Volume of emergency department admissions for sepsis is related

to inpatient mortality: Results of a nationwide cross-sectional

analysis
Emilie S. Powell, MD, MBA; Rahul K. Khare, MD; D. Mark Courtney, MD; Joe Feinglass, PhD

Objectives: Emergency department resuscitation plays a sig-
nificant role in sepsis care, and it is unknown if patient outcomes
vary by institution based on the level of sepsis experience of the
emergency department. This study examines whether there is an
association between the annual volume of patients admitted via
the emergency department with sepsis and inpatient mortality.
Design: Cross-sectional analysis of the 2007 Nationwide Inpa-
tient Sample.
Setting and Patients: We included 87,166 adult emergency
department sepsis admissions from 551 hospitals.
Measurements: Hospitals were categorized into quartiles by 2007
emergency department sepsis volume. Univariate associations of patient
characteristics, hospital characteristics, and inpatient mortality with
sepsis volume level were evaluated by chi-square test. A population-
averaged logistic regression model of inpatient mortality was used to
estimate the effects of age, gender, comorbid conditions, payer status,
median zip code income, hospital bed size, teaching status, and emer-
gency department sepsis volume.
Main Results: Overall inpatient sepsis mortality was 18.0% and
early mortality (2 days after admission) was 6.9%. The risk-adjusted
odds ratios of mortality were 0.73 (95% confidence interval, 0.64 –
0.83; p < .001) in quartile 4 (highest volume), 0.83 in quartile 3 (95%
confidence interval, 0.74 – 0.93; p ⴝ .001), and 0.90 in quartile 2 (95%
confidence interval, 0.82– 0.99; p < .05) when compared to quartile
1 (lowest volume). Adjusted results were similar for early mortality:
0.69 (95% confidence interval, 0.61– 0.76; p < .001) in quartile 4, 0.83
in quartile 3 (95% confidence interval, 0.74 – 0.93; p < .05), and 0.85
in quartile 2 (95% confidence interval, 0.77– 0.94; p < .05) when
compared to quartile 1.
Conclusions: After adjustment for comorbidity and hospital-
level factors, there was a significant relationship between emer-
gency department sepsis case volume and overall and early
inpatient mortality among patients admitted through the emer-
gency department with sepsis. Patients admitted to hospitals in
the highest-volume quartile had 27% lower odds of inpatient
mortality in this large heterogeneous sample. (Crit Care Med 2010;
38:2161–2168)
KEY WORDS: sepsis; outcome assessment; mortality; health ser-
vices research

S epsis is a growing problem in
the United States, with mortality
rates that reach 50%– 60% once
the disease progresses to septic
shock (1–3). Although not all sepsis pa-
tients are admitted through, or initially re-
suscitated in, the emergency department
(ED), care in this setting plays a significant
role in the treatment of sepsis nationwide,
and care delivered in the ED can signifi-
cantly reduce mortality (4 –9). The impor-
From the Department of Emergency Medicine
(ESP, RKK, MC), Feinberg School of Medicine, North-
western University, Chicago, IL; Institute for Healthcare
Studies and Division of General Internal Medicine (ESP,
RKK, JF), Northwestern University, Feinberg School of
Medicine, Chicago, IL.
Dr. Powell and Dr. Khare are supported by National
Research Service Award postdoctoral fellowship grant
through the Institute for Healthcare Studies at North-
western University under institutional awards from
Agency for Healthcare Research and Quality (T-32 HS
000078 and F-32 HS 17876-01). Dr. Courtney was
supported by grant 5K23HL077404-04 from the Na-
tional Heart, Lung, and Blood Institute.
tance of ED sepsis care has resulted in ac-
cepted, validated approaches to immediate
and aggressive resuscitation that have been
adopted in many hospitals in quality im-
provement efforts and as a standard of care
(10 –12). These approaches can be time and
resource intensive, and successful imple-
mentation and patient outcome could be
dependent on the level of experience at
both the clinician and institution levels.
However, little is known about whether
Dr. Khare received a year-long fellowship training
grant from Agency for Healthcare. The remaining au-
thors have not disclosed any potential conflicts of
interest.
Supplemental digital content is available for this ar-
ticle. Direct URL citations appear in the printed text and
are provided in the HTML and .pdf versions of this article
on the journal’s Web site (www.ccmjournal.com).
For information regarding this article,
E-mail: emilie-powell@md.northwestern.edu
Copyright © 2010 by the Society of Critical Care
Medicine and Lippincott Williams & Wilkins
DOI: 10.1097/CCM.0b013e3181f3e09c
EDs with greater experience in sepsis re-
suscitation actually perform better and
have improved patient outcomes.
One method of assessing the potential
contribution of ED experience in sepsis re-
suscitation is to examine if there is a rela-
tionship between annual case volume and
inpatient mortality among patients admit-
ted to the hospital via the ED with sepsis.
The volume– outcome relationship has
been explored extensively in the surgical
literature. Higher case volume, reflecting
experience, has been demonstrated to cor-
relate with superior quality of care (13, 14).
It is possible that higher annual ED sepsis
case volume may reflect greater depart-
ment-level experience in treating sepsis
and superior quality of care. If experience
and case volume are related to superior
quality of care, as has been shown in the
surgical literature, then we would expect to
see lower early inpatient mortality (within
the first 2 days of admission), as well as
lower overall inpatient mortality at higher-
volume institutions.

