact of other socioeconomic variables, such as household income, age, education, race, gender and insurance status, on sepsis-attributable mortality.

METHODS

The Baltimore City data was extracted from the 2017 Neighborhood Health Profiles, a data set available through the Baltimore City Health Department. The data set compiles a variety of demographic (individual and community level) and outcome data from several sources, resulting in community statistical areas.¹¹ Community statistical areas aim to define demographically homogenous areas (similar in social, demographic and economic characteristics) with a total population between 5,000 to 20,000 in order to allow for the aggregation of a wide range of data for a stable geography over time. Data for the Neighborhood Health Profiles comes from the 2010 US Census and the American Community Survey (2011–2015) and the Maryland State Vital Statistic Administration (2011–2015), where sepsis-attributable mortality is listed as a distinct category. Sepsis-attributable mortality is defined as the total mortality of the population minus the mortality associated with sepsis.¹¹ At the time of the completion of the 2017 Neighborhood Health Profiles, the total Baltimore City population was 622,454 with a high minority population (African American 62.8%, Hispanic/Latino 4.6%, Asian 2.6%) in 55 total neighborhoods.

We reviewed demographic data including age (percentage of the population 65 years and older), gender, and race (African American percentage). Socioeconomic characteristics included median household income, education status (specifically, percent of residents age 25 and older who have completed a bachelor degree or more) and poverty rate. In regards to poverty, the most recent year whereby US Census data was available (2015) has the poverty threshold for a family of four set at $24,257.³⁰ Poverty was determined through several variables (pre-tax income adjusted for family size, composition and age) set against a threshold three times the cost of a minimum food diet in 1963, updated annually for inflation using the Consumer Price Index.¹² The official national poverty rate in 2015 was 13.5% with 43.1 million people in poverty.³⁰

The 55 communities were dichotomized at the median household income into lower income (n=28) and higher income (n=27) as well as by percentage of poverty (20% or greater versus less than 20%). Dichotomizing by the percentage of poverty in the neighborhoods allows the data to be generalizable beyond the Baltimore City population.³⁴ Results are presented as median with 25^th to 75^th percentiles. Neighborhood characteristics and health outcomes between the two income categories were compared by Mann Whitney U test for continuous variables and Fisher’s exact test for categorical variables. Univariate and multivariable (race, income, poverty, education, gender, insurance status and age) linear regression models were evaluated for best-fit. Variance inflation factors (VIFs) were calculated to evaluate high correlation between variables. Statistical analyses were conducted with SigmaPlot 11.0 (San Jose, CA) and R software (Version 0.99.903) with a statistical significance level of 0.05.

RESULTS

Neighborhood Characteristics

The collective median income was $38,660 (IQR $32,530, $54,480). The median poverty rate was 28.4% (IQR 13.5%, 38.8%). Median percentage of uninsured adults was 12.2% (IQR 9.2%, 14.6%). Median percentage of educational status of residents 25 years or older with a bachelor degree or more was 19.5% (IQR 11.2%, 39.2%) Median sepsis-attributable death rate was 3.1 (IQR 2.6, 4.1) per 10,000 residents. Table 1 provides summary of demographics and socioeconomic variables of the 55 communities of Baltimore City.

Table 1.

Demographics of the 55 neighborhoods of Baltimore City.

Variable	Median (25th, 75th percentile)
Median Household Income (US Dollars)	38660 (32530, 54480)
Poverty Rate (%)	28.4 (13.5, 38.8)
African Americans (%)	74.3 (31.3, 90.5)
Females (%)	53.8 (50.8, 55.0)
Educationa (%)	19.5 (11.2, 39.2)
Ageb (%)	12.0 (9.1, 14.1)
Uninsuredc (%)	12.2 (9.2, 14.6)
Death from Sepsis (per 10,000 residents)	3.1 (2.6, 4.1)

^aUninsured implies adults 18 years of age or older.

^bEducation implies residents 25 years or older with a bachelor degree or more.

^cAge is defined as percentage of the population 65 years of age or older.

