Outcomes of Primary and Catheter-related Bacteremia
A Cohort and Case–Control Study in Critically Ill Patients
BERTRAND RENAUD and CHRISTIAN BRUN-BUISSON, for the ICU–Bacteremia Study Group
Service de Réanimation Médicale and Unité d’Hygiène et Prévention de l’Infection, Assistance Publique Hôpitaux de Paris and Université Paris XII,
Hôpital Henri Mondor, Créteil, France
We studied a cohort of 2201 patients hospitalized in 15 French in-
tensive care units (ICUs) for  48 h during a 4-mo period to assess
the incidence and outcomes of primary and definite catheter-related
bloodstream (CRB) or secondary nosocomial bloodstream infec-
tion (NBSI). Variables associated with ICU death and duration of
stay were determined by logistic regression, and attributable mor-
tality and length of stay (LOS) from a nested matched case–con-
trol (96 pairs) study, stratified on the source of bacteremia. Bacte-
remia occurred in 5% (95% CI 4.1–6%) of patients with  48 h ICU
stay. Primary, CRB, and secondary NBSI accounted for 29%, 26%,
and 45% of the 111 episodes, respectively. NBSI was associated
with a markedly increased risk of death (OR  4.6; 95% CI 2.9–7.1)
and an attributable mortality of 35% (95% CI, 28%–47%). In the
case–control study, the excess mortality was 20% (p  0.03) in pa-
tients with primary bacteremia and CRB, and 55% (p  0.001) for
secondary bacteremia; in patients with CRB only, the excess mor-
tality was 11.5%. The median excess ICU LOS in survivors of NBSI
was 9.5 d, and was similar, irrespective of its source. The risk of
mortality associated with primary and catheter-related bacteremia
appears much lower than that of secondary bacteremia, but is siz-
able, and the excess LOS incurred by the various categories of
bacteremia is comparable. Differentiating catheter-related bacter-
emia from both primary and other secondary bacteremia appears
warranted in studies conducted in critically ill patients.
Nosocomial bloodstream infection (NBSI) is a leading cause
of death in critically ill patients (1, 2). The overall incidence of
nosocomial bacteremia has been increasing recently (3–5). At
one institution, a 3-fold increase in NBSI rates has been noted
over the past decade, from 6.7/1000 to 18.4/1000 discharges
(4). Much of this increase appears to be due to primary bacte-
remia (3), which accounts for 20% to 30% of all nosocomial
bacteremias (3, 6). Changes in the distribution of organisms
have also been noted, with increasing rates of coagulase-nega-
tive staphylococci, Staphylococcus aureus, Enterococci, and
Candida sp. (3, 7). Most of these changes appear related to in-
travascular device-associated infection, likely reflecting the in-
creasing proportion of intensive care beds in hospitals and as-
sociated increased illness severity of patients, and possibly
increased awareness of coagulase-negative staphylococci as a
significant pathogen (3, 8). Intravascular device-associated in-
fections are therefore of major concern to both infection con-
trol professionals and health care providers, especially in the
intensive care unit (ICU) setting.
Several studies using various approaches have examined the
impact of NBSI on outcome of patients (1, 6, 9). In surgical ICU
(Received in original form December 16, 1999 and in revised form December 6, 2000)
Members of the ICU–Bacteremia study group are provided in the APPENDIX.
Correspondence and requests for reprints should be addressed to Christian Brun-
Buisson, M.D., Service de Réanimation Médicale and Unité d’Hygiène et Préven-
tion de l’Infection, Assistance Publique Hôpitaux de Paris and Université Paris XII,
Hôpital Henri Mondor, 94010 Créteil, France. E-mail: christian.brun-buisson@
hmn.ap-hop-paris.fr
Am J Respir Crit Care Med Vol 163. pp 1584–1590, 2001
Internet address: www.atsjournals.org
patients, Pittet and coworkers found a 35% higher mortality in
bacteremic patients compared with matched controls (1). In a
large European prevalence study of nosocomial infection (EPIC),
bacteremia or sepsis was associated with a 1.7 odds ratio for
death, using multiple regression techniques (10). Using a simi-
lar analytic strategy, Fagon and coworkers found that NBSI
was associated with a 2.5 odds ratio of mortality in a cohort
of 1978 patients from one medical ICU (11). However, the
respective impact of the different categories of bacteremia,
whether primary bacteremia, bacteremia secondary to cathe-
ter infection, or secondary to other sources, remains unclear.
In hospital-wide and multiinstitutional surveillance systems of
nosocomial bacteremia, it is difficult to differentiate catheter-
related bacteremia from true primary bacteremia (3, 4, 12).
Thus, estimates of the overall consequences of NBSI may not
apply to catheter-related bacteremia. Two recent case–control
studies have suggested that primary or central venous cathe-
ter-related bacteremia occurring in critically ill patients was
not associated with a significantly higher mortality than in
matched control patients (13, 14); both these studies used
matching on variables reflecting severity of illness recorded
just before the occurrence of bacteremia to adjust for severity
of patients. However, this approach may result in underesti-
mating the impact of infection on outcome of patients.
Determining the impact of nosocomial infections on out-
come of patients is important to design future studies, and to
assess the impact of preventive measures. We therefore con-
ducted this study to determine the incidence of nosocomial
bacteremia in ICU patients, and the relative proportion of sec-
ondary, primary, or definite catheter-related bacteremia; and
to estimate the influence of the source of NBSI on mortality
and ICU length of stay. For these purposes, we examined a co-
hort of patients from 15 ICUs in France, assessed factors asso-
ciated with mortality and length of stay in the cohort, and used
both logistic regression and a nested case–control study of
bacteremic cases to estimate attributable mortality and length
of stay of bacteremia, comparing catheter-related and primary
bacteremia with secondary bacteremia.
METHODS
Setting
This prospective cohort study was conducted between February 15
and June 15, 1998 in 15 adult ICUs of university and university-affili-
ated hospitals in the Paris area. There were 11 medical and 4 surgical-
trauma ICUs, for a total of 222 beds. Participating ICUs had an ongo-
ing continuous monitoring of nosocomial infections, including routine
culture of all central venous catheters using a quantitative culture
technique (15). No antimicrobial-coated catheters were used during
the study period.
Patients newly admitted to the participating ICUs were included in
the cohort if they were aged  16 yr and stayed in the ICU for  48 h;
readmissions were excluded.
Definitions
A “case” was defined as any new patient in whom ICU-acquired bac-
teremia was diagnosed, as identified by one or more positive blood
Renaud and Brun-Buisson: Primary and Catheter-related Bacteremia 1585

