The Effect of Inodilatory Agents on Mortality
The Effect of Inodilatory Agents on Mortality
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 trials were included in the final analysis (Table 1). The study characteristics of included trials are detailed in Supplementary Appendix S3 http://bja.oxfordjournals.org/content/114/5/746/suppl/DC1. Briefly, the 46 trials randomised 2647 patients, including 1044 (39%) patients receiving placebo, 893 (34%) receiving levosimendan, 390 (15%) receiving milrinone, 273 (10%) receiving dobutamine, and 47 (1.8%) receiving enoximone.
(Enlarge Image)
Figure 1.
Flow diagram - description of selection process of included studies.
Nineteen (41%) of the randomised controlled trials were at low risk of bias while 52 and 7% were at moderate and high risk of bias respectively, lacking important details on the method used to generate random sequence, on allocation or on the intention to treat analyses (Supplementary Table S1 http://bja.oxfordjournals.org/content/114/5/746/suppl/DC1). The lack of blinding was the main bias and the main reason of low quality for the included trials.
The network configuration in Figure 2 connects treatments that have been directly compared in one or more RCTs and shows that almost half of the studies compared levosimendan vs placebo.
(Enlarge Image)
Figure 2.
Network configuration - for each comparison is displayed the estimate effect and number of studies analysed.
We also performed eight simple pairwise meta-analyses comparing each agent against each other agent (Supplementary Figures S1 and S2 http://bja.oxfordjournals.org/content/114/5/746/suppl/DC1). Only the use of levosimendan was associated with a reduction in mortality. The reduction in mortality was present when levosimendan was compared with placebo (35/725 [4.8%] in the levosimendan group vs 64/698 [9.2%] in the placebo group, OR=0.54, 95% CI 0.35 to 0.83, P for effect=0.005, P for heterogeneity=0.9, I=0%, NNT=23, with 20 studies included) and when compared with dobutamine (8/139 [5.8%] in the levosimendan group vs 18/138 [13%] in the dobutamine group, OR=0.39, 95% CI 0.16 to 0.95, P for effect=0.039, P for heterogeneity=0.6, I=0% NNT=14, with four studies included). For each pairwise comparison, the overall funnel plot did not show an important asymmetrical shape of estimate distribution (Supplementary Figure S3 http://bja.oxfordjournals.org/content/114/5/746/suppl/DC1). The superiority of levosimendan vs placebo was confirmed when sequentially removing each study while the superiority of levosimendan vs dobutamine was lost when removing one large trial.
The Bayesian network meta-analysis (Table 2) found that only the use of levosimendan was associated with a decrease in mortality. The survival benefit was limited to the comparison with placebo (posterior mean of OR=0.48, 95% CrI 0.28 to 0.80) with the model having a good fit (Dres=155.8 and DIC=183.7). The validity of consistency analysis were confirmed by a low probability in favour of inconsistency model (probability=0.01). The similarity between the effect estimated by both consistency and inconsistency model, was shown in the Supplementary Figure S4 http://bja.oxfordjournals.org/content/114/5/746/suppl/DC1.
Furthermore, the Bayesian meta-regressions of average follow-up against log-risk of mortality showed no significant time-related effects on mortality (regression coefficient=0.005 CrI −0.001 to 0.011, days).
Table 3 show the results of sensitivity analyses: the beneficial effect of levosimendan vs placebo was confirmed changing the prior distribution, with studies not supported by a drug company and removing the largest study, however, analysing the 19 low risk of bias trials, this superiority was lost.
We also performed a second analysis combining the enoximone and milrinone groups (Dres=155.9 and DIC=183.7) to overcome the zero-cell count problem (Table 2) and confirmed that levosimendan was better than placebo.
Table 4 reports the posterior distribution of the probability for each inodilator to be the best and the worst drug, showing that levosimendan is the best agent to improve survival after cardiac surgery. Figure 3 shows the cumulative rank curves of the probability to be the first, the second, the third and the worst treatment, confirming that levosimendan is the best treatment.
(Enlarge Image)
Figure 3.
Cumulative rank curves - probability to be the first, the second, the third and the worst for each treatment.
