Impact of socio-economic status and race on emergency hospital admission outcomes: Analysis from hospital admissions between 2001 and 2012

Abstract

Socio-economic and racial/ethnic disparities in healthcare quality have been the point of huge discussion and debate. There is currently a public debate over healthcare legislation in the United States to eliminate the disparities in healthcare. We reviewed the literature and critically examined standard socio-economic and racial/ethnic measurement approaches. As a result of the literature review, we identified and discussed the limitations in existing quality assessment for identifying and addressing these disparities. The aim of this research was to investigate the difference between health outcomes based on patients’ ability to pay and ethnic status during a single emergency admission. We conducted a multifactorial analysis using the 11-year admissions data from a single hospital to test the bias in short-term health outcomes for length of stay and death rate, based on ‘payment type’ and ‘race’, for emergency hospital admissions. Inconclusive findings for racial bias in outcomes may be influenced by different insurance and demographic profiles by race. As a result, we found that the Self-Pay (no insurance) category has the shortest statistically significant length of stay. While the differences between Medicare, Private and Government are not significant, Self-Pay was significantly shorter. That ‘Whites’ have more Medicare (older) patients than ‘Blacks’ might possibly lead to a longer length of stay and higher death rate for the group.

Keywords

Emergency admissions payment types MIMIC III hospital treatment financial bias racial bias treatment impact US Hospital Admission socio-economic status self-pay Medicare and Medicaid

Introduction

For several decades, researchers have found differences in the health outcomes of patients in the United States based on their racial and socio-economic status.^1–3 Studies have usually looked at this through the lens of a single condition or an individual and separate instances^4–10 or over geographically large study areas.¹¹ Most studies reported specific and predictable differences in health outcomes based on racial and socio-economic status. However, researchers draw modest results back to education, income, comorbidity levels, access to services and general lower socio-economic status traits common in low education/income populations. There is conflicting information³ in the literature of discrepant treatment of patients based on the race or income by the health system and health providers other than to acknowledge that it may exist in some cases and may have an effect.^4–6

Studies of individual discrimination and residential segregation are evident at the micro and meso levels and contribute to the overall health inequalities. Phelan and Link¹² have pioneered in a fundamental-cause-of-disease theory seeking to explain socio-economic and racial disparities. Phelan and Link also highlighted that, people with high socio-economic status utilize superior resources for better health, while individuals with low status have historically been denied such resources.¹² Health researchers need to specify the racially advantaged identities and advantages of privileged ‘individuals’ who control the differential allocation of relevant resources. This accent on resource inequality along with socio-economic and racial lines marks an important shift but draws little on the critical-race tradition and does not explicitly articulate the racial and socio-economic factors contributing to the health inequalities. Some of the literature reviews in this area are presented here.^13–20

The passage of the ‘Emergency Medical Treatment and Active Labor Act’ (EMTALA) by the US Congress in 1986 (42 CFR 489.24)⁷ made illegal the widespread practice at the time of hospitals ‘dumping’ indigent emergency patients by transferring them unsafely to public hospitals for care.^7,8 This allowed private hospitals to unburden themselves of non-paying patients even when it was not in the patient’s best medical interest to be moved. The concept of the law at the time was relatively simple, but its implementation based on subsequent rule interpretations and court decisions has been far reaching.^8,9 The definitions of ‘emergency’, ‘presenting at hospital’ and ‘medically sound transfer’ had become subjective, so the courts have taken a patient-centric and patient-wellbeing view of these definitions. As a result, in an effort to reduce their potential legal liability, hospitals now accept non-paying patients as a cost of doing business and write-off the expense as a charitable donation.^7–9 Moreover, one of the industry drivers of the recent Affordable Care Act legislation (Obamacare, 2010) was an effort to get more patients insured so hospitals could improve their profitability.^9–11,21

Individual experiences of healthcare and broader public health issues should be considered together when examining racial and socio-economic factors. Morbidity rates may be effected by patient–physician interactions and sadly by unjust distribution of healthcare resources based on the socio-economic and racial factors.