cultures obtained  48 h following ICU admission, and unrelated to Kruskal–Wallis test otherwise. Paired analysis was conducted using
an infection present on ICU admission. Bacteremia related to an in- the McNemar’s test for qualitative variables, and the Wilcoxon signed
fection acquired outside the ICU was excluded. Only the first episode rank test for continuous data.
of bacteremia in a given patient was included in the analysis. Bactere- Stepwise logistic regression analyses of variables associated with
mia occurring after a “do not resuscitate” order had been issued was ICU mortality and length of stay were performed according to Hos-
excluded from analysis. mer and Lemeshow (20), using the forward procedure. A p value 
Nosocomial (ICU-acquired) bacteremia was defined as either one 0.10 by univariate analysis was selected for entering variables tested in
or more positive blood culture of a known pathogen; at least two the model. Quantitative variables such as age and SAPS II were intro-
blood cultures positive with the same microorganism taken from duced as continuous variables, and qualitative variables were catego-
blood samples obtained at least 2 h apart within a 48-h period were re- rized. For the latter variables, those that had only two classes were
quired for the following microorganisms: coagulase-negative staphy- coded as binary variables (0, 1); variables represented by more than
lococci, Corynebacterium sp., Micrococcus sp., Bacillus sp., Propioni- two classes were transformed into dummy variables and arbitrarily
bacterium sp., or strictly aerobic gram-negative organisms other than coded, using the clinically least severe as the reference category,
Pseudomonas aeruginosa and Acinetobacter baumannii. Polymicro- which was ascribed a score of 1. Testing for significant interaction be-
bial bacteremia was identified when at least two different microorgan- tween variables was introduced into the model at the last step of the
isms were recovered from blood within a 48-h period (16), and at least regression analysis. The time to occurrence of bacteremia was forced
two of these microorganisms fulfilled the above criteria for ICU- into the model (as a continuous variable) in an attempt to adjust for
acquired bacteremia. risk factors appearing during ICU stay. Multivariate analysis of mor-
Episodes of ICU-acquired bacteremia were classified as secondary tality was first performed in the whole population and then in patients
to an identified source when both blood and a clinically identified
source grew the same microorganism (i.e., identical species, pheno-
type, and antimicrobial resistance profile). Bacteremia was defined as TABLE 1. CHARACTERISTICS OF PATIENTS INCLUDED IN THE
definitely catheter related in the presence of catheter exit site inflam- PROSPECTIVE MULTICENTER COHORT STUDY OF THE IMPACT
mation with or without signs of systemic inflammatory response, and OF NBSI ON ICU MORTALITY AND LENGTH OF STAY
recovery of at least 103 colony-forming units of the same organism
Whole Cohort NBSI
from blood and the intravascular tip of a central venous or arterial
(n  2201) (n  111)
catheter (15, 17). Bacteremia was classified as primary in the absence
of an identified source of infection growing the same organism(s) as re- Number of Cases Number of Cases
covered from blood. For purpose of analysis, definite catheter-related (%) or (%) or
bacteremias and primary bacteremias were grouped together as “cath- Variables Mean (SD)* Mean (SD)* p Value†
eter-related and primary bacteremia” and compared with “secondary”
Age, yr 58.9 (19.0) 61.2 (17.0) 0.16
bacteremias, unless otherwise stated. Antimicrobial therapy was deemed
Sex ratio, male/female 1.47 2.3 0.07
appropriate when at least one antibiotic active in vitro on all organisms Patient origin 0.32
causing ICU-acquired bacteremia was administered in appropriate Direct admission 1162 (52.8) 54 (48.6)
dosage within the first 48 h of occurrence of bacteremia (16). Intrahospital transfer 807 (36.7) 41 (36.9)
Other hospital 227 (10.3) 16 (14.4)
Data Collection Admission category  0.001
One of us (B.R.) prospectively collected surveillance data in each Medical 1560 (76.1) 65 (58.6)
ICU. The following sources of information were used: nosocomial Unscheduled surgery 223 (10.9) 19 (17.1)
Scheduled surgery 149 (7.3) 8 (7.2)
surveillance data from each unit, medical and nursing records, com-
Multiple trauma 118 (5.8) 18 (16.2)
puterized admission and diagnostic coding of all patients admitted in
Leading diagnostic 0.047
each ICU, and listings of all positive blood and catheter cultures re-
Medical 1923 (87.4) 78 (70.3)
corded in each hospital’s clinical microbiology laboratory. For bacter- Cardiac failure 354 (16.8) 11 (9.9)
emic patients, the entire medical record was reviewed, using daily and Respiratory failure 723 (34.3) 36 (32.4)
summary charts for review of relevant clinical events and antibiotics Digestive tract 116 (5.5) 5 (4.5)
administered, and interview with the physicians in charge of the pa- Central nervous system 180 (8.5) 9 (8.1)
tient and with the microbiologist when appropriate. Original microbi- Metabolic disorders 101 (4.8) 6 (5.4)
ological results, results from other laboratories, and all data needed Renal failure 35 (1.7) 1 (0.9)
for scoring calculations were checked. Drug intoxication 143 (6.8) 1 (0.9)
For all patients included in the cohort, we prospectively collected Obstetrical 12 (0.6) 0 (0.0)
information on patients’ age, sex, and underlying diseases as described Other medical 165 (7.8) 9 (8.1)
by Kreger and coworkers (stratified as rapidly or ultimately fatal, or Surgical 278 (12.6) 33 (29.7)
nonfatal) (18), admission category (classified as medical, scheduled or Abdominal 76 (3.6) 8 (7.2)
unscheduled surgery, or trauma), and location prior to ICU admission Cardiac 15 (0.7) 2 (1.8)
(stratified in three classes: direct admission, when the patient was ad- Vascular 11 (0.5) 3 (2.7)
mitted to the ICU from home or the emergency room; intrahospital Other surgery 58 (2.8) 2 (1.8)
transfer, when the patient was transferred from another unit of the COPD 227 (10.5) 3 (2.7) 0.007
Immunodepression 389 (19.3) 29 (26.4) 0.055
same hospital; or interhospital transfer, when the patient was trans-
Underlyling disease
ferred from another hospital). In addition, we recorded the Simplified
Nonfatal 1194 (61.3) 62 (55.9)
Acute Physiology Score II (SAPS II) in the 24 h following ICU admis-
Ultimately fatal 492 (25.3) 32 (28.8)
sion (19), and the main reason for ICU admission and secondary diag- Rapidly fatal 262 (13.4) 17 (15.3)
noses. For patients having ICU-acquired bacteremia, we recorded the Admission SAPS II score 35.1 (17.9) 45.9 (17.9)  0.001
date of occurrence of the first episode of bacteremia and its source, ICU length of stay, d 11.1 (16.6) 34.3 (28.4)  0.001
the microorganisms involved and their antimicrobial resistance pat- ICU deaths 407 (18.5) 60 (54.0)  0.001
tern, and the appropriateness of antimicrobial therapy administered
within the first 48 h of occurrence of bacteremia. Definition of abbreviations: COPD  chronic obstructive pulmonary disease; ICU  in-
tensive care unit; NBSI  nosocomial bloodstream infection; SAPS II  Simplified Acute
Physiology Score II.
Data Analysis * Except for the sex ratio, data are shown as number (percent) of patients for qualita-
Data were analyzed using the Stata (Stata Corp., College Station, TX) tive variables and as mean (standard deviation) for quantitative variables.
†
p Values (chi-square) are given for comparison between patients with ICU-acquired
software program. Proportions were compared using the chi-square bacteremia and all other ICU patients hospitalized for at least 48 h. When variables are
test of homogeneity or the Fisher exact test; continuous variables categorized in several classes, the p value refers to the global test for that specific vari-
were compared using the t test when normally distributed, and the able.
1586 AMERICAN JOURNAL OF RESPIRATORY AND CRITICAL CARE MEDICINE VOL 163 2001

