Академический Документы
Профессиональный Документы
Культура Документы
Department of Anaesthesiology and Intensive Care, IRCCS San Raffaele Scientific Institute, Via Olgettina 60, Milan 20132, Italy
Department of Anaesthesiology and Intensive Care, La Sapienza University, Rome, Italy
& The Author 2015. Published by Oxford University Press on behalf of the British Journal of Anaesthesia. All rights reserved.
For Permissions, please email: journals.permissions@oup.com
Background. Inodilators are commonly used in critically ill patients, but their effect on survival
has not been properly studied to date. The objective of this work was to conduct a network
meta-analysis on the effects of inodilators on survival in adult cardiac surgery patients, and
to compare and rank drugs that have not been adequately compared in head-to-head trials.
BJA
surgery that reported mortality data. We therefore performed
a Bayesian network meta-analysis of randomised trials
to grade all the inodilators in order to allow physicians to optimize cost-effectiveness analyses in their centres. A network
meta-analysis is a statistical technique for comparison of different treatments that were never properly studied through
adequately powered randomised controlled trials, but that
can be compared through a third common comparator.6 7 On
the basis of statistical inference, it is possible to identify the
superior treatment, reaching, through indirect comparison, reliable conclusions otherwise hard or impossible to achieve.
Methods
Study selection
Statistical analysis
Page 2 of 11
Greco et al.
BJA
Table 1 Description of the 46 studies included in the meta-analysis. CPB, cardiopulmonary bypass; OP, off-pump; CABG, coronary artery bypass
graft; ICU, intensive care unit; NA, not available. *Study published as abstract only
Year
Surgery
Contrast
Number of
patients
Longest follow-up
Abacilar AF9
2013
CABG
100 vs 100
5 days
Al-Shawaf E10
2006
CABG
Levosimendan vs milrinone
14 vs 16
48 h
Alvarez J11
2005
CPB
Levosimendan vs dobutamine
15 vs 15
Hospital stay
Alvarez J12
2006
CABG, valve
Levosimendan vs dobutamine
25 vs 25
15 days
Arbeus M13
2009
CABG
22 vs 22
Hospital stay
Barisin S14
2004
OPCABG
21 vs 10
Hospital stay
Bautin A*15
2011
NA
33 vs 31
Hospital stay
Baysal A16
2014
67 vs 66
30 days
Carmona MJ17
2010
NA
Milrinone vs dobutamine
10 vs 10
2 years
Couture P18
2007
CABG
25 vs 25
Hospital stay
De Hert SG19
2007
CPB
Levosimendan vs milrinone
15 vs 15
38 days
Doolan LA20
1997
CPB
15 vs 15
30 days
Ensinger H21
1999
NA
8 vs 8
3 days
Erb J22
2014
19 vs 18
6 months
Eriksson HI23
2009
CABG
30 vs 30
33 days
Ersoy O24
2013
Valve surgery
10 vs 10
24 h
Feneck RO25
2001
NA
Milrinone vs dobutamine
60 vs 60
3 days
Gandham R26
2013
Levosimendan vs dobutamine
30 vs 30
36 h
Hachenberg T27
1997
Enoximone vs dobutamine
10 vs 10
72 h
Hadadzadeh M28
2013
OPCABG
40 vs 40
ICU stay
Hayashida N29
1999
CABG
12 vs 12
72 h
Husedzinovic I30
2005
OPCABG
12 vs 13
60 min
Jarvela K31
2008
12 vs 12
1 year
Jebeli M32
2010
CABG
35 vs 35
Hospital stay
Jo HR33
2010
OPCABG
20 vs 20
Hospital stay
Kodali RK34
2013
OPCABG
15 vs 15
24 h
Kwak YL35
2004
OPCABG
29 vs 33
Hospital stay
Lahtinen P36
2009
103 vs 104
Hospital stay
Lee JH37
2006
OPCABG
24 vs 26
Hospital stay
Leppikangas H38
2011
12 vs 12
4 days
Levin R39
2008
NA
Levosimendan vs dobutamine
69 vs 68
Hospital stay
Levin R40
2012
CABG
127 vs 125
48 h
Continued
Page 3 of 11
First author
BJA
Greco et al.
