MR1.VOL1:Nejm:NEJM:Layout:front_030417:4mart.oa
Fri Mar 21 03 12:13:16
original article
The
new england journal
of
medicine
n engl j med
348;16
www.nejm.org april
17, 2003
1546
The Epidemiology of Sepsis in the United States
from 1979 through 2000
Greg S. Martin, M.D., David M. Mannino, M.D., Stephanie Eaton, M.D.,
and Marc Moss, M.D.
From the Division of Pulmonary, Allergy,
and Critical Care, Department of Medi-
cine, Emory University School of Medi-
cine (G.S.M., S.E., M.M.); and the National
Center for Environmental Health, Centers
for Disease Control and Prevention (D.M.M.)
— both in Atlanta. Address reprint re-
quests to Dr. Martin at the Division of Pul-
monary, Allergy, and Critical Care, Depart-
ment of Medicine, Emory University, Grady
Memorial Hospital, 69 Jesse Hill Jr. Dr.,
SE, Rm. 2D-004, Atlanta, GA 30303, or at
greg_ martin@ emory.org.
N Engl J Med 2003;348:1546-54.
Copyright © 2003 Massachusetts Medical Society.
background
Sepsis represents a substantial health care burden, and there is limited epidemiologic
information about the demography of sepsis or about the temporal changes in its inci-
dence and outcome. We investigated the epidemiology of sepsis in the United States,
with specific examination of race and sex, causative organisms, the disposition of pa-
tients, and the incidence and outcome.
methods
We analyzed the occurrence of sepsis from 1979 through 2000 using a nationally rep-
resentative sample of all nonfederal acute care hospitals in the United States. Data on
new cases were obtained from hospital discharge records coded according to the
Inter-
national Classification of Diseases, Ninth Revision, Clinical Modification
.
results
Review of discharge data on approximately 750 million hospitalizations in the United
States over the 22-year period identified 10,319,418 cases of sepsis. Sepsis was more
common among men than among women (mean annual relative risk, 1.28 [95 percent
confidence interval, 1.24 to 1.32]) and among nonwhite persons than among white per-
sons (mean annual relative risk, 1.90 [95 percent confidence interval, 1.81 to 2.00]). Be-
tween 1979 and 2000, there was an annualized increase in the incidence of sepsis of
8.7 percent, from about 164,000 cases (82.7 per 100,000 population) to nearly 660,000
cases (240.4 per 100,000 population). The rate of sepsis due to fungal organisms in-
creased by 207 percent, with gram-positive bacteria becoming the predominant patho-
gens after 1987. The total in-hospital mortality rate fell from 27.8 percent during the
period from 1979 through 1984 to 17.9 percent during the period from 1995 through
2000, yet the total number of deaths continued to increase. Mortality was highest among
black men. Organ failure contributed cumulatively to mortality, with temporal improve-
ments in survival among patients with fewer than three failing organs. The average
length of the hospital stay decreased, and the rate of discharge to nonacute care medi-
cal facilities increased.
conclusions
The incidence of sepsis and the number of sepsis-related deaths are increasing, although
the overall mortality rate among patients with sepsis is declining. There are also dispar-
ities among races and between men and women in the incidence of sepsis. Gram-positive
bacteria and fungal organisms are increasingly common causes of sepsis.
abstract
MR1.VOL1:Nejm:NEJM:Layout:front_030417:4mart.oa
Fri Mar 21 03 12:13:16
n engl j med
348;16
www.nejm.org april
17, 2003
epidemiology of sepsis in the u.s.
1547
are of patients with sepsis costs
as much as $50,000 per patient,
1
resulting in
an economic burden of nearly $17 billion
annually in the United States alone.
2
Sepsis is often
lethal, killing 20 to 50 percent of severely affected pa-
tients.
3
It is the second leading cause of death among
patients in noncoronary intensive care units (ICUs)
4
and the 10th leading cause of death overall in the
United States.
5
Furthermore, sepsis substantially re-
duces the quality of life of those who survive.