The focus of this work was to look for single institution bias in short-term health outcomes, as defined by length of stay (LOS) and death rate (DR), based on (a) Payment Type and (b) Race, for emergency admissions. Political conservatives are pushing for less government involvement in healthcare even though millions of Americans do not have any insurance.^22,23 Whereas liberal politicians are proposing the other extreme which is a taboo ‘single payer’ system where the government pays for all healthcare, thus levelling the playing field for everyone seeking care.^22,23 Some analysis of actual data shows bias against people without any means of payment; while this is not surprising, it would quantify health-related outcomes of real patients at a real hospital of this discrepant treatment and thus points to the total economic impact to society for this bias.^24,25

This research investigated the financial and racial factors from 50,000 patient admissions over an 11-year period from a single hospital site. This study looks exclusively at emergency patients (covered by EMTALA)⁷ and looked at data for the LOS, that is, the period they were in the hospital. Information on race and payment type is available in the selected dataset, so an analysis of two measures of outcome, LOS and DR, was conducted to examine differences in LOS and DR based on the patient’s status. By doing this, we attempt to remove socio-economic status to an extent by limiting the data to a single hospital. In addition, hundreds of illnesses and conditions were included to study the broader impact of bias by diluting some of the effects that socio-economic status can have on some long-term conditions associated with poverty.²²

The following section describes the dataset we have used and how we selected a subset for analysis of this project.

Materials and methods

Data: Medical information mart for intensive care (MIMIC-III)

With the appropriate approvals for research, we used the MIMIC-III (Medical Information Mart for Intensive Care) data set for our study, as it is a large, publicly available database comprising de-identified health-related data associated with 53,423 distinct hospital admissions for adult patients (aged 16 years or above) admitted to critical care units at Beth Israel-Deaconess Hospital between 2001 and 2012.²⁶ Beth Israel-Deaconess Medical Center is one of the largest teaching and research hospitals in the United States and is one of the teaching hospitals of the Harvard University Medical School. It is a private, not-for-profit hospital. The database includes information such as demographics, vital sign measurements made at the bedside (∼1 data point per hour), laboratory test results, procedures, medications, nurse and physician notes, imaging reports and out-of-hospital mortality. There are several variables (data tables) available in the MIMIC-III data set. The data collected in the ‘Admission Table’ of the MIMIC database used in the analysis of this work are shown in Table 1.²⁷

Table 1.

MIMIC-III admission table entries.²⁷

Data entry	Postgres data type
ROW_ID	INT
SUBJECT_ID	INT
HADM_ID	INT
ADMITTIME	TIMESTAMP (0)
DISCHTIME	TIMESTAMP (0)
DEATHTIME	TIMESTAMP (0)
ADMISSION_TYPE	VARCHAR (50)
ADMISSION_LOCATION	VARCHAR (50)
DISCHARGE_LOCATION	VARCHAR (50)
INSURANCE	VARCHAR (255)
LANGUAGE	VARCHAR (10)
RELIGION	VARCHAR (50)
MARITAL_STATUS	VARCHAR (50)
ETHNICITY	VARCHAR (200)
EDREGTIME	TIMESTAMP (0)
EDOUTTIME	TIMESTAMP (0)
DIAGNOSIS	VARCHAR (300)
HOSPITAL_EXPIRE_FLAG	TINYINT
HAS_CHARTEVENTS_DATA	TINYINT

For this study, we selected the data variables highlighted in Table 1 – ‘ADMITTIME’, ‘DISCHTIME’, ‘ADMISSION_TYPE’, ‘INSURANCE’, ‘ETHNICITY’ and ‘HOSPITAL_EXPIRE_FLAG’ – from the Admissions data table of the MIMIC III database (Table 1). For the payment type, there were five payment types recorded in the data set, shown in Table 2.²⁷

Table 2.

Payment type categories.

Payment type	Description
Government	Includes Worker’s Compensation, Federal Government Employees, Civilian Military Personnel
Medicaid	Health cover for poor and indigent patients, provided jointly by State and Federal Government. Certain eligibility criteria need to be met to qualify.
Medicare	Federal government insurance cover for major hospitalizations, in most cases, for those above 65 years old. Not based on income.
Private	Private commercial insurance, paid for by the patient and/or their employer.
Self-pay	Patients with no health insurance.

