Intensive Care Med

DOI 10.1007/s00134-017-4782-4

ORIGINAL

Effect of a multifaceted educational

intervention for anti‑infectious measures
on sepsis mortality: a cluster randomized trial
Frank Bloos1,2, Hendrik Rüddel1,2, Daniel Thomas‑Rüddel1,2, Daniel Schwarzkopf1, Christine Pausch3,
Stephan Harbarth4, Torsten Schreiber5, Matthias Gründling6, John Marshall7, Philipp Simon8, Mitchell M. Levy9,
Manfred Weiss10, Andreas Weyland11, Herwig Gerlach12, Tobias Schürholz13,14, Christoph Engel3,
Claudia Matthäus‑Krämer1, Christian Scheer6, Friedhelm Bach15, Reimer Riessen16, Bernhard Poidinger1,2,
Karin Dey17, Norbert Weiler18, Andreas Meier‑Hellmann19, Helene H. Häberle20, Gabriele Wöbker21,
Udo X. Kaisers8,22 and Konrad Reinhart1,2*, for the MEDUSA study group

© 2017 Springer-Verlag Berlin Heidelberg and ESICM

Abstract
Purpose: Guidelines recommend administering antibiotics within 1 h of sepsis recognition but this recommendation
remains untested by randomized trials. This trial was set up to investigate whether survival is improved by reducing
the time before initiation of antimicrobial therapy by means of a multifaceted intervention in compliance with guide‑
line recommendations.
Methods: The MEDUSA study, a prospective multicenter cluster-randomized trial, was conducted from July 2011 to
July 2013 in 40 German hospitals. Hospitals were randomly allocated to receive conventional continuous medical
education (CME) measures (control group) or multifaceted interventions including local quality improvement teams,
educational outreach, audit, feedback, and reminders. We included 4183 patients with severe sepsis or septic shock in
an intention-to-treat analysis comparing the multifaceted intervention (n = 2596) with conventional CME (n = 1587).
The primary outcome was 28-day mortality.
Results: The 28-day mortality was 35.1% (883 of 2596 patients) in the intervention group and 26.7% (403 of 1587
patients; p = 0.01) in the control group. The intervention was not a risk factor for mortality, since this difference was
present from the beginning of the study and remained unaffected by the intervention. Median time to antimicrobial
therapy was 1.5 h (interquartile range 0.1–4.9 h) in the intervention group and 2.0 h (0.4–5.9 h; p = 0.41) in the control

*Correspondence: [email protected]‑jena.de
1
Center for Sepsis Control and Care (CSCC), Jena University Hospital,
Jena, Germany
Full author information is available at the end of the article

Members of the MEDUSA study group are listed in the

“Acknowledgements”.

Take-home message: This first cluster-randomized controlled

multicenter clinical trial addressing the effect of time to empirical
antimicrobial therapy on 28-day mortality was unable to reduce time
to antimicrobial therapy in the intervention group. Although the
intervention itself did not affect survival, time to first antimicrobial
therapy and time to source control were both associated with mortality.
 group. The risk of death increased by 2% per hour delay of antimicrobial therapy and 1% per hour delay of source con‑
trol, independent of group assignment.
Conclusions: Delay in antimicrobial therapy and source control was associated with increased mortality but the
multifaceted approach was unable to change time to antimicrobial therapy in this setting and did not affect survival.
Keywords: Critical illness, Septic shock/drug therapy, Antimicrobial agents, Quality improvement

Introduction Methods
Severe sepsis and septic shock are life-threatening con- This prospective cluster-randomized MEDUSA trial was
ditions with a high risk of mortality [1]. Several retro- the second study phase after an observational pre-study
spective studies suggest that timely and appropriate [25]. Each hospital of this pre-study was eligible for par-
antimicrobial therapy (AT) is an important predic- ticipation and formed a cluster. Participating hospitals
tor of survival [2–6]. Delayed surgery is also associated were involved in primary sepsis care and senior repre-
with lower survival rates [7–9], but the optimal timing sentatives of the participating ICUs were required to
remains poorly defined. The Surviving Sepsis Campaign involve other hospital divisions and stakeholders in the
guidelines recommend administration of adequate AT quality improvement process. We identified 45 hospitals
within 1 h and surgical source control within 6–12 h of eligible for the study of which 44 agreed to participate.
the onset of severe sepsis [10]. These recommendations To minimize the risk of randomization failure, the clus-
are based only on non-randomized studies. Such studies ters were stratified according to time to antimicrobial
are prone to overestimation of the measured effects, bias therapy observed in the pre-study and then randomized
due to secular trends, or the Hawthorne effect [11, 12]. 1:1 to either a control group or an intervention group.
The concept that anti-infectious measures should be Randomization was computer generated. Blinding was
initiated rapidly is both intuitively rational and indi- not possible. Study interventions were solely performed
rectly supported by a recent randomized trial on early on hospital level.
goal-directed therapy [13]. However, a recent meta- All consecutive patients between July 1, 2011 and
analysis showed no increase in mortality with a delay in June 30, 2013 with proven or suspected infection and
AT therapy up to 5 h after severe sepsis recognition [14]. acute onset of infection-related organ dysfunction were
Furthermore, directing physicians’ behavior by protocols eligible. Patients for whom there was no full commit-
might increase the risk for inappropriate prescriptions of ment to therapy at onset of sepsis, who were not treated
broad-spectrum antibiotics for noninfectious conditions on a participating ICU after sepsis diagnosis, or whose
[15]. To date, no randomized clinical trial has directly sepsis therapy was started in another hospital were
examined the effect of rapid antibiotic administration on excluded. The study was registered at ClinicalTrials.gov
outcomes. (NCT01187134) and was approved by the local ethics
Compliance with current guideline recommendations committees at each participating institution and by the
on early antimicrobial use is low [1]. A multifaceted responsible state data protection boards. The need for
intervention including audit and feedback is an effective informed consent was waived since the interventions
strategy to facilitate guideline implementation [6, 16–19]. comprised quality improvement measures, and randomi-
Indeed, multifaceted educational programs have resulted zation was performed on the cluster level.
in increased compliance with guideline recommenda-
tions and improved quality of care in critically ill patients Intervention
[20–24]. Hospitals in the control group were offered standard
In this cluster-randomized clinical trial, we sought to lectures about sepsis care twice a year and regular news-
prospectively confirm the hypothesis that shortening letters with current sepsis-related publications or con-
the time to initiation of antimicrobial therapy improves ference proceedings. The intervention group received
patient survival and to test the secondary hypothesis that a multifaceted implementation strategy in addition to
the effects of a multifaceted intervention accompanied by the measures of the control group. Training included
audit and feedback to facilitate compliance with guide- primary sepsis care with focus on timing and strategy
line recommendations of anti-infectious measures are of empiric anti-infectious management. The strategy
superior to conventional continuous medical education consisted of formation of interprofessional local qual-
(CME). ity improvement teams (ICU and non-ICU members),
educational outreach, audit, and feedback as well as pas- using the variance inflation factor (VIF) to account for
sive and active reminders (see electronic supplement for clustering [27]. VIFs were calculated individually for each
detailed description). variable separately for each group. Missing values were
treated as such. Hierarchical (mixed) regression mod-
Data collection els with center as random effect were used to adjust for