TABLE 2. SOURCES OF BACTEREMIA AND MICROORGANISMS associated with a longer than median ICU length of stay as matching
INVOLVED IN 111 PATIENTS WITH ICU-ACQUIRED BACTEREMIA variables to select appropriate controls for bacteremic cases. In addi-
tion, controls were selected as having a duration of ICU stay at least
Source of Bacteremia
as long as that of the case up to the occurrence of bacteremia. The
Catheter Other case–control study was conducted on the whole population of patients
Primary Related Secondary All Episodes having ICU-acquired bacteremia, and in the subgroups of patients
(n  32) (n  29) (n  50) (n  111) having secondary, or primary and catheter-related bacteremia.
Gram-positive cocci 15 12 17 44
S. aureus 4 4 11 20 RESULTS
Coagulase-negative Population Studied
staphylococci 7 7 1 15
Enterococci 4 0 4 8 Between February 15 and June 15, 1998, 3281 patients were ad-
Streptococci 0 0 1 1 mitted to the 15 participating ICUs. Medical ICUs accounted
Gram-negative bacilli 10 8 26 44 for 82.4% of admissions, and surgical/trauma ICUs for 17.6%.
Enterobacteriaceae 4 5 19 28 Of these, 891 (27.2%) ICU admissions were excluded because
Aerobic gram-negative 6 3 7 16
they were of  48 h duration, and 90 (2.7%) patients were ex-
Anaerobes 1 0 2 3
Candida 4 2 2 8 cluded due to readmission. The proportion of missing values for
Polymicrobial 2 7 3 12 all variables recorded did not exceed 10%, except in one ICU,
Total organisms 34 36 53 123
where 31% of the patients had missing values; the 99 patients
from this ICU (5 of whom had ICU-acquired bacteremia) were
therefore excluded from further analysis, and 2201 patients re-
mained in the cohort. Table 1 shows the clinical characteristics
with bacteremia. Because of some missing values, 1989 patients (90.3% of these patients and of those having ICU-acquired bacteremia.
of all patients in the cohort) were included in these analyses.
Because the distribution of length of ICU stay was skewed and Incidence of ICU-acquired Bacteremia
bacteremia occurred after a median of 10 d, we first determined the
median ICU length of stay in the whole cohort, and then used a step-
Nosocomial bacteremia occurred in 111 (3.5%; 95% confidence
wise multiple logistic regression to determine variables associated interval [CI] 2.9%–4.2%) of all ICU admissions, after a median
with a longer than median length of ICU stay. Thereafter, we per- of 10 d following admission (interquartile range [IQR], 6 to 18
formed a pairwise case–control (1:1) analysis, first in the whole popu- d). In the 2201 patients staying for  48 h, the incidence of
lation and then in surviving pairs only, using factors previously found NBSI was 5% (95% CI [4.1%–6.0%]) or 4.5/1000 patient-ICU