Crit Care Med 2010 Vol. 38, No. 11 2161

 The primary goal of this investigation
was to use population-based national hos-
pital discharge data to examine whether
there is an association between inpatient
mortality and the annual volume of pa-
tients admitted to the hospital via the ED
with a principal diagnosis of sepsis. We ex-
amined this question before and after con-
trolling for other key patient and hospital
characteristics. We hypothesized that hos-
pitals with higher ED sepsis case volume
would have superior outcomes as defined
by lower early inpatient (within 2 days pos-
tadmission) and overall inpatient mortality.
METHODS
Study Design
We performed a cross-sectional analysis of
the 2007 Healthcare Cost and Utilization Project
Nationwide Inpatient Sample (NIS). The NIS is
the largest publicly available all-payer inpatient
database in the United States and is provided by
the U.S. Agency for Healthcare Research and
Quality. Data are weighted to result in a sample
that is representative of all admissions to non-
federal U.S. hospitals on an annual basis. Addi-
tional details of the NIS can be found on the
Healthcare Cost and Utilization Project Web site
(http://www.hcup-us.ahrq.gov/). The Northwest-
ern University Institutional Review Board found
this study of deidentified data exempt.
Sample Selection
The 2007 NIS contains 8,043,415 discharge
records from 1,044 hospitals in 40 states. Dis-
charge data include International Classification
of Diseases ninth revision coded diagnoses. Ad-
missions for sepsis were identified based on the
principal diagnosis International Classification
of Diseases ninth revision codes previously de-
Figure 1. Study population sample selection. NIS, nationwide inpatient sample; ED, emergency
department.
2162
fined by Martin et al (3) to detect patients with
sepsis from hospital discharge data: 038 (septi-
cemia), 020.2 (septicemic plague), 790.7 (bacte-
remia), 117.9 (disseminated fungal infection),
112.5 (disseminated Candida infection), or
112.81 (disseminated fungal endocarditis).
Of 134,824 NIS adult sepsis admissions in
2007, 42,212 (31.3%) were admitted through a
source other than the ED and were therefore
excluded from this analysis. To ensure that
hospitals with extremely low frequencies of
sepsis admissions did not bias the analysis, we
excluded admissions from hospitals with ⬍25
sepsis admissions in 2007. This applied to
2,468 admissions (2.7% of the ED sepsis pa-
tient sample) from 287 hospitals. This exclu-
sion allowed comparison of the remaining in-
stitutions to more likely reflect process rather
than purely structural quality of care. Finally,
patients transferred to another acute care hos-
pital after admission through the ED (3.5% of
all ED admissions) were also excluded because
their survival could not be attributed to the
admitting hospital (Figure 1).
Data Synthesis: Patient and
Hospital Variables
NIS data included admission source, dis-
charge status (alive or dead), gender, age (di-
vided into quintiles), and race/ethnicity (black,
Hispanic, white, other, or missing). Race and
ethnicity data were missing for 28.1% of study
patients from states where this information was
not mandated; therefore, this demographic in-
formation was not included in the multivariate
analysis. NIS data also included length of stay,
primary payer source, and patients’ zip code me-
dian incomes (divided into quartiles). As an ad-
ditional proxy for lower socioeconomic status in
the analyses, because individual income level is
not available from the NIS, we coded patients
with a primary payer status of Medicaid (a fed-
erally and state-funded program providing med-
ical care for those with low incomes) or self-pay
(uninsured) into a single category.
To account for confounding attributable to
comorbid conditions present on admission that
may have influenced mortality, we used the Elix-
hauser International Classification of Diseases
ninth revision secondary diagnosis codes to
identify conditions likely to have been present on
admission. These conditions often have been
used in mortality models to adjust for the sever-
ity of preexisting illness (15). Patients were clas-
sified as having or not having each of 30 Elix-
hauser chronic comorbid conditions. We first
tested the univariate association of each Elix-
hauser comorbid condition with inpatient mor-
tality and included conditions found to have
significant associations in the final multivariate
models of inpatient mortality.
Hospital-level data available in the NIS in-
cluded bed size (categorized by NIS as small,
medium, or large; based on the number of
short-term acute beds in the hospital and spe-
cific to the hospital region, location, and
teaching status), teaching status (member of
the Council of Teaching Hospitals or not), and
location (rural vs. urban). To assess whether
ED sepsis case volume was an independent
predictor of mortality, these hospital charac-
teristics were also included in multivariate
models of inpatient mortality.
ED Sepsis Case Volume
The primary predictor variable of interest,
hospital ED sepsis case volume, was defined as
the number of admissions originating in the ED
to a given hospital with a principal diagnosis of
sepsis in 2007. In addition to testing the signif-
icance of annual ED sepsis volume as a contin-
uous measure, ED sepsis admissions to each
hospital were empirically divided equally into
four volume-level quartiles to improve interpre-
tation of multivariate model odds ratios (ORs).
Key Outcome Measures
The key outcome measures in this study
were overall inpatient mortality (death at any
time during hospital admission) and early inpa-
tient mortality (death within the first 2 days of
admission). We chose to evaluate early mortality
in addition to overall inpatient mortality to more
directly examine the impact of ED sepsis care
while minimizing possible delayed confounding
events that may affect mortality independent of
ED or early resuscitation events (e.g., ventilator-
associated pneumonia, establishment of do-not-
