Download Early combination antibiotic therapy yields improved survival

Early combination antibiotic therapy yields improved survival compared
with monotherapy in septic shock: A propensity-matched analysis*
Anand Kumar, MD; Ryan Zarychanski, MD; Bruce Light, MD; Joseph Parrillo, MD; Dennis Maki, MD; Dave
Simon, MD; Denny Laporta, MD; Steve Lapinsky, MD; Paul Ellis, MD; Yazdan Mirzanejad, MD;
Greg Martinka, MD; Sean Keenan, MD; Gordon Wood, MD; Yaseen Arabi, MD; Daniel Feinstein, MD;
Aseem Kumar, PhD; Peter Dodek, MD; Laura Kravetsky, BSc; Steve Doucette, MSc; the Cooperative Antimicrobial
Therapy of Septic Shock (CATSS) Database Research Group
Background: Septic shock represents the major cause of infection-associated mortality in the intensive care unit. The possibility
that combination antibiotic therapy of bacterial septic shock improves outcome is controversial. Current guidelines do not recommend combination therapy except for the express purpose of broadening
coverage when resistant pathogens are a concern.
Objective: To evaluate the therapeutic benefit of early combination therapy comprising at least two antibiotics of different
mechanisms with in vitro activity for the isolated pathogen in
patients with bacterial septic shock.
Design: Retrospective, propensity matched, multicenter, cohort study.
Setting: Intensive care units of 28 academic and community
hospitals in three countries between 1996 and 2007.
Subjects: A total of 4662 eligible cases of culture-positive,
bacterial septic shock treated with combination or monotherapy
from which 1223 propensity-matched pairs were generated.
Measurements and Main Results: The primary outcome of
study was 28-day mortality. Using a Cox proportional hazards
model, combination therapy was associated with decreased 28-
*See also pp. 1905 and 1906.
From Section of Critical Care Medicine (AK, RZ, BL,
LK), Health Sciences Centre/St. Boniface Hospital, University of Manitoba, Winnipeg, Manitoba, Canada; Cancer
Care Manitoba (RZ), University of Manitoba, Winnipeg,
Manitoba, Canada; Cooper Hospital/University Medical
Center (AK, JP), Robert Wood Johnson Medical School,
UMDNJ, Camden, New Jersey; Section of Infectious Diseases (DM), University of Wisconsin, Madison, Madison,
WI; Section of Infectious Diseases (DS), Rush University,
Chicago, Illinois; Section of Critical Care Medicine (DL),
Jewish General Hospital, McGill University, Montreal,
Quebec City, Canada; Section of Critical Care Medicine
(SL), Mount Sinai Hospital, University of Toronto, Toronto,
Ontario Canada; Department of Emergency Medicine (PE),
University Health Network, University of Toronto, Toronto,
Ontario, Canada; Surrey Memorial Hospital (YM), Surrey,
British Columbia, Canada; Richmond General Hospital
(GM), Vancouver, British Columbia, Canada; Royal Columbian Hospital (SK), Vancouver, British Columbia, Canada;
Royal Jubilee Hospital/Victoria General Hospital (GW), University of British Columbia, Victoria, British Columbia,
Canada; Intensive Care Department (YA), King Saud Bin
Abdulaziz University for Health Sciences, Riyadh, Saudi
Crit Care Med 2010 Vol. 38, No. 9
day mortality (444 of 1223 [36.3%] vs. 355 of 1223 [29.0%]; hazard
ratio, 0.77; 95% confidence interval, 0.67– 0.88; p ⴝ .0002). The
beneficial impact of combination therapy applied to both Grampositive and Gram-negative infections but was restricted to patients
treated with ␤-lactams in combination with aminoglycosides, fluoroquinolones, or macrolides/clindamycin. Combination therapy was
also associated with significant reductions in intensive care unit (437
of 1223 [35.7%] vs. 352 of 1223 [28.8%]; odds ratio, 0.75; 95%
confidence interval, 0.63– 0.92; p ⴝ .0006) and hospital mortality
(584 of 1223 [47.8%] vs. 457 of 1223 [37.4%]; odds ratio, 0.69; 95%
confidence interval, 0.59 – 0.81; p < .0001). The use of combination
therapy was associated with increased ventilator (median and [interquartile range], 10 [0 –25] vs. 17 [0 –26]; p ⴝ .008) and pressor/
inotrope-free days (median and [interquartile range], 23 [0 –28] vs.
25 [0 –28]; p ⴝ .007) up to 30 days.
Conclusion: Early combination antibiotic therapy is associated
with decreased mortality in septic shock. Prospective randomized
trials are needed. (Crit Care Med 2010; 38:1773–1785)
KEY WORDS: antibiotic; combination; monotherapy; mortality;
sepsis; septic shock
Arabia; Moses H. Cone Memorial Hospital (DF), Greensboro, North Carolina; Laurentian University (AK), Biomolecular Sciences Program and Department of Chemistry
and Biochemistry, Sudbury, Ontario, Canada; Section of
Critical Care Medicine (PD), St. Paul’s Hospital, University
of British Columbia, Vancouver, British Columbia, Canada;
Ottawa Health Research Centre (SD), Ottawa, Ontario,
Canada.
Dr. Kumar has received grants from Wyeth, AstraZeneca, Pfizer, and Roche. Dr. Parrillo consulted with
Sangart, Artisan, Philips, and Immunetrics. He received
a grant from the Robert Wood Johnson Foundation. Dr.
Mirjanezad consulted for the advisory boards of Schering-Plough Corporation and Pfizer. He received honoraria/speaking fees from Merck, Schering-Plough Corporation, Bayer, and Wyeth. He also received grants
from Par101, C.diff, and INC Research. All other authors have no potential conflicts of interest to disclose.
None of the authors have financial or personal
relationships or affiliations that could influence (or
bias) the decision regarding the analysis or manuscript
in any regard.
Funding for this work was provided by the Manitoba Health Research Council, Health Sciences Centre
Foundation, and the Alfred Deacon Foundation. Additional support was provided by through unrestricted
grants from Eli-Lilly, Pfizer, Astellas Pharma, Merck,
Wyeth, Bayer, Bristol-Myers-Squibb, and AstraZeneca. Funding sources had no role in the design and
conduct of the study; collection, management, analysis, and interpretation of the data; and preparation,
review, or approval of the manuscript.
Dr. Kumar had full access to all the data in the
study is responsible for the integrity of the database
and the accuracy of the data analysis.
This specific research concept, the septic shock
database, and manuscript were developed by Dr. Kumar. Dr. Kumar and Mr. Doucette were responsible for
the methodological design issues and data analysis. All
authors assisted with data interpretation and manuscript revisions.
For information regarding this article, E-mail:
[email protected]
Copyright © 2010 by the Society of Critical Care
Medicine and Lippincott Williams & Wilkins
DOI: 10.1097/CCM.0b013e3181eb3ccd
1773
S
everal studies have shown that
appropriate antimicrobial therapy, defined as the use of at
least one agent with in vitro
activity for the isolated pathogen, leads to
lower mortality rates in life-threatening
infections associated with sepsis (1–5).
The incremental benefit of combination
as opposed to single-agent antimicrobial
therapy in such situations is controversial (6 –10).
Some clinical studies of bacterial infection, including endocarditis, Gramnegative bacteremia, and neutropenic
sepsis (11–13), and animal models of severe infection (14 –16) have supported
the possibility of clinically relevant antimicrobial synergism with appropriate
combinations of antibiotics. However,
two separate meta-analyses have failed to
demonstrate any consistent benefit with
combination therapy in immunocompetent patients with sepsis and/or Gramnegative bacteremia (17, 18).
The beneficial effect of early appropriate antimicrobial therapy appears to be
most important in critically ill patients,
particularly those with septic shock (5).
Given these data and in view of a recent
meta-regression study that suggests that
the beneficial effect of combination therapy may be restricted to patients with
life-threatening infection at highest risk
for death (19), we performed a propensity-matched study examining the impact
of early combination antibiotic therapy in
adult patients with bacterial septic shock
using a multinational septic shock database. The hypothesis of this study was
that early combination antimicrobial
therapy of septic shock as defined by the
use of at least two antibiotics (of different
mechanistic classes) with in vitro activity
against the isolated pathogen is associated with reduced mortality.
specific criteria for septic shock as described
by the 1991 Society of Critical Care Medicine/
American College of Chest Physicians Consensus Statement on Sepsis Definitions (20). The
process used to identify the final study population is outlined in Figure 1.
Data Elements and Definitions
Clinical infection definitions were adapted
from previous recommendations or studies
(21–23). To qualify as potential pathogens
causing shock, isolates from anatomical sites
and/or blood cultures were required to have
been obtained within 48 hrs of onset of shock.
A priori criteria were developed to uniformly determine the primary pathogen/
pathogens and to assess the appropriateness of
antimicrobial therapy across participating institutions (Appendices 1 and 2). The first use
of any appropriate antimicrobial therapy (i.e.,
with in vitro activity for the primary isolated
pathogen or pathogens) was determined for all
cases. For the purposes of this study, antibiotic monotherapy was defined as the administration of any single, appropriate, intravenous,
preferably bactericidal (see Appendix 2 for exceptions) antibiotic at any point after the onset of recurrent or persistent hypotension.