TABLE 3. VARIABLES ASSOCIATED WITH MORTALITY OF 111 PATIENTS WITH ICU-ACQUIRED BACTEREMIA

Number of Number of Adjusted Odds

Cases Deaths (%) p Value* Ratio [95% CI] p Value*

Age 61.2 (17) 0.006 1.026 [1.0–1.05] 0.045

Sex 0.8
Male 77 41 (53) —
Female 34 19 (56) —
Patient location prior to ICU 0.017 0.19
Direct admission 54 22 (41) 1
Intrahospital transfer 41 26 (63) 1.7 [0.6–4.3]
Other hospital 16 12 (75) 2.2 [0.5–9.2]
Admission category 0.001 0.49
Medical 63 38 (60) 1
Unscheduled surgery 22 15 (68) 1.9 [0.6–6.1]
Scheduled surgery 8 5 (62) 0.8 [0.2–4.3]
Trauma 18 2 (11) 0.5 [0.1–3.1]
Immunodepression 0.34 —
Yes 81 42 (52) —
No 29 18 (62) —
Underlying disease
Absent/nonfatal 62 25 (40) 1
Ultimately/rapidly fatal 49 35 (71) 0.001 3.1 [1.3–7.3] 0.010
SAPS II on admission 45.9 (17.9) 0.005 1.3 [1.0–1.06] 0.041
Microbiological category 0.54 —
CoNS 15 7 (47) —
Other 96 53 (55) —
Source of bacteremia 0.008
Primary and catheter-related 61 26 (43) 1
Secondary 50 34 (68) 2.2 [0.92–5.15] 0.08
Appropriate antimicrobial therapy† 0.85 —
Yes 66 37 (56)
No 43 23 (53) —

Definition of abbrevitions: CoNS  coagulase-negative staphylococci; ICU  intensive care unit; SAPS II  Simplified Acute Physiology
Score II.
* p Values are given using chi-square test or Fisher exact test when appropriate. Variables were considered in the multivariate analysis
when associated with mortality at p  0.10 in the univariate analysis. Continuous variables (age, SAPS II) are shown as mean (  SD) and
corresponding odds ratios are given for one unit.
†
Within 48 h of the occurrence of bacteremia.
Renaud and Brun-Buisson: Primary and Catheter-related Bacteremia 1587

days, and varied between 4.3% (95% CI [3.4%–5.2%]) and
9.5% (95% CI [6.2%–12.8%]), respectively, in medical and sur-
gical ICUs (p  0.001). Patients admitted following trauma had
the highest incidence (16.2%) of bacteremia (Table 1).
Source of Bacteremia and Microorganisms Involved
Of the 111 episodes analyzed, 61 (55%) were primary bactere-
mia and catheter related, and 50 (45.0%) were secondary to
other sources (a single source in 46 patients and several likely
sources in 4 patients); 29 (26% of all episodes) were classified as
definite intravascular catheter-related infection. Thirty-three pa-
tients (30%) had several positive blood cultures with the same
organism. Forty-four (39.6%) episodes involved gram-positive
cocci only; almost half (45.5%) of these episodes were caused by
Staphylococcus aureus, 34.1% by coagulase negative staphylo-
cocci, and 18.2% by enterococci (Table 2). Gram-negative bacilli
were recovered in only 44 (39.6%) episodes; Enterobacteriaceae
accounted for two-thirds of these episodes, and one-third of
these episodes was caused by strict aerobic gram-negative bacilli
(mostly Pseudomonas aeruginosa). Gram-negative anaerobes
were recovered in only three cases (2.7%), and in eight cases
(7.2%) blood cultures grew Candida sp. Bacteremia was polymi-
crobial in 12 cases (10.8%), of which 58% were catheter-related
bacteremia. A larger proportion of Enterobacteriaceae was re-
covered from episodes of secondary bacteremia (Table 2). Ap-
propriate antimicrobial therapy was administered to 66 (59.5%)
patients within 48 h of the first positive blood culture.
Mortality of Patients Included in the Cohort
and of Bacteremic Cases
The crude ICU mortality rate was 18.5% among patients in
the whole cohort, and 54% in patients with ICU-acquired bac-
teremia (Table 1). In the logistic regression analysis, bactere-
mia was an independent risk factor for mortality (OR  4.6
[95% CI 2.9–7.1]; p  0.0001), as well as age (p  0.0001), the
admission SAPS II (p  0.0001), and being transferred from
another hospital (p  0.0001) or from a ward (p  0.03). The
time elapsed from ICU admission to occurrence of NBSI was
not associated with mortality.
Variables associated by univariate analysis with death in
bacteremic patients were (Table 3) age, the patient’s location
prior to ICU, the admission category, the prognosis of under-
lying disease, the admission SAPS II, and the source of bacter-
emia. After multiple logistic regression, the independent vari-
ables associated with ICU death in bacteremic patients (Table
3) were a rapidly or ultimately fatal underlying disease (OR 
3.1, p  0.01), the admission SAPS II (p  0.04), and age (p 
0.045). Secondary bacteremia was associated with a poorer
prognosis (OR  2.2; 95% CI 0.92–5.15, p  0.08) than pri-
mary and catheter-related bacteremia. Of the 60 deaths re-
corded among bacteremic patients, 24 (40%) occurred early
( 72 h) after bacteremia, and 36 (60%) deaths occurred
among the 87 patients who survived  72 h after bacteremia.
The independent risk factors for early ( 72 h) mortality were
an inappropriate initial antibiotic therapy (OR  2.9 [1.1–7.9],
p  0.031) and a rapidly or ultimately fatal underlying disease
(OR  3.5 [1.3–9.4], p  0.016).
Case–Control Study
The median ICU length of stay was 5 (IQR, 3 to 12) d for the en-
tire cohort and 27 (IQR, 14 to 46) d for bacteremic cases. The in-
dependent variables associated with a longer than median length
of stay in the whole cohort, as identified by logistic regression
analysis, were age, the patient location prior to ICU admission,
the admission category, the SAPS II score, a rapidly or ultimately
fatal underlying disease, and ICU-acquired bacteremia (Table 4).
To determine the excess mortality and length of stay attrib-
utable to bacteremia, we used a pairwise (1:1) case–control study,
selecting controls matched to bacteremic cases on the vari-
ables identified above. Control patients had to fulfill all the
following six criteria: the same admission category, location
prior to ICU admission, age ( 5 yr), severity of underlying

TABLE 4. FACTORS ASSOCIATED WITH A LONGER THAN MEDIAN (5 d) LENGTH OF STAY IN THE
COHORT OF PATIENTS WITH ICU STAY OF OVER 48 h (n  2197)