resuscitate status, procedure complications).
Evaluating early inpatient mortality at a fixed
interval from admission also avoids the potential
pitfall of higher overall inpatient mortality re-
lated to longer lengths of stay.
Crit Care Med 2010 Vol. 38, No. 11
Statistical Analysis
Analyses were conducted with SPSS version
17.0 (SPSS, Chicago, IL) and STATA version
10.0 (STATA, College Station, TX). Patient and
hospital characteristic data were evaluated
across sepsis case volume quartiles with chi-
square tests for categorical variables and analysis
of variance for continuous measures. The unad-
justed univariate association of ED sepsis case
volume with mortality was evaluated continu-
ously with a t test and by quartiles with the
chi-square test. A population-averaged logistic
regression model with an exchangeable working
correlation matrix was used to analyze the si-
multaneous effects of age, gender, comorbid
conditions, payer status, hospital size, location,
and ED sepsis case volume (measured either
continuously or as quartile levels) on mortality.
This approach accounts for nesting of admis-
sions within hospitals (16). Forward censoring
was not used in developing this regression
model. National population estimates were made
using the discharge-level weights provided with
the NIS. Results are reported as ORs and 95%
confidence intervals (95% CIs). Significant asso-
ciations were interpreted as p ⬍ .05. We used the
discharge-level weights and the postestimation ad-
justment command in STATA to simulate the es-
timated number of lives saved had all hospitals
performed at the level of the highest ED volume
quartile. The resulting difference between pre-
dicted and actual 2007 mortality was used to esti-
mate potential lives saved across all U.S. nonfederal
hospital ED admissions for sepsis in 2007.
RESULTS
ED Sepsis Patient Sample
The final study population included
87,166 patients from 551 hospitals,
with an overall inpatient mortality rate
of 18.0% and an early inpatient mor-
Figure 2. Inpatient sepsis mortality. Distribution of hospital inpatient mortality rates for patients
admitted through the emergency department with a principal diagnosis of sepsis.
Crit Care Med 2010 Vol. 38, No. 11
tality of 6.9%. This represents approxi-
mately 428,000 estimated admissions to
all U.S. nonfederal hospitals and ap-
proximately 77,000 deaths nationwide.
The median annual ED sepsis case vol-
ume per hospital was 249 sepsis admis-
sions (range, 25–1251). The four annual
hospital volume quartiles ranged from 25
to 145, from 146 to 248, from 249 to 371,
and ⬎371 admissions per year. Figure 2
displays the distribution of hospital inpa-
tient mortality rates for the 551 hospitals
in our sample.
On preliminary analysis, 96% of the
patients in our study population had a
principal diagnosis of septicemia (038).
We found that 13 of the 30 Elixhauser
comorbid conditions were significantly
associated with higher mortality (p ⬍
.05). These conditions included alcohol
abuse, congestive heart failure, chronic
pulmonary disease, coagulopathy, cancer
(combination of patients with lymphoma,
solid tumor without metastasis, or meta-
static cancer), fluid and electrolyte disor-
ders, liver disease, peripheral vascular
disorders, pulmonary circulation disor-
ders, renal failure, and weight loss (see
Supplemental Table 1 [Supplemental
Digital Content 1, http://links.lww.com/
CCM/A167]). We included indicators for
these 11 (after the three oncologic co-
morbidities were combined) conditions
in the final analysis.
Patient and Hospital
Characteristics Across the ED
Sepsis Case Volume Quartiles
Table 1 displays patient and hospital
characteristic data for the 2007 ED sepsis
patient sample, including differences be-
tween patients in each of the sepsis case
volume quartiles. All admissions had a
mean patient age of 69.9 yrs (95% CI,
69.8 –70.1). There was a slight trend to-
ward older age in the highest-volume
quartile hospitals as compared to the low-
est-volume quartile (p ⬍ .001). Just more
than half (52.9%) of the sample was fe-
male. For states with available data on
race and ethnicity (71.8% of the sample
population), we found a heterogeneous
distribution of race and ethnicity, with
48.1% of the patient sample being black,
Hispanic, or of other nonwhite races or
ethnicities. Patients admitted through
the higher-volume EDs had a relatively
greater number of coded comorbidities
(p ⬍ .05). Patients admitted through EDs
in the highest quartile had a greater pro-
portion of patients with chronic heart
failure, cancer, coagulopathy, fluid and
electrolyte disorders, pulmonary circula-
tion disorders, peripheral vascular dis-
ease, and renal failure as compared with
patients admitted through EDs in the
lowest sepsis case volume quartile (p ⬍
.05). Admissions to the lowest sepsis case
volume hospitals had a higher proportion
of the lowest income quartile patients
(p ⬍ .05). Each of the sepsis case volume
quartiles had a similar percentage of pa-
tients with Medicaid or without insur-
ance as compared with patients with an-
other form of insurance, including
patients age 65 or older insured by Medi-
care (p ⫽ .53).
The sample was primarily from urban
hospitals (88.2%); 36.7% of admissions
were from small or medium hospitals.
More than half of ED admissions for sep-
sis were to nonteaching hospitals
(58.5%). Hospitals in the highest-volume
quartile were nearly all located in urban
areas (98.2%). Only 70.6% of the lowest-
volume quartile hospitals were urban.
The highest-volume quartile also in-
cluded a significantly higher percentage
of hospitals with large bed size (80.1% vs.
39.3% in the lowest quartile; p ⬍ .05).
Last, the highest-volume quartile hospi-
tals were also more likely to be teaching
hospitals (60.8% vs. 23.1% in the lowest
quartile; p ⬍ .05).
Univariate Association of ED
Sepsis Case Volume and
Mortality
There was a significant inverse asso-
ciation between ED sepsis case volume
and both early and overall inpatient
2163
Table 1. Unadjusted analyses