Combination therapy was defined as the concomitant use of two or more such antibiotics
of different mechanistic classes for at least 24
hrs after the onset of hypotension or until
death if the patient survived ⬍24 hrs after
hypotension documentation. The second
agent had to be started within 24 hrs of the
first antibiotic or within 24 hrs of the onset of
hypotension (if the first agent was initiated
before hypotension was documented). The
combination of two ␤-lactams or a ␤-lactam
and a glycopepide was not considered to represent antibiotic combination therapy (because all are cell-wall–active agents with similar mechanisms of action) for purposes of this
analysis. Typical permutations of antibiotics
meeting criteria for combination therapy include any two of a cell-wall–active agent (i.e.,
a ␤-lactam or glycopeptide), aminoglycoside,
fluoroquinolone, or macrolide/clindamycin.
Questionable cases or data elements were
reviewed and adjudicated by the principal investigator. Cases of septic shock associated
with negative or absent cultures and those
caused by yeast/fungi, anaerobes, and atypical
pathogens, such as Mycobacterium tuberculosis and Legionella species, were excluded (Fig.
1). Patients who did not receive any appropriate antimicrobial therapy before death were
also excluded.
Subsets of subject data examined in this
study have been utilized for several earlier
publications (5, 21, 24, 25). Data collection
methods have been described in those previous studies (5, 21). Data were collected by
trained research nurses/medical students us-
SUBJECTS AND METHODS
A retrospective review of adult (age 18 yrs
or older) patients with septic shock was performed. A waived consent protocol was approved by the Health Ethics Board of the University of Manitoba and at each individual
participating center. Consecutive adult septic
shock patients from 28 medical institutions in
Canada, United States, and Saudi Arabia for
periods between 1996 and 2007 were retrospectively identified using internal intensive
care unit registries/databases and/or International Classification of Diseases (9 or 10) coding strategies. Each institution contributed a
minimum of 50 cases. Each potential case was
screened to determine whether the case met
1774
Figure 1. Subject selection flow diagram. SCCM, Society of Critical Care Medicine; ACCP, American
College of Chest Physicians; ICU, intensive care unit. *Clostridia species, Bacteroides species, peptostreptococci, Clostridium difficile-associated septic shock, miscellaneous anaerobes. **Mycobacterium tuberculosis, Legionella species, Listeria, Bacillus species, Corynebacterium jeikeium.
Crit Care Med 2010 Vol. 38, No. 9
ing a standardized and piloted data extraction
template. Variables collected included patient
demographics, baseline comorbidities, Acute
Physiology and Chronic Health Evaluation II
score (26), physiologic/laboratory parameters,
and the use of hemodynamic or ventilatory
support. The site of infection, microbiological
culture results, and the time to appropriate
antimicrobial therapy from the onset of hypotension were also recorded.
Outcome Measures
The primary outcome variable was mortality over 28 days. Mortality stratified by severity
of illness (Acute Physiology and Chronic
Health Evaluation II score), time from initial
documentation of hypotension to first appropriate antimicrobial, and time between initiation of the first and second appropriate antibiotic were identified a priori as secondary
outcome measures. Other predetermined secondary end points included hospital and intensive care unit mortality and vasopressor/
inotropic support-free days in the first 30 days
after shock. Exploratory analyses were performed to examine the association of any potential benefit of combination therapy according to antibiotics utilized, clinical syndrome,
and primary pathogen.
Statistical Analysis
Baseline characteristics between patients
receiving monotherapy and combination antibiotic therapy were compared using Student’s
t test or Wilcoxon’s rank sum test for continuous variables, or the chi-square test for categorical variables. All reported p values were
two-tailed. Because combination therapy was
not randomly assigned, a propensity analysis
was undertaken to account for potential confounding factors and selection biases. The propensity matching and analytic methods used
in this study incorporated aspects from several
reference sources (24, 27, 28). A propensity
score for combination antibiotic therapy use
was developed using multivariable logistic regression. This score represents the probability
that a patient would receive combination therapy based on variables that were known or
suspected to influence group assignment
(monotherapy or combination therapy) or to
affect mortality risk. Variables included in the
derivation of the propensity score are shown
in Table 1. Among these were age, gender,
Acute Physiology and Chronic Health Evaluation II score, the number of day 1 organ failures, occurrence of shock before initiation or
during therapy with an appropriate antibiotic,
time to the initial dose of the first appropriate
antibiotic (after documentation of sepsisassociated hypotension), site of infection acquisition (community, nosocomial), preexisting medical conditions, infecting organism,
Crit Care Med 2010 Vol. 38, No. 9
presence of bacteremia, primary antibiotic
therapy, anatomical site of infection, volume
of fluid resuscitation in the first hour of hypotension, use of therapies including source
control, activated protein C, corticosteroids,
mechanical ventilation, and a variety of laboratory data including white blood cell count,
platelet count, the international normalized
ratio, serum creatinine, and serum bicarbonate. To account for temporal and geographic
practice variability, the date of intensive care
unit admission and hospital sites (region/
academic vs. nonacademic) were also incorporated as matching variables.
One of the two antibiotics in each combination therapy regimen had to be designated as the
primary agent in the combination for matching
purposes. The required monotherapy match was
prioritized as follows: ␤-lactam/vancomycin,
fluoroquinolones, macrolides/clindamycin, and
other. The priority-matched drug was considered the primary antibiotic and the additional
drug of the combination regimen was considered the secondary/supplemental antibiotic.
Aminoglycosides were always considered secondary/supplemental.
Propensity scores were used to match patients who received combination therapy to a
control patient receiving monotherapy using a
SAS macro (SAS, Cary, NC). A greedy matching procedure selected match pairs initially
identical to five decimal places of probability
(29). If no match existed at five decimal places,
then matching would occur at four decimal
places, and so on. If no match existed at one
decimal place, then that patient receiving
combination therapy was excluded from the
study. To compensate for immortal time bias
(30), matching was restricted so that the minimum duration of survival (from hypotension)
of the matched monotherapy case was consistently longer than the duration of time between hypotension onset and administration
of the second appropriate antibiotic for the
combination therapy patient (i.e., the matching monotherapy patient always lived long
enough to have had the same or greater opportunity to have received a second appropriate antibiotic). This was accomplished by ensuring the monotherapy cases utilized for
matching always lived at least a day (range, 1
min to 48 hrs) beyond the point that the
combination therapy patient received the
first dose of the second appropriate antibiotic. Using this strategy, 1223 of 1714
(71.4%) patients who received combination
therapy were able to be suitably matched
using propensity scores.
Mortality over 28 days was assessed using a
Cox proportional hazards model. Hazard models incorporated survival data over the complete duration of the study period (28 days) or
until the time of censoring (i.e., death). Mortality estimates stratified by the delay from
hypotension to the first antibiotic and the delay from the first to second antibiotic (in combination regimens) were assessed by the addition
of an interaction term to the hazard model (31).
A hazard or odds ratio ⬍1 signifies decreased
mortality in the combination therapy group
compared to the monotherapy group. Statistical
analyses were conducted using SAS version 9.1
(SAS Institute, Cary, NC). The confidence limits
and p values reported reflect ␣ level of 0.05.
RESULTS
Baseline Characteristics Before
Propensity Matching
Antibiotic combination therapy was
administered to 1714 of 4662 (36.8%)
patients with eligible bacterial septic
shock (Fig. 1). The remaining 63.2% received monotherapy. Baseline demographics, preexisting medical conditions,
and relevant clinical, physiologic, and
laboratory parameters in the unmatched
study population are summarized in Table 1. Males comprised 56.8% and 58.7%
of the combination and monotherapy
groups. The age and admission Acute
Physiology and Chronic Health Evaluation II score in the unmatched study population was 61.9 (⫾16.2) and 23.7 (⫾8.0),
respectively. Age was significantly
younger in the combination therapy
group compared with the monotherapy
group; Acute Physiology and Chronic
Health Evaluation II scores also trended
lower. The median time to appropriate
antibiotic therapy was significantly
shorter in the combination therapy
group, although there were no differences in fluid resuscitation volumes between groups in the first hour of hypotension.
Several epidemiologic and clinical differences between the groups existed in
the unmatched cohort. The baseline
prevalence of liver failure/cirrhosis, diabetes, and chronic renal failure was
higher in the monotherapy group,
whereas the prevalence of invasive/
metastatic malignancy, neutropenia, and
other immunosuppression was higher in
the combination therapy group (Table 1).
Recent elective or emergency surgery/
trauma was associated with an increased
probability of receiving monotherapy.