Lower income neighborhoods had a larger proportion of African Americans, greater rates of poverty, lower formal education, more uninsured adults and more females (Table 2) compared to higher income neighborhoods. There was no statistically significant difference in the percentage of adults ≥65 years of age between low- and high-income neighborhoods. In regard to sepsis-attributable mortality, low income neighborhoods had a higher rate of death from sepsis at 3.65 (IQR 2.78, 4.40) per 10,000 persons versus high income neighborhoods (2.80 per 10,000 persons (IQR 2.05, 3.55)) (p = 0.020) (Figure 1).

Table 2.

Demographic and socioeconomic variables between low and high income neighborhoods. Data were compared by Mann Whitney U test for continuous variables and Fisher’s exact test for categorical variables where appropriate.

Variable	Low Income (N=28)	High Income (N=27)	p-value
Median Household Income (US Dollars)	32530 (25540, 36100)	54870 (42810, 73960)	<0.001
Poverty Rate (%)	37.65 (30.05, 46.02)	13.70 (10.25, 22.60)	<0.001
African Americans (%)	90.10 (75.87, 94.95)	37.80 (14.05, 70.90)	<0.001
Females (%)	54.35 (52.88, 55.50)	51.20 (50.10, 54.45)	0.007
Residents ≥25 with a bachelor degree or more (%)	11.90 (8.53, 16.60)	36.70 (24.45, 63.25	<0.001
Population ≥65 years of age (%)	12.95 (9.10, 14.80)	11.50 (8.90, 13.75)	0.775
Uninsured adults (≥ 18 years old) (%)	13.80 (11.68, 15.35)	9.80 (7.30, 13.10)	0.001
Death from Sepsis (per 10,000 residents)	3.65 (2.78, 4.40)	2.80 (2.05, 3.55)	0.02

Figure 1.

Comparison of the rate of death from sepsis between low-income and high-income neighborhoods in Baltimore City (p=0.02)

Given that dichotomizing data based on low and high income neighborhoods is highly sample specific and may prove challenge to replicate, the neighborhoods were further grouped based on the rate of poverty in the neighborhoods (less than 20% versus 20% or greater) (Table 3). Findings were similar to the dichotomization in regards to median income, especially with regards to sepsis-attributable mortality (4.20 per 10,000 persons (IQR 2.90, 5.30) in high poverty neighborhoods versus 2.90 per 10,000 persons (IQR 2.00, 3.60) in low poverty neighborhoods, p=0.013).

Table 3.

Demographic and socioeconomic variables between neighborhoods stratified by poverty rate (poverty rate less than 20% versus poverty rate 20% or greater). Data were compared by Mann Whitney U test for continuous variables and Fisher’s exact test for categorical variables where appropriate.

Variable	≥ 20% Poverty (N=37)	<20% Poverty (N=18)	p-value
Median Household Income (US Dollars)	34520 (27450, 38910)	58240 (49150, 77140)	<0.001
Poverty Rate (%)	35.10 (28.40, 43.60)	10.55 (7.73, 13.18)	<0.001
African Americans (%)	87.60 (58.50, 94.50)	27.90 (12.18, 61.90)	<0.001
Females (%)	54.20 (51.60, 55.50)	51.75 (50.22, 54.62)	0.080
Residents ≥25 with a bachelor degree or more (%)	15.40 (8.60, 22.79)	53.80 (26.35, 66.40)	<0.001
Population ≥65 years of age (%)	11.40 (9.00, 13.50)	12.85 (10.38, 14.25)	0.311
Uninsured adults (≥ 18 years old) (%)	13.90 (11.60, 13.75)	8.35 (7.08, 10.68)	<0.001
Death from Sepsis (per 10,000 residents)	4.20 (2.90, 5.30)	2.90 (2.00, 3.60)	0.013

Sepsis-Attributable Mortality Incidence Rates

In unadjusted models, lower median household income (β = −2.23, p = 0.02) and higher poverty rate (β = 2.82, p = 0.03) were associated with higher sepsis-attributable mortality. Neighborhoods with higher proportion African American residents, lower formal education and higher rates of lack of insurance were also associated with higher sepsis mortality rates (Table 4). Gender and age did not achieve statistically significant associations in unadjusted models.

Table 4.

Regression models for sepsis-attributable mortality associated socioeconomic variables, both unadjusted and adjusted.