ED Sepsis Case Volume (Quartiles)

Quartile 1: 25–145 Quartile 2: 146–248 Quartile 3: 249–371 Quartile 4: ⬎371

Sample % Cases/yr Cases/yr Cases/yr Cases/yr
(n ⫽ 87,166) (n ⫽ 21,566) (n ⫽ 21,774) (n ⫽ 21,695) (n ⫽ 22,131) p

Patient characteristics
Age (yr) ⬍.001
18–50 13.7 13.5 14.2 13.2 14.0
51–65 21.0 21.3 21.4 21.3 20.2
66–74 17.8 18.0 18.5 17.8 17.0
75–84 27.3 27.4 26.4 27.3 28.2
⬎84 20.1 19.9 19.4 20.4 20.6
Female 52.9 52.9 53.2 52.7 52.9 .74
Race/ethnicitya ⬍.001
White 51.9 54.8 51.6 54.5 46.6
Black 10.1 7.9 10.5 9.7 12.0
Hispanic 7.3 4.1 8.6 9.5 7.0
Other 2.6 1.8 1.8 3.6 3.2
Mean comorbid conditions (SD) 3.5 (⫾1.86) 3.3 (⫾1.74) 3.5 (⫾1.83) 3.6 (⫾1.89) 3.7 (⫾1.94) ⬍.001
Selected comorbid conditions
Alcohol abuse 3.4 3.3 3.6 3.4 3.1 .05
Cancerb 10.7 9.7 11.3 10.8 11.0 ⬍.001
Chronic heart failure 25.7 24.6 25.2 26.4 26.6 ⬍.001
Chronic pulmonary disease 25.8 25.8 26.0 25.8 25.4 .54
Coagulopathy 11.3 10.0 11.8 12.1 11.2 ⬍.001
Fluid and electrolyte disorder 57.0 55.0 56.3 59.4 57.4 ⬍.001
Liver failure 4.6 4.1 5.2 4.7 4.3 ⬍.001
Peripheral vascular disease 7.1 6.4 6.5 7.3 8.3 ⬍.001
Pulmonary circulation disorder 3.1 2.6 3.1 3.2 3.3 ⬍.001
Renal failure 23.9 21.6 24.5 25.3 24.3 ⬍.001
Weight loss 11.7 9.8 10.6 13.2 13.1 ⬍.001
Zip code median income quartilea ⬍.001
1 (lowest) 28.4 34.4 29.2 26.7 23.6
2 27.1 27.4 27.6 25.8 27.5
3 24.0 21.2 22.7 24.9 27.0
4 (highest) 20.5 16.9 20.4 22.7 21.9
Primary insurance .53
Medicaid/self-pay 11.5 11.3 11.7 11.5 11.4
Other insurance 88.5 88.7 88.3 88.5 88.6
Hospital characteristics
Location ⬍.001
Rural 11.8 29.4 13.7 2.6 1.8
Urban 88.2 70.6 86.3 97.4 98.2
Bed size ⬍.001
Small 11.5 20.9 10.4 5.2 9.5
Medium 25.2 39.8 24.5 26.5 10.4
Large 63.3 39.3 65.1 68.2 80.1
Teaching status ⬍.001
Teaching 41.5 23.1 36.1 45.3 60.8
Nonteaching 58.5 76.9 63.9 54.7 39.2
Inpatient mortality
1-day early mortality 4.9 5.6 5.1 5.0 4.1 ⬍.001
2-day early mortality 6.9 7.8 7.1 7.1 5.8 ⬍.001
3-day early mortality 8.3 9.2 8.6 8.5 7.0 ⬍.001
Overall mortality 18.0 18.2 18.9 18.7 16.4 ⬍.001

Unadjusted univariate associations of patient and hospital characteristics of patients admitted through the emergency department with a principal
diagnosis of sepsis across sepsis case volume quartiles. Unadjusted univariate associations of inpatient mortality across the emergency department sepsis
case volume quartiles.
a
All variables had ⬍1% missing data with the exception of race and ethnicity (28.1% missing) and median household income by zip code quartile (2.5%
missing); bcancer is a combination of three Elixhauser comorbid conditions: lymphoma, metastatic cancer, or solid tumor without metastasis.

mortality when ED sepsis case volume
was evaluated continuously (p ⬍ .001;
data not shown). Table 1 presents 1 day,
2 days (early inpatient mortality), and 3
days after admission, as well as overall
inpatient mortality rates by volume
quartiles. Early inpatient mortality was
significantly lower across ED sepsis vol-
ume quartiles; case volume increased
for up to 3 days after admission. How-
ever, when examining overall inpatient
mortality, we found that although the
highest ED sepsis volume quartile
(quartile 4) had significantly lower
mortality as compared to the lowest-
volume quartile (quartile 1; p ⫽ .03),
admissions in quartiles 2 and 3 with the
highest inpatient mortality rates had
higher overall inpatient mortality than
patients in both quartile 1 and quartile
4 (both comparisons: p ⬍ .001).