The documented presence of bacteremia,
in contrast, was associated with combination therapy. The platelet count was significantly lower and the international
normalized ratio trended lower in the
1775
Table 1. Unmatched and propensity score-matcheda baseline characteristics
Unmatched Cohort
Male, n
Age, yr, mean ⫾ SD
Shock date, median
Time to first antibiotic, hrs, median (IQR)
Acute Physiology and Chronic Health Evaluation
II Score, mean ⫾ SD
Total day 1 organ failures, median (IQR)
Duration of hospitalization before shock (IQR)
Infection acquisition site, n (%)
Community
Hospital (nosocomial)
Hospital/regional distribution
Central Canada (academic)
Central Canada (community)
Eastern Canada (mixed)
West Coast Canada (academic)
West Coast Canada (community)
East Coast United States (mixed)
Central United States (academic)
Outside North America (academic)
Preexisting medical conditions, n (%)
Liver failure/cirrhosisb
Chronic obstructive pulmonary diseasec
Diabetes mellitusc
Chronic renal insufficiencyd
Dialysis dependence
Malignancye
Immunosuppressionf
Neutropenia (⬍1000 neutrophils/␮L)
Recent surgical history, n (%)
Elective surgery
Emergency surgery
No elective or emergency surgery
Physiologic and laboratory parameters on
admission, median (IQR)
White blood cells, ⫻108 cells/L
Platelet count, ⫻1011 cells/L
Serum creatinine (␮mol/L)
Serum bicarbonate (mEq/L)
Serum bilirubin (␮mol/L)
International normalized ratio
Bacteremia, n (%)
Cointerventions, n (%)
Activated protein C
Steroids
Source control
Ventilator support (admission day)
First hr fluid resuscitation, L, mean ⫾ SD
Site of Infection, n (%)
Primary bloodstream infection
Catheter-related infection
Respiratory infection
Urinary tract infection
Intra-abdominal infection
Central nervous system infection
Soft tissue infection
Surgical site infection
Nonrespiratory intrathoracic infection
Other infection
1776
Propensity Matched Cohort
Monotherapy,
n ⫽ 2948
Combined Therapy,
n ⫽ 1714
p
Monotherapy,
n ⫽ 1223
Combined Therapy,
n ⫽ 1223
p
1730 (58.7%)
62.8 ⫾ 16
May 21, 2001
3.75 (0.4–10.3)
23.9 ⫾ 10
974 (56.8%)
60.3 ⫾ 16.4
September 5, 2000
1.53 (0–5.1)
23.3 ⫾ 9.2
.22
⬍.0001
⬍.0001
⬍.0001
.07
697 (57.0%)
61.6 ⫾ 16.3
October 31, 2001
2.15 (0.02–6)
23.4 ⫾ 9.5
686 (56.1%)
61.8 ⫾ 16.1
November 13, 2001
2.0 (0.08–6.2)
23.7 ⫾ 9.6
.65
.75
.65
.94
.46
3 (2–5)
1 (0–9)
3 (2–4)
1 (0–3)
.03
⬍.0001
3 (2–5)
1 (0–5)
3 (2–4)
1 (0–5)
.97
.85
1659 (56.3%)
1289 (43.7%)
1247 (72.8%)
467 (27.2%)
⬍.0001
841 (68.8%)
382 (31.2%)
822 (67.2%)
401 (32.8%)
.41
924 (31.3%)
323 (11%)
447 (15.2%)
375 (12.7%)
386 (13.1%)
172 (5.8%)
144 (4.9%)
177 (6%)
568 (33.1%)
221 (12.9%)
192 (11.2%)
171 (10%)
181 (10.6%)
131 (7.6%)
164 (9.6%)
86 (5%)
⬍.0001
407 (33.3%)
163 (13.3%)
146 (11.9%)
107 (8.7%)
145 (11.9%)
89 (7.3%)
85 (7%)
81 (6.6%)
407 (33.3%)
160 (13.1%)
154 (12.6%)
121 (9.9%)
131 (10.7%)
86 (7%)
93 (7.6%)
71 (5.8%)
.9
264 (9%)
417 (14.1%)
850 (28.8%)
462 (15.7%)
228 (7.7%)
424 (14.4%)
374 (12.7%)
95 (3.2%)
87 (5.1%)
219 (12.8%)
444 (25.9%)
227 (13.2%)
128 (7.5%)
300 (17.5%)
271 (15.8%)
96 (5.6%)
⬍.0001
.19
.03
.02
.74
.005
.003
⬍.0001
72 (5.9%)
165 (13.5%)
341 (27.9%)
178 (14.6%)
101 (8.3%)
209 (17.1%)
185 (15.1%)
61 (5%)
76 (6.2%)
169 (13.8%)
332 (27.1%)
184 (15%)
104 (8.5%)
205 (16.8%)
181 (14.8%)
55 (4.5%)
.73
.81
.68
.73
.83
.83
.82
.57
491 (16.7%)
258 (8.8%)
2231 (75.7%)
216 (12.6%)
88 (5.1%)
1424 (83.1%)
.0002
⬍.0001
⬍.0001
172 (14.1%)
82 (6.7%)
981 (80.2%)
182 (14.9%)
76 (6.2%)
977 (79.9%)
.57
.62
.84
13.9 (5.9–21.3)
162 (88–257)
141.4 (80–247)
16.2 (0–22)
14 (6–30.8)
1.4 (1.1–1.8)
1305 (44.3%)
13.5 (4.6–21.3)
151 (80–237)
141.4 (88–239)
16.7 (0–21.4)
14 (6.8–27)
1.3 (1.1–1.7)
951 (55.5%)
.23
.001
.24
.58
.96
.06
⬍.0001
14 (4.8–21.9)
159 (87–251)
141 (83–245)
16.0 (0–21.2)
14 (6–29)
1.3 (1.1–1.7)
622 (50.9%)
13.7 (5.0–21.0)
155 (84–244)
139 (85–240)
16.4 (0–21.6)
14 (6.8–28)
1.3 (1.1–1.7)
627 (51.3%)
.30
.39
.61
.87
.77
.57
.84
132 (4.5%)
937 (31.8%)
1141 (38.7%)
2183 (74.1%)
0.67 ⫾ 0.96
100 (5.8%)
518 (30.2%)
662 (38.6%)
1160 (67.7%)
0.68 ⫾ 0.94
.04
.27
.96
⬍.0001
.81
50 (4.1%)
355 (29%)
478 (39.1%)
853 (69.7%)
0.69 ⫾ 0.87
60 (4.9%)
361 (29.5%)
475 (38.8%)
852 (69.7%)
0.67 ⫾ 0.93
.21
.79
.90
.96
.71
152 (5.2%)
124 (4.2%)
1214 (41.2%)
379 (12.9%)
681 (23.1%)
41 (1.4%)
281 (9.5%)
48 (1.6%)
17 (0.6%)
11 (0.4%)
113 (6.6%)
72 (4.2%)
639 (37.3%)
304 (17.7%)
332 (19.4%)
13 (0.8%)
201 (11.7%)
21 (1.2%)
8 (0.5%)
11 (0.6%)
⬍.0001
68 (5.6%)
50 (4.1%)
449 (36.7%)
224 (18.3%)
272 (22.2%)
13 (1.1%)
119 (9.7%)
16 (1.3%)
7 (0.6%)
5 (0.4%)
70 (5.7%)
52 (4.3%)
439 (35.9%)
229 (18.7%)
275 (22.5%)
12 (1%)
116 (9.5%)
16 (1.3%)
8 (0.7%)
6 (0.5%)
1.00
Crit Care Med 2010 Vol. 38, No. 9
Table 1.—Continued
Unmatched Cohort
Primary pathogen, n (%)
Streptococcus pyogenes (group A streptococci)
Non-group A b-hemolytic streptococci
Viridans streptococci
Streptococcus pneumoniae
Staphylococcus aureus
Enterococcus species
Other Gram-positivesg
Escherichia coli
Klebsiella species
Enterobacter species
Other Enterobacteriaceaeh
Pseudomonas aeruginosa
Haemophilus species
Other non-Enterobacteriaceaei
Neisseria meningitidis
Moraxella catarrhalis
Primary antibiotic, n (%)
Penicillinsj
Anti-staphylococcal penicillinsk
␤-lactam/␤-lactamase inhibitorsl
Cephalosporins, 1st generationm
Cephalosporins, 2nd generationn
Cephalosporins, 3rd generationo
Cephalosporins, anti-pseudomonal/monobactamp
Carbapenemsq
Vancomycin
Fluorquinolonesr
Macrolidess/clindamycin
Othert
Secondary antibiotic, n (%)
Aminoglycosides
Fluoroquinolonesr
Macrolidess/clindamycin
Othert
Monotherapy,
n ⫽ 2948
Combined Therapy,
n ⫽ 1714
47 (1.6%)
50 (1.7%)
56 (1.9%)
170 (5.8%)
803 (27.2%)
179 (6.1%)
4 (0.1%)
681 (23.1%)
269 (9.1%)
118 (4%)
150 (5.1%)
226 (7.7%)
76 (2.6%)
81 (2.7%)
27 (0.9%)
11 (0.4%)
141 (8.2%)
57 (3.3%)
39 (2.3%)
281 (16.4%)
129 (7.5%)
33 (1.9%)
3 (0.2%)
490 (28.6%)
179 (10.4%)
68 (4%)
97 (5.7%)
125 (7.3%)
32 (1.9%)
28 (1.6%)
6 (0.4%)
6 (0.4%)
86 (2.9%)
117 (4%)
603 (20.5%)
51 (1.7%)
146 (5%)
513 (17.4%)
172 (5.8%)
437 (14.8%)
483 (16.4%)
266 (9%)
55 (1.9%)
19 (0.6%)
188 (11%)
40 (2.3%)
357 (20.8%)
28 (1.6%)
77 (4.5%)
520 (30.3%)
173 (10.1%)
172 (10%)
85 (5%)
67 (3.9%)
6 (0.4%)
1 (0.1%)
Propensity Matched Cohort
Monotherapy,
n ⫽ 1223
Combined Therapy,
n ⫽ 1223
⬍.0001
45 (3.7%)
33 (2.7%)
30 (2.5%)
141 (11.5%)
139 (11.4%)
29 (2.4%)
2 (0.2%)
373 (30.5%)
144 (11.8%)
55 (4.5%)
67 (5.5%)
102 (8.3%)
32 (2.6%)
22 (1.8%)
4 (0.3%)
5 (0.4%)
49 (4%)
30 (2.5%)
30 (2.5%)
141 (11.5%)
128 (10.5%)
30 (2.5%)
2 (0.2%)
386 (31.6%)
139 (11.4%)
51 (4.2%)
71 (5.8%)
98 (8%)
31 (2.5%)
25 (2%)
6 (0.5%)
6 (0.5%)
1.00
⬍.0001
62 (5.1%)
40 (3.3%)
289 (23.6%)
21 (1.7%)
60 (4.9%)
339 (27.7%)
119 (9.7%)
154 (12.6%)
82 (6.7%)
50 (4.1%)
6 (0.5%)
1 (0.1%)
71 (5.8%)
36 (2.9%)
293 (24%)
18 (1.5%)
62 (5.1%)
332 (27.1%)
116 (9.5%)
152 (12.4%)
76 (6.2%)
60 (4.9%)
6 (0.5%)
1 (0.1%)
1.00
p
683 (39.9%)
651 (38.0%)
350 (20.4%)
30 (1.8%)
p
526 (43.0%)
498 (40.7%)
174 (14.2%)
25 (2.0%)
IQR, interquartile range.