Table 1 Continued
Year
Surgery
Contrast
Number of
patients
Longest follow-up
Lilleberg J41
1998
CABG
15 vs 8
1h
Modine T42
2005
OPCABG
16 vs 16
30 days
Mollhoff T43
1999
CABG
11 vs 11
Hospital stay
Nijhawan N44
1999
CPB
12 vs 6
8h
Parviainen I45
1995
CABG
11 vs 11
90 min
Ristikankare A46
2012
CABG
30 vs 30
Hospital stay
Shah B47
2014
OPCABG
25 vs 25
30 days
Sharma P48
2013
20 vs 20
Hospital stay
Song JW49
2011
OPCABG
31 vs 31
Tritapepe L50
2006
CABG
12 vs 12
Hospital stay
Tritapepe L51
2009
CABG
53 vs 53
30 days
2007
17 vs 17
18 h
this Bayesian network meta-analysis were 1) placebo, 2) levosimendan, 3) enoximone, 4) milrinone, 5) dobutamine. The pairwise
association between each treatment (also called direct estimate)
was delineated by a graphical representation of the network. In
the diagram each node represents a single anaesthetic agent
and each edge connects treatments that have been directly compared in one or more RCTs.
We performed pairwise meta-analyses to look at the direct
comparisons. In the standard meta-analyses, a magnitude
of heterogeneity between-studies was represented by the
degree of inconsistency (I 2), while uncertainty over whether
apparent heterogeneity due to the effects of chance (random
error) was expressed using p-value from Cochrane Q statistic.
The presence of effect-modifiers due heterogeneity was considered acceptable with one chi-square P-value .0.10. Mortality data from individual studies were analysed to compute
pooled OR with pertinent 95% confidence intervals (CI) by
means of the inverse of variance method with fixed effect
model in case of low statistical inconsistency (I 2 25%) or by
means of DerSimonian-Laird method with random effect
model (which better accommodates clinical and statistical variations) in case of moderate or high statistical inconsistency
(I 2 .25%). The number need to treat (NNT) was computed
for the significant associations. The robustness of the standard
meta-analysis was assessed by sequentially removing each
study from the overall dataset and reanalysing the remaining
datasets.
The Bayesian network analysis was carried out modeling the
outcome mortality with the Bayesian hierarchical model (binomial model with logit link function) using Markov chain Monte
Page 4 of 11
First author
BJA
Results
Description of included trials
Figure 1 shows the flowchart for the selection of randomised
trials. The references of major exclusions are available after contacting the authors. Forty-six randomised clinical trials9 53 were
included in the final analysis (Table 1). The study characteristics
of included trials are detailed in Supplementary Appendix S3.
Page 5 of 11
BJA
Greco et al.
Placebo
OR
)
83
0.
5
.3 s
(0 die
4
5 tu
0. 0 s
R= 2
.0
udy
1 st
OR
=0.
Dobutamine
OR=0.39 (0.160.95)
4 studies
y
tud
1s
Enoximone
Milrinone
Fig 2 Network configuration - for each comparison is displayed the estimate effect and number of studies analysed.
Page 6 of 11
Discussion
The main finding of this study is that only treatment with levosimendan is associated with mortality reduction in cardiac
surgery. This Bayesian network meta-analysis suggested that
levosimendan is associated with the lowest risk of mortality
among the comparators included in the study: the inodilators
PDE-3 inhibitors enoximone and milrinone, the classic dobutamine and placebo. The superiority of levosimendan vs placebo
was confirmed both in the simple pairwise and in Bayesian
network meta-analysis while the superiority of levosimendan
vs dobutamine was present in the pairwise but not confirmed
in the network meta-analysis. These findings should be treated with caution because the statistical significance was lost
when sensitivity analyses including only low risk of bias study
were performed and because no multicentre randomised trial
to confirm these findings exists. Notably, the Bayesian metaanalysis failed to rank inodilators: with the notable exception
of levosimendan, the other studied drugs are either similar
one each other or not properly studied so far not allowing to
reach a statistically significant difference among them in mortality, even using indirect comparisons.