2164 Crit Care Med 2010 Vol. 38, No. 11

Table 2. Multivariate logistic regression Multivariate Analyses: ED
Sepsis Case Volume and
Early Inpatient Mortality Overall Inpatient Mortality
Odds Ratio Odds Ratio Mortality
(95% Confidence Interval) (95% Confidence Interval)
There was a highly significant associ-
Patient characteristics ation between both early and overall in-
Age (yr) patient mortality and ED sepsis case vol-
18–50 Reference Reference ume when included in the model as a
51–65 1.41 (1.26–1.58)a 1.41 (1.31–1.51)a continuous variable (p ⬍ .001; data not
66–74 1.75 (1.55–1.98)a 1.71 (1.59–1.85)a
75–84 2.36 (2.10–2.64)a 2.21 (2.05–2.37)a
shown). Patients admitted via the ED to
⬎84 3.30 (2.93–3.70)a 2.90 (2.68–3.13)a hospitals in the highest annual ED sepsis
Female 1.05 (0.99–1.11) 0.99 (0.96–1.03) case volume quartile had 27% lower odds
Selected comorbid conditions of inpatient death (OR, 0.73; 95% CI,
Alcohol abuse 1.42 (1.23–1.64)a 1.42 (1.29–1.56)a 0.64 – 0.83; p ⬍ .001) and 31% lower odds
Cancer 1.74 (1.61–1.87)a 1.99 (1.89–2.10)a
Chronic heart failure 0.94 (0.88–1.00)b 1.38 (1.33–1.44)a of early inpatient death (OR, 0.69; 95%
Chronic pulmonary disease 0.85 (0.79–0.90)a 0.99 (0.95–1.03) CI, 0.61– 0.80; p ⬍ .001) when compared
Coagulopathy 1.20 (1.11–1.30)a 1.92 (1.82–2.01)a to patients admitted to hospitals in the
Fluid and electrolyte disorder 1.20 (1.14–1.27)a 1.39 (1.34–1.44)a lowest ED sepsis case volume quartile.
Liver failure 1.78 (1.58–2.00)a 1.88 (1.73–2.03)a
Peripheral vascular disease 1.31 (1.19–1.44)a 1.19 (1.11–1.27)a
Unlike the unadjusted results presented
Pulmonary circulation disorder 0.97 (0.83–1.14) 1.26 (1.15–1.39)a in Table 1, the multivariate model results
Renal failure 0.97 (0.91–1.03) 1.19 (1.14–1.24)a indicated that patients admitted to the
Weight loss 0.48 (0.43–0.54)a 1.20 (1.14–1.27)a second highest ED sepsis case volume
Zip code median income quartile quartile (quartile 3) also had 17% lower
Quartile 1 (lowest) Reference Reference
Quartile 2 0.96 (0.89–1.04) 0.98 (0.93–1.03) odds of inpatient mortality (OR, 0.83;
Quartile 3 0.96 (0.89–1.04) 0.97 (0.92–1.03) 95% CI, 0.74 – 0.93; p ⫽ .001) and 17%
Quartile 4 (highest) 1.00 (0.92–1.09) 0.96 (0.90–1.02) lower odds of early mortality (OR, 0.83;
Medicaid/self–pay 1.21 (1.09–1.34)a 1.09 (1.02–1.16)b 95% CI, 0.74 – 0.93; p ⬍ .05) when com-
Hospital characteristics
Rural location 0.91 (0.81–1.02) 0.75 (0.67–0.84)a
pared to patients admitted to the lowest
Bed size quartile hospitals. The ED volume–
Small Reference Reference outcome relationship continued to hold
Medium 1.00 (0.89–1.14) 1.09 (0.97–1.22) for early mortality and overall inpatient
Large 1.07 (0.94–1.20) 1.20 (1.07–1.35) b mortality in quartile 2 (OR, 0.85; 95% CI,
Teaching hospital 1.05 (0.97–1.14) 1.21 (1.12–1.31)a
Emergency department sepsis 0.77– 0.95; p ⬍ .05 and OR, 0.90; 95% CI,
case volume/yr (quartile) 0.82– 0.99; p ⬍ .05, respectively). Ad-
25–145 (1) Reference Reference justed mortality results based on the
146–248 (2) 0.85 (0.77–0.95)b 0.90 (0.82–0.99)b multivariate model translate to an esti-
249–371 (3) 0.83 (0.74–0.93)b 0.83 (0.74–0.93)b mated overall inpatient mortality of
⬎371 (4) 0.69 (0.61–0.80)a 0.73 (0.64–0.83)a
17.2% and estimated early inpatient mor-
Multivariate logistic regression (adjusted) analyses of the likelihood of overall and early (within 2 tality of 6.4%.
days of hospital admission) inpatient mortality in 87,166 admissions from the emergency department The model of overall inpatient mortal-
with a principal diagnosis of sepsis to 551 hospitals in 2007. ity had a C-statistic of 0.64 (95% CI,
a
p ⬍ .001; bp ⬍ .05. 0.63– 0.64), whereas the model of early
mortality had a C-statistic of 0.61 (95%
CI, 0.61– 0.62). Results were virtually
Multivariate Analyses: Patient Medicaid or self-pay patients (OR, 1.09; identical when the analyses were per-
and Hospital Factors and 95% CI, 1.02–1.16; p ⬍ .05 and OR, formed including hospitals with ⬍25 ad-
1.21; 95% CI, 1.09 –1.34; p ⬍ .001, re- missions included (data not shown).
Mortality
spectively).
Many hospital and patient factors Patients admitted to teaching hospi- DISCUSSION
were significantly associated with mor- tals had 21% greater odds of inpatient
tality when simultaneously controlling This is the first work to our knowledge
death when compared to nonteaching
for other risk factors (Table 2). The risk to demonstrate a relationship between
hospitals (OR, 1.21; 95% CI, 1.12–1.31;
of inpatient mortality and early inpa- ED sepsis volume and hospital mortality
tient mortality increased with age. Most p ⬍ .001) and patients admitted to larger and is one of the few analyses of the
of the comorbid conditions included in hospitals had a similar 20% increased association between ED case volume and
the adjusted model were associated odds of inpatient mortality (OR, 1.20; outcome (17, 18). We first demonstrate
with an increased risk of mortality. Af- 95% CI, 1.07–1.35; p ⬍ .05). These rela- interhospital variability in mortality
ter adjustment for other factors, there tionships held for early mortality, but the among patients admitted to the hospital
was no significant association between differences were smaller and nonsignifi- via the ED with sepsis. Even after exclud-
zip code median income level and mor- cant (OR, 1.05; 95% CI, 0.97–1.14; p ⫽ ing extreme outlier hospital EDs with ex-
tality, but there was an increased odds .238 and OR, 1.07; 95% CI, 0.94 –1.20; tremely low sepsis volumes of ⬍25 cases
of both overall and early mortality for p ⫽ .302, respectively). per year, our results demonstrate a wide