a
Variables included in propensity derivation but not shown include the occurrence of shock while using appropriate antimicrobial therapy,
appropriateness of initial antibiotic therapy, the presence of congestive heart failure, and admission serum lactate and albumin concentrations; bdefined
by appropriate history in context of clinical symptoms/signs of liver dysfunction per attending physicians; cmedication-dependent; dcreatinine 1.5⫻ normal
value; einvasive or metastatic; facquired immune deficiency syndrome, malignancy or autoimmune-related chemotherapy, ⬎20 mg/day chronic prednisone
equivalent or major organ transplant; gStaphylococcus lugdenesis, Leuconostoc, and Micrococcus species; hSerratia, Proteus, Citrobacter, Morganella,
Salmonella, Providencia, and Hafnia species; iAcinetobacter, Stenotrophomonas, Aeromonas, Burkholderia, and Acaligenes; jpenicillin, ampicillin,
piperacillin, ticarcillin, and mezlocillin; kcloxacilin, nafcillin, and oxacillin; lpiperacilln/tazobactam, ticarcilin/clavulanate, and ampicillin/sulbactam;
m
cefazolin; ncefuroxime, cefoxitin, and cefotetan; ocefotaxime, and ceftriaxone; pceftazidime, cefepime, and aztreonam; qmeropenem, imipenem/cilastatin,
and ertapenem; rlevofloxacin, ciprofloxacin, gatifloxacin, trovofloxacin, and ofloxacin; sazithromycin, erythromycin, and clarithromycin; tcolistin, trimethoprim/sulfamethoxazole, linezolid, daptomycin, quinipristin/dalfopristin, and rifampin. Acute Physiology and Chronic Health Evaluation. Patients
were assessed on the day of onset of shock. The range of scores for this test is 0 to 71.
combination therapy group. All patients
required vasoactive medications because
of hypotension. The need for mechanical
ventilation on admission was higher in
the monotherapy group. Nosocomial and
healthcare-associated infections were
more likely than community-acquired
infections to be in the monotherapy
groups. There were no statistically significant differences in the use of stress
dose steroids and the provision of
source control between groups, but the
use of activated protein C was significantly higher with combination therCrit Care Med 2010 Vol. 38, No. 9
apy. There were significant differences
between groups in distribution of the
anatomical site of infection, primary
pathogenic organism, primary antibiotic used, and hospital sites (Table 1).
Baseline Characteristics After
Propensity Matching
Suitable propensity matches were
found for 1223 (71.4%) of 1714 patients
receiving combination therapy. The c
statistic for the propensity derivation
model was 0.785. The range of propen-
sity scores was similar in both the combination and the monotherapy groups
(each, 0.01– 0.96). The matching process eliminated all significant differences that existed between the combination and monotherapy groups
regarding patient demographics, epidemiologic factors, preexisting medical
conditions, or relevant clinical, physiologic and laboratory parameters (Table
1). Penicillin or carbapenem therapy of
enterococci accounted for 29 of the 33
cases in which a static regimen was
included in the matched cohorts.
1777
Figure 2. Adjusted Cox proportional hazards of mortality associated with combination antibiotic
therapy of septic shock.
Combination Antibiotic Therapy
and Mortality
In the propensity-adjusted Cox model,
mortality over 28 days was significantly
reduced with combination therapy (444
of 1223 [36.3%] vs. 355 of 1223 [29.0%];
hazard ratio, 0.77; 95% confidence interval, 0.67– 0.88; p ⫽ .0002) (Fig. 2). Stratification by Acute Physiology and Chronic
Health Evaluation II score also revealed
consistently reduced 28-day mortality in
each assessed tertile, with significant differences in the middle and highest risk
tertile groups (Table 2).
Stratification by time from hypotension to first appropriate antimicrobial
demonstrated evidence of an increased
effect with greater delays (test of interaction p ⫽ .39). Stratification by the delay
between the first and second antimicrobial in combination therapy revealed reduced efficacy of combination therapy,
with increasing delays out to 24 hrs (test
for interaction p ⫽ .03) (Table 2). The
absolute reduction in mortality in patients with the most rapid administration
of the second drug (delay of 0.0 –1.0 hr)
was 10.1%, with a corresponding hazard
ratio of 0.68 (95% confidence interval,
0.53– 0.89; p ⫽ .004), whereas the absolute reduction in mortality for the most
delayed second drug (maximum delay, 24
hrs) was 2.5%, with a hazard ratio of 0.89
(95% confidence interval, 0.69 –1.15; p ⫽
.37) (Table 2). Combination therapy reduced mortality associated with death attributable to refractory shock, sepsisrelated organ failure, and apparent
nonsepsis-related causes (Table 3).
In the propensity matched cohort, the
proportion of ventilated patients success1778
fully liberated from mechanical ventilation
was higher in the combination therapy
group compared with the monotherapy
group (67.8% vs. 61.8%; odds ratio, 1.34;
95% confidence interval, 1.12–1.61; p ⫽
.001). Successful discontinuation of vasopressor/inotrope support was also higher in
the combination therapy group (80.1% vs.
75.3%; odds ratio, 1.32; 95% confidence
interval, 1.09- 1.61; p ⫽ .005). Similarly the
number of ventilator and pressor/inotropefree days (in the first 30 days) was significantly greater in patients receiving combination therapy (Table 4). The median
hospital length of stay in survivors (total
n ⫽ 1647) was significantly shorter in the
combination therapy group (22 days; interquartile range [IQR], 13– 44; vs. 26 days;
IQR, 14 – 49; p ⫽ .01), but the median intensive care unit length of stay was not
(monotherapy: 7 days; IQR, 4 –15; vs. combination therapy: 7 days; IQR, 4 –13; p ⫽
.29). Log rank analysis of the fraction of
surviving patients in each group remaining
on pressor/inotropic support to 28 days
showed a significant (p ⫽ .03) advantage
for combination therapy (Fig. 3). An approximate 5% divergence became apparent
by day 5.
Of the eight hospital/regional groupings
shown in Table 1, six demonstrated evidence of combination therapy benefit in the
unadjusted analysis, with five retaining significance (p ⬍ .05) after propensity matching. The remaining three did not demonstrate a significant advantage to either
combination therapy or monotherapy.
The beneficial effect of combination
therapy was predominantly seen with
␤-lactams as primary therapy including
both penicillins and cephalosporins (Fig.
4). When ␤-lactams were included as the
primary agent, combination therapy with
aminoglycosides, fluoroquinolones, and
macrolides/clindamycin as secondary/
supplemental agents was associated with
a superior outcome compared to monotherapy with the ␤-lactam (Fig. 5).
The benefit of combination therapy
was significant for bacteremic and nonbacteremic infections (Fig. 6). Similarly,
both pneumonia and nonpneumonia infections demonstrated superiority of
combination therapy (Fig. 6). A significant benefit was also seen in both Grampositive and Gram-negative organisms
and in Streptococcus pneumoniae and
Enterobacteriaceae infections in particular (Fig. 7). For both clinical infections
and specific organisms, most groups
trended in favor of combination therapy.
DISCUSSION
In this retrospective, propensitymatched cohort study of septic shock, the
use of combination antibiotic therapy was
associated with reduced 28-day mortality.
Intensive care unit and hospital mortality
were similarly reduced with combination
therapy. The benefit of combination therapy was greatest with shorter periods between administration of the two drugs.
Mortality differences narrowed with increasing delay from the first to the second
antibiotic (Table 2). A shorter period between documentation of hypotension and
administration of the first drug was associated with improved survival but a relative
reduction in the benefit of combination
therapy (Table 2). Combination therapy
was associated with more pressor-free and
ventilator-free days, an increase in successful liberation of ventilatory support, and
discontinuation of vasoactive medications.
The benefit of combination therapy appeared to be associated primarily with combinations of ␤-lactams with aminoglycosides, fluoroquinolones, and macrolides/
clindamycin (Fig. 5). However, the effect
was consistent across most clinical infections and pathogens.