Number of Number of LOS Adjusted Odds

Patients  5 d (%) p Value Ratio [95% CI] p Value

Age* 2199 —  0.001 1.0 [1.0–1.01] 0.026

Sex
Male 1312 749 (57.1) 0.08 0.6
Female 887 473 (53.3)
Patient origin  0.001
Direct admission 1163 574 (40.4) 1
Intrahospital transfer 809 487 (60.2) 1.4 [1.3–177] 0.002
Other hospital 227 161 (70.9) 2.4 [1.7–3.4]  0.0001
Admission category  0.001
Medical 1562 823 (52.7) 1
Unscheduled surgery 224 161 (71.9) 2.0 [1.4–2.9]  0.0001
Scheduled surgery 149 83 (55.7) 0.95 [0.6–1.4] 0.8
Trauma 118 73 (61.9) 1.8 [1.15–2.85] 0.01
Immunodepression
Yes 390 237 (60.8) 0.03 0.40
No 1627 890 (54.7)
Underlying disease
Absent/nonfatal 1195 595 (49.8) 1
Ultimately/rapidly fatal 756 489 (64.7)  0.001 1.5 [1.2–1.8]  0.0001
SAPS II* 2047 —  0.001 1.02 [1.01–1.02]  0.0001
ICU-acquired bacteremia
Yes 111 60 (54.0)  0.001 21.7 [6.8–69.3]  0.0001
No 2086 347 (16.6)

Definition of abbreviations: ICU  intensive care unit; LOS  length of stay; NBSI  ICU-acquired bloodstream infection; SAPS II  Sim-
plified Acute Physiology Score II.
* Age and SAPS II were analyzed as continuous variables.
1588 AMERICAN JOURNAL OF RESPIRATORY AND CRITICAL CARE MEDICINE VOL 163 2001

TABLE 5. CHARACTERISTICS OF CASES (n  96) WITH ries (Table 5). There were 50 (52.1%) deaths among cases and
ICU-ACQUIRED BACTEREMIA AND OF MATCHED 16 (16.7%) among their matched control patients. Thus, the ex-
CONTROLS (n  96) cess overall mortality of ICU-acquired bacteremia was 35.4%.
Number (%) or Number (%) or The 42 cases having secondary (non-catheter-related) bactere-
Mean (SD) Mean (SD) mia had an excess ICU mortality of 55%, whereas in the 54
for Cases for Controls p Value patients with primary or catheter-related bacteremia, the mor-
Age 60.3 (19.2) 62.9 (18.2) 0.23 tality attributable to bacteremia was 20.4% (95% CI [3.7%–
Sex, male 65 (67.7) 60 (62.5) 0.45 37%]; p  0.03). In the subgroup of 28 patients with definite
Patient location prior or ICU 0.99 catheter-related bacteremia, the attributable mortality was es-
Direct admission 46 (48.0) 47 (43.5) timated at 11.5% (95% CI [14%–37%]; p  0.3) (Table 6).
Other wards 37 (38.5) 36 (37.5)
Case patients with ICU-acquired bacteremia stayed in the
Other hospital 13 (13.5) 13 (13.5)
Admission category 0.99
ICU a median of 5.5 d longer than control patients (Table 6).
Medical 59 (61.5) 58 (60.4) In surviving pairs only (n  41), the median ICU length of
Unscheduled surgery 15 (15.6) 15 (15.6) stay was 9.5 d longer in bacteremic cases than in matched con-
Scheduled surgery 5 (5.2) 6 (6.2) trols (31 [IQR 23.5–56] versus 21.5 [13–36] d, p  0.01). In pa-
Trauma 17 (17.7) 17 (17.7) tients with primary or catheter-related bacteremia, the median
Admission diagnosis 0.39
attributable ICU LOS was 9.5 d (28 [IQR 14–44] d versus 18.5
Medical 67 (69.8) 65 (67.7)
Cardiac failure 8 (8.3) 9 (9.4)
[IQR 13–39] d; p  0.34); and in surviving pairs only (n  26),
Respiratory failure 32 (33.3) 24 (25.0) the median ICU length of stay was 12.5 d longer in bacteremic
Digestive tract 3 (3.1) 6 (6.3) cases (31.5 [IQR 21–52] d) than in matched controls (19 [IQR
Central nervous system 8 (8.3) 10 (10.4) 11–32] d, p  0.025).
Metabolic disorders 7 (7.3) 2 (2.1)
Renal failure 2 (2.1) 2 (2.1)
Other medical 7 (7.3) 12 (12.5) DISCUSSION
Surgical 29 (30.2) 31 (32.3)
Because the impact of primary and catheter-related bactere-
Abdominal 4 (4.2) 5 (5.2)
Cardiac 3 (3.1) 3 (3.1) mia on outcome of patients is still debated, we designed this
Vascular 3 (3.1) 0 (0.0) multicenter study to assess the morbidity and mortality associ-
Other surgery 2 (2.1) 6 (6.3) ated with ICU-acquired bacteremia, and contrast primary and
Immunodepression 25 (26.0) 25 (26.0) 1 catheter-related bacteremia to secondary bacteremia. Our study
McCabe score 1 confirms others demonstrating that nosocomial bacteremia is
Nonfatal 55 (57.3) 55 (57.3)
an independent risk factor for mortality of ICU patients (OR 
Ultimately or rapidly fatal 45 (46.7) 45 (46.7)
SAPS II on admission 45.1 (17.3) 46.2 (17.2) 0.60 4.6), with an attributable mortality of about 35%. In the case–
ICU length of stay 32.2 (27.4) 26.6 (21.5) 0.023 control study, primary and catheter-related bacteremia had a
ICU deaths 50 (52.1) 16 (16.7) 0.001 significant, but more than 2-fold lower attributable mortality
Definition of abbreviations: ICU  intensive care unit; SAPS II  Simplified Acute Phys-
than secondary bacteremia (20.5% versus 55%); definite cath-
iology Score II. eter-related bacteremia was associated with an approximately
2-fold lower excess mortality (estimated at 11.5%) than pri-
mary bacteremia. The median excess length of ICU stay of pa-
disease as compared with the corresponding case, and a simi- tients surviving the bacteremic episode was 9.5 d overall, and
lar ( 5 points) SAPS II score; in addition, controls had to was similar in patients with primary or catheter-related bacte-
have an ICU length of stay at least as long as that of the corre- remia (12.5 d) and those with secondary bacteremia (7 d).
sponding case up to the occurrence of bacteremia. When sev- We estimated the mean attack rate of ICU-acquired BSI at
eral control patients were available, the patient having the 5 per 100 admissions (or 4.5/1000 patient-days) in ICU pa-
highest and closest admission SAPS II to the case patient was tients admitted for  48 h. In recent studies, the incidence of
selected. Adequately matched controls were available for 96 ICU-acquired bacteremia ranged from 2.7 to 7.4 per 100 ad-
of 111 (86.5%) bacteremic patients. missions (1, 9, 11, 21, 22). In a prior multicenter study from 24
hospitals in France (6), we found an overall incidence of NBSI
Excess Mortality and ICU Length of Stay in ICU patients of 4.1 (95% CI 3.3–5.0) per 100 ICU admis-
Attributable to Bacteremia sions, a figure about 10-fold higher than in the general hospital
The characteristics of cases and controls did not differ in terms population (0.44 bacteremia per 100 hospital admissions). How-
of variables used for matching and of major diagnostic catego- ever, ICUs participating in the present survey all belonged to