Variables	Unadjusted Model		Adjusted Model
	β (95% CI)	p-value	β (95% CI)	p-value
Household Income	−2.23 (−4.14 to −3.20)	0.02	−8.42 (−14.7 to −3.04)	0.006
Poverty Rate	2.82 (0.23 to 5.40)	0.033	2.71 (0.40 to 4.07)	0.013
African Americans	1.22 (0.05 to 2.38)	0.039	1.38 (−1.15 to 3.90)	0.287
Female	−0.55 (−15.65 to 14.55)	0.745	−11.33 (−29.91 to 7.26)	0.230
Uninsureda	8.55 (0.30 to 17.4)	0.038	6.35 (0.91 to 9.67)	0.040
Educationb	−1.88 (−3.69 to −0.08)	0.043	−1.27 (−3.69 to −0.06)	0.011
Agec	−7.62 (−17.07 to 1.83)	0.111	−7.76 (−17.77 to 2.15)	0.127

CI = confidence intervals.

^aUninsured implies adults 18 years of age or older.

^bEducation implies residents 25 years or older with a bachelor degree or more.

^cAge is defined as percentage of the population 65 years of age or older.

In adjusted models, lower median household income (β = −8.42, p = 0.006) and higher poverty rate (β = 2.71, p = 0.01) continued to have an association with higher sepsis-attributable mortality (Table 4). Lower formal education (β = −1.27, p = 0.01) and lack of insurance (β = 6.35, p = 0.04) were also associated with sepsis-attributable mortality. However, in the adjusted model, race was no longer statistically associated with sepsis-related deaths (β = 1.38, p = 0.28). VIFs were calculated for each variable in the adjusted model, with race demonstrating the largest VIF of 3.93. A sensitivity analysis excluding race from the adjusted model yielded similar results to the adjusted model with race (data not shown).

DISCUSSION

We found that sepsis-related mortality was higher in the low-income neighborhoods of Baltimore City compared to the higher income neighborhoods. Neighborhood poverty rate, lack of insurance status and lower formal education status were also independently associated with sepsis-related mortality. The neighborhood proportion of African American residents was associated with sepsis-related mortality in unadjusted models, but after accounting for other factors, race was no longer statistically significantly associated with mortality rates. Interestingly, we did not find a statistically significant association with gender or age and sepsis-related mortality.

Prior studies have shown that disparities in sepsis-related health outcomes (incidence and mortality) exist, with race-attributable differences often reported and highlighted.^{2,3,5,13- 15} However, it is unclear how race effects the development and natural course of sepsis and its outcomes, with recent studies and reports calling into question the influence of race on sepsis.^16,17 While we show that race appears to be associated with sepsis-related mortality, its significance was overshadowed when we adjusted our models for other socioeconomic variables at the community level. This is consistent with reports that health outcome disparities demonstrated greater alignment with social class as compared to race alone.^18,19 Therefore, to sufficiently address the gap in sepsis outcomes, more studies are needed to explore how non-biological factors influence sepsis disparities while pursuing the optimization of inpatient sepsis care.^20,21

One non-biological factor that helps capture immediate socioeconomic disparities is poverty. Goodwin et al found that poverty impacted sepsis-related outcomes in patients living in South Carolina.²² The authors reported that persons living in medically underserved areas had higher odds of being admitted with severe sepsis and were more likely to have severe sepsis-related deaths. Medically underserved areas were defined by the ratio of primary care physicians per 1000 people, infant mortality rate, proportion of population with income below the poverty rate and percentage of population ≥ 65 year of age.²² We evaluated poverty in both unadjusted and adjusted models and found it had a significant association with sepsis-attributable death. It should be noted that our definition of poverty differed from Goodwin et al given the different time points of data collection: we used the US Census 2015 data, while Goodwin et al used data from 2014 from the South Carolina Department of Health and Environmental Control. Further, median household income had a statistically significant inverse association with sepsis-attributable death.

Moore et al found that regional variations in sepsis incidence may be partly explained by community poverty.²³ Performing a retrospective analysis of data from 2003 to 2012, the authors found that among community characteristics, poverty explained a significant portion of variation in regards to sepsis incidence. A key difference between both the Goodwin et al and Moore et al study is how communities were defined in our study. Zip-code and county-level data were used to help with the prior studies investigate community-related variables.^22,23 However, as discussed in the methods section, we used community statistical areas, and creating a more demographically homogenous community as compared to zip codes and counties. Therefore, we believe that the continued identification of poverty in community statistical areas in our study strengthens prior findings of communities defined by zip codes or county.