Although the use of combinations of
antibiotics is relatively common in septic
shock, the rationale is typically to ensure
that the pathogenic organism is covered
by at least one active drug in the initial
regimen (as per current guidelines and
recommendations) (32, 33). An effort to
ensure that the probable pathogen or
pathogens are covered by more than one
active agent (effective combination therCrit Care Med 2010 Vol. 38, No. 9
Table 2. Mortality outcomes
Mortality Rate
n of Deaths/Total n of Patients (%)
Sample
Size, n
Monotherapy
Combination Rx
Hazard Ratio (95%
Confidence Interval)
p
4662
2446
1277/2948 (43.3%)
444/1223 (36.3%)
461/1714 (26.9%)
355/1223 (29.0%)
0.57 (0.51–0.63)
0.77 (0.67–0.88)
⬍.0001
.0002
755
747
818
66/366 (18.0%)
114/393 (39.0%)
242/398 (60.8%)
52/389 (13.4%)
78/354 (22.0%)
213/420 (50.7%)
0.72 (0.50–1.04)
0.73 (0.55–0.98)
0.78 (0.64–0.93)
.08
.03
.007
600
610
642
57/297 (19.2%)
98/312 (31.4%)
177/316 (56.0%)
57/303 (18.8%)
69/298 (23.2%)
150/326 (46.0%)
0.97 (0.67–1.40)
0.71 (0.52–0.96)
0.78 (0.62–0.96)
.87
.03
.02
1186
1147
1147
1147
332/925 (35.8%)
332/925 (35.8%)
332/925 (35.8%)
332/925 (35.8%)
67/261 (25.7%)
68/222 (30.6%)
67/222 (30.2%)
74/222 (33.3%)
0.68 (0.53–0.89)
0.84 (0.65–1.09)
0.82 (0.63–1.07)
0.89 (0.69–1.15)
.004
.19
.14
.37
1663
783
278/841 (33.1%)
166/382 (43.5%)
221/822 (26.9%)
134/401 (33.4%)
0.79 (0.66–0.94)
0.73 (0.58–0.91)
.008
.006
594
1852
112/298 (37.6%)
332/925 (35.9%)
79/296 (26.7%)
276/927 (29.8%)
0.67 (0.50–0.89)
0.80 (0.69–0.94)
.007
.007
28-Day mortality
Unadjusted
Propensity score adjusted
Matched cohort 28-day mortality analysis
stratified by Acute Physiology and
Chronic Health Evaluation II
5–20
21–27
28⫹
Matched cohort 28-day mortality
stratified by delay between
documented hypotension and 1st
appropriate antibiotica
0.000–1.99
2–5.99
6⫹
Matched cohort 28-day mortality
stratified by time from 1st to 2nd
appropriate antibiotic in
combination therapya
0.000–1
1.001–4
4.001–10
⬎10
Matched cohort 28-day mortality
stratified by community/nosocomial
infection acquisition
Community
Nosomial
Matched cohort 28-day mortality
stratified by whether the initial
appropriate antibiotic was
administered before or after first
documented hypotension
Before
After
a
Excluding those where first appropriate antibiotic was administered before hypotension.
Table 3. Mortality outcomes
Mortality Rate by Therapy
n of Deaths/Total n of Patients (%)
Intensive care unit mortality
Hospital mortality
Death from:
Refractory shock
Sepsis-related organ failure
Nonsepsis-related organ failure
Sample
Size, n
Monotherapy
Combination Rx
Odds Ratio (95%
Confidence Interval)
p
2446
2446
437/1223 (35.7%)
584/1223 (47.8%)
352/1223 (28.8%)
457/1223 (37.4%)
0.75 (0.63–0.88)
0.69 (0.59–0.81)
.0006
⬍.0001
2446
2446
2446
311/1223 (25.4%)
184/1223 (15.0%)
89/1223 (7.3%)
258/1223 (21.1%)
137/1223 (11.2%)
62/1223 (5.1%)
0.78 (0.65–0.95)
0.71 (0.56–0.90)
0.68 (0.49–0.95)
.01
.005
.02
apy) is not advised based on several metaanalyses that have failed to show a benefit
of combination therapy in severe infections, including sepsis and Gram-negative bacteremia (17, 18).
This study and a recently completed
meta-regression/meta-analysis (19) suggest
that combination antibiotic therapy may be
advantageous in septic shock. The basis on
which combination therapy provides a surCrit Care Med 2010 Vol. 38, No. 9
vival benefit at higher levels of mortality
risk can potentially be related to several
mechanisms. These include an increased
likelihood that the infecting pathogen will
be susceptible to at least one of the components of the regimen; prevention of
emergence of resistant superinfection (34 –
36); potential beneficial immunomodulatory effect of the secondary agent (37, 38);
and generation of an additive or even syn-
ergistic antimicrobial effect of the combination (i.e., more rapid killing of the pathogen) (11, 15, 39 – 41).
With regard to this analysis, the possibility of increased breadth of coverage is
immaterial. The study was restricted to
culture-positive infections with standard
sensitivity testing so that all patients
were known to have been treated with
one (monotherapy) or more (combina1779
Table 4. Secondary outcomes
Within 30 days
Nonmechanical intubation days
Nonvasoactive medication days
Days alive out of hospital
Days alive out of intensive care unit
Sample
Size
Monotherapy
Combined
Therapy
p
2446
2446
2446
2446
10 (0–25)
23 (0–28)
0 (0–9)
14 (0–24)
17 (0–26)
25 (0–28)
0 (0–13.6)
19 (0–25)
.008
.007
⬍.0001
.0003
Data displayed as median (interquartile range).
Figure 3. Log-rank assessment of persistence of pressor/inotrope dependence associated with combination therapy of septic shock. Combination therapy was associated with more rapid liberation from
pressor/inotrope support.
Figure 4. Primary antibiotic and outcome. The use of ␤-lactams as part of combination therapy was
associated with reduced hazard ratio of death. A significant association with survival was seen with use
of most penicillins and cephalosporins in such therapy. The benefit did not extend to antipseudomonal
␤-lactamase inhibitors, cephalosporins, and carbapenems. Combinations in which an antibiotic other
than a ␤-lactam was the primary agent also did not show evidence of benefit. Monotherapy represents
the reference group. MT, monotherapy; CT, combination therapy; staph, Staphylococcus; gen, generation; ceph, cephalosporin; Ps, pseudomonal; HR, unadjusted hazard ratio; CI, confidence interval.
For antibiotics belonging to each group, refer to Table 1.
tion therapy) antibiotics appropriate to
their primary pathogen. Although it is
difficult to rule out the possibility that
increased late development of infections
1780
with resistant organisms could contribute to increased mortality in the monotherapy group, the magnitude of the effect required to produce such a large
change in mortality appears improbable.
Finally, immunomodulatory activity has
been ascribed to macrolides (38, 42– 45)
and, to a lesser extent, fluoroquinolones
(37, 38, 46). Data supporting direct immunomodulatory effects of aminoglycosides are very limited (47, 48). Even so,
clinically relevant immunomodulatory
effects of the secondary agent would seem
to be improbable as a cause of the beneficial effect because such effects would
seem unlikely to exist across all groups of
secondary agents (aminoglycosides, fluoroquinolones, and macrolides/clindamycin).
In addition, an immunomodulatory basis of
benefit of combination therapy might also
be expected to extend across all ␤-lactams
rather than only the least powerful agents
(in terms of standard pharmacokinetic indices, i.e., time above minimal inhibitory
concentration).
Given the known benefit of synergistic
therapy in certain infections, including
bacterial endocarditis, it would seem
likely that the beneficial effect is related
to faster clearance of organisms in the
combination therapy patients. Antimicrobial synergy appears to be best-established for ␤-lactam/aminoglycoside
combinations (49 –52). However, similar data on synergistic activity have
emerged for combinations of a ␤-lactam and fluoroquinolone (53–59).
There are also data suggesting additive
effects (60) or even potential synergism
(61– 64) for ␤-lactam/macrolides combinations in certain circumstances. Although
speculative, synergistic antimicrobial combinations may eradicate the underlying
bacterial pathogens more quickly than a
single drug, resulting in more rapid resolution of infection-associated physiologic
instability (Fig. 3), leading to improved
clinical cure and survival.
Interestingly, carbapenems (almost
entirely imipenem/cilastatin and meropenem), extended-spectrum ␤-lactam/␤lactamase inhibitor combinations, and
anti-pseudomonal cephalosporins, which
tend to demonstrate optimal pharmacokinetic indices (with presumably maximal kill rates) for most septic shock
pathogens, yielded the weakest evidence
of benefit of combination therapy. This
effect cannot be explained on the basis of
immune modulation but may be explainable on the basis of a second agent’s inability to substantially further increase
the rate of bacterial clearance (cidality)
when the most potent ␤-lactams are part
of a combination regimen.
Crit Care Med 2010 Vol. 38, No. 9
Figure 5. Supplemental antibiotic and outcome. The use of aminoglycoside (AG), fluoroquinolone
(FQ), or a macrolide/clindamycin (ML/CL) in addition to a ␤-lactam was associated with a reduced
hazard ratio for death compared to ␤-lactam alone. No other drug combinations demonstrated
evidence of significant benefit. Primary antibiotic group in italics. Supplemental agents listed with “⫹”
in normal text. Monotherapy with primary antibiotic represents reference. The numbers of deaths/total
number of cases in the primary monotherapy group and in each supplemental combination therapy
group are denoted. HR, unadjusted hazard ratio; CI, confidence interval. For antibiotics belonging to
each group, refer to Table 1.