There were over 50 different race categories listed, including – ‘Unknown’, ‘Not Specified’, ‘Refused to Answer’ or ‘Unable to Obtain’. It is important to recognize that Medicaid and Self-Pay patients represent a similar lower socio-economic subset of the US population and that many patients move between the two categories frequently.

Data preparation, inclusion and exclusion criteria

The data set contains several types of admission; elective, newborn, urgent and emergency. For our analysis, we included only the emergency admissions which comprise all medical conditions. There are over 1000 diagnoses for emergency admission. We excluded the Information on religion, marital status, language and other categories as it is unrelated to the scope of the study. Patient data with ‘Unknown’, ‘Not Specified’, ‘Refused to Answer’ or ‘Unable to Obtain’ as their ‘race type’ was also excluded. Furthermore, there were over 50 different categories for ‘Race’ in the data set. To manage this, some races with lower numbers were removed like Asian (often a separate category for each Asian country) and Middle Eastern. The remaining categories were collapsed into three – Black, Hispanic and White. For example, there is a separate Hispanic category for nearly every country in Central and South America. For simplicity, these were grouped as simply Hispanic. The remainder of the patients were similarly able to be grouped into Black and White. In summary, we examined three racial categories – Black, White and Hispanic; and five Payment Type categories – Government, Medicaid, Medicare, Private (Insurance) and Self Pay.

Data analysis

To calculate LOS, the discharge date minus admission date was used, but only for patients who did not expire during their stay. This was calculated for each Race and for each Payment Type. We also investigated the LOS for the largest Racial Group, that is, White only, by Payment Type in an attempt to remove racial bias. As well, LOS was evaluated for only the largest Payment Type, Private, in an attempt to look at racial bias without regard to Payment Type.

To determine the Death Rate (DR), all patients in the reduced dataset were considered with the rate of deaths by Race and Payment Type measured. Similar to LOS, in an attempt to eliminate racial bias, DR was evaluated by Race within the Private payment type only. Similarly, in an attempt to look at Payment Type bias, DR was analysed by Race for only the Private payment type.

We used average values and standard deviations (SDs) for LOS for evaluation. To compare the data, a two-tailed, unpaired, unequal size T-test was performed.

Results and evaluation

Medical payment type by race

An analysis of Payment Type by Race shows some interesting characteristics to explain some inconsistencies that result from looking at bias based on race. Table 3 shows the number and percentages of patient admissions for different Payment Types sorted by Race. The reduced dataset contains 31,431 distinct patient admissions, and Figure 1 shows a breakdown of Payment Types by race verification when compared with other races. When compared to White and Black, Hispanics have fewer Medicare and more Medicaid patients. This generally implies that Hispanic patients are younger and poorer, and this is in general agreement with larger demographic data for the country as a whole.^27,28

Table 3.

Number of patients in each group studied, actual (percentage).

Race	Government (%)	Medicaid (%)	Medicare (%)	Private (%)	Self-pay (%)	Total (%)
Black	212 (0.7)	758 (2.4)	2175 (6.9)	867 (2.8)	50 (0.2)	4062 (12.9)
Hispanic	156 (0.5)	361 (1.1)	523 (1.7)	304 (1.0)	48 (0.2)	1392 (4.4)
White	509 (1.6)	1947 (6.2)	14,810 (47.1)	8484 (27.0)	227 (0.7)	25,977 (82.6)
Total	877 (2.8)	3066 (9.7)	17,508 (55.7)	9655 (30.7)	325 (1.0)	31,431 (100)

Figure 1.

Plot of payment type count for each racial type in the reduced data set.

LOS by insurance type

Table 4 shows average LOS in days along with the SD in days. The actual data’s average and standard deviation (STD) show LOS values of less than zero for many patients. In fact, the data is not perfectly normal as there are some outliers to the high side, with a few over 150 days and one over 200 days. It is worth further investigating the study effects of these outliers on the conclusions of this work, including the appropriateness of the T-tests used.