capture application OpenClinica® (OpenClinica, LLC,

The data set was collected using the electronic data imbalances between the groups at baseline. As a sensitiv-
ity analysis, hierarchical regression models were repeated
Waltham, MA, USA). Data integrity was confirmed by after multilevel multiple imputation of missing data in
source data monitoring of a random sample of 10% of the 100 data sets (see electronic supplement for detailed
patients and by data checks within the database. Onset of description) [28]. Data were analyzed using R, version
severe sepsis or septic shock was defined as the time of 3.02.
first infection-related organ dysfunction as documented
in the patient file. Time and type of first AT as well as Results
preexisting AT were also recorded from the medical Participants
records. Time to AT was negative if AT was prescribed Three intervention group hospitals and one control group
up to 24 h before the onset of sepsis-related organ dys- hospital withdrew from the study as they were unable to
function. Inadequacy of initial AT was defined as esca- contribute patients to the study resulting in 19 participat-
lation of AT within the first 5 days [25]. A set of rules ing hospitals in the intervention group and 21 hospitals
depending on focus of infection and location of infection in the control group (Fig. 1). Hospital characteristics are
acquisition assessed the compliance of empiric AT with shown in Electronic Supplementary Material Table E1.
German guidelines. Time to source control was defined Frequency of public, private, and non-profit hospital
as duration from onset of sepsis to start of first surgical operators were equally distributed between the groups.
or interventional source control. De-escalation of AT was We enrolled 4183 patients during the study. Compared
defined as switch to a drug class resulting in a less broad to the control group, allocation to the intervention group
spectrum of coverage. was associated with a 1.6 times higher recruitment rate.
The primary endpoint of this study was all-cause mor- Information on 28-day mortality was missing in 155
tality within 28 days after onset of severe sepsis or sep- patients. All patients were analyzed. Patient character-
tic shock, which was obtained by telephone contact if istics are shown in Table 1. In 1241 patients (47.8%) of
the patient was no longer in the hospital. Secondary the intervention group and 672 patients (42.3%) of the
endpoints were initiation of AT within 1 h, time to first control group, onset of severe sepsis was detected out-
empirical antimicrobial therapy, fraction of patients with side of the ICU. Patients in the intervention group were
at least two sets of blood cultures taken, adequacy of AT, descriptively more severely ill than patients in the control
time to source control, ICU and hospital length of stay, group as reflected in higher SAPS II scores, higher ini-
and ICU and hospital mortality. tial SOFA scores, higher initial lactate, and lower arterial
pH, although these differences were not significant after
Statistical analyses adjusting for clustering. There were more community-
We calculated the sample size based on the reported acquired infections in the control than in the interven-
increases by 7.5% for each 1 h delay in starting AT [2], tion group.
a 28-day mortality of 34.8%, and an intracluster cor-
relation coefficient (ICC) of 0.024 (95% CI −0.004 to Effects of intervention on infection management
0.051) observed in the pre-study [25]. To detect an abso- In the intervention group, a total of 104 quality improve-
lute mortality reduction of 7.5% with a power of 0.8 at ment team visits were performed by the study coordina-
α = 0.05, 22 hospitals per group with a mean recruitment tors; all intervention centers received at least four visits,
rate of 100 patients per hospital were required (see elec- 89% (16/19) received at least five visits, and 67% (12/19)
tronic supplement for detailed description). Hence, we were visited six times. At the beginning of the study, the
aimed for 2200 patients per group [26]. proportion of patients receiving AT within 1 h was well
The primary endpoint was analyzed by comparing balanced between the groups (Fig. 2a). There was no
28-day all-cause mortality between the two groups using clear consistent effect of the multifaceted intervention
the Chi-squared test. The secondary endpoints were on compliance with early AT. Median duration of AT
assessed likewise, if binary. For continuous data, two- remained unaffected by the intervention [intervention
sided t tests were applied. If variables were not normally group: 8 (interquartile range: 4–13) days, control group
distributed, appropriate transformations were applied. 8 (5–13) days, p = 0.91]. Compliance with blood culture
Test statistics were analyzed on patient level and adjusted increased to nearly 70% in the intervention group while
 Assessed for eligibility (45 ICUs)

Excluded (1 ICU)
no longer involved in
primary sepsis care

Randomized (44 ICUs)

Allocated to intervention group (22 ICUs) Allocated to control group (22 ICUs)
Received allocated intervention Received allocated intervention
19 ICUs, median ICU size: 14 beds, IQR: 11-24 21 ICUs, median ICU size: 20 beds, IQR: 10-38
2596 patients 1587 patients
Did not receive allocated intervention Did not receive allocated intervention
No study activity (1 ICU); withdrawn after Withdrawn after randomization: 1 ICU (6 beds)
randomization (2 ICUs): 12, 13, and 42 beds

Lost to follow up: Lost to follow up:

0 ICUs 0 ICUs
28-day mortality not available 28-day mortality not available
in 80 patients (3.1 %) in 75 patients (4.7 %)

Clusters: Clusters:
Analyzed Analyzed
19 ICUs, median ICU size: 14 beds, IQR: 11-24 21 ICUs, median ICU size: 20 beds, IQR: 10-38
Excluded from analysis Excluded from analysis
No ICUS without documentation No ICUS without documentation
Participants: Participants:
2596 (100%) analysed 1587 (100%) analysed

