Risk of Mortality with a Bloodstream Infection
is Higher in the Less Severely Ill at Admission
Peter W. Kim1, Trish M. Perl4,5, Eithne F. Keelaghan2, Patricia Langenberg2, Eli N.
Perencevich2,3 Anthony D. Harris2,3, Xiaoyan Song4,5 and Mary-Claire Roghmann2,3
1. Food and Drug Administration, Division of Anti-Infective Drug Products, Rockville,
2. Department of Epidemiology and Preventive Medicine, University of Maryland School
3. Epidemiology Section, Veterans Affairs Maryland Health Care System
4. Department of Hospital Epidemiology and Infection Control, The Johns Hopkins
5. Division of Infectious Diseases, Department of Medicine, Johns Hopkins Medical
*The views expressed in this paper do not necessarily reflect those of the U.S. Food and
Author for Correspondence and Reprints:
AJRCCM Articles in Press. Published on December 10, 2004 as doi:10.1164/rccm.200407-916OC
Copyright (C) 2004 by the American Thoracic Society.
Mary-Claire Roghmann, M.D., M.S.
VA Maryland Health Care System
100 N. Greene Street (Lower level)
Baltimore, MD 21201
Phone (410) 706-0062
Fax (410) 706-0098
Financial support: This research was supported by # UR8/CCU315092-03 from the Center for
Disease Control and Prevention for Epicenters for the Prevention of Health care–associated
Running Title: Impact of Bloodstream Infections in the ICU
Subject Category: 37- Non pulmonary nosocomial infections
Word count for Text: 2657
This article has an online data supplement, which is accessible from this issue's table of content
online at www.atsjournals.org.
Rationale Health care–associated bloodstream infections are common in critically ill patients;
however, investigators have had difficulty in quantifying the clinical impact of these infections
given the high-expected mortality among these patients.
Objective To estimate the impact of health care–associated bloodstream infections on in-hospital
mortality after adjusting for severity of illness at critical care admission.
Method A cohort of medical and surgical intensive care unit patients.
Measurements Severity of illness at admission, bloodstream infection, and inhospital mortality
Main Results Among the 2,783 adult patients, 269 developed unit-associated bloodstream
infections. After adjusting for severity of illness, patients with a lower initial severity of illness
who developed an infection had a greater than two-fold higher risk for in-hospital mortality
(HR= 2.42, 95%CI 1.70, 3.44) when compared to patients without infection and with a similar
initial severity of illness. In contrast, patients with a higher initial severity of illness who
subsequently developed an infection did not have an increased risk for in-hospital mortality
(HR=0.96, 95%CI 0.76, 1.23) when compared to patients without infection but with a similar
initial severity of illness.
Conclusions These results suggest that these infections in less ill patients have a higher
attributable impact on subsequent mortality than in more severely ill patients. Focusing
interventions to prevent bloodstream infections in less severely ill patients would be expected to
have a greater benefit in terms of mortality reduction.
Word count: 228
Key Words: bloodstream infection; mortality; intensive care unit; adults; cohort study; health
Bloodstream infections (BSIs) are a common health care–associated infection among intensive
care unit (ICU) patients.(1, 2) Most BSIs are categorized as being due to intravascular catheters;
and catheter-associated bloodstream infections represent approximately 15% of all ICU
infections.(2) The use of intravascular catheters has become more common in hospitalized
patients both inside and outside the ICU.(3) The Center for Disease Control and Prevention’s
Hospital Infection Control Practices Advisory Committee recently published guidelines for
preventing intravascular catheter associated bloodstream infections.(4) These recommendations
include, but are not limited to, surveillance for bloodstream infections, the use of full barrier
drapes with catheter insertion and appropriate skin preparation, and if infection control measures
are not successful, the use of anti-septic or anti-microbial impregnated catheters. These and other
interventions are time-consuming and increase costs; therefore, it is important to quantify the
impact of BSIs in order to justify implementing their use.