6,7
Accurate national data on sepsis may be used to
establish health care policy and to allocate health
care resources. It is impractical to attempt to obtain
national epidemiologic estimates prospectively, and
data from a limited population or a short period
may be inaccurate, making national administrative
data sets an essential tool for such investigations.
8-10
Epidemiologic estimates are equally dependent on
consistent defining criteria. By consensus, sepsis is
defined as the combination of pathologic infection
and physiological changes known collectively as the
systemic inflammatory response syndrome.
11
Pa-
tients with acute organ dysfunction are considered
to have severe sepsis. The usefulness of these def-
initions remains contentious,
12,13
although their
application allows the identification of patients in
whom a response to effective therapy is possible.
14
These consensus criteria have been applied in
five epidemiologic surveys of sepsis.
2,15-18
Brun-
Buisson et al. and Alberti et al. focused on microbi-
al patterns and the ICU-specific incidence of severe
sepsis in Europe.
15,16
Rangel-Frausto et al. described
the natural history of the systemic inflammatory
response syndrome in a single-institution cohort
during a nine-month period.
17
Sands et al., in a study
involving a sample of inpatients from eight hospi-
tals during a 16-month period, observed that sep-
sis accounted for 2.0 percent of all hospitalizations,
with 59 percent of patients with sepsis requiring
intensive care and accounting for 10.4 percent of
admissions to the ICU.
18
Angus et al. quantified
severe sepsis in 1995, using state-hospital discharge
records with codes from the
International Classifica-
tion of Diseases, Ninth Revision, Clinical Modification
(ICD-9-CM) that are indicative of infection and or-
gan dysfunction.
2
On the basis of the use of ICD-
9-CM codes, a 1990 report from the Centers for
Disease Control and Prevention suggested that the
incidence of sepsis was increasing.
19
We undertook a study of nationally collected data
in order to provide a broad characterization of sep-
sis for use in epidemiologic estimates, as well as to
identify specific groups with an altered propensity
for sepsis.
15,20-23
We sought to evaluate temporal
changes in the incidence and outcome of sepsis in
the United States from 1979 through 2000, with spe-
cific examination of race and sex, causative organ-
isms, and outcome, including the disposition of pa-
tients at discharge and the effect of organ failure.
data source
The National Center for Health Statistics has con-
ducted the National Hospital Discharge Survey
(NHDS) continuously since 1965. Since 1979, the
NHDS has conformed to the guidelines of the Uni-
form Hospital Discharge Data Set for consistency
of reporting in records. The NHDS is composed of
a sample of all nonfederal acute care hospitals in
the United States, including approximately 500 hos-
pitals, with equal representation of all geographic
regions. The data base is constructed through the
surveying of discharge records for inpatients from
each participating hospital, representing approxi-
mately 1 percent of all hospitalizations, or 350,000
discharges annually in the United States. Discharge
records are abstracted for demographic information
(age, sex, ethnic background, geographic location,
and marital status), seven diagnostic codes (from
ICD-9-CM), four procedural codes (from
Current Pro-
cedural Terminology
[CPT]), dates of hospital admis-
sion and discharge, sources of payment, and dispo-
sition at discharge.
definitions
Cases were identified from discharge records in
the NHDS that included a code for sepsis. Sepsis
was defined by the presence of any of the follow-
ing ICD-9-CM codes: 038 (septicemia), 020.0 (sep-
ticemic), 790.7 (bacteremia), 117.9 (disseminated
fungal infection), 112.5 (disseminated candida in-
fection), and 112.81 (disseminated fungal endocar-
ditis). Organ failure was defined by a combination
of ICD-9-CM and CPT codes, as outlined in the Ap-
pendix.
validation
The ICD-9-CM coding system for identifying pa-
tients with sepsis was validated by nested case–con-
trol analysis. The patients with sepsis were patients
admitted to a large university hospital during a six-
month period with a 038 code in their discharge
records. The controls were patients admitted im-
c
methods
MR1.VOL1:Nejm:NEJM:Layout:front_030417:4mart.oa
Fri Mar 21 03 12:13:16
n engl j med
348;16
www.nejm.org april
17
,
2003
The
new england journal
of
medicine
1548
mediately before or after each identified patient with
sepsis, who were included if there was no 038 code
in their discharge records. Sepsis was deemed to be
present when the consensus-conference definition
of sepsis was met.