TABLE 6. CASE–CONTROL STUDY (96 PAIRS) OF MORTALITY AND LENGTH OF STAY (LOS) ATTRIBUTABLE TO ICU-ACQUIRED
BACTEREMIA, ACCORDING TO ITS SOURCE

Catheter-related
Bacteremia Primary Bacteremia Secondary Bacteremia All Bacteremias

Cases Controls Cases Controls Cases Controls Cases Controls

Mortality, n (%) 10/26 (38.5) 7/26 (26.9) 14/28 (50) 6/28 (21.4) 26/42 (61.9) 3/42 (7.1) 50/96 (52.1) 16/96 (16.7)
Attributable, % [95 CI] 11.5 [14.5, 37.5] 28.6 [5.6, 51.6] 54.8 [35.9, 73.6] 35.4 [23, 48]
p Value 0.095 0.11 0.006 0.0002
LOS, d
All pairs, median 32.5 17.5 21.5 20 24 21 24.5 19
Surviving pairs, [IQR] 33 [21–58] 19 [11–40] 29 [24–50] 21 [11–32] 29 [24–54] 22 [13–52] 31 [23.5–56] 21.5 [13–36]
p Value 0.23 0.016 0.13 0.01

Definition of abbreviations: IQR  interquartile range; LOS  length of stay.