Estimates of the excess mortality due to bloodstream infections fall most often between 35% and
60%, but range from 5% to 80%.(5-14) This wide range of mortality estimates illustrates the
difficulty in separating the BSI mortality from mortality due to the severity of patients’
underlying disease processes.(15, 16) However, the use of well validated severity of illness
measures which predict hospital or ICU mortality prior to the BSI can minimize this
confounding.(17) The objective of our study was to quantify the attributable impact of ICU–
associated BSIs on the in-hospital mortality of medical and surgical ICU patients while adjusting
for the severity of their illness upon admission to the ICU. Identifying a group of patients that
would benefit most from a reduction in BSI incidence could lead to cost-effective use of
measures to reduce the risk of BSIs. Some of the results of this study have been previously
reported in the form of an abstract. (18)
CONDENSED METHODS (Word count 501)
This prospective cohort study was conducted in a population of adult patients from 5 ICUs [3
surgical (n=30 beds, 19 beds, and 10 beds) and 2 medical (n=25 beds and 10 beds)] in 3 tertiary
care institutions in Baltimore, Maryland: Johns Hopkins Hospital (~1100 beds), University of
Maryland Medical System (~500 beds) and Baltimore Veterans Affairs Medical Center (~150
beds) from August 1998 to April 2000. Patients admitted to participating ICUs were included in
the study if they were aged > 18 years and stayed in the ICU for more than 48 hours.
Research assistants prospectively collected clinical data on all patients via chart review and were
unaware of patient outcomes at the time of data collection. Certified infection control
professionals prospectively identified all BSIs as a part of infection control surveillance in the
ICUs. Data on disposition at discharge were obtained retrospectively from administrative
Eligible patients with cultured blood growing organisms (a positive blood culture) obtained more
than 48 hours after admission to the ICU, and which was unrelated to infection present on ICU
admission, had an ICU –associated BSI. BSIs were then classified based on suspected etiology
as either “primary” or “secondary” using CDC National Nosocomial Infection Surveillance
System definitions.(19) See online-data repository for further details.
APACHE II scores were calculated on all patients in the participating ICUs during the first 24
hours of admission to the unit. (20) Age, gender, date(s) of hospital and ICU admission and
discharge, admitting diagnosis, surgical procedures, and history of underlying medical conditions
were collected prospectively during the patient’s admission.
Mortality was assessed at the time of discharge from the acute care hospital. Patient discharge
disposition was noted as: discharged home, discharged to a long-term care facility, deceased, or
other disposition (not otherwise specified).
We compared categorical variables using chi-square and Fisher’s exact tests. We compared
continuous variables using student’s t-tests. We categorized patient APACHE II scores based on
precedents set in previous literature. (21) We performed unstratified and stratified analyses to
determine the unadjusted and adjusted associations between the development of a bloodstream
infection in the ICU and in-hospital mortality. We tested for effect modification using the
Breslow-Day test for homogeneity (p<.05 level). If we did not detect effect modification, we
evaluated for confounding. We suspected confounding if the Mantel-Haenszel common relative
risk differed by 10% or more from the unadjusted relative risk found in the unstratified analysis
of the association between BSI status and in-hospital mortality.
We then constructed Cox proportional hazard models based on time in the ICU until death or
censoring, defined as discharge from the hospital. We included BSI status as a time-dependent
variable, which identified BSI status at the time cultured blood grew organisms. Thus, for those
with BSIs, the time variable is time from cultured blood with organisms to death or censoring;
for those without, time is from admission to the ICU to death or censoring. Additional models
evaluated the effect of previously identified effect modifying and confounding variables.