Table 4.

LOS by payment type.

Payment type	LOS	SD
Government	9.6	11.5
Medicaid	10.8	13.1
Medicare	9.9	9.3
Private	10.1	11.2
Self-pay	7.1	7.5

Even though Medicaid and Self-Pay have similar socio-economic demographic profiles,^2,27,28 and often patients move back and forth between the two categories regularly depending on their fluctuating eligibility, their health outcomes are dramatically different (P = 2.3E-14) based on their ability to pay for this particular hospital admission. Medicaid has statistically the longest stay, indicating a higher level of care and possibly lower general health of the group, while Self-Pay has statistically the shortest stay by far, indicating that, for a similar population, inability to pay means 34% shorter stays (less access to care).

Death rate by insurance type

Table 5 shows the DR by Payment Type. The data show that only Self-Pay and Medicare have statistically similar DRs, and they have the largest values of the five categories. All others are significantly (P < 0.05) different from each other.

Table 5.

DR by payment type.

Payment type	Death rate (%)
Government	6.1
Medicaid	7.8
Medicare	14.8
Private	9.0
Self-pay	13.3

This also shows evidence of statistically disparate outcome (death) based on ability to pay. Self-Pay has a 1.7× higher DR than the similar Medicaid group (13.3% vs. 7.8%. In fact, the only statistically similar comparison is that Self-Pay DRs are the same as the much older Medicare group who would be expected to have a higher DR due to their age.

LOS by race

In the Race category, Table 6 shows LOS and SD based on the Race. The significant difference (P = 0.016) was between Black and White, with White having a longer LOS than Black.

Table 6.

LOS by race.

Race	LOS	Standard deviation
Black	9.7	10.5
Hispanic	9.6	9.8
White	10.1	10.4

We found that the LOS for White is statistically similar to Hispanic, and Hispanic and Black is also similar, but White versus Black is statistically different with White LOS is longer. It appears to show racial bias between Black and White groups, but subtleties in Payment Type and age variances between the groups could have an impact. This is explored further where this relationship is evaluated within a single Payment Type.

Death rate by race

Table 7 shows the death by Race for all Payment Types. This highlights the highest, and most insured race (White) have the highest DR. It is possible that a lower group age leads to lower Hispanic DRs.

Table 7.

Death by race.

Race	Death rate (%)
Black	9.0
Hispanic	7.5
White	12.8

LOS by payment type within race (White)

In order to isolate LOS by Payment Type from Race interactions, LOS was evaluated for one Race, White, as it was the largest group. A similar analysis could be done for the Black and Hispanic groups but is omitted here for simplicity. The results are shown in Table 8.

Table 8.

LOS by payment type for White group only.

Payment type	LOS	Standard deviation
Government	9.8	11.3
Medicaid	11.2	10.6
Medicare	9.9	9.2
Private	10.3	11.4
Self-pay	7.0	7.6

Discrepancy between Medicaid LOS and Self-Pay LOS remains (P = 1.1E-11), but the distinction between Private and Medicare now becomes significant (P = 0.0176). It is unclear why private patients not limited to age >65 would now be significantly different than the result of the entire group since the numbers look similar to the results that include all races. It is possible that the Payment/Insurance Types themselves set rules on LOS for full reimbursement.

Death rate by payment type (White)

To try to isolate Payment Type results from Race interactions, DR was evaluated within the White group only, chosen because it was the largest group. A similar analysis could be done for the Black and Hispanic groups but is omitted here for simplicity. Table 9 show DR averages and statistical comparison, respectively.

Table 9.

DR by payment type for White group only.

Payment type	Death rate (%)
Government	6.6
Medicaid	8.4
Medicare	15.5
Private	9.0
Self-pay	14.0

As before, the DR for the Self-Pay population is statistically indifferent (P = 0.467) to the older Medicare population. Also, as before, the DR for Self-Pay patients is much higher (P = 0.0122) than their socio-economically similar Medicaid group.