Renaud and Brun-Buisson: Primary and Catheter-related Bacteremia
large university or university-affiliated hospitals, which may
account for the relatively high bacteremia rate (3, 6).
As noted above, much of the increase in incidence of noso-
comial bacteremia in hospitalized patients (3, 4) appears to be
due to ICU-acquired cases, particularly to intravascular cathe-
ter-associated infections. Overall, primary and definite cathe-
ter-related bacteremia accounted for 55% of all episodes re-
corded in this study, with primary bacteremia accounting for
29% and definite catheter-related bacteremia for 26%. In a
recent Spanish multicenter study, catheter-related and pri-
mary bacteremia accounted, respectively, for 37% and 28% of
all ICU-acquired bacteremia (9). In another study from the
United States (4), 20% of all episodes of NBSI were associ-
ated with catheter infection and 22% were considered primary
(non-catheter-related infection). In our study, the incidence of
definite catheter-associated bacteremia was 0.88 per 100 ICU
admissions, comparable to the 0.75 rate reported from hospi-
tals participating to the National Nosocomial Infections Sur-
veillance System (23). Although we did not use molecular sub-
typing of organisms to confirm the source of bacteremia, we
used strict definitions for catheter-related infections, and it is
unlikely that we overestimated the incidence rate of bactere-
mia. Accordingly, the incidence rate and distribution of noso-
comial and catheter-related bacteremia in our study are com-
parable with those recorded in several recent studies.
In the series focusing on ICU patients, the crude mortality
of bacteremia ranged between 25% and 62% (4, 6, 24). How-
ever, most prior studies were conducted at a single institution,
and few have specifically addressed the epidemiology and prog-
nosis of nosocomial bacteremia in ICU patients or examined
the influence of primary and catheter-related bacteremia on
the outcome of patients (1, 6, 25). To further examine the spe-
cific outcomes of primary and catheter-related bacteremia, we
included 2201 consecutive patients from 15 ICUs of large uni-
versity-affiliated hospitals. We found in both our cohort analy-
sis and case–control study a 35% overall mortality attributable
to ICU-acquired bacteremia. This figure is remarkably similar
to the 35% overall mortality attributable to nosocomial bacte-
remia reported by Pittet and coworkers in surgical ICU pa-
tients (1). In one study, the risk of death was 5-fold higher in
bacteremic patients compared with nonbacteremic patients
(22), comparable to our estimate of a 4-fold higher risk of
death. Secondary bacteremia was associated with a 2.7-fold higher
risk of death than primary or catheter-related bacteremia. In
one study, secondary bacteremia had a 1.6-fold higher risk of
mortality than primary bacteremia (12); this study, however,
was not restricted to ICU patients. Our results are consistent
with those of several previous studies showing a higher mor-
tality associated with secondary bacteremia (e.g., from an ab-
dominal or pulmonary source) than with primary bacteremia
(5, 6, 12, 16, 21, 26).
Two recent studies have addressed the specific question of
the outcome of patients with primary and/or catheter-related
bacteremia (13, 14). DiGiovine and coworkers studied 68 patients
with primary (including catheter-related) bacteremia matched
to 68 control patients on diagnosis and severity of illness using
variables recorded on the day prior to the occurrence of bacte-
remia (13); they found no excess mortality attributable to bac-
teremia (35.3% versus 30.9%, p  0.51). Similarly, in a case–
control study of 38 patients with catheter-related bacteremia,
matched to 75 control patients on variables recorded on ICU
admission, Soufir and coworkers found that bacteremia was
associated with a 2-fold higher risk of mortality; however, the
relative risk of death was of only 1.3 (95% CI 0.69–2.46) after
controlling for the severity of patients assessed 3 d before the
occurrence of bacteremia (14). Therefore, although their con-
1589
clusions were slightly different, both studies found a nonsignif-
icant impact of primary or catheter-related bacteremia on sur-
vival of patients. However, the methods used in these two
studies (i.e., matching cases and controls on the severity of ill-
ness just before the occurrence of bacteremia) may result in
overfitting, and actually underestimate the true influence of
bacteremia on outcome of patients.
We used both logistic regression and a nested case–control
study to assess the impact of the various categories of bactere-
mia on the outcome of patients. Our findings suggest that pri-
mary and catheter-related bacteremia do have a sizable attrib-
utable mortality. In our case–control analysis, primary and
catheter-related bacteremia were associated with an attribut-
able mortality of 20% (95% CI 3.7%–37%). When differenti-
ating definite catheter-related from primary bacteremia, we
estimated the mortality associated with the former at 11.5%.
This figure is substantially (although not significantly) lower
than that of non-catheter-related primary bacteremia (29%),
itself having an almost 2-fold lower attributable mortality than
that of secondary bacteremia (55%). Our estimate for cathe-
ter-related NBSI is very close to the 13% attributable mortal-
ity reported in a prior case–control study focusing on bactere-
mia caused by coagulase-negative staphylococci (8), the major
pathogen currently involved in catheter-associated bactere-
mia. However, there was a wide confidence interval around our
estimate (95% CI [14%–37%]; p  0.3), and a larger sample
of catheter-associated bacteremia would have been necessary
to estimate more accurately the mortality associated with this
infection. Although we used a different approach and did not
adjust on severity variables recorded just before the infection,
our findings appear consistent with those of Soufir and co-
workers, who estimated the attributable mortality associated
with catheter-related bacteremia at 10% to 20% (14).
In surveillance programs of nosocomial bacteremia, it is of-
ten difficult to differentiate catheter-related bacteremia from
true primary bacteremia (3, 4, 12), in the absence of a system-
atic policy for diagnosing catheter-associated bacteremia includ-
ing routine culture of all central venous catheters suspected of
infection. In many reports, primary bacteremia is therefore
pooled with bacteremia identified as secondary to an intravas-
cular catheter, on the premise that most cases are in fact asso-
ciated with an intravascular line infection. This pragmatic ap-
proach to surveillance may be suboptimal in the ICU, where
the incidence of both primary and secondary NBSI is highest,
and the diagnosis of catheter-related infection may be easier
to confirm. Our analysis of outcomes of bacteremia, contrast-
ing its various sources, would justify routinely differentiating
true catheter-associated bacteremia from primary bacteremia
in epidemiological and outcome studies in ICU patients.
After controlling for variables associated with a longer than
median ( 5 d) ICU stay in the whole cohort (the patient ori-
gin, admission category, age, SAPS II, and McCabe score), we
estimated the excess ICU length of stay incurred by bactere-
mia at 5.5 d overall and 9.5 d in surviving patients. The esti-
mated excess length of ICU stay in survivors (9 to 12 d) was
comparable for all sources of bacteremia. In a previous study
of bacteremic patients from one surgical ICU, the excess