Between August 1998 and April 2000, 2811 adult patients were admitted to the ICU and stayed
for greater than 48 hours. Twenty-eight of these patients were admitted with BSIs and thus
excluded from further analysis. Of the remaining 2783 patients, 269 had documented ICU–
associated bloodstream infections during their ICU stay. Table 1 provides descriptive
information about the study population. Ten percent of the study population (n=269) developed
ICU–associated BSIs a median of 10 days from ICU admission (range 3-89 days). When
stratified by severity of illness of admission (APACHE <20; >= 20), there were no significant
differences in duration of time to BSI (median 10 days vs. 10 days, p=0.57, Wilcoxon). Sixty-
eight percent of these patients had primary BSIs, out of which 95% were noted to have had a
central line in place at the time the bloodstream infection was diagnosed. The major pathogens
causing BSIs (both primary and secondary) were as follows: Enterococcus species (29%),
coagulase-negative staphylococci (24%), gram-negative organisms (17%), Candida species
(13%), and Staphylococcus aureus (13%).
Table 2 provides a comparison of ICU patients with and without BSIs. Patients with BSIs were
significantly more likely to die during their hospital admission. In addition, they were more ill on
admission as judged by higher APACHE II scores. BSI patients were more likely to have been
admitted to a hospital floor for greater than 48 hours prior to admission to the ICU. Furthermore,
patients with BSIs were significantly more likely to: be younger, have cirrhosis, have received a
solid organ transplant, and be immunosuppressed than patients without BSIs.
Table 3 shows variables significantly associated with in-hospital mortality. They included: older
age, female gender, hospital admission for greater than 48 hours prior to ICU admission,
admission to a medical ICU (vs. a surgical ICU), not having surgery during the hospital
admission and higher APACHE II score. Four percent of ICU patients with an APACHE score
of 0-9 died during hospitalization; 13% of patients with a score of 10-19 died; 29% of patients
with a score of 20-29 died and 47% of patients with a score > =30 died. Co-morbid conditions
associated with in-hospital mortality included: end-stage renal disease, cirrhosis, HIV positive
status, being on immunosuppressive therapy, being on dialysis, and not having coronary artery
disease/peripheral vascular disease.
Patients with BSIs were three times more likely to die in-hospital than patients without BSIs (RR
3.12, 95% CI 2.72-3.56). Stratified analyses of the association between BSI status and in-
hospital mortality were performed on patient variables significantly associated with both BSI
status and in-hospital mortality, or solely associated with in-hospital mortality in order to test for
effect modification and confounding (see Table 1 in Online Data Supplement). Effect
modification is the ability of a variable to change the relationship, as measured by the relative
risk, between a risk factor (BSI) and an outcome (in-hospital mortality); for example, asbestos
exposure may be an effect modifying factor for the relationship between cigarette smoking and
lung cancer. Cigarette smokers exposed to asbestos have a higher risk for lung cancer than
cigarette smokers not exposed to asbestos. Effect modification in our analysis was noted for the
APACHE II score (Stratum 1 APACHE II <20: RR 6.44 95% CI (5.00-8.34); Stratum 2
APACHE II >20: RR 1.91 95% CI (1.63-2.24) Breslow Day test p<0.01) and HIV infection (see
on-line data supplement). We chose to exclude the patients with HIV infection from further
analysis because of the small number of patients with HIV infection. Confounding was noted for
two variables: being admitted to the hospital greater than 48 hours prior to ICU admission and
cirrhosis (see Table 1 in Online Data Supplement).
We compared time to death in HIV-negative patients with BSI to those without BSI stratified by
their APACHE score at ICU admission. In those with an APACHE score <20, the curves
significantly diverged (p=<.01, Log Rank test). In those with an APACHE score >=20, the
curves were not significantly different (p=0.57, Log Rank test). Table 4 displays the results of
the proportional hazard models for time to in-hospital mortality in HIV negative patients; BSI is
used as a time dependent variable in the model. Patients with a BSI and a low admission
APACHE II score (<20) had 2.42 times the risk of death in-hospital compared with patients
without a BSI and a low admission APACHE II score. However, in patients with a high
admission APACHE II score (> 20), the hazard ratio did not show an increased risk of in-
hospital mortality (HR 0.96). For each increase in age by one year, the mortality rate increased
by 1% in the model. Cirrhosis and admission to the hospital for more than two days prior to ICU
admission were also found to significantly increase the risk of mortality.