Finally, evaluating racial factors within a single payment type, LOS and DR results are not as clear. For LOS by Race within Insurance Type, within Private insurance type, Whites have a statistically longer stay (P = 0.015) than Hispanics. This may reflect the different demographic ages. All others are not statistically different. One would expect Hispanic LOS within this insurance to be lower since generally, the Hispanic demographic is younger in the US, but that may not truly be the case for the privately insured population. For the DR by Race within Private Insurance Type, the DR for Whites is higher than for Blacks (P = 0.0055), but unlike before, it is not different from Hispanic DR. Currently, it is difficult to explain, is the White Private insurance population older than the Black or Hispanic? Why do Whites die at a higher rate given the same insurance cover? These issues all merit further examination.

Discussion

Payment type bias

The Self-Pay (no insurance) category has the shortest statistically significant LOS by far as shown in Table 4. While the differences between Medicare, Private and Government are not significant, Self-Pay is significantly shorter. This is even more damning when looking at the Medicaid group, a similar socio-economic group to Self-pay. The Medicaid group has a significantly longer LOS than other groups, possibly reflecting the poorer general health condition of that category while the similar Self-Pay group is at the other end of the spectrum, a 34% shorter stay than the Medicaid group, indicating that those of similar backgrounds who cannot pay receive much less access to care.

Table 5 shows the clear evidence of disparate DRs (outcomes) based on ability to pay. While nearly all categories are significantly different, of particular interest is again the comparison between the socio-economically similar Medicaid and Self Pay groups, where the Self Pay group’s DR is nearly twice the Medicaid group. The only groups that show no significant difference are the Medicare and Self-Pay groups. This itself is significant because the Medicare group, being generally limited to those 65 years of age and older, would be expected to have a higher DR. This shows a shocking result – that the uninsured are facing a similar DR as the elderly when in hospital. Our analysis shows the same LOS and DR data but is limited to the White ethnic category (to remove racial bias). Tables 8 and 9 show the same pattern of bias in health care access and outcomes with respect to ability to pay, for patient emergency admissions at a single hospital.

Implicit in the current debate on health care insurance in the US is the question: Is health care a right or a commercial commodity? These results are highlighting the social and economic impact on society when healthcare delivery systems are influenced a by patient’s ability to pay for their care (beyond the well understood socio-economic impact factors). While this debate has been long settled in other countries, it is very much current in the US today.

Racial bias

Our analysis shows a more mixed result with Whites having more Medicare (older) patients than Blacks. That could lead to a longer LOS and higher DR for the group. It is also possible that this is evidence of racial bias in access to care once admitted to hospital. However, the higher DR for Whites seems to discount that, on the other hand, Whites show a longer LOS, and similar dynamic for LOS was found when limited to just the Private Insurance group (attempt to eliminate ability to pay bias). We showed a statistically significant difference in a higher DR than Black and Hispanic, which is unexpected. A question for further work would be to look for hospital ‘milking’ of the potentially more lucrative private insurance payments. This is not conclusive and merits further work to try to understand the other interacting variables at work within the race category, whether they are age, income, bias or others.

Footnotes

Declaration of conflicting interests

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The author(s) received no financial support for the research, authorship, and/or publication of this article.

ORCID iD

Mirza M Baig

References

McDavid

Tucker

Sloggett

, et al. Cancer survival in Kentucky and health insurance coverage. Arch Intern Med 2003; 163: 2135–2144.

Roetzheim

Pal

Gonzalez

, et al. Effects of health insurance and race on colorectal cancer treatments and outcomes. Am J Public Health 2000; 90: 1746–1754.

Yergan

Flood

LoGerfo

, et al. Relationship between patient race and the intensity of hospital services. Med Care 1987; 25: 592–603.

National Center for Health Statistics (US). Health, United States, 2011. With special feature on socio-economic status and health, 2012.

Fiscella

Franks

Gold

, et al. Inequality in quality: addressing socioeconomic, racial, and ethnic disparities in health care. JAMA 2000; 283: 2579–2584.

Braveman

Cubbin

Egerter

, et al. Socio-economic status in health research: one size does not fit all. JAMA 2005; 294: 2879–2888.