Fig. 1 Study flow diagram. ICU intensive care unit, IQR interquartile range

compliance in the control group varied between 45 and mortality in the intervention group was already observed
58% (Electronic Supplementary Material Fig. E1). De- at the beginning of the study and did not show a distinct
escalation of AT after microbiological results occurred in change in this difference attributable to the study inter-
416 (16.0%) patients of the intervention group and in 227 vention (Fig. 2a). When retrospectively analyzing the
(14.4%) of the control group (p = 0.668). patients from the run-in phase according to the later
group assignment for this trial, the difference in recruit-
Outcomes ment rate and 28-day mortality was already present
Twenty-eight day mortality was 35.1% in the intervention there (Electronic Supplementary Material Table E4). In
group and 26.7% in the control group (p = 0.01; Table 2); a hierarchical model with 28-day mortality as dependent
the ICC was 0.027 (95% CI 0.011–0.043). A higher variable, age, SAPS II score, initial lactate levels, initial
Table 1 Patients characteristics
Intervention group Control group Missing p value p value ICC
n = 2596 n = 1587 valuesa (unadjusted) (adjusted)b

Age, median (IQR), years 70 (59–77) 70 (59–77) 0/1 0.71 0.88 0.02
Male gender 1600 (61.6%) 1011 (63.7%) 0/0 0.19 0.33 0.004
Type of admission
Medical 1153 (44.4%) 678 (42.7%) 0/0 0.22 – –e
Unscheduled surgery 1061 (40.9%) 689 (43.4%)
Scheduled surgery 285 (11.0%) 177 (11.2%)
Trauma 67 (2.6%) 27 (1.7%)
Other 30 (1.3) 16 (1.0%)
Location of first documentation of sepsis ICU
Emergency room 1355 (52.2%) 915 (57.7%) 0/0 <0.001 – –e
General ward 377 (14.5%) 205 (12.9%)
Operating room 356 (13.7%) 189 (11.9%)
Prehospital 309 (11.9%) 115 (7.2%)
Emergency physician 71 (2.7%) 27 (1.7%)
IMC 128 (4.9%) 136 (8.6%)
Organ dysfunction, first 12 h
Encephalopathy 992 (38.2%) 687 (43.6%) 0/11 0.001 0.54 0.14
Respiratory 1717 (66.2%) 1116 (70.8%) 3/11 0.002 0.40 0.06
Vasopressor need 1951 (75.2%) 1231 (78.1%) 0/11 0.03 0.51 0.05
Renal dysfunction 753 (29.0%) 535 (33.9%) 2/11 0.001 0.45 0.09
Thrombocytopenia 474 (18.3%) 353 (22.4%) 2/11 0.001 0.27 0.04
Metabolic acidosis 1338 (51.6%) 654 (41.5%) 4/11 <0.001 0.07 0.06
Origin of infection
Community acquired 1016 (39.2%) 715 (45.1%) 1/0 <0.001 – –e
Nosocomial (ICU/IMC) 580 (22.4%) 415 (26.1%)
Nosocomial (general 936 (36.1%) 432 (27.2%)
ward)
Nosocomial (nursing 63 (2.4%) 25 (1.6%)
home)
Focus of infection
Pneumonia 918 (35.4%) 597 (37.7%) 1/2 0.14 0.76 0.10
Respiratory tract, other 160 (6.2%) 91 (5.7%) 0.62 0.90 0.09
Intra-abdominal 974 (35.7%) 568 (35.8%) 0.28 0.81 0.09
Cardiovascular 39 (1.5%) 35 (2.2%) 0.12 0.43 0.02
Urogenital 314 (12.1%) 216 (13.6%) 0.16 0.64 0.04
Bones/soft tissue 207 (8.0%) 148 (9.3%) 0.14 0.48 0.02
CNS 43 (1.7%) 22 (1.4%) 0.58 0.66 0.01
Primary bacteremia 42 (1.6%) 23 (1.5%) 0.77 0.80 0.01
Intravascular device 57 (2.2%) 52 (3.3%) 0.04 0.34 0.02
related
Thoracic 56 (2.2%) 36 (2.3%) 0.89 –f 0.00
Wound infection 108 (4.2%) 37 (2.3%) 0.002 0.24 0.03
Other 19 (0.7%) 17 (1.1%) 0.33 –f 0.00
Unknown 96 (3.7%) 50 (3.3%) 0.40 0.56 0.01
SAPS II scorec, median 50 (39–62) 46 (36–59) 261/216 <0.001 0.48 0.17
(IQR)
SOFA scored at day 1, 10 (7–12) 9 (7–11) 558/460 <0.0001 0.29 0.05
median (IQR)
Lactate at day 1, mmol/l, 2.8 (1.6–5.2) 2.4 (1.6–4.3) 102/73 <0.001 0.13 0.03
median (IQR)
Table 1 continued
Intervention group Control group Missing p value p value ICC
n = 2596 n = 1587 valuesa (unadjusted) (adjusted)b
Procalcitonin at day 1, 5.9 (1.5–23.9) 5.9 (1.7–24.9) 182/265 0.30 0.78 0.06
ng/ml, median (IQR)
All p values are calculated on patient level
CNS central nervous system, ICC intra-cluster correlation coefficient (a higher ICC reflects a higher between-cluster variance, see electronic supplement for detailed
explanation), ICU intensive care unit, IMC intermediate care unit, IQR interquartile range, SAPS II simplified acute physiology score II, SOFA sequential organ failure
assessment
a
Missing values for intervention/control group
b
Test statistics were adjusted for clustering (see electronic supplement for detailed explanation)
c
SAPS II is calculated from 12 routinely measured physiological and biochemical variables within the first 24 h of intensive care unit admission. The range varies from
0 to 163 points where values reflect a higher risk of in-hospital death
d
SOFA score is calculated from 6 routinely measured physiological and biochemical variables at a given day on the ICU. The range varies from 0 to 24 where higher
values reflect a higher degree of organ dysfunction
e
No readily available test adjusting for clustering for contingency tables with more than two rows or columns
f
Too few cases for reliable adjustment