14
statistical analysis
Incidences were normalized to the population dis-
tribution in the 2000 U.S. Census, and all estimates
are presented according to accepted guidelines for
the accuracy of NHDS data. That is, only absolute,
unweighted samples of more than 60 patients with
relative standard error (RSE) measures of less than
30 percent were included in data analyses. The RSE
was calculated as a first-order Taylor-series approx-
imation with the use of SUDAAN software,
24
as
outlined in the RSE tables of the 2000 NHDS docu-
mentation.
25
The standard error was calculated by
multiplying the RSE by the estimated incidence or
mortality rate, and 95 percent confidence intervals
were calculated from these standard errors with the
use of Excel software (Microsoft). Data for contin-
uous variables were compared by analysis of vari-
ance, and data for categorical variables were com-
pared by the chi-square test or Fisher’s exact test, as
appropriate for the size of the sample, with the use
of SAS software (SAS Institute). Annual data are di-
vided into four subperiods (1979 through 1984,
1985 through 1989, 1990 through 1994, and 1995
through 2000) for the assessment of temporal
changes or comparison of samples of limited size.
Since information on race was missing for some
persons (the rate of missing data on race ranged
from 1 to 20 percent for any given year), these per-
sons were excluded from the calculations of race-
specific rates but were included in all other calcu-
lations of rates. Reported P values are two-sided.
demographics
During the study period, there were a total of approx-
imately 750 million hospitalizations in the United
States. The demographic characteristics of and co-
existing conditions in the population of patients
with sepsis in each of the four subperiods are shown
in Table 1. The average age of patients with sepsis
increased consistently over time, from 57.4 years in
the first subperiod to 60.8 years in the last subperi-
od (the mean change between these subperiods was
an increase of 3.5 years [95 percent confidence in-
terval, 2.1 to 4.9 years]). Sepsis developed later in
life in female patients than in male patients — the
mean age among women was 62.1 years, as com-
pared with 56.9 years among men (difference, 5.2
years [95 percent confidence interval, 4.1 to 6.0
years]). There was a similar pattern to the increases
in incidence among men and among women, al-
though the incidence among women increased
more rapidly during the study period (an annualized
increase of 8.7 percent vs. 8.0 percent). Although
men accounted for 48.1 percent of cases of sepsis
on average per year, adjustment for sex in the popu-
lation of the United States reveals that in every year,
men were more likely to have sepsis than women
(mean annual relative risk, 1.28 [95 percent confi-
dence interval, 1.24 to 1.32]) (Fig. 1).
Whites had the lowest rates of sepsis during the
study period, with both blacks and other nonwhite
groups having a similarly elevated risk as compared
with whites (mean annual relative risk, 1.89 [95 per-
cent confidence interval, 1.80 to 1.98] and 1.90 [95
percent confidence interval, 1.80 to 2.00], respec-
tively) (Fig. 2). Black men had the highest rate of
sepsis during the study period (330.9 cases per
100,000), the youngest age at onset (mean age, 47.4
years), and the highest mortality (23.3 percent).
incidence
During the 22-year study period, there were
10,319,418 reported cases of sepsis (accounting for
1.3 percent of all hospitalizations). The number of
patients with sepsis per year increased from 164,072
in 1979 to 659,935 in 2000 (an increase of 13.7 per-
cent per year). After normalization to the population
distribution in the 2000 U.S. Census, the incidence
of sepsis increased over the 22-year period from
82.7 cases per 100,000 population to 240.4 cases
per 100,000 population, for an annualized increase
of 8.7 percent. The increasing incidence was most
apparent during the first two subperiods, from 1979
through 1989.