The real effects of inodilators on mortality in cardiac surgery
have been the white whale for many years. Our study summarises the data on mortality from all randomised trials on
inodilators in a single picture, including direct and indirect
comparisons. Levosimendan was superior to placebo and,
overall, it was the best agent to reduce perioperative mortality.
1 stu
dy
43
2 s (0.05
tud
ies 3.53)
15
)
.85
02
(0.4 es
i
07
=1. 2 stud
1
Levosimendan
0
(
stu 0.10
die 1
s 0.
OR
=1
BJA
Table 2 Posterior distribution of mean and 95% credible interval, for the inodilator difference effects, derived by Bayesian hierarchical model with
Markov Chain Monte Carlo algorithm. * Indirect treatment difference effect calculated from consistency equation. Significant treatment difference
effect
Contrast
OR
OR
Levosimendan vs placebo
0.48
Enoximone vs placebo
1.53
0.28 to 0.80
0.48
,0.01 to .100
1.12
0.39 to 3.25
0.28 to 0.80
Milrinone vs placebo
1.12
0.40 to 3.25
Dobutamine vs placebo
1.45
0.44 to 4.68
1.47
0.44 to 4.66
Enoximone vs levosimendan*
3.20
,0.01 to .100
2.32
0.71 to 7.62
Milrinone vs levosimendan*
2.35
0.73 to 7.59
3.06
0.85 to 10.98
3.03
0.83 to 11.04
0.73
,0.01 to .100
Dobutamine vs enoximone*
0.95
,0.01 to .100
1.32
0.27 to 6.41
Dobutamine vs milrinone*
1.29
0.27 to 6.25
Table 3 Sensitivity analyses. Posterior distribution of mean and 95% credible interval, for the inodilator difference effects, derived by Bayesian
hierarchical model with Markov Chain Monte Carlo algorithm. * Indirect treatment difference effect calculated from consistency equation.
Levosimendan vs placebo
Enoximone vs placebo
Inverse-Gamma as prior
distribution (46 trials)
OR
95% credible
interval
OR
OR
OR
0.48
0.28 to 0.81
95% credible
interval
95% credible
interval
0.32
0.01 to 4.49
0.36
0.18 to 0.71
1.06
,0.01 to .100
1.60
,0.001 to .100
0.54
,0.01
95% credible
interval
0.29 to 0.95
,0.01 to .100
Milrinone vs placebo
1.14
0.40 to 3.24
3.33
0.16 to .100
1.02
0.33 to 3.26
1.19
0.41 to 3.55
Dobutamine vs placebo
1.46
0.45 to 4.83
0.03
,0.01 to .100
1.12
0.27 to 4.22
1.65
0.46 to 5.47
3.34
,0.01 to .100
4.39
,0.001 to .100
Enoximone vs
levosimendan*
,0.01
,0.01 to .100
Milrinone vs
levosimendan*
2.36
0.74 to 7.54
10.44
0.11 to .100
2.79
0.75 to 10.45
2.20
0.64 to 7.54
Dobutamine vs
levosimendan*
3.03
0.83 to 11.05
0.08
,0.01 to .100
3.06
0.67 to 13.96
3.04
0.79 to 11.74
Milrinone vs enoximone*
3.13
,0.01 to .100
0.64
,0.001 to .100
,0.01
,0.01 to .100
Dobutamine vs
enoximone*
0.03
,0.01 to .100
0.70
,0.001 to .100
,0.01
,0.01 to .100
Dobutamine vs milrinone*
1.29
0.01
,0.01 to .100
1.10
0.19 to 6.41
0.27 to 6.18
1.39
0.27 to 7.08
Page 7 of 11
Dobutamine vs levosimendan*
Milrinone vs enoximone*
BJA
Greco et al.