length of ICU stay of bacteremic patients was 8 d, both in the
whole cohort studied and in surviving pairs (1). Our findings
are also similar to those recently reported by DiGiovine and
coworkers, who found a mean excess length of ICU stay of
about 10 d in survivors of primary bacteremia (13).
Understanding the risk factors and evaluating the impact of
nosocomial infection are important to allocate resources de-
voted to infection control and for the design of clinical trials
evaluating therapeutic or preventive strategies. As intravascu-
1590 AMERICAN JOURNAL OF RESPIRATORY AND CRITICAL CARE MEDICINE
lar device-associated infections are increasingly prevalent in
ICU patients, and new preventive strategies are emerging (27),
it is necessary to assess more accurately their impact on pa-
tients’ outcomes. Primary bacteremia—about half of which
appears in fact secondary to definite catheter-related infec-
tion—accounts for 25% to 50% of all nosocomial bacteremia
in ICUs. Compared with other sources of bacteremia, such in-
fections have a more modest, but significant impact on mortal-
ity of ICU patients, and a comparable influence on the length
of ICU stay. Our results further suggest that differentiating
definite catheter-related bacteremia from bacteremia with no
identified source may be useful in studies conducted in criti-
cally ill patients.
References
1. Pittet D, Tarara D, Wenzel RP. Nosocomial bloodstream infection in
critically ill patients: excess length of stay, extra costs, and attributable
mortality. JAMA 1994;271:1598–1601.
2. Mermel LA. Prevention of intravascular catheter-related infections. Ann
Intern Med 2000;132:391–402.
3. Banerjee SN, Emori TG, Culver DH, Gaynes RP, Jarvis WR, Horan TC,
Edwards JR, Tolson JS, Handerson T, Martone WJ, National Nosoco-
mial Infections Surveillance (NNIS) system. Secular trends in primary
bloodstream infections in the United States, 1980-1989. Am J Med
1991;91:86S–89S.
4. Pittet D, Wenzel RP. Nosocomial bloodstream infections: secular trends
in rates, mortality, and contribution to total hospital deaths. Arch In-
tern Med 1995;155:1177–1184.
5. Gatell JM, Trilla A, Latorre X, Almela M, Mensa J, Moreno A, Miro
JM, Martinez JA, Jimenez de Anta MT, Soriano E, San Miguel JG.
Nosocomial bacteremia in a large Spanish teaching hospital: analysis
of factors influencing prognosis. Rev Infect Dis 1988;10:203–210.
6. Brun-Buisson C, Doyon F, Carlet J, and the French Bacteremia-Sepsis
Study Group. Bacteremia and severe sepsis in adults: a multicenter
prospective survey in ICUs and wards of 24 hospitals. Am J Respir
Crit Care Med 1996;154:617–624.
7. Schaberg DR, Culver DH, Gaynes RP. Major trends in the microbial eti-
ology of nosocomial infections. Am J Med 1991;91:72S–75S.
8. Martin MA, Pfaller MA, Wenzel RP. Coagulase-negative staphylococcal
bacteremia: mortality and hospital stay. Ann Intern Med 1989;110:9–16.
9. Vallés J, Leon C, Alvarez-Lerma F, and The Spanish Collaborative
Group for Infections in Intensive Care Units of SEMIUC. Nosoco-
mial bacteremia in critically ill patients: a multicenter study evaluating
epidemiology and prognosis. Clin Infect Dis 1997;24:387–395.
10. Vincent JL, Bihari D, Suter PM, Bruining HA, White J, Nicolas MH,
Wolff M, Spencer RC. The prevalence of nosocomial infection in in-
tensive care in Europe: results of the European prevalence of infec-
tion in intensive care (EPIC) study. JAMA 1995;274:639–644.
11. Fagon JY, Chastre J, Vuagnat A, Trouillet J-L, Novara A, Gibert C. Noso-
comial pneumonia and mortality among patients in intensive care
units. JAMA 1996;275:866–869.
12. Pittet D, Li N, Wenzel RP. Association of secondary and polymicrobial
nosocomial bloodstream infections with higher mortality. Eur J Clin
Microbiol Infect Dis 1993;12:813–819.
13. DiGiovine B, Chenoweth C, Watts C, Higgins M. The attributable mor-
tality and costs of primary nosocomial bloodstream infections in the
intensive care unit. Am J Respir Crit Care Med 1999;160:976–981.
14. Soufir L, Timsit JF, Mahe C, Carlet J, Regnier B, Chevret S. Attribut-
able morbidity and mortality of catheter-related septicemia in criti-
VOL 163 2001
cally ill patients: a matched, risk-adjusted, cohort study. Infect Control
Hosp Epidemiol 1999;20:396–401.
15. Brun-Buisson C, Abroug F, Legrand P, Huet Y, Larabi S, Rapin M. Di-
agnosis of central venous catheter-related sepsis: critical level of quan-
titative tip cultures. Arch Intern Med 1987;147:873–877.
16. Roberts FJ, Geere IW, Coldman A. A three-year study of positive blood
cultures, with emphasis on prognosis. Rev Infect Dis 1991;13:34–46.
17. Raad II, Bodey GP. Infectious complications of indwelling vascular
catheters. Clin Infect Dis 1992;15:197-210.
18. Kreger BE, Craven DE, McCabe WR. Gram-negative bacteremia: IV.
Reevaluation of clinical features and treatment in 612 patients. Am J
Med 1980;68:344–355.
19. Le Gall JR, Lemeshow S, Saulnier F. A new simplified acute physiology
score based on a European-North American multicenter study. JAMA
1993;270:2957–2963.
20. Hosmer DW, Lemeshow S. Applied logistic regression. New York: John
Wiley & Sons; 1989.
21. Rello J, Ricart M, Mirelis B, Quintana E, Gurgui M, Net A, Prats G.
Nosocomial bacteremia in a medical-surgical intensive care unit: epi-
demiologic characteristics and factors influencing mortality in 111 epi-
sodes. Intensive Care Med 1994;20:94–98.
22. Pittet D, Thiévent B, Wenzel RP, Li N, Auckenthaler R, Suter PM. Bed-
side prediction of mortality from bacteremic sepsis: a dynamic analy-
sis of ICU patients. Am J Respir Crit Care Med 1996;153:684–693.
23. Jarvis WR, Edwards JR, Culver DH, Hughes JM, Horan T, Emori TG,
Banerjee S, Tolson J, Henderson T, Gaynes RP. Nosocomial infection
rates in adult and pediatric intensive care units in the United States.
Am J Med 1991;91:185S–191S.
24. Pittet D, Thiévent B, Wenzel RP, Li N, Gurman G, Suter PM. Impor-
tance of preexisting comorbidities for prognosis of septicemia in criti-
cally ill patients. Intensive Care Med 1993;19:265–272.
25. Smith RL, Meixler SM, Simberkoff MS. Excess mortality in critically ill
patients with nosocomial bloodstream infections. Chest 1991;100:164–167.
26. Pittet D, Li N, Woolson RF, Wenzel RP. Microbiological factors influ-
encing the outcome of nosocomial bloodstream infections: a 6-year
validated, population-based model. Clin Infect Dis 1997;24:1068–1078.
27. Veenstra DL, Saint S, Sullivan SD. Cost-effectiveness of antiseptic-
impregnated central venous catheters for the prevention of catheter-
related bloodstream infection. JAMA 1999;282:554–560.
APPENDIX: LIST OF PARTICIPANTS IN THE
ICU–BACTEREMIA STUDY GROUP
Hôpital de Bicêtre: C. Richard, A. Mercat, K. Samii, O. Mimoz,
P. Gillard, K. Amal, S. Leotard, and P. Nordmann; Hôpital
Bichat-Caude Bernard: C. Paugam, J. M. Desmonts, M.
Thuong, B. Régnier, C. Gibert, J. L. Trouillet, A. Scanvic, and
A. Andremont; Hôpital Broussais: A. Novara, J. Y. Fagon, L.
Gutmann, and A. Buu Hoï; Hôpital Cochin: Y. Ozier, F. Bau-
din, F. Bellenfant, J. F. Dhainaut, H. Blanchard, and A. Philip-
pon; Hôpital Henri Mondor: G. Dhonneur, A. Boyer, E. Gi-
rou, C. Brun-Buisson, P. Legrand, and C. Soussy; Hôpital de la
Pitié-Salpétrière: P. Derennes, T. Similowsky, J. Robert, J.
Nguyen, and V. Jarlier; Hôpital Rotschild: W. Rosenbaum and
J. C. Nicolas; Hôpital Saint-Antoine: E. Maury, G. Offenstadt,
D. Lesage, and J. C. Petit; Hôpital Saint-Louis: J. R. Legall, G.
Leleu, and P. Lagrange; Hôpital Saint-Joseph: J. Carlet, B.
Misset, M. Ben Ali, and J. Acar; Hôpital Tenon: C. Mayaud,
M. Denis, A. Parrot, and G. Arlet.