Crit Care Med 2010 Vol. 38, No. 11 2165

variation in mortality rates across a nor-
mal distribution (Fig. 2). This variation in
outcome alone could support that there
are likely significant quality-of-care dif-
ferences in sepsis treatment across hos-
pitals and EDs in this nationwide sample.
However, our findings further demon-
strate a significant association between
an ED-specific factor, ED sepsis case vol-
ume, and both overall and early inpatient
mortality after adjustment for comorbid
conditions and hospital factors such as
bed size and teaching status. Patients ad-
mitted through hospital EDs in the high-
est sepsis volume quartile had 27% lower
risk-adjusted odds of inpatient mortality.
These findings were robust with respect
to early inpatient mortality: there was a
31% lower risk-adjusted odds of early in-
patient mortality. Early mortality would
presumably be more closely related to the
ED resuscitation of a septic patient. This
early mortality reduction gives even
greater support to the potential positive
effects of ED experience and quality of
care. We believe that these differences in
mortality across institutions, after adjust-
ment for comorbid conditions and hospi-
tal factors, likely reflect improved ED
quality of care in higher sepsis volume
EDs with more experience in sepsis re-
suscitation. This higher quality of care
can lead to potentially important conse-
quences for patient survival.
Differences Between the ED
Sepsis Volume Quartiles
Although the four sepsis case volume
quartiles are quite clinically similar with
respect to patients’ age distribution and
gender, patients in the highest sepsis case
volume quartile have a significantly
greater number of comorbid conditions
recorded when compared to the lowest
sepsis volume quartile. There are also sig-
nificant differences between patients seen
at hospitals in the highest vs. the lowest
quartiles with respect to the incidence of
specific comorbid conditions, such as
cancer, chronic heart failure, peripheral
vascular disease, and renal failure. We
hypothesize that other significant differ-
ences, such as fluid and electrolyte disor-
ders and coagulopathy, indicate higher
sepsis severity at higher-volume centers
(Table 1).
This could occur if secondary diagno-
sis codes were assigned in relation to the
primary sepsis disease process. Although
patients in the lowest sepsis case volume
quartile have lower unadjusted overall
2166
mortality as compared with patients in
the two middle quartiles, this difference
disappears in the fully adjusted analysis
after adjustment for these comorbid con-
ditions and other patient and hospital fac-
tors.
Large Hospitals vs. High-
Volume Sepsis Centers
The adjusted results demonstrate that
although patients admitted through hos-
pitals with a high volume of ED sepsis
admissions have lower mortality, patients
admitted through hospitals with large
bed size have significantly higher overall
mortality. There were 38 hospitals in the
highest ED sepsis volume quartile: 37 of
these are unsurprisingly also classified by
the NIS as large-bed-size hospitals. There
are 275 total large-bed-size hospitals, and
a significant proportion of large-bed-size
hospitals have EDs in the second and
third sepsis volume quartiles (43.9% of
large-bed-size hospitals). The overall un-
adjusted mortality rates at large-bed-size
hospitals in ED volume quartiles 2 and 3
were 19.3% and 19.0%, respectively, as
compared to 16.9% at the 37 large-bed-
size hospitals with EDs in the highest ED
sepsis volume quartile.
Volume–Outcome Relationship
and Quality of Care
The relationship between volume and
outcome has been explored extensively in
other disciplines, particularly elective
surgery (13). There are two major hy-
potheses that have been presented to ex-
plain the volume– outcome relationship
and its relationship to quality of care:
selective referral patterns (patients dis-
proportionately seek care at, and physi-
cians refer to, hospitals known for high
quality) and the practice-makes-perfect
theory (experience through higher vol-
ume improves quality of care) (19). In
high-acuity emergent conditions, such as
sepsis, the practice-makes-perfect ratio-
nale is much more likely to be the dom-
inant reason for better quality of care
(19). Septic patients are acutely and se-
verely ill and often do not have the luxury
of choosing to which ED they present.
However, in our sample population, high
sepsis volume centers were more likely to
be associated with academic institutions
(60.8% in highest-volume quartile vs.
23.1% in the lowest-volume quartile; p ⬍
.001), the volume– outcome relationship
remained significant after adjustment for
hospital teaching status. Thus, although
selective referral patterns could play a
small role, it is likely that greater experi-
ence in sepsis resuscitation confers a
quality-of-care advantage.
Using the NIS nationally representa-
tive discharge weights and extrapolating
to all U.S. hospital ED admissions for
sepsis, if all hospitals were to perform at
the level of the highest sepsis volume
quartile hospitals, then we would achieve
a reduction in the baseline adjusted mor-
tality rate from 17.2% to 15.0%, and this
would result in approximately 9,400 lives