Table 4 Posterior distribution of the probability to be the best and the worst for each inodilators drug, derived by Bayesian hierarchical model with
Markov Chain Monte Carlo algorithm
Drugs
Levosimendan
0.5146
0.0001
0.9088
0.0002
Enoximone
0.4424
0.5421
0.0674
0.2874
Milrinone
0.0302
0.1370
Dobutamine
0.0114
0.2605
0.0028
0.1274
Placebo
0.0014
0.0603
0.0028
0.1274
Cumulative prob to be
the best
Cumulative prob to be
the second
Levosimendan
Cumulative prob to be
the third
Enoximone
Placebo
Cumulative prob to be
the fourth
Milrinone
Cumulative prob to be
the worst
Dobutamine
1
0.9
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0
Cumulative prob to be
the best
Levosimendan
Cumulative prob to be
the second
Placebo
Cumulative prob to be
the third
Cumulative prob to be
the worst
Enoximone-Milrinone
Dobutamine
Fig 3 Cumulative rank curves - probability to be the first, the second, the third and the worst for each treatment.
Page 8 of 11
1
0.9
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0
BJA
performed. To confirm the findings of this study, a large multicentre factorial randomised controlled trial comparing levosimendan vs PDE-3 inhibitors, dobutamine and placebo should
be conducted. Alternatively, we should wait for the results of
the two large multicentre randomised trials should comparing
levosimendan vs placebo on top of the best available treatment that are running so far in Europe (NCT00994825) and in
the United States (NCT02025621).
Limitations
Conclusions
This manuscript is the most updated and methodologically
strongest meta-analysis on the effect of inodilators on mortality in cardiac surgery. Using Bayesian network meta-analyses
techniques, we attempted ranking among different inodilators
that have never properly been compared one each other, and
identified levosimendan to be the best inodilator to improve
survival in adult cardiac surgery. Since meta-analyses are hypothesis generating and our findings were not confirmed
when limiting the analysis to low risk of bias studies, these
results should be confirmed by large multicentre RCTs.
Supplementary material
Supplementary material is available at British Journal of
Anaesthesia online.
Authors contributions
T.G.: definition of intellectual content, Literature search, Data
acquisition, Data analysis, Statistical analysis, Manuscript
preparation, Manuscript editing. M.G.C.: literature search,
Data acquisition, Manuscript review. R.D.C.: literature search,
Data acquisition, Manuscript editing. M.G.: design, Literature
Page 9 of 11
The major weakness of indirect meta-analysis is that confounding can be a major bias. That is, the risk status of one
study cohort may be markedly different of another. Furthermore, several RCTs included in our meta-analysis were of suboptimal quality. Nonetheless, our conclusions are based on
41% of studies of high quality and all the sub-analyses performed confirmed that levosimendan treatment improves
survival when compared with placebo. Notably, thirty-six
further studies were excluded because they lacked data on
mortality outcome and authors did not answer to our request
to provide these data. This limited the power of the analysis
and the precision of results. At the same time, we focused on
mortality, the most clinically relevant outcome and the one
less subject to interpretation, we used several databases to
identify the pertinent studies and we focused on randomised
trials only. The comparisons included different numbers of
trials depending on study drugs, while trials differed by
number of included patients, surgical settings and mortality
risk of their populations. Some studies only included patients
with low cardiac output syndrome after cardiac surgery,
without other specifications. A large part of included studies
did not report on funding, thus making more difficult to draw
conclusion on the possible influence of drug companies on
the results of the original papers. Data from trials on enoximone had zero-cell counts and we had to repeat a second analysis combining PDE-3 inhibitors. Dose and length of drugs
administration were heterogeneous and not reported in our
tables, nonetheless they corresponded to local routine practice. At the same time, since inotropic drugs are often considered to be lifesaving in cardiac surgery most trials had
placebo on top of standard treatment as comparator and not
placebo only. Bayesian network meta-analysis incorporates
both direct and indirect comparisons between treatments.