platelet count, inadequate empiric antimicrobial therapy, but not in the control group. This was associated with a
acute renal failure, and number of organ dysfunctions shorter time to AT in the interventions groups but not
were independent predictors (Table 3). Assignment of with improved survival (Electronic Supplementary Mate-
the hospital to the intervention group remained a statisti- rial Figs. E3 and E4). Other subgroups (patients with sep-
cally significant risk factor for mortality in the multiple tic shock, patients who developed sepsis on the ICU) did
imputation analysis but not in the complete data analysis. not differ from the main analysis (Electronic Supplemen-
There was a significant increase of the risk of death of 2% tary Material Table E5).
per hour delay of empiric AT independent of the group
assignment when time to AT was added to the model Discussion
(Electronic Supplementary Material Table E2). This multifaceted intervention compared to standard
Only those 1830 patients requiring surgical source CME neither had a significant impact on time to empiric
control were included into another hierarchical model. AT nor on 28-day mortality. Secondly, independent of
Time to source control was significantly associated with the study arm, every hour delay of antimicrobial therapy
28-day mortality (Electronic Supplementary Material and measures for source control was associated with an
Table E3) corresponding to an increased risk of death of increased risk of patient death of 2% and 1%, respectively.
1% per hour delay of surgical source control (Electronic Several studies achieved an improvement in compli-
Supplementary Material Fig. E2). Patients who had surgi- ance with guideline recommendations associated with
cal source control delayed for more than 6 h had a higher marked reductions in hospital mortality [6, 17, 19, 29].
28-day mortality [202/568 (35.6%), vs. 337/1207 (27.9%), The failure of our multifaceted study intervention to
p < 0.001]. reduce 28-day mortality contrasts results of these previ-
Secondary outcomes are shown in Table 2. The median ous studies. However, we also were unable to significantly
time to AT tended to be shorter by 30 min in the inter- reduce time to AT or affect AT de-escalation. The imple-
vention group compared to the control group (p = 0.41). mentation of the multifaceted approach was delivered
There was no significant difference between the two and supported as intended by the central study coordi-
groups in the fraction of patients receiving AT within nation team. Semi-structured expert interviews with the
the first hour. Compliance with guidelines regarding local quality improvement team leaders at the end of the
blood culturing tended to be higher in the intervention intervention blamed failure to establish an effective mul-
than in the control group without reaching statistical tidisciplinary quality improvement team [30]. The inter-
significance. viewees criticized insufficient staff time dedicated to the
We excluded university hospitals to visualize effects project, lack of human resources, lack of support by other
unaffected by the large academic centers resulting in department heads and hospital leadership, and especially
1278 patients from 20 hospitals in the control group failure to achieve consistent involvement of emergency
and 1289 patients from 17 hospitals in the interven- departments and involvement of regular wards in the
tion group. Administration of early AT was maintained quality improvement efforts. Interestingly, a hospital with
during the whole study only in the intervention group a comprehensive sepsis quality improvement program
 A

Fig. 2 Proportion and 95% confidence intervals of patients receiving antimicrobials within the first hour after onset of severe sepsis during the trial
(a) and 28-day mortality during the trial (b) for the control and the intervention group. In the control group, 28-day mortality changed from 24.2%
(69/285) in the first quarter of the trial to 24.2% (31/128, p = 1.00) in the last quarter compared to 33.1% (123/372, first quarter) vs. 34.9% (87/249,
p = 0.69, last quarter) in the intervention group

including financing and staff support was successful in rapid response teams—a concept very common in other
improving time to AT and decreasing sepsis mortality countries [32, 33]. In addition, the study centers did not
[31]. receive any financial support for the quality initiative
More than 45% of the patients in the MEDUSA trial within this trial. All these factors may have contributed
had developed sepsis outside the ICU. It is therefore rea- to the failure of this quality improvement approach. To
sonable to assume that difficulty to roll out the change overcome barriers may be more difficult in larger hospi-
process to all pertinent stakeholders was an important tals where more personnel are involved. Our finding, that
barrier to change. None of the participating hospitals had time to AT was successfully affected in smaller hospitals
Table 2 Primary and secondary outcomes
Intervention n = 2596 Control n = 1587 Missing valuesa Adjusted p valueb ICC

28-day mortality 883 (35.1%) 403 (26.7%) 80/75 0.01 0.03

ICU mortality 848 (32.7%) 412 (26.0%) 2/4 0.08 0.03
Hospital mortality 1068 (41.2%) 517 (32.7%) 3/5 0.02 0.03
ICU length of stay, median (IQR), day 11 (5–25) 13 (6–29) 2/6 0.31 0.07
Hospital length of stay, median (IQR), day 29 (16–47) 32 (17–52) 5/5 0.37 0.05
AT ≤1 h 1019 (39.6%) 563 (36.0%) 20/24f 0.55 0.06
AT after onset of sepsis onlyc 509 (24.6%) 308 (23.5%) 0.83 0.06
Time to AT, median (IQR), h 1.5 (0.1–4.9) 2.0 (0.4–5.9) 206/107g 0.41 0.1
c
AT after onset of sepsis only 2.5 (1.0–6.1) 3.0 (1.0–7.2) 0.62 0.1
Time to focus controld, median (IQR), h 2.0 (−0.9 to 10.7) n = 1171 2.0 (−1.8 to 9.0) n = 708 1448/905h 0.83 0.09
≥2 sets of blood cultures 1675 (64.8%) 832 (52.7%) 13/9 0.06 0.08
Adequate empiric ATe 1453 (56.0%) 860 (54.4%) 2/5 0.82 0.09
AT antimicrobial therapy, ICC intra-cluster correlation coefficient (a higher ICC reflects a higher between-cluster variance, see electronic supplement for detailed
explanation), ICU intensive care unit, IQR interquartile range
a
Missing values for intervention/control group
b
p values on patient level adjusted for clustering (see electronic supplement for detailed explanation)
c
Time to AT was negative if AT was prescribed before onset of infection-related organ dysfunction
d
Only patients with source control
e
Inadequate AT was defined as AT escalation within 5 days after onset of sepsis
f
Cases with no new AT given for sepsis-related organ dysfunction are defined as not receiving AT within first hour; cases with no information regarding new AT are
considered as missing
g
Only patients where new AT was given for sepsis-related organ dysfunction are considered; 20 cases of intervention group and 24 cases of control group had no
information on new AT
h
Only patients who received focus control; 23 cases of intervention group and 28 cases of control group had no information on focus control

Table 3 Hierarchical model with endpoint 28-day mortality and center as random effect
Complete data analysis Multiple imputation analysis
Odds ratio (95% CI) p value Odds ratio (95% CI) p value