If BSIs are causally associated with death, then BSIs due to more virulent bacteria (e.g.,
Staphylococcus aureus and Gram-negative rods) would have a stronger association with
mortality than BSIs due to less virulent bacteria (e.g., Coagulase-negative Staphylococci).
Among those patients with a low admission APACHE II score (< 20), those with a virulent
organism were over seven times more likely to die compared to patients without BSIs (RR 7.43,
95% CI 5.74-9.62, p=<0.01). In comparison, those with BSIs due to less virulent bacteria and a
low admission APACHE score had a lower risk of dying (RR 4.10, 95% CI 2.34-7.13, p=<0.01).
This increase in risk with an increase in virulence of the organism was not seen among patients
with a high admission APACHE score (>= 20). Among those patients with a high admission
APACHE II score (>= 20), those with a virulent organism were 1.93 times more likely to die
compared to those without BSIs (RR 1.93, 95% CI 1.63-2.28, p=<0.01). Those with BSIs due to
less virulent bacteria were 1.86 times more likely to die compared to those without BSIs (RR
1.86, 95% CI 1.38-2.52, p=<0.01).
In this large cohort of patients admitted to ICUs, mortality was up to seven times higher among
patients who developed BSIs than those who did not. In contrast to traditional thinking, we
found that less severely ill ICU patients who develop an ICU–associated BSI had a higher
mortality than their counterparts who did not develop a BSI. This risk was significantly higher
than that seen among more ill ICU patients who developed a BSI. Age, cirrhosis, and being in
the hospital longer than 48 hours prior to ICU admission were also independent risk factors for
Our study findings that mortality is higher among patients with ICU–associated BSIs are
consistent with other studies of BSIs and mortality in intensive care unit patients.(6, 10-14)
Smith et al. demonstrated a one and a half times increased risk among medical ICU patients at a
Veterans Affairs Medical Center.(13) Pittet et al. found a greater than three times increased risk
of death among their BSI patients.(6) Renaud et al. showed that the odds of death among their
BSI patients were more than four times those of BSI-negative patients.(11) Our study findings
are unique because we describe an interesting interaction between ICU patients’ severity of
illness at admission and their risk of in-hospital mortality if they develop a BSI. Other studies
looking at the impact of ICU complications on in-hospital mortality may consider testing for this
type of interaction.
Measuring the impact of an ICU complication on an outcome such as mortality is difficult
because the measurement of the patient’s underlying risk of mortality is imprecise. Still, we
believe that our findings are biologically plausible. Patients with higher severity of illness may
have many reasons for increased mortality. The addition of a BSI may not greatly increase
severely ill patients’ overall risk of mortality. Clinically we know a cirrhotic patient with a
gastrointestinal bleed who requires an ICU admission has a high risk of mortality. The addition
of a BSI to this patient’s list of life-threatening problems may be inconsequential. One may
postulate that severely ill patients have an impaired cytokine response to microbial antigens such
that pro-inflammatory cytokines, normally induced during acute bacteremia, are less able to
wreak their havoc. This could occur either through a reduction in cytokine-producing cells
following chemotherapy or immunosuppressive therapy, or the induction of tolerance to
microbial pathogens secondary to a permeable or “leaky” gut. Although the pathophysiologic
mechanism is not clear, this observation merits further investigation.
Our study is methodologically strong. To our knowledge, this is the largest cohort study to assess
the association between mortality and ICU–associated BSIs. The closest in size was Renaud et
al.’s study of 2201 patients of whom 96 unique patients developed ICU–associated bloodstream
infections.(11) We used a standardized severity of illness scoring system (the APACHE II
system) that is well-validated in predicting ICU and hospital mortality.(20) We measured
severity of illness at admission to the ICU prior to development of the BSI. Given the fact that
we derived our cohort from three tertiary care, teaching hospitals, we believe that this study is
generalizable to most tertiary care ICU populations.