saved annually.
This study provides support for the
potential role of studying the operations
of high-volume institutions to discover
sepsis resuscitation delivery best prac-
tices, ultimately to be disseminated to all
types of EDs, regardless of experience and
volume. The solution may lie in targeted
efforts toward EDs with low sepsis vol-
umes to increase education of care pro-
viders with respect to early recognition,
diagnosis, resuscitation, and technical as-
pects of sepsis care. Implementation of
sepsis best practices is likely to be differ-
ent and require alterations in smaller
low-volume institutions as compared to
higher-volume institutions. Future
work should examine the success
achieved at higher-volume institutions
to identify best operational practices in
sepsis care delivery and then translate
and adapt these best practices to lower-
volume institutions through methodol-
ogies in translational and dissemina-
tion research (20).
Limitations
There are several limitations to this
study, many of which are inherent to
secondary data analysis of population-
based hospital discharge datasets. The
principal diagnosis is widely accepted as
the diagnosis chiefly responsible for ad-
mission to the hospital (21). We were
unable, in this deidentified dataset, to go
back to source medical records to con-
firm this for each case. In an effort to
increase the specificity of our patient co-
hort search strategy, we excluded all pa-
tients with a secondary diagnosis of sepsis
and therefore likely missed some patients
admitted through the ED with sepsis but
were admitted for other significant con-
ditions. Ideally, we would have performed
further analyses on patients with varying
severity of illness, i.e., sepsis, severe sep-
sis, and septic shock. However, we were
Crit Care Med 2010 Vol. 38, No. 11
also unable to define the population of
patients admitted via the ED with severe
sepsis because the diagnosis of organ fail-
ure as a secondary diagnosis could have
been based on a subsequent in-hospital
complication that might not have been
present at ED evaluation. For this reason,
the reported mortality rate of 18.0% for
sepsis admissions is lower than that com-
monly reported for patients with severe
sepsis or septic shock (1, 2).
It is also important to recognize that
hospitals may have varying tendencies to
diagnose sepsis at different stages of the
disease process. For example, hospitals
with a greater number of sepsis patients
may be more likely to diagnose and treat
sepsis early in the disease process and
therefore maintain patients at a lower
severity of illness overall and subse-
quently have lower mortality rates. Lower
sepsis volume centers could be more con-
servative in their approach to diagnosing
sepsis and therefore care for a population
with higher severity of illness. Further
research is needed to better understand
this potential confounder. Alternatively,
as discussed, we hypothesize that patients
in the highest sepsis volume quartile
might have a higher severity of illness. If
we were to have limited our analysis to
only patients with severe sepsis or septic
shock, then we hypothesize that this
would reduce this potential confounder
and we would see a more pronounced
relationship between ED sepsis volume
and mortality.
Use of administrative hospital data
also guided the information that we are
able to analyze and study. We are limited
to the comorbid conditions and severity
of illness classification methods provided
by International Classification of Diseases
ninth revision codes. We hypothesize that
some of the differences in association be-
tween certain comorbid conditions and
early mortality vs. overall mortality are
secondary to coding bias. For example,
weight loss is protective in the adjusted
model of early inpatient mortality but
hazardous in the model of overall mor-
tality. We hypothesize that this is because
of more intensive coding of acute com-
plications for patients with early hospital
death, as opposed to coding of more
chronic conditions for patients who re-
mained in the hospital longer and then
died. We did not have access to more
refined severity measures such as detailed
admission laboratory or imaging data to
confirm the presence or absence of severe
sepsis on admission to the hospital.
Crit Care Med 2010 Vol. 38, No. 11
The logistic regression model discrim-
ination is relatively low, as indicated by
the C-statistics. This is a function in part
of the high sepsis mortality rate that pre-
cludes high accuracy for a model predict-
ing that all patients survive. Models of
lower-death-rate conditions (⬍5%) gen-
erally achieve C-statistics between 0.7
and 0.8. Model C-statistics are also lower
because we lacked access to detailed clin-
ical information to better adjust for the
severity of illness. The inclusion of more
detailed information on admission sever-
ity of illness, such as Acute Physiology
and Chronic Health Evaluation scores,
would add to the calibration and discrim-
ination of the mortality models that nev-
ertheless will have large remaining un-
measured elements of chance.
Finally, the results of this study are lim-
ited to demonstrating that there was an
association between ED sepsis case volume