Treated in hospital of interventions group 1.28 (0.91, 1.80) 0.15 1.40 (1.03, 1.90) 0.03
Age, years 1.02 (1.02, 1.03) <0.001 1.02 (1.02, 1.03) <0.001
SAPS IIa 1.03 (1.02, 1.03) <0.001 1.03 (1.02, 1.04) <0.001
Surgical source control necessary 0.88 (0.73, 1.07) 0.20 0.89 (0.75, 1.06) 0.19
Unscheduled surgery 0.82 (0.67, 1.00) 0.05 0.84 (0.70, 1.00) 0.06
log (lactate), mmol/l 1.69 (1.50, 1.91) <0.001 1.71 (1.52, 1.91) <0.001
Platelets (square root) 0.96 (0.94, 0.98) <0.001 0.97 (0.96, 0.99) 0.002
Base excess, mmol/l 1.01 (0.99, 1.02) 0.30 1.01 (1.00, 1.02) 0.22
Vasopressor required 0.94 (0.76, 1.18) 0.61 0.97 (0.79, 1.19) 0.76
Inadequate AT 1.41 (1.19, 1.67) <0.001 1.41 (1.21, 1.64) <0.001
Renal replacement therapy 1.61 (1.19, 2.17) 0.002 1.52 (1.16, 2.00) 0.002
Number of organ dysfunctions 1.08 (0.99, 1.17) 0.08 1.10 (1.02, 1.18) 0.01
The complete data analysis without imputation of missing values was built with 3353 patients from 38 hospitals; multiple imputation analysis used 4183 cases and 40
hospitals in 100 imputed data sets
AT antimicrobial therapy, CI confidence interval, SAPS II simplified acute physiology score II
a
SAPS II is calculated from 12 routinely measured physiological and biochemical variables within the first 24 h of intensive care unit admission. The range varies from
0 to 163 points where higher values reflect a higher risk of in-hospital death

only, supports this hypothesis and is in accordance with Our median time to initial antimicrobials was already
the finding that patients treated in academic hospitals are short compared to other studies with a greater impact of
at higher risk of receiving a delayed AT [34]. quality improvement efforts [6, 19]. Educational efforts
may be less effective if baseline performance is already The multifaceted approach in the intervention group
good [16]. However, the wide interquartile range for might have facilitated motivation differences between
time to initial AT indicates that there was great room for the groups. However, later intervention centers already
improvement also in our study. documented more patients in the previous run-in phase
The increase in 28-day mortality by 2% for each 1 h than the later centers of the control group [25]. Thus, the
delay in initiating antimicrobial therapy in this study difference in recruitment also suggests randomization
is considerably lower than the reported 7.5% mortality failure rather than attrition bias. Cluster-randomized tri-
increase [2]. However, it is almost identical to the results als are prone to randomization failure when the number
from a large retrospective study [35]. This contrasts the of clusters is low [36]. We tried to minimize this risk by
conclusion of a meta-analysis which failed to demon- stratifying the randomization by the median time to AT
strate that antimicrobial therapy within 1 h after severe of each center in the run-in phase. Other predictors of
sepsis recognition improves survival and therefore ques- mortality might have been more suitable for stratifica-
tioned the evidence base for the respective SSC guideline tion. However, time to AT as the intervention target was
recommendations [10]. Our study confirms similar find- supposed to be equally distributed between the groups.
ings from the run-in phase of this trial on the impact of The number of hospitals limited further stratification.
surgical source control on 28-day mortality [25]. Findings Fourthly, we did not assess adequacy of AT by microbio-
from other studies underline the importance of timely logical susceptibility testing to limit the workload of the
surgical source control [8, 9] and suggest reconsideration participating hospitals. This pragmatic definition might
of the actual SSC guideline recommendation to allow a have partly been responsible for the low rate of adequacy
time window up to 12 h for source control [10]. of only 54–56% as physicians may empirically escalate AT
Strengths of our study include the cluster-randomized, when patients start to deteriorate. Fifthly, patients that
controlled design and the high external validity as sites were never treated on the ICU were not included into
of all levels of care had participated in the trial. All of the study. Promoting early AT might reduce the number
this contributes to the generalizability of the findings to of ICU admissions. We consider the chance of this bias
similar healthcare settings. Our study has several limi- as low as patients with acute organ dysfunction are fre-
tations. Firstly, the study was underpowered to detect quently admitted to the ICU.
the expected reduction of 7.5% mortality because of the In conclusion, our findings confirm that delay in anti-
dropout of four centers. Our results and recently pub- microbial therapy and source control was associated
lished studies suggest a far smaller effect of reduced time with increased mortality but the multifaceted approach
to AT [35]. The cluster-randomized trial design might improved time to empiric antibiotics only in non-aca-
therefore not be feasible to study this issue since too demic hospitals without affecting outcome. Our study
many clusters would be necessary. Different designs like results suggest that the success of quality improvement
the cluster-randomized cross-over trial or the stepped in sepsis care depends on the existence of an institutional
wedge design might be more appropriate. The lack of patient safety-centered culture and the capability for
power does not limit the conclusion of our study since interdisciplinary cooperation.
the intervention did not significantly affect time to AT
Electronic supplementary material
and could therefore also not affect mortality. Secondly, The online version of this article (doi:10.1007/s00134-017-4782-4) contains
overall 28-day mortality was higher in the intervention supplementary material, which is available to authorized users.
group at baseline than in the control group. This differ-
Author details
ence might be attributable to higher disease severity 1
Center for Sepsis Control and Care (CSCC), Jena University Hospital, Jena,
scores and other important differences in patient charac- Germany. 2 Department of Anesthesiology and Intensive Care Medicine, Jena
teristics (e.g., higher proportion of healthcare-associated University Hospital, Jena, Germany. 3 Institute for Medical Informatics, Statistics
and Epidemiology, University of Leipzig, Leipzig, Germany. 4 Service Préven‑
than community-associated infections). Allocation bias tion et Contrôle de l’Infection, Hôpitaux Universitaires de Genève, Geneva,
by randomization failure might explain the observed dif- Switzerland. 5 Department of Anaesthesia and Intensive Care Medicine, Zen‑
ferences. The higher risk of death associated with alloca- tralklinik Bad Berka GmbH, Bad Berka, Germany. 6 Department of Anesthesiol‑
ogy and Intensive Care Medicine, University Hospital Greifswald, Greifswald,
tion to the intervention group only disappeared in the Germany. 7 Department of Surgery and the Li Ka Shing Knowledge Institute,
multivariate analysis of the complete data set but not in St Michael’s Hospital, University of Toronto, Ontario, Canada. 8 Department
the multiple imputation analysis. Thirdly, the study cent- of Anesthesiology and Intensive Care Medicine, University Hospital Leipzig,
Leipzig, Germany. 9 Division of Pulmonary and Critical Care Medicine, Warren
ers of the intervention group included considerably more Alpert Medical School of Brown University, Providence, RI, USA. 10 Depart‑
patients into the study than control centers, although ment of Anesthesiology, University Hospital Ulm, Ulm, Germany. 11 Depart‑
hospital size was similar between the groups. This finding ment of Anesthesiology and Intensive Care Medicine, University Hospital
Oldenburg, Oldenburg, Germany. 12 Department of Anesthesiology, Surgical
suggests that the motivation and adherence to the study Intensive Care Medicine and Pain Therapy, Vivantes Hospital Neukölln, Berlin,
inclusion criteria differed between the two study groups. Germany. 13 Department of Intensive Care Medicine, University Hospital RWTH
Aachen, Aachen, Germany. 14 Department of Anesthesiology and Intensive Care and Emergency Medicine: Rüdinger Sinz; Thüringen-Kliniken Saalfeld,
Care Medicine, University Hospital Rostock, Rostock, Germany. 15 Depart‑ Department of Anesthesiology, Intensive Care Medicine, and Pain Therapy:
ment of Anesthesiology, Intensive Care, Transfusion and Emergency Medicine Petra Bautz, Annemarie Fischer; Ev. Jung-Stilling Hospital Siegen, Depart‑
and Pain Therapy, Bethel Hospital Bielefeld, Bielefeld, Germany. 16 Depart‑ ment of Anesthesiology, Intensive Care, and Emergency Medicine: Armin
ment of Internal Medicine, University Hospital Tübingen, Tübingen, Germany. Seibel, Christoph Fleischhacker; University Hospital Tübingen; Department
17
Department of Anesthesiology and Intensive Care Medicine, Hospital of Anesthesiology: Helene Häberle, Philipp Henn, Friederike Mezger, Peter
of the Bundeswehr Berlin, Berlin, Germany. 18 Department of Anesthesiology Rosenberger; University Hospital Tübingen; Dept of Internal Medicine: Reimer
and Intensive Care Medicine, University Medical Center Kiel, Kiel, Germany. Riessen, Silvia Ziegler; University Hospital Medical School Ulm, Clinic of Anaes‑
19
Department of Anesthesiology, Intensive Care Medicine and Pain Therapy, thesiology: Eberhard Barth, Hendrik Bracht, I. Heymann, A. Hinder, R. Sens,
Helios Hospital Erfurt, Erfurt, Germany. 20 Department of Anesthesiology Manfred Weiss; Hufeland Hospital Weimar; Department of Anesthesiology and
and Intensive Care Medicine, University Hospital Tübingen, Tübingen, Ger‑ Intensive Care Medicine: Christof Lascho, Henriette Micke, Falk Schmidt; Helios
many. 21 Department of Intensive Care Medicine, Helios Hospital Wuppertal, Hospital Wuppertal. Department of Intensive Care Medicine: Stefanie Schilling,
Wuppertal, Germany. 22 University Hospital Ulm, Ulm, Germany. Gabriele Wöbker.