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 trials were included in the final analysis (Table 1). The study characteristics of included trials are detailed in Supplementary Appendix S3 http://bja.oxfordjournals.org/content/114/5/746/suppl/DC1. Briefly, the 46 trials randomised 2647 patients, including 1044 (39%) patients receiving placebo, 893 (34%) receiving levosimendan, 390 (15%) receiving milrinone, 273 (10%) receiving dobutamine, and 47 (1.8%) receiving enoximone.
(Enlarge Image)
Figure 1.
Flow diagram - description of selection process of included studies.
Nineteen (41%) of the randomised controlled trials were at low risk of bias while 52 and 7% were at moderate and high risk of bias respectively, lacking important details on the method used to generate random sequence, on allocation or on the intention to treat analyses (Supplementary Table S1 http://bja.oxfordjournals.org/content/114/5/746/suppl/DC1). The lack of blinding was the main bias and the main reason of low quality for the included trials.
Quantitative Data Synthesis
The network configuration in Figure 2 connects treatments that have been directly compared in one or more RCTs and shows that almost half of the studies compared levosimendan vs placebo.
(Enlarge Image)
Figure 2.
Network configuration - for each comparison is displayed the estimate effect and number of studies analysed.
We also performed eight simple pairwise meta-analyses comparing each agent against each other agent (Supplementary Figures S1 and S2 http://bja.oxfordjournals.org/content/114/5/746/suppl/DC1). Only the use of levosimendan was associated with a reduction in mortality. The reduction in mortality was present when levosimendan was compared with placebo (35/725 [4.8%] in the levosimendan group vs 64/698 [9.2%] in the placebo group, OR=0.54, 95% CI 0.35 to 0.83, P for effect=0.005, P for heterogeneity=0.9, I=0%, NNT=23, with 20 studies included) and when compared with dobutamine (8/139 [5.8%] in the levosimendan group vs 18/138 [13%] in the dobutamine group, OR=0.39, 95% CI 0.16 to 0.95, P for effect=0.039, P for heterogeneity=0.6, I=0% NNT=14, with four studies included). For each pairwise comparison, the overall funnel plot did not show an important asymmetrical shape of estimate distribution (Supplementary Figure S3 http://bja.oxfordjournals.org/content/114/5/746/suppl/DC1). The superiority of levosimendan vs placebo was confirmed when sequentially removing each study while the superiority of levosimendan vs dobutamine was lost when removing one large trial.
The Bayesian network meta-analysis (Table 2) found that only the use of levosimendan was associated with a decrease in mortality. The survival benefit was limited to the comparison with placebo (posterior mean of OR=0.48, 95% CrI 0.28 to 0.80) with the model having a good fit (Dres=155.8 and DIC=183.7). The validity of consistency analysis were confirmed by a low probability in favour of inconsistency model (probability=0.01). The similarity between the effect estimated by both consistency and inconsistency model, was shown in the Supplementary Figure S4 http://bja.oxfordjournals.org/content/114/5/746/suppl/DC1.
Furthermore, the Bayesian meta-regressions of average follow-up against log-risk of mortality showed no significant time-related effects on mortality (regression coefficient=0.005 CrI −0.001 to 0.011, days).
Table 3 show the results of sensitivity analyses: the beneficial effect of levosimendan vs placebo was confirmed changing the prior distribution, with studies not supported by a drug company and removing the largest study, however, analysing the 19 low risk of bias trials, this superiority was lost.
We also performed a second analysis combining the enoximone and milrinone groups (Dres=155.9 and DIC=183.7) to overcome the zero-cell count problem (Table 2) and confirmed that levosimendan was better than placebo.
Table 4 reports the posterior distribution of the probability for each inodilator to be the best and the worst drug, showing that levosimendan is the best agent to improve survival after cardiac surgery. Figure 3 shows the cumulative rank curves of the probability to be the first, the second, the third and the worst treatment, confirming that levosimendan is the best treatment.
(Enlarge Image)
Figure 3.
Cumulative rank curves - probability to be the first, the second, the third and the worst for each treatment.
Source...