However indirect evidence is susceptible to confounding,6
and thus should be interpreted with caution since it does not
always agree with the corresponding direct estimates.7 Since
the indirect estimate derives results from the direct effects
of two treatments vs a third common comparator, it can confound the overall results because it not always agrees with
the corresponding direct estimates. This bias results in a
greater pooled estimate variability.7 Like heterogeneity in traditional random-effects models of meta-analysis, the comparison between the consistency and the inconsistency model is
used to quantify variation among estimates and is incorporated into the estimate of the confidence interval.60 The discrepancy between consistency and inconsistency estimates
BJA
search, Data analysis, Statistical analysis, Manuscript review.
L.P.: definition of intellectual content, Data acquisition, Manuscript preparation. A.D.: concepts, Design, Definition of intellectual content, Manuscript preparation, Manuscript review. G.L.:
concepts, Design, Definition of intellectual content, Data
analysis, Manuscript preparation, Manuscript review, Guarantor. A.Z.: concepts, Design, Definition of intellectual content,
Manuscript editing, Manuscript review.
Acknowledgements
Greco et al.
14
15
16
Declaration of interest
References
1 Gillies M, Bellomo R, Doolan L, Buxton B. Bench-to-bedside review:
Inotropic drug therapy after adult cardiac surgery -- a systematic
literature review. Crit Care 2005; 9: 266 79
2 Fellahi JL, Fischer MO, Daccache G, Gerard JL, Hanouz JL. Positive
inotropic agents in myocardial ischemia-reperfusion injury: a
benefit/risk analysis. Anesthesiology 2013; 118: 14605
3 Fellahi JL, Parienti JJ, Hanouz JL, Plaud B, Riou B, Ouattara A. Perioperative use of dobutamine in cardiac surgery and adverse
cardiac outcome: propensity-adjusted analyses. Anesthesiology
2008; 108: 979 87
4 Shahin J, DeVarennes B, Tse CW, Amarica DA, Dial S. The relationship
between inotrope exposure, six-hour postoperative physiological
variables, hospital mortality and renal dysfunction in patients
undergoing cardiac surgery. Crit Care 2011; 15: R162
5 Nielsen DV, Hansen MK, Johnsen SP, Hansen M, Hindsholm K,
Jakobsen CJ. Health Outcomes with and without Use of Inotropic
Therapy in Cardiac Surgery: Results of a Propensity Score-matched
Analysis. Anesthesiology 2014; 120: 1098108
6 Baker SG, Kramer BS. The transitive fallacy for randomized trials: if A
bests B and B bests C in separate trials, is A better than C? BMC Med
Res Methodol 2002; 2: 13
7 Song F, Altman DG, Glenny AM, Deeks JJ. Validity of indirect comparison for estimating efficacy of competing interventions: empirical evidence from published meta-analyses. BMJ 2003; 326: 472
8 Biondi-Zoccai GG, Agostoni P, Abbate A, Testa L, Burzotta F. A simple
hint to improve Robinson and Dickersins highly sensitive PubMed
search strategy for controlled clinical trials. Int J Epidemiol 2005;
34: 2245
9 Abacilar AF, Dogan OF. Levosimendan use decreases atrial fibrillation in patients after coronary artery bypass grafting: a pilot
study. Heart Surg Forum 2013; 16: E287 94
10 Al-Shawaf E, Ayed A, Vislocky I, Radomir B, Dehrab N, Tarazi R. Levosimendan or milrinone in the type 2 diabetic patient with low ejection fraction undergoing elective coronary artery surgery.
J Cardiothorac Vasc Anesth 2006; 20: 353 7
11 Alvarez J, Taboada M, Rodrguez J, et al. Hemodynamic effects
of levosimendan following cardiac surgery. Rev Esp Anestesiol
Reanim 2005; 52: 389 94
12 Alvarez J, Bouzada M, Fernandez AL, et al. Hemodynamic effects of
levosimendan compared with dobutamine in patients with low
cardiac output after cardiac surgery. Rev Esp Cardiol 2006; 59: 33845
13 Arbeus M, Axelsson B, Friberg O, Magnuson A, Bodin L, Hultman J.
Milrinone increases flow in coronary artery bypass grafts after
Page 10 of 11
18
19
20
21
22
23
24
25
26
27
28
29
30
BJA
31
32
33
34
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
Page 11 of 11
35