Acknowledgements Clinical data collection and management

Matthias Löbe, Frank Meineke, Christine Pausch, Christoph Engel (Institute
MEDUSA contributing sites and site investigators, alphabetical by site of Medical Informatics, Statistics and Epidemiology, University of Leipzig);
University Hospital RWTH Aachen, Department of Intensive Care Medicine: statistical support for protocol development: Heike Hoyer (Institute for Medi‑
Gernot Marx, Achim Schindler, Tobias Schürholz; Ilm-Kreis-Kliniken Arnstadt, cal Statistics, Information, and Documentation: Jena University Hospital);
Department of Anesthesiology and Intensive Care Medicine: Heike Schlegel- statistical support for implementation of the multiple imputation method: S.
Höfner, Gunther Lehmann, Annett Sander, Steffen Friese, Christian Scholz; Grund, A. Robitzsch (Leibnitz Institute for Science and Mathematics Education,
Helios Hospital Aue, Department of Anesthesiology and Intensive Care Medi‑ University Kiel).
cine: Pia Fischer; Zentralklinik Bad Berka GmbH, Department of Anaesthesia
and Intensive Care Medicine: Christina Fuchs, Lutz Becher, Norbert Salewsky, Compliance with ethical standards
Torsten Schreiber; Charité Berlin, Department of Anesthesiology and Operative
Intensive Care Medicine: Anton Goldmann, Didier Keh, Katrin Schmid; Hufe‑ Conflicts of interest
land-Klinikum Bad Langensalza, Department of Anesthesiology and Intensive All authors have completed and submitted the ICMJE Form for Disclosure of
Care Medicine: Winfried Menning, Renate Steuckart; Bundeswehrkrankenhaus Potential Conflicts of Interest. Dr. Bloos reported receiving lecture honoraria
Berlin, Department of Anesthesiology and Intensive Care Medicine: Robert from biosyn, Gilead, and CSL Behring. Dr. Harbarth reported receiving personal
Barz, Karin Dey, Meike Fahrenholz, Martin Müller; Vivantes Klinikum Neukölln- fees from Johnson & Johnson and Novartis and grants from Pfizer and BioMer‑
Berlin, Department of Anesthesiology, Surgical Intensive Care Medicine, and ieux. Dr. Levy reported receiving grants and personal fees from ImmuneEx‑
Pain Therapy: Herwig Gerlach, Susanne Toussaint; Helios Hospital Berlin-Buch, press. Prof. Reinhart reported receiving grants and non-financial support from
Department of Intensive Care Medicine: Jörg Brederlau; Ev. Hospital Bielefeld, biosyn AG and Thermofischer/BRAHMS, being a shareholder of InflaRx Jena,
Department of Anesthesiology, Emergency and Intensive Care Medicine, and and receiving personal fees from Adrenomed. Dr. Schuerholz reported grants
Pain Therapy: Friedhelm Bach, Dirk Buschmann, Ingo Gummelt, J. Hoeschen, and personal fees from Astellas Pharma, grants from B Braun Melsungen and
Marion Klaproth, Ina Vedder; HELIOS-Hospital St. Josefs-Hospital Bochum-Lin‑ Köhler Chemie, personal fees from Bayer Healthcare. Dr. Weyland reported
den, Department of Anesthesiology: Ulrike Bachmann-Holdau; St. Georg Hos‑ receiving recruitment fees from Jena University Hospital for another trial.
pital Eisenach, Department of Anesthesiology and Intensive Care Medicine: Dr. Bach, Dr. Dey, Dr. Engel, Dr. Gerlach, Dr. Gründling, Dr. Häberle, Dr. Kaisers,
Jürgen Eiche, Rolf Hauschild; Hospital Rudolf Elle, Eisenberg, Department of Dr. Marshall, C. Matthäus-Krämer, Dr. Meier-Hellmann, Dr. Poidinger, Dr.
Anesthesiology and Intensive Care Medicine: Martina Lange, Davia Herrmann- Riessen, Dr. Rüddel, Dr. Scheer, Dr. Schreiber, D. Schwarzkopf, Dr. Simon, Dr.
Karbaum; Helios-Hospital Emil-von Behring, Department of Interdisciplinary Thomas-Rüddel, Dr. Weiler, Dr. Weiss, and Dr. Woebker reported no conflicts of
Intensive Care and Emergency Medicine: Annette Lubasch, Marcus Rücker; interest.
Helios-Hospital Erfurt, Department of Anesthesiology and Intensive Care
Medicine: Christian Icke, Alexander Lucht, Andreas Meier-Hellmann, Jan Wag‑ Funding/support
ner; Catholic Hospital St. Johann Nepomuk Erfurt, Department of Anesthesiol‑ The study was funded by the German Federal Ministry of Education and
ogy and Intensive Care Medicine: Olaf Arnold, Steffen Kästner, Tobias Clausen; Research via the integrated research and treatment center “Center for Sepsis
Hospital Friedberg, Department of Internal Medicine: Michael Sternkopf, Control and Care” (FKZ 01EO1002).
Robert Voswinckel; SRH Waldklinikum Gera, Department of Anesthesiology
and Intensive Care Medicine: T. Benndorf, Christel Eiserloh, Gerhard Kuhnle,
Mathias Koch; University Hospital Greifswald, Department of Anesthesiol‑ Received: 5 October 2016 Accepted: 21 March 2017
ogy and Intensive Care Medicine: Manuela Gerber, Matthias Gründling, Liane
Guderian, Sven-Olaf Kuhn, Christian Scheer; Hospital Ilmenau; Department of