causative organisms
From 1979 through 1987, gram-negative bacteria
were the predominant organisms causing sepsis,
whereas gram-positive bacteria were reported most
commonly in each subsequent year (Fig. 3). Among
the organisms reported to have caused sepsis in
2000, gram-positive bacteria accounted for 52.1
percent of cases, with gram-negative bacteria ac-
counting for 37.6 percent, polymicrobial infections
for 4.7 percent, anaerobes for 1.0 percent, and fungi
for 4.6 percent. Specific organisms causing sepsis
results
MR1.VOL1:Nejm:NEJM:Layout:front_030417:4mart.oa
Fri Mar 21 03 12:13:16
n engl j med
348;16
www.nejm.org april
17, 2003
epidemiology of sepsis in the u.s.
1549
were recorded in 51 percent of all discharge records
over the 22-year period, with the rate increasing dur-
ing the first subperiod and remaining static there-
after. The greatest relative changes were observed
in the incidence of gram-positive infections, which
increased by an average of 26.3 percent per year. The
number of cases of sepsis caused by fungal organ-
isms increased by 207 percent, from 5231 cases in
1979 to 16,042 cases in 2000.
disposition of patients
In 1979, 78.5 percent of surviving patients were dis-
charged home; the rate decreased to 56.4 percent
in 2000. Concurrently, the rate of discharge to other
health care facilities (i.e., rehabilitation centers or
other long-term care facilities) increased from 16.8
percent to 31.8 percent of all survivors of sepsis-
related hospitalizations (P<0.001). Over time, signif-
icantly more patients had hospitalizations of fewer
than 7 days, and significantly fewer patients stayed
in the hospital more than 30 days (P<0.001 for both
trends).
organ failure and mortality
Mortality rates for the entire cohort declined over
the 22-year period, averaging 27.8 percent during
the first subperiod and 17.9 percent during the fi-
nal subperiod (P<0.001) (Fig. 4). Despite the im-
proved survival rates, the increasing incidence of
sepsis resulted in nearly a tripling of the number of
* Plus–minus values are means ±SE. HIV denotes human immunodeficiency virus.
† Data are normalized for race in the 2000 U.S. Census.
‡ HIV-specific coding appeared in 1986.
Table 1. Characteristics of Patients with Sepsis, According to Subperiod.*
Characteristic
1979–1984
(N=1,332,468)
1985–1989
(N=2,220,659)
1990–1994
(N=2,697,472)
1995–2000
(N=4,068,819)
Demographic characteristics
Age — yr
57.4±28.9 59.3±22.9 60.8±16.2 60.8±13.7
Male sex — % 49.6 48.9 46.8 48.0
Race — no./100,000 population (% of patients)†
White
Black
Other
92.1 (81.2)
163.0 (15.2)
187.3 (3.6)
166.4 (80.3)
301.7 (16.0)
298.0 (3.7)
167.8 (78.5)
322.8 (17.2)
300.6 (4.3)
186.3 (76.3)
378.2 (17.7)
370.5 (6.0)
Length of hospital stay — days 17.0±8.5 15.6±6.0 15.3±4.0 11.8±2.6
Coexisting conditions — % of patients
Chronic obstructive pulmonary disease
5.7 7.3 9.3 12.1
Congestive heart failure 8.6 9.9 13.6 15.2
Cancer 17.1 17.9 18.0 14.5
HIV infection‡ — 1.0 2.1 2.0
Cirrhosis 2.4 2.5 2.2 2.3
Diabetes 12.2 14.5 16.9 18.7
Hypertension 7.0 9.2 13.6 18.6
Pregnancy 0.6 0.5 0.4 0.3
No. of organs with failure — % of patients
0
83.2 78.1 74.0 66.4
1 13.6 17.9 20.1 24.6
2 2.7 3.5 4.8 7.1
≥3 0.5 0.5 1.1 1.9