Figure 6. Association of combination therapy with outcome stratified by clinical syndrome and
characteristics. Among clinical syndromes, a significantly reduced hazard ratio for death was seen
among respiratory tract infections treated with combination therapy; similar, but nonsignificant,
trends were seen with septic shock attributable to other infections. In aggregate, all nonrespiratory
infections also demonstrated a reduced hazard ratio with combination therapy. Similar results were
found when the analysis was restricted to septic shock patients who were either bacteremic or
non-bacteremic and whether or not source control was required. MT, monotherapy; CT, combination
therapy; PBSI, primary bloodstream infection (no apparent anatomical source); CRI, intravascular
catheter-related infection; RTI, respiratory tract infection; UTI, urinary tract infection; IAI, intraabdominal infection; CNSI, central nervous system infection; SSTI, skin and soft tissue infection; SSI,
surgical site infection; ITI, intrathoracic infection (exclusive of respiratory tract); SC⫹ inf, source
control requiring infection; SC⫺ inf, nonsource control requiring infection; bact⫹, infection associated with documented bacteremia; bact⫺, infection not associated with documented bacteremia; HR,
unadjusted hazard ratio; CI, confidence interval.
Crit Care Med 2010 Vol. 38, No. 9
As with any retrospective analysis, this
study has limitations and weaknesses that
merit attention. There are at least two
potential time-dependent biases associated with this study. Immortal time bias
is, in essence, a mathematical problem in
which survival duration may be linked to
an inappropriate reference point yielding
an inaccurate time-dependent survival
probability (30). A retrospective study of
critically ill patients with a very high
early mortality risk may also demonstrate
a survival duration-related selection bias.
In this case, the group of patients who are
known to have lived long enough to receive a second antibiotic during septic
shock may be selected to more likely live
to a given temporal end point than the
group who cannot be known to have lived
long enough to receive the second drug.
Several statistical approaches can be used
to control for immortal time bias. One
conservative approach to control for both
types of potential biases is to ensure that
subjects in both groups live long enough
to potentially have received the second
antibiotic. For that reason, each combination therapy case was matched to a
monotherapy patient who lived, at a minimum, to the equivalent time point (after
initial documentation of hypotension) at
which the patient receiving combination
therapy was given the initial dose of the
second appropriate antibiotic, i.e., the
matched patient always lived long
enough to have received the same or better opportunity to have been administered a second drug. In fact, because the
matching of each monotherapy case patient was limited to subjects who lived at
least 1 day past the point at which the
combination therapy patients received
the second drug (to be conservative), a
consistent survival bias in favor of monotherapy is created.
Despite sophisticated methods to account for individual patient differences,
retrospective studies also may be confounded by indication. Some comorbid
conditions may indicate or contraindicate
the addition of a second antibiotic and
also could be associated with clinical outcomes. For example, aminoglycosides
and, less so, fluoroquinolones may be
problematic and, therefore, less frequently used in patients with acute renal
failure. These patients would also be expected to have an increased mortality
risk. Alternately, increased severity of illness might increase the probability of administration of a second agent generating
a spurious association between combina1781
# deaths/total (%)
MT
CT
3.
4.
5.
6.
0.1
1
10
Hazard Ratio
Figure 7. Association of combination therapy with outcome stratified by bacterial pathogen. In aggregate,
both Gram-positive and Gram-negative pathogens yielded evidence of reduced hazard ratio for death with
combination therapy. Similarly, combination therapy of enterobacteriaceae group pathogens was associated with improved survival, whereas nonenterobacteriaceae group pathogens trended in a similar direction. The only individual pathogen that yielded significant evidence of benefit of combination therapy was
Streptococcus pneumoniae. GAS, group A ␤-hemolytic streptococci; nonGAS strep, ␤-hemolytic streptococci other than group A; sp, species; Ps, Pseudomonas; EB, Enterobacteriaceae; HR, unadjusted hazard
ratio; CI, confidence interval. For organisms in each group, refer to Table 1.
tion therapy and mortality. Other biases
may be less obvious. We attempted to
control for potential confounders
through propensity matching of a variety
of variables (epidemiologic factors including hospital site, laboratory measures, severity of illness, specific pathogen, clinical infection syndrome, and
antibiotic used). Nonetheless, it is possible that residual confounders not recorded in the dataset could exist. Propensity methods are unable to account for
these unknown factors. An obvious weakness of any retrospective design is that
the allocation of patients and the use of
combination therapy cannot be randomized, nor can the antibiotic regimens be
standardized. Further, in this study, indications for the use of combination therapy could not be clearly defined.
This study also has important strengths.
A comprehensive clinical database allowed
for the identification of a large number of
patients eligible for analyses in this study.
Baseline differences between the combination and monotherapy groups existed,
which could have the potential to bias mortality estimates; however, the large sample
size allowed for a rigorously conducted propensity matched analyses whereby patients
were successfully matched for ⬎30 clinically
relevant parameters. Although no retrospective method can replace the advantage of prospective randomization, propensity analyses
1782
have been demonstrated to be an effective
means of reducing bias in baseline characteristics when assessing treatment effects (28,
65). In this analysis, all significant baseline
differences between study groups were adequately reconciled using this method. The
inclusion of patients from multiple hospitals
with a variety of clinical infections and pathogens adds further to the applicability of the
findings as they relate to septic shock.
Our data strongly suggest that early empirical combination antibiotic therapy with
two antibiotics of different mechanisms of
action is associated with superior outcomes
compared with monotherapy in the treatment of bacterial septic shock. These observational data point to the need for a large,
randomized, controlled trial to examine
this issue in patients with septic shock.
7.
8.
9.
10.
11.
12.
13.
14.
15.
ACKNOWLEDGMENTS
We thank Christine Mendez, Sheena
Ablang, Debbie Friesen, and Lisa Halstead
for data entry.
16.
REFERENCES
17.
1. McCabe WR, Jackson GG: Gram-negative
bacteremia II. Clinical, laboratory and therapeutic observations. Arch Intern Med 1962;
110:856 – 864
2. Kreger BE, Craven DE, McCabe WR: Gramnegative bacteremia IV. Re-evaluation of
18.
clinical features and treatment in 612 patients. Am J Med 1980; 68:344 –355
Kollef MH, Sherman G, Ward S, et al: Inadequate antimicrobial treatment of infections: A
risk factor for hospital mortality among critically ill patients. Chest 1999; 115:462– 474
Parkins MD, Sabuda DM, Elsayed S, et al:
Adequacy of empirical antifungal therapy and
effect on outcome among patients with invasive Candida species infections. J Antimicrob
Chemother 2007; 60:613– 618
Kumar A, Ellis P, Arabi Y, et al: Initiation of
inappropriate antimicrobial therapy results
in a five-fold reduction of survival in human
septic shock. Chest 2009; 136:1237–1248
Chow JW, Yu VL: Combination antibiotic
therapy versus monotherapy for gramnegative bacteraemia: A commentary. Int J
Antimicrob Agents 1999; 11:7–12
Mufson MA, Stanek RJ, Mufson MA, et al:
Revisiting combination antibiotic therapy for
community-acquired invasive Streptococcus
pneumoniae pneumonia. Clin Infect Dis
2006; 42:304 –306
Patel SM, Saravolatz LD: Monotherapy versus combination therapy. Med Clin North
Am 2006; 90:1183–1195
Paul M, Leibovici L: Combination antibiotic
therapy for Pseudomonas aeruginosa bacteraemia. Lancet Infect Dis 2005; 5:192–193
Weiss K, Tillotson GS: The controversy of
combination vs monotherapy in the treatment of hospitalized community-acquired
pneumonia. Chest 2005; 128:940 –946
Anderson ET, Young LS, Hewitt WL: Antimicrobial synergism in the therapy of Gramnegative rod bacteremia. Chemotherapy
1978; 24:45–54
De Jongh CA, Joshi JH, Newman KA, et al:
Antibiotic synergism and response in gramnegative bacteremia in granulocytopenic
cancer patients. Am J Med 1986; 80:96 –100
Bouza E, Munoz P: Monotherapy versus
combination therapy for bacterial infections.
Med Clin North Am 2000; 84:1357–1389
Darras-Joly C, Bedos JP, Sauve C, et al: Synergy between amoxicillin and gentamicin in
combination against a highly penicillinresistant and -tolerant strain of Streptococcus pneumoniae in a mouse pneumonia
model. Antimicrob Agents Chemother 1996;
40:2147–2151
Calandra T, Glauser MP: Immunocompromised animal models for the study of antibiotic combinations. Am J Med 1986; 80:45–52
Kumar A, Mensing J, Zelenitsky S, et al:
Effect of antibiotic sequence on blood bacterial counts in a rat model of E. coli peritonitis/septic shock. ICAAC Proceedings, 2004,
pp 26, A-1296
Safdar N, Handelsman J, Maki DG, et al: Does
combination antimicrobial therapy reduce
mortality in Gram-negative bacteraemia? A
meta-analysis. Lancet Infect Dis 2004;
4:519 –527
Paul M, Benuri-Silbiger I, Soares-Weiser K,
et al: ␤lactam monotherapy versus ␤lactamaminoglycoside combination therapy for sep-
Crit Care Med 2010 Vol. 38, No. 9
19.