Insurance status and access to health care have been shown to mediate the effects of other socioeconomic determinants of health.²⁴ In regards to sepsis-specific health issues, prior studies have shown that uninsured patients have higher risk-adjusted odds of sepsis-associated hospitalizations and increased odds of sepsis-associated mortality, compared to patients with insurance.^25,26 Our findings are consistent with these studies, showing an association between lack of insurance and sepsis-attributable mortality, both in unadjusted and adjusted models. Having insurance may reduce sepsis-related mortality through prevention of infections (e.g. insurance may allow easier access to vaccinations),²⁵ maintenance of comorbid illness (e.g. diabetes), and/or prevention of organ dysfunction (e.g. expedited medical attention before organ injury has taken place). Our data further show that low-income neighborhoods have higher rates of uninsured persons. Therefore, future prospective analyses should evaluate if policy and legislation in the United States aimed at decreasing the number of persons without insurance attenuates the health disparities gap in sepsis-related health outcomes.

Interestingly, we found that higher level of education – specifically obtaining a college degree – was inversely related with sepsis-attributable mortality. While education has been linked to other health outcomes, such as life expectancy,^27,28 fewer studies have focused on an association between educational status and sepsis-related outcomes ³². Education may reflect certain non-socioeconomic characteristics that impact health outcomes, such as problem solving skills or health literacy, which may be able to offset the adversities posed by other socioeconomic variables, such as lack of insurance.²⁹ Exploring how education, and potentially other socioeconomic variables, may protect individuals and communities from negative health outcomes, such as sepsis-related mortality, is warranted for future studies.

We did not take into account the baseline health of the population of Baltimore City, which may play a role in the incidence of and mortality from sepsis. For instance, many studies have shown that certain diseases that alter the immune system function are more prevalent in minority populations (e.g. diabetes).^3,33 However, sepsis disparities have been show to occur at young ages, well before certain disease consequences associated with chronicity have taken place. Barnato et al showed that sepsis disparities, specifically regarding incidence, begin to occur at the age of 20 between blacks and whites,⁴ while other studies have shown that dramatic contrasts in sepsis disparities occur between 35–44 years of age (especially in males)^2,3. Further, Mayr et al showed that the most dramatic contrasts between black and white patients in sepsis rates occurred in patients younger than 65 years of age.⁵ It is unclear if disparities in sepsis-attributable mortality occur in the pediatric population, which cannot be determined by our data and warrants future investigations. This is similar to our findings where the population of Baltimore City is young (median percentage of the population 65 or older is 12.0%). Therefore, chronic medical conditions alone are unlikely to explain the disparities found in sepsis-attributable mortality.

This study has several important limitations. First, data from the Neighborhood Profiles only evaluated mortality due to sepsis, and not the overall incidence of sepsis among individual neighborhoods. Therefore, we are unable to determine what the overall burden of sepsis is within these communities and relative mortality rates in the context of overall sepsis incidence. Second, we cannot distinguish whether sepsis-attributable deaths were due to community-acquired versus hospital-acquired sepsis, which would have different implications in approaching sepsis disparities. Third, we did not take into account the baseline health of the Baltimore City population, especially in regards to the prevalence of certain non-communicable diseases that are known to alter the immune system (e.g. diabetes) and future studies should evaluate such associations. Finally, we do not know the rate of HIV present in the communities, as HIV itself has been shown to impact sepsis disparities.^3,30 However, the Neighborhood Profile Data does distinguish between deaths due to HIV from sepsis-attributable deaths, minimizing the concern for HIV itself being a potential confounder in sepsis-attributable deaths.

CONCLUSION

Several socioeconomic variables, including poverty, insurance status, and education were associated with sepsis-attributable mortality in Baltimore City neighborhoods. These findings suggest that sepsis outcomes may be determined even before a patient enters a hospital, and influenced by local and regional rather than only individual (medical) factors. Given our findings regarding socioeconomic variables and sepsis, we believe further studies are warranted to evaluate for any potential causation, while urging future policies that impact socioeconomic status to consider the potential consequences towards sepsis-related outcomes.