Berardi

KL.

Emergency medical treatment and active labor act. West N Eng Law Rev 2007; 30: 515.

Carlisle

Leake

BD.

Differences in the effect of patients’ socioeconomic status on the use of invasive cardiovascular procedures across health insurance categories. Am J Public Health 1998; 88: 1089–1092.

Ayanian

Kohler

Abe

, et al. The relation between health insurance coverage and clinical outcomes among women with breast cancer. N Engl J Med 1993; 329: 326–331.

10.

Whittle

Conigliaro

Good

, et al. Racial differences in the use of invasive cardiovascular procedures in the department of veterans affairs medical system. N Engl J Med 1993; 329: 621–662.

11.

Boxer

, MD, Dimick

, MD, Wainess

, BSet al. Payer status is related to differences in access and outcomes of abdominal aortic aneurysm repair in the United States. Surgery 2003; 134: 142–145.

12.

Link

Phelan

Social conditions as fundamental causes of disease. J Health Social Behav 1995 ; 35: 80–94.

13.

Feinstein

JS.

The relationship between socioeconomic status and health: a review of the literature. Milbank Quart 1993; 71: 279–322.

14.

Addo

Ayerbe

Mohan

, et al. Socioeconomic status and stroke: an updated review. Stroke 2012; 43: 1186–1191.

15.

Callander

McDermott

Measuring the effects of CVD interventions and studies across socioeconomic groups: a brief review. Int J Cardiol 2017; 227: 635–643.

16.

Van Hecke

Heinen

Fernández ‐ Ortega

, et al. Systematic literature review on effectiveness of self‐management support interventions in patients with chronic conditions and low socio‐economic status. J Adv Nurs 2017; 73: 775–793.

17.

Barnett

Hixon

Okwiri

, et al. Factors involved in access and utilization of adult hearing healthcare: a systematic review. Laryngoscope 2017; 127: 1187–1194.

18.

Dietrichson

Bøg

Filges

, et al. Academic interventions for elementary and Middle school students with low socioeconomic status: a systematic review and meta-analysis. Rev Educ Res 2017; 87: 243–282.

19.

Attanasio

Hardeman

Kozhimannil

, et al. Prenatal attitudes toward vaginal delivery and actual delivery mode: variation by race/ethnicity and socioeconomic status. Birth 2017; 44: 306–314.

20.

James

SA.

The strangest of all encounters: racial and ethnic discrimination in US health care. Cadernos Saude Publ 2017 ; 33: e00104416.

21.

Braveman

Schaaf

Egerter

, et al. Insurance-related differences in the risk of ruptured appendix. N Engl J Med 1994; 331: 444–449.

22.

Kelz

Gimotty

Polsky

, et al. Morbidity and mortality of colorectal carcinoma surgery differs by insurance status. Cancer 2004; 101: 2187–2194.

23.

Lemaire

Chad

Sean

, et al. The impact of race and insurance type on the outcome of endovascular abdominal aortic aneurysm (AAA) repair. In: Presented at the 2007 vascular annual meeting of the society for vascular surgery, Philadelphia, PA, 6–10 June 2007.

24.

Herring

Wilper

Himmelstein

, et al. Increasing length of stay among adult visits to U.S. Emergency departments, 2001–2005. Acad Emerg Med 2009; 16: 609–616.

25.

Zibulewsky

The emergency medical treatment and active labor act (EMTALA): what it is and what it means for physicians. Proc (Bayl Univ Med Cent) 2001; 14: 339–346.

26.

Johnson

AEW

Pollard

Shen

, et al. MIMIC-III, a freely accessible critical care database. Sci Data 2016; 3:160035.

27.

Goldberger

Amaral

LAN

Glass

, et al. PhysioBank, PhysioToolkit, and PhysioNet: components of a new research resource for complex physiologic signals. Circulation 2000; 101: e215–e220.

28.

Feder

Hadley

Mullner

Falling through the cracks: poverty, insurance coverage, and hospital care for the poor, 1980 and 1982. Milbank Mem Fund Quart Health Soc 1984; 62: 544–566.