20.
21.
22.
23.
24.
25.
26.
27.
28.
29.
30.
31.
32.
33.
34.
sis in immunocompetent patients: Systematic review and meta-analysis of randomised
trials. BMJ 2004; 328:668 – 679
Kumar A, Safdar N, Reddy S, et al: The survival benefit of combination antibiotic therapy for serious infections associated with
sepsis and septic shock is contingent only on
the risk of death: A meta-analytic/metaregression study. Crit Care Med 2010; 38:
1651–1664
Bone RC, Balk R, Cerra FB, et al: ACCP/
SCCM Consensus Conference: Definitions for
sepsis and organ failure and guidelines for
use of innovative therapies in sepsis. Chest
1992; 101:1644 –1655
Kumar A, Roberts D, Wood KE, et al: Duration of hypotension before initiation of effective antimicrobial therapy is the critical determinant of survival in human septic shock.
Crit Care Med 2006; 34:1589 –1596
McGeer A, Campbell B, Emori TG, et al:
Definitions of infection for surveillance in
long-term care facilities. Am J Infect Control
1991; 19:1–7
Garner JS, Jarvis WR, Emori TG, et al: CDC
definitions for nosocomial infections, 1988.
Am J Infect Control 1988; 16:128 –140
Zarychanski R, Doucette S, Fergusson D, et
al: Early intravenous unfractionated heparin
and mortality in septic shock. Crit Care Med
2008; 36:2973–2979
Bagshaw SM, Lapinsky S, Dial S, et al: Acute
kidney injury in septic shock: Clinical outcomes and impact of duration of hypotension
prior to initiation of antimicrobial therapy.
Intensive Care Med 2009; 35:871– 881
Knaus WA, Draper EA: APACHE II: A severity
of disease classification system. Crit Care
Med 1985; 13:818 – 829
Glynn RJ, Schneeweiss S, Sturmer T: Indications for propensity scores and review of
their use in pharmacoepidemiology. Basic
Clin Pharmacol Toxicol 2006; 98:253–259
Luellen JK, Shadish WR, Clark MH: Propensity scores: An introduction and experimental test. Eval Rev. 2005; 29:530 –558
Parsons LS: SUGI 26: Reducing Bias in a
Propensity Score Matched-pair Sample Using
Greedy Matching Techniques, SAS Institute,
Cary, NC, 2001
Suissa S: Immortal time bias in observational studies of drug effects Pharmacoepidemiol Drug Saf 2007; 16:241–249
Breslow NE, Day NE: Statistical methods in
cancer research. Analysis of case-control
studies. Volume 1. Lyon, France, International Agency for Research on Cancer, IARC
Scientific Publications No. 32, 1980, pp
138 –146
Bochud PY, Glauser MP, Calandra T, et al:
Antibiotics in sepsis, Intensive Care Med
2001; 27(Suppl 1):S33–S48
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
Gribble MJ, Chow AW, Naiman SC, et al:
Crit Care Med 2010 Vol. 38, No. 9
35.
36.
37.
38.
39.
40.
41.
42.
43.
44.
45.
46.
47.
48.
Prospective randomized trial of piperacillin
monotherapy versus carboxypenicillinaminoglycoside combination regimens in the
empirical treatment of serious bacterial infections. Antimicrob Agents Chemother
1983; 24:388 –393
EORTC International Anti-microbial Therapy
Cooperative Group: Ceftazidime combined
with a short or long course of amikacin for
empirical therapy of gram-negative bacteremia in cancer patients with granulocytopenia. The EORTC International Antimicrobial
Therapy Cooperative Group. N Engl J Med
1987; 317:1692–1698
Milatovic D, Braveny I: Development of resistance during antibiotic therapy. Eur J Clin
Microbiol 1987; 6:234 –244
Pasquale TR, Tan JS: Nonantimicrobial effects of antibacterial agents. Clin Infect Dis
2005; 40:127–135
Labro MT: Interference of antibacterial
agents with phagocyte functions: immunomodulation or “immuno-fairy tales”? Clin
Micro Rev 2000; 13:615– 650
De Jongh CA, Joshi JH, Thompson BW, et al:
A double ␤lactam combination versus an
aminoglycoside-containing regimen as empiric antibiotic therapy for febrile granulocytopenic cancer patients. Am J Med 1986; 80:
101–111
Klastersky J, Zinner SH: Synergistic combinations of antibiotics in gram-negative bacillary infections. Rev Infect Dis 1982;
4:294 –301
Giamarellou H: Aminoglycosides plus betalactams against gram-negative organisms.
Evaluation of in vitro synergy and chemical
interactions. Am J Med 1986; 80:126 –137
Guchelaar HJ, Schultz MJ, van D, et al: Pharmacokinetic-pharmacodynamic modeling of
the inhibitory effect of erythromycin on tumour necrosis factor-alpha and interleukin-6
production. Fund Clin Pharmacol 2001; 15:
419 – 424
Labro MT: Cellular and molecular effects of
macrolides on leukocyte function. Curr
Pharm Des 2004; 10:3067–3080
Khan AA, Slifer TR, Araujo FG, et al: Effect of
clarithromycin and azithromycin on production of cytokines by human monocytes. Int J
Antimicrob Agents 1999; 11:121–132
Oda H, Kadota J, Kohno S, et al: Erythromycin inhibits neutrophil chemotaxis in bronchoalveoli of diffuse panbronchiolitis. Chest
1994; 106:1116 –1123
Khan AA, Slifer TR, Remington JS: Effect of
trovafloxacin on production of cytokines by
human monocytes. Antimicrob Agents Chemother 1998; 42:1713–1717
Goscinski G, Lundholm M, Odenholt I, et al:
Variation in the propensity to release endotoxin after cefuroxime exposure in different
gram-negative bacteria: Uniform and dosedependent reduction by the addition of tobramycin. Scand J Infect Dis 2003; 35:40 – 46
Goscinski G, Lipcsey M, Eriksson M, et al:
Endotoxin neutralization and anti-inflammatory effects of tobramycin and ceftazidime in
49.
50.
51.
52.
53.
54.
55.
56.
57.
58.
59.
60.
61.
62.
63.
porcine endotoxin shock. Crit Care 2004;
8:R35–R41
Comber KR, Basker MJ, Osborne CD, et al:
Synergy between ticarcillin and tobramycin
against Pseudomonas aeruginosa and Enterobacteriaceae in vitro and in vivo. Antimicrob Agents Chemother 1977; 11:956 –964
Archer G, Fekety FR Jr: Experimental endocarditis due to Pseudomonas aeruginosa II.
Therapy with carbenicillin and gentamicin.
J Infect Dis 1977; 136:327–335
Yoshikawa TT, Shibata SA: In vitro antibacterial activity of amikacin and ticarcillin,
alone and in combination, against Pseudomonas aeruginosa. Antimicrob Agents Chemother 1978; 13:997–999
Kluge RM, Standiford HC, Tatem B, et al:
Comparative activity of tobramycin, amikacin, and gentamicin alone and with carbenicillin against Pseudomonas aeruginosa. Antimicrob Agents Chemother 1974; 6:442– 446
Gimeno C, Borja J, Navarro D, et al: In vitro
interaction between ofloxacin and cefotaxime
against gram-positive and gram-negative
bacteria involved in serious infections. Chemotherapy 1998; 44:94 –98
Gradelski E, Kolek B, Bonner DP, Valera L, et
al: Activity of gatifloxacin and ciprofloxacin
in combination with other antimicrobial
agents. Int J Antimicrob Agents 2001; 17:
103–107
Zembower TR, Noskin GA, Postelnick MJ, et
al: The utility of aminoglycosides in an era of
emerging drug resistance. Int J Antimicrob
Agents 1998; 10:95–105
Neu HC: Synergy and antagonism of fluoroquinolones with other classes of antimicrobial agents. Drugs 1993; 45(Suppl 3):54 –58
Pohlman JK, Knapp CC, Ludwig MD, et al:
Timed killing kinetic studies of the interaction between ciprofloxacin and beta-lactams
against gram-negative bacilli. Diagnostic Microbiol Infect Dis 1996; 26:29 –33
Milatovic D, Wallrauch C: In vitro activity of
trovafloxacin in combination with ceftazidime, meropenem, and amikacin. Eur J Clin
Microbiol Infect Dis 1996; 15:688 – 693
Pankuch GA, Lin G, Seifert H, et al: Activity
of meropenem with and without ciprofloxacin and colistin against Pseudomonas aeruginosa and Acinetobacter baumannii. Antimicrob Agents Chemother 2008; 52:333–336
Lin E, Stanek RJ, Mufson MA: Lack of synergy of erythromycin combined with penicillin or cefotaxime against Streptococcus
pneumoniae in vitro. Antimicrob Agents
Chemother 2003; 47:1151–1153
MacGowan AR, Bowker K, Bedford KA, et al:
Synergy testing of macrolide combinations
using the chequerboard technique. J Antimicrob Chemother 1993; 32:913–915
Furuya R, Nakayama H, Kanayama A, et al: In
vitro synergistic effects of double combinations
of beta-lactams and azithromycin against clinical isolates of Neisseria gonorrhoeae. J Infect
Chemother 2006; 12:172–176
Maioli E, Debbia EA, Gualco L, et al: In vitro
interaction between mecillinam and pipera-
1783
cillin-tazobactam in the presence of azithromycin against members of the Enterobacteriaceae family and Pseudomonas aeruginosa.