The major limitation of our study is that it is observational. We attempted to control for
important confounding variables such as severity of illness at ICU admission, but there may still
be hidden bias due to the observational nature of the study. Because of the difficulty in
determining whether an infection was directly or indirectly related to a patient’s death, we can
only infer that the bloodstream infection was the important factor that increased the risk of
mortality. In support of our supposition, we found that the risk of death was greater among
patients with bloodstream infections caused by more virulent organisms. It is important to note
that certain research questions can only be answered with observational studies for both ethical
and practical reasons. For the question at hand, we feel that we used the best methodology
available to us.
In conclusion, we found that ICU patients who have a lower severity of illness and develop a BSI
are at significantly higher increased risk for in-hospital mortality compared with ICU patients
with higher severity of illness who develop a BSI. Although only a third of the BSIs occurred in
patients with a lower severity of illness, the impact of the BSI was twice as high in this
population. This study further clarifies the relationship between bloodstream infections and in-
hospital mortality among ICU patients. Our study supports implementing interventions to
prevent bloodstream infections in less severely ill patients in the ICU, as well as at risk patients
on general medical and surgical floors. Infection control efforts have long been focused in the
intensive care unit because of the higher risk of infection in ICU patients. The results of this
study suggest that decisions on using infection control interventions might be more cost-
effectively driven by using both the impact of infections on specific patient populations and the
overall risk of infection in that patient population.
*Acknowledgement: We would like to thank Dr. Alan S. Cross for contributing his ideas on the
biological plausibility of our findings.
1. Donowitz LG, Wenzel RP, and Hoyt JW. High risk of hospital-acquired infection in the
ICU patient. Crit Care Med 1982; 10:355-357.
2. National Nosocomial Infections Surveillance (NNIS) System Report, data summary from
January 1992 through June 2003, issued August 2003. Am J Infect Control 2003; 31:481-498.
3. Climo M, Diekema D, Warren DK, Herwaldt LA, Perl TM, Peterson L, Plaskett T, Price
C, Sepkowitz K, Solomon S, Tokars J, Fraser VJ, and Wong E. Prevalence of the use of central
venous access devices within and outside of the intensive care unit: results of a survey among
hospitals in the prevention epicenter program of the Centers for Disease Control and Prevention.
Infect Control Hosp Epidemiol 2003; 24:942-945.
4. O'Grady N P, Alexander M, Dellinger EP, Gerberding JL, Heard SO, Maki DG, Masur
H, McCormick RD, Mermel LA, Pearson ML, Raad, II, Randolph A, and Weinstein RA.
Guidelines for the prevention of intravascular catheter-related infections. Am J Infect Control
5. Pittet D. Nosocomial bloodstream infections. In R. P. Wenzel, editor. Prevention and
Control of Nosocomial Infections, 2nd ed. Williams & Wilkins, Baltimore, MD. 512-555 1993.
6. Pittet D, Tarara D, and Wenzel RP. Nosocomial bloodstream infection in critically ill
patients. Excess length of stay, extra costs, and attributable mortality. Jama 1994; 271:1598-
7. Kreger BE, Craven DE, Carling PC, and McCabe WR. Gram-negative bacteremia. III.
Reassessment of etiology, epidemiology and ecology in 612 patients. Am J Med 1980; 68:332-
8. Gatell JM, Trilla A, Latorre X, Almela M, Mensa J, Moreno A, Miro JM, Martinez JA,
Jimenez de Anta MT, Soriano E, and et al. Nosocomial bacteremia in a large Spanish teaching
hospital: analysis of factors influencing prognosis. Rev Infect Dis 1988; 10:203-210.
9. Weinstein MP, Murphy JR, Reller LB, and Lichtenstein KA. The clinical significance of
positive blood cultures: a comprehensive analysis of 500 episodes of bacteremia and fungemia in
adults. II. Clinical observations, with special reference to factors influencing prognosis. Rev
Infect Dis 1983; 5:54-70.
10. Soufir L, Timsit JF, Mahe C, Carlet J, Regnier B, and Chevret S. Attributable morbidity