Anesthesiology and Intensive Care Medicine: Gerd Scheiber; Jena University
Hospital, Center for Sepsis Control & Care/Department of Anesthesiology and
Intensive Care Medicine: Frank Bloos, Susann Christink, Martina Kortegast,
Claudia Matthäus-Krämer, Marcel Mücke, Bernhard Poidinger, Hendrik-Rüddel, References
Ulrike Redlich, Daniel Schwarzkopf, Daniel Thomas-Rüddel, Christel Volkmer; 1. Levy MM, Artigas A, Phillips GS et al (2012) Outcomes of the Surviv‑
University Hospital Kiel, Department of Anesthesiology and Intensive Care ing Sepsis Campaign in intensive care units in the USA and Europe: a
Medicine: Stefanie D’Aria, Thees Lemke, Birgit Michaelsen, Dirk Schädler, Nina prospective cohort study. Lancet Infect Dis 12:919–924. doi:10.1016/
Schulz-Ruhtenberg, Norbert Weiler; Hospital Landshut-Achdorf, Department S1473-3099(12)70239-6
of Anesthesiology and Surgical Intensive Care Medicine: Martin Anetseder, 2. Kumar A, Roberts D, Wood KE et al (2006) Duration of hypotension
Zoran Textor; University Hospital Leipzig, Department of Anesthesiology and before initiation of effective antimicrobial therapy is the critical deter‑
Intensive Care Medicine: Udo Kaisers, Philipp Simon; Hospital Meiningen, minant of survival in human septic shock. Crit Care Med 34:1589–1596.
Department of Intensive Care and Emergency Medicine: Georg Braun, Nicole doi:10.1097/01.CCM.0000217961.75225.E9
Jensen, Werner Gegenfurtner, Alexander Meinhardt, Robert Schmitt, Andrea 3. Larché J, Azoulay E, Fieux F et al (2003) Improved survival of critically ill
Teichert; Saale-Unstrut-Hospital Naumburg, Department of Anesthesiology cancer patients with septic shock. Intensive Care Med 29:1688–1695.
and Intensive Care Medicine: Klaus-Dieter Becker; Hospital Oldenburg, Depart‑ doi:10.1007/s00134-003-1957-y
ment of Anesthesiology, Intensive Care Medicine; Emergency Medicine, and 4. Gaieski DF, Mikkelsen ME, Band RA et al (2010) Impact of time to antibiot‑
Pain Therapy: Anja Diers, Florian Jelschen, Andreas Weyland; Thüringen-Klinik ics on survival in patients with severe sepsis or septic shock in whom
Pößneck, Department of Anesthesiology and Intensive Care Medicine: Frieder early goal-directed therapy was initiated in the emergency department.
Knebel, Thomas Kupfer; Asklepios Hospital Radeberg, Department of Intensive Crit Care Med 38:1045–1053. doi:10.1097/CCM.0b013e3181cc4824
5. Clec’h C, Timsit JF, De Lassence A et al (2004) Efficacy of adequate 22. van der Veer SN, de Vos MLG, van der Voort PHJ et al (2013) Effect of a
early antibiotic therapy in ventilator-associated pneumonia: influence multifaceted performance feedback strategy on length of stay compared
of disease severity. Intensive Care Med 30:1327–1333. doi:10.1007/ with benchmark reports alone: a cluster randomized trial in intensive
s00134-004-2292-7 care. Crit Care Med 41:1893–1904. doi:10.1097/CCM.0b013e31828a31ee
6. Noritomi DT, Ranzani OT, Monteiro MB et al (2014) Implementation of a 23. Marsteller JA, Sexton JB, Hsu Y-J et al (2012) A multicenter, phased,
multifaceted sepsis education program in an emerging country setting: cluster-randomized controlled trial to reduce central line-associated
clinical outcomes and cost-effectiveness in a long-term follow-up study. bloodstream infections in intensive care units. Crit Care Med 40:2933–
Intensive Care Med 40:182–191. doi:10.1007/s00134-013-3131-5 2939. doi:10.1097/CCM.0b013e31825fd4d8
7. Boyer A, Vargas F, Coste F et al (2009) Influence of surgical treatment tim‑ 24. Writing Group for the CHECKLIST-ICU Investigators and the Brazilian
ing on mortality from necrotizing soft tissue infections requiring inten‑ Research in Intensive Care Network (BRICNet), Cavalcanti AB, Bozza FA
sive care management. Intensive Care Med 35:847–853. doi:10.1007/ et al (2016) Effect of a quality improvement intervention with daily round
s00134-008-1373-4 checklists, goal setting, and clinician prompting on mortality of critically
8. Buck DL, Vester-Andersen M, Møller MH, Danish Clinical Register of Emer‑ ill patients: a randomized clinical trial. JAMA 315:1480–1490. doi:10.1001/
gency Surgery (2013) Surgical delay is a critical determinant of survival in jama.2016.3463
perforated peptic ulcer. Br J Surg 100:1045–1049. doi:10.1002/bjs.9175 25. Bloos F, Thomas-Rüddel D, Rüddel H et al (2014) Impact of compliance
9. Azuhata T, Kinoshita K, Kawano D et al (2014) Time from admission to with infection management guidelines on outcome in patients with
initiation of surgery for source control is a critical determinant of survival severe sepsis: a prospective observational multi-center study. Crit Care