and mortality and a variation in sepsis in-
patient mortality and early mortality across
hospitals based on annual ED sepsis case
volume. We are unable to comment with
certainty on why EDs with higher sepsis
case volumes performed better than other
hospital EDs, but we postulate that it is less
likely a patient selection effect and more
likely to be related to better sepsis resusci-
tation experience and quality of care. Fu-
ture research should examine processes-of-
care differences between EDs with varying
experience levels to determine whether
specific factors in sepsis resuscitation pro-
cess are associated with improved out-
comes.
CONCLUSION
The results of this model demonstrate
a wide variation in mortality rates among
patients admitted to hospitals across the
United States via the ED with sepsis and
support a significant association between
ED sepsis experience and mortality. After
adjustment for comorbid conditions and
hospital-level factors, there was a signif-
icant relationship between ED sepsis case
volume and individual overall and early
inpatient mortality in patients admitted
through the ED with sepsis. When com-
pared with hospitals with the lowest an-
nual ED volume of sepsis cases, patients
cared for in the highest quartile hospitals
had 27% lower odds of overall inpatient
mortality in this large heterogeneous
sample. Efforts to optimize sepsis resus-
citation experience and translate sepsis
resuscitation best practices to lower-
volume EDs should be considered and
may result in a significant mortality re-
duction nationwide.
REFERENCES
1. Angus DC, Linde-Zwirble WT, Lidicker J, et
al: Epidemiology of severe sepsis in the
United States: Analysis of incidence, out-
come, and associated costs of care. Crit Care
Med 2001; 29:1303–1310
2. Dombrovskiy VY, Martin AA, Sunderram J, et
al: Rapid increase in hospitalization and
mortality rates for severe sepsis in the United
States: A trend analysis from 1993 to 2003.
Crit Care Med 2007; 35:1244 –1250
3. Martin GS, Mannino DM, Eaton S, et al: The
epidemiology of sepsis in the United States
from 1979 through 2000. N Engl J Med 2003;
348:1546 –1554
4. Gao F, Melody T, Daniels DF, et al: The
impact of compliance with 6-hour and 24-
hour sepsis bundles on hospital mortality in
patients with severe sepsis: A prospective ob-
servational study. Crit Care 2005;
9:R764 –R770
5. Jones AE, Focht A, Horton JM, et al: Prospec-
tive external validation of the clinical effec-
tiveness of an emergency department-based
early goal-directed therapy protocol for se-
vere sepsis and septic shock. Chest 2007;
132:425– 432
6. Nguyen HB, Corbett SW, Steele R, et al:
Implementation of a bundle of quality indi-
cators for the early management of severe
sepsis and septic shock is associated with
decreased mortality. Crit Care Med 2007; 35:
1105–1112
7. Puskarich MA, Marchick MR, Kline JA, et al:
One year mortality of patients treated with
an emergency department based early goal
directed therapy protocol for severe sepsis
and septic shock: A before and after study.
Crit Care 2009; 13:R167
8. Sands KE, Bates DW, Lanken PN, et al: Epi-
demiology of sepsis syndrome in 8 academic
medical centers. JAMA 1997; 278:234 –240
9. Wang HE, Shapiro NI, Angus DC, et al: Na-
tional estimates of severe sepsis in United
States emergency departments. Crit Care
Med 2007; 35:1928 –1936
10. Dellinger RP, Levy MM, Carlet JM, et al:
Surviving Sepsis Campaign: international
guidelines for management of severe sepsis
and septic shock: 2008. Crit Care Med 2008;
36:296 –327
11. The Surviving Sepsis Campaign and Institute
for Healthcare Improvement Sepsis Bundles:
1009 [cited October 25, 2009]. Available at:
http://www.ihi.org/IHI/topics/CriticalCare/
Sepsis. Accessed August 26, 2010
12. Townsend SR, Schorr C, Levy MM, et al:
Reducing mortality in severe sepsis: The Sur-
viving Sepsis Campaign. Clin Chest Med
2008; 29:721–733, x
13. Birkmeyer JD, Siewers AE, Finlayson EV, et
al: Hospital volume and surgical mortality in
2167
 the United States. N Engl J Med 2002; 346:
1128 –1137
14. Elixhauser A, Steiner C, Fraser I: Volume
thresholds and hospital characteristics in the
United States. Health Aff (Millwood) 2003;
22:167–177
15. Elixhauser A, Steiner C, Harris DR, et al:
Comorbidity measures for use with adm-
inistrative data. Med Care; 1998:
8 –27
16. Zeger SL, Liang KY: Longitudinal data anal-
2168
ysis for discrete and continuous outcomes.
Biometrics 1986; 42:121–130
17. Chase M, Hollander JE: Volume and out-
come: The more patients the better? Ann
Emerg Med 2006; 48:657– 659
18. Schull MJ, Vermeulen MJ, Stukel TA: The
risk of missed diagnosis of acute myocardial
infarction associated with emergency depart-
ment volume. Ann Emerg Med 2006; 48:
647– 655
19. Luft HS, Hunt SS, Maerki SC: The volume-
outcome relationship: Practice-makes-
perfect or selective-referral patterns? Health
Serv Res 1987; 22:26
20. Westfall JM, Mold J, Fagnan L: Practice-
based research–“Blue Highways” on the NIH
roadmap. JAMA 2007; 297:403– 406
21. International Classification of Diseases-9-CM
Official Guidelines for Coding and Reporting
2005. [cited December 1, 2009]. Available at:
http://www.cdc.gov/nchs/data/icd9/icdguide09.
pdf. Accessed August 26, 2010
Crit Care Med 2010 Vol. 38, No. 11