New Microbiol 2008; 31:37– 46
64. Allen NE, Epp JK: Mechanism of penicillinerythromycin synergy on antibiotic-resistant
Staphylococcus aureus. Antimicrob Agents
Chemother 1978; 13:849 – 853
65. Connors AF Jr, Speroff T, Dawson NV, et al:
The effectiveness of right heart catheterization in the initial care of critically ill patients. JAMA 1996; 276:889 – 897
APPENDIX 1
Rules to assign clinical significance to
microbial isolates:
1) Clinically significant isolates from either local site and/or blood cultures
were required to have been obtained
within 48 hrs of onset of shock.
2) The following were considered to represent clinically significant isolates:
a) Positive blood culture for any pathogen other than coagulase-negative
staphylococci or other skin contaminants.
b) Any growth from a normally sterile
site (e.g., gall bladder, bronchial lavage, peritoneal, pleural fluid, operative tissue specimen) apart from
coagulase-negative staphylococci
and other skin contaminants.
c) Growth of a pathogen in sputum
sample of a patient with respiratory signs and symptoms or a new
infiltrate on chest radiography,
with no other likely source of infection.
d) Growth of a pathogen in a urine
sample (⬎108 organisms/L) with
either local clinical symptoms or
in the absence of a more plausible
clinical infection site.
e) Growth from a deep biopsy or a
deep aspirate of a finding in soft
tissue or skin.
f) Concurrent, congruent, positive,
semiquantitative catheter colonization (⬎15 colonies) with blood
culture or clinical evidence of site
infection.
g) A positive direct measurement of
Streptococcus pneumoniae, Neisseria meningitidis, or Haemophilus influenzae antigen in the sputum or cerebrospinal fluid.
3) Enterococci were considered to be clinically significant only in the absence of
other more plausible pathogens.
4) Staphylococcus epidermidis was uniformly considered to be incapable of
1784
causing septic shock. Other coagulase-negative staphylococci similarly
were considered to be unlikely to
cause septic shock unless present as a
sole isolate in multiple blood cultures
in the absence of evidence of endovascular infection.
APPENDIX 2
Rules regarding antimicrobial therapy
1) The following organisms were considered sensitive to listed antibiotics (i.e.,
appropriate therapy) even in the absence of specific sensitivity testing:
group A, B, and G streptococcus to all
␤-lactams; all Gram-positive bacteria
except enterococci to vancomycin;
and organisms to ␤-lactamase inhibitor combinations if sensitive to the
␤-lactam alone.
2) The following organisms were considered resistant to listed antibiotics (i.e.,
inappropriate therapy): Enterococci to
all cephalosporins and trimethoprim/
sulfamethoxazole and Enterococcus
faecalis to quinupristin-dalfopristin.
3) Clindamycin, macrolides, and thirdgeneration cephalosporins were not
considered appropriate for Staphylococcus aureus septic shock irrespective of listed sensitivity.
4) Cefotaxime and ceftriaxone were not
considered appropriate therapy of
Pseudomonas aeruginosa irrespective
of listed sensitivity.
5) Treatment with bacteriostatic antibiotics was considered appropriate therapy only in the case of sensitive enterococci treated with a penicillin or
carbapenem and for any pathogen for
which no commonly available bactericidal primary or supplemental agents
existed.
6) Use of aminoglycosides as the supplemental/secondary antibiotic for treatment of group A streptococci (Streptococcus pyogenes) or viridans
streptococci infections with an appropriate ␤-lactam or vancomycin was
considered to be appropriate combination therapy even in the absence of
specific sensitivity data.
7) In case of multiple isolates at a local
site, appropriate therapy was considered to have been delivered if the
densest pathogen was covered. If multiple pathogens were isolated at similar density, then all pathogens were
required to have been covered.
8) For multiple simultaneous blood isolates, appropriate therapy had to cover
all pathogens.
APPENDIX 3
Additional Members of the Cooperative Antimicrobial Therapy of Septic
Shock (CATSS) Database Research Group
include: Kenneth E. Wood, MD, University of Wisconsin Hospital and Clinics,
Madison, WI, USA; Kevin Laupland, MD,
Foothills Hospital, Calgary AB, Canada;
Andreas Kramer, MD, Brandon General
Hospital, Brandon MB, Canada; Charles
Penner, MD, Brandon General Hospital,
Brandon MD, Canada; Bruce Light, MD,
Winnipeg Regional Health Authority,
Winnipeg MB, Canada; Satendra Sharma,
MD, Winnipeg Regional Health Authority, Winnipeg MB, Canada; Steve Lapinsky, MD, Mount Sinai Hospital, Toronto
ON, Canada; John Marshall, MD, St. Michael’s Hospital, Toronto ON, Canada;
Sandra Dial, MD, Jewish General Hospital, Montreal QC, Canada; Sean Bagshaw,
MD, University of Alberta Hospital, Edmonton AB, Canada; Ionna Skrobik, MD,
Hôpital Maisonneuve Rosemont, Montreal QC, Canada; Gourang Patel,
PharmD, Rush-Presbyterian-St. Luke’s
Medical Center, Chicago IL, USA; Dave
Gurka, MD, Rush-Presbyterian-St. Luke’s
Medical Center, Chicago, IL, USA; Sergio
Zanotti, MD, Cooper Hospital/University
Medical Center, Camden, NJ, USA; Phillip
Dellinger, MD, Cooper Hospital/University Medical Center, Camden, NJ, USA;
Dan Feinstein, MD, St. Agnes Hospital, Baltimore, MD, USA; Jorge Guzman, MD,
Harper Hospital, Detroit, MI, USA; Nehad
Al Shirawi, MD, King Abdulaziz Medical
City, Riyadh, Saudi Arabia; Ziad Al Memish,
MD, King Abdulaziz Medical City, Riyadh,
Saudi Arabia; John Ronald, MD, Nanaimo
Regional Hospital, Nanaimo BC, Canada.
Associate Members of the CATSS Database Research Group include: Mustafa
Suleman, MD, Concordia Hospital, Winnipeg, MB Canada; Harleena Gulati, MD,
University of Manitoba, Winnipeg, MB Canada; Erica Halmarson, MD, University of
Manitoba, Winnipeg MB, Canada; Robert
Suppes, MD, University of Manitoba, Winnipeg MB, Canada; Cheryl Peters, University of Manitoba, Winnipeg MB, Canada;
Katherine Sullivan, University of Manitoba,
Winnipeg MB, Canada; Rob Bohmeier, University of Manitoba, Winnipeg MB, Canada;
Sheri Muggaberg, University of Manitoba,
Winnipeg MB, Canada; Laura Kravetsky,
University of Manitoba, Winnipeg MB, CanCrit Care Med 2010 Vol. 38, No. 9
ada; Muhammed Wali Ahsan, MD, Winnipeg MB; Canada, Amrinder Singh, MD,
Winnipeg, MB Canada; Lindsey Carter, BA,
Winnipeg MB, Canada; Kym Wiebe, RN, St.
Boniface Hospital, Winnipeg MB, Canada;
Laura Kolesar, RN, St. Boniface Hospital,
Winnipeg MB, Canada; Jody Richards,
Camosun College, Victoria BC, Canada;
Danny Jaswal, MD, University of British Columbia, Vancouver BC, Canada; Harris
Chou, BSc, of British Columbia, Vancouver
BC, Canada; Tom Kosick, MD, University of
British Columbia, Vancouver BC, Canada;
Winnie Fu, University of British Columbia,
Crit Care Med 2010 Vol. 38, No. 9
Vancouver BC, Canada; Charlena Chan,
University of British Columbia, Vancouver
BC, Canada; Jia Jia Ren, University of British Columbia, Vancouver BC, Canada; Mozdeh Bahrainian, MD, Madison WI, Ziaul
Haque, MD, Montreal QC, Canada; Omid
Ahmadi Torshizi, MD, Montreal QC, Canada; Heidi Paulin, University of Toronto,
Toronto ON, Canada; Farah Khan, MD, Toronto ON, Canada; Runjun Kumar, University of Toronto, Toronto ON, Canada; Johanne Harvey, RN, Hôpital Maisonneuve
Rosemont, Montreal QC, Canada; Christina
Kim, McGill University, Montreal QC, Can-
ada; Jennifer Li, McGill University, Montreal QC, Canada; Latoya Campbell, McGill
University, Montreal QC, Canada; Leo
Taiberg, MD, Rush Medical College, Chicago, IL, USA; Christa Schorr, RN, Cooper
Hospital/University Medical Center, Camden, NJ, USA; Ronny Tchokonte, MD,
Wayne State University Medical School, Detroit, MI, USA; Catherine Gonzales, RN,
King Abdulaziz Medical City, Riyadh, Saudi
Arabia, Norrie Serrano, RN, King Abdulaziz
Medical City, Riyadh, Saudi Arabia, Sofia
Delgra, RN, King Abdulaziz Medical City,
Riyadh, Saudi Arabia.
1785