in patients with gastrointestinal perforation with associated septic shock. 18:R42. doi:10.1186/cc13755
Crit Care 18:R87. doi:10.1186/cc13854 26. Hemming K, Girling AJ, Sitch AJ et al (2011) Sample size calculations for
10. Rhodes A, Evans LE, Alhazzani W et al (2017) Surviving Sepsis Campaign: cluster randomised controlled trials with a fixed number of clusters. BMC
international guidelines for management of sepsis and septic shock: Med Res Methodol 11:102. doi:10.1186/1471-2288-11-102
2016. Intensive Care Med. doi:10.1007/s00134-017-4683-6 27. Wears RL (2002) Advanced statistics: statistical methods for analyzing
11. Grimshaw J, Campbell M, Eccles M, Steen N (2000) Experimental and cluster and cluster-randomized data. Acad Emerg Med 9:330–341
quasi-experimental designs for evaluating guideline implementation 28. Grund S, Lüdtke O, Robitzsch A (2016) Multiple imputation of multilevel
strategies. Fam Pract 17(Suppl 1):S11–S16 missing data. SAGE Open. doi:10.1177/2158244016668220
12. Ioannidis JP, Haidich AB, Pappa M et al (2001) Comparison of evidence 29. Miller RR, Dong L, Nelson NC et al (2013) Multicenter implementation of
of treatment effects in randomized and nonrandomized studies. JAMA a severe sepsis and septic shock treatment bundle. Am J Respir Crit Care
286:821–830. doi:10.1001/jama.286.7.821 Med 188:77–82. doi:10.1164/rccm.201212-2199OC
13. Lilly CM (2014) The ProCESS trial—a new era of sepsis management. N 30. Matthaeus-Kraemer CT, Thomas-Rueddel DO, Schwarzkopf D et al (2015)
Engl J Med 370:1750–1751. doi:10.1056/NEJMe1402564 Barriers and supportive conditions to improve quality of care for criti‑
14. Sterling SA, Miller WR, Pryor J et al (2015) The impact of timing of cally ill patients: a team approach to quality improvement. J Crit Care
antibiotics on outcomes in severe sepsis and septic shock: a systematic 30:685–691. doi:10.1016/j.jcrc.2015.03.022
review and meta-analysis. Crit Care Med 43:1907–1915. doi:10.1097/ 31. Scheer CS, Fuchs C, Kuhn S-O et al (2017) Quality improvement initia‑
CCM.0000000000001142 tive for severe sepsis and septic shock reduces 90-day mortality: a
15. Rhee C, Gohil S, Klompas M (2014) Regulatory mandates for sepsis 7.5-year observational study. Crit Care Med 45:241–252. doi:10.1097/
care—reasons for caution. N Engl J Med 370:1673–1676. doi:10.1056/ CCM.0000000000002069
NEJMp1400276 32. Cross G, Bilgrami I, Eastwood G et al (2015) The epidemiology of sepsis
16. Ivers N, Jamtvedt G, Flottorp S et al (2012) Audit and feedback: effects on during rapid response team reviews in a teaching hospital. Anaesth
professional practice and healthcare outcomes. Cochrane Database Syst Intensive Care 43:193–198
Rev 6:CD000259. doi:10.1002/14651858.CD000259.pub3 33. Jäderling G, Bell M, Martling C-R et al (2013) ICU admittance by a rapid
17. Shiramizo SCPL, Marra AR, Durão MS et al (2011) Decreasing mortality response team versus conventional admittance, characteristics, and out‑
in severe sepsis and septic shock patients by implementing a sepsis come. Crit Care Med 41:725–731. doi:10.1097/CCM.0b013e3182711b94
bundle in a hospital setting. PLoS One 6:e26790. doi:10.1371/journal. 34. Amaral ACKB, Fowler RA, Pinto R et al (2016) Patient and organizational
pone.0026790 factors associated with delays in antimicrobial therapy for septic shock.
18. Tipler PS, Pamplin J, Mysliwiec V et al (2013) Use of a protocolized Crit Care Med 44:2145–2153. doi:10.1097/CCM.0000000000001868
approach to the management of sepsis can improve time to first dose of 35. Ferrer R, Martin-Loeches I, Phillips G et al (2014) Empiric antibiotic treat‑
antibiotics. J Crit Care 28:148–151. doi:10.1016/j.jcrc.2012.08.021 ment reduces mortality in severe sepsis and septic shock from the first
19. Ferrer R, Artigas A, Levy MM et al (2008) Improvement in process of care hour: results from a guideline-based performance improvement pro‑
and outcome after a multicenter severe sepsis educational program in gram. Crit Care Med 42:1749–1755. doi:10.1097/CCM.0000000000000330
Spain. JAMA 299:2294–2303. doi:10.1001/jama.299.19.2294 36. Leon AC, Demirtas H, Li C, Hedeker D (2013) Subject-level matching for
20. Garrouste-Orgeas M, Soufir L, Tabah A et al (2012) A multifaceted pro‑ imbalance in cluster randomized trials with a small number of clusters.
gram for improving quality of care in intensive care units: IATROREF study. Pharm Stat 12:268–274. doi:10.1002/pst.1580
Crit Care Med 40:468–476. doi:10.1097/CCM.0b013e318232d94d
21. Scales DC, Dainty K, Hales B et al (2011) A multifaceted intervention for
quality improvement in a network of intensive care units: a cluster rand‑
omized trial. JAMA 305:363–372. doi:10.1001/jama.2010.2000