Anaesthetic Drugs and Survival in Cardiac Surgery
Anaesthetic Drugs and Survival in Cardiac Surgery
Figure 1 shows the flow chart for the selection of randomized trials. Database searches, snowballing, and contacts with experts yielded a total of 2630 citations. Excluding 2518 non-pertinent titles or abstracts, we retrieved in complete form and assessed according to the selection criteria 112 studies. A total of 74 studies were further excluded because of their non-experimental design, including the use of historical controls, or because of duplicate publication. Specifically, we excluded 48 studies because there were no outcome data and further details could not be obtained by the authors, 16 studies because of overlapping populations, 7 observational studies, 3 studies performed in non-cardiac surgery settings. We finally identified 38 eligible randomized clinical trials, which were included in the final analysis (Table 1, Supplementary Tables S1 and S2).
(Enlarge Image)
Figure 1.
Flow diagram for selection of articles.
The 38 included trials enrolled 3996 patients, including 1648 (41%) receiving TIVA and 2348 (59%) receiving volatile agents. Specifically, 1086 (27%) patients received propofol, 622 (16%) received isoflurane, 701 (17%) received desflurane, and 1025 (26%) received sevoflurane. The trials had a median sample size of 60 (range 20–414) and were published between 1991 and 2012. Clinical heterogeneity was mostly because of control treatment and the follow-up duration. Most studies [24/38 (63%)] were performed on CABG patients with CPB and only five studies were performed in valve surgery patients (two aortic valve replacement and three mitral surgery). Baseline and procedural features were largely similar across the included studies.
The most common pairwise comparison was sevoflurane vs propofol (11 studies) followed by isoflurane vs propofol (six studies) and desflurane vs propofol (six studies). Three- and four-arm studies represented altogether 16% of the trials. Studies appeared to be of medium quality. Particularly, 19 (50%) of the randomized controlled trials (RCTs) were regarded to have low risk of bias, while the other studies lacked important details on the method used for random sequence generation and allocation (Supplementary Table S3). Many studies did not have blinding of the anaesthesiologists but of the staff collecting the outcome data.
The overall standard meta-analysis (Fig. 2) showed that the use of volatile agents (isoflurane, desflurane, or sevoflurane) was associated with a reduction in mortality when compared with TIVA at the longest follow-up available [25/1994 (1.3%) in the volatile group vs 43/1648 (2.6%) in the TIVA arm, OR=0.51, 95% CI 0.33–0.81, P-value for effect=0.004, number need to treat=74, P-value for heterogeneity=0.9, I=0% with 35 studies included and the three studies comparing a volatile agent vs another volatile agent excluded].
(Enlarge Image)
Figure 2.
Forest plot of volatile agents (isoflurane, desflurane, or sevoflurane) vs TIVA for the risk of mortality at the longest follow-up available. CI, confidence interval; OR, odds ratio.
Visual inspection of the funnel plot did not identify an important skewed or asymmetrical shape (Supplementary Fig. S1). Nonetheless, as quantitative evaluation suggested a possible presence of publication bias, as measured by Peters' test (P=0.02) and Begg's test (P=0.18), we used the trim-and-fill approach (adding the missing studies as suggested by the computer) to confirm the results of our meta-analysis after adjusting for the theoretical presence of unpublished studies (OR=0.43, 95% CI 0.28–0.64, P-value for effect <0.001, P-value for heterogeneity=0.9, I=0%, with 13 studies added).
The results of secondary and sensitivity analysis are reported in Table 2. Volatile agents were associated with a reduced time of mechanical ventilation, and duration of ICU and hospital stay. Furthermore, of 17 studies with troponin I analysis, 7 significantly favoured the volatile regimen, in 6 we observe a trend in favour of volatile agents and in 4 a trend in favour of TIVA.
When comparing TIVA, isoflurane, desflurane, and sevoflurane through direct comparisons, we found non-significant differences in mortality: (i) isoflurane [3/449 (0.7%)] vs TIVA [10/504 (2.0%)], OR=0.71, 95% CI 0.29–1.75, P-value for effect=0.5, I=0% with 13 studies included; (ii) desflurane [12/680 (1.8%)] vs TIVA [31/771 (4.0%)], OR=0.64, 95% CI 0.35–1.18, P-value for effect=0.15, I=0% with 10 studies included; (iii) sevoflurane [10/865 (1.2%)] vs TIVA [25/833 (3.0%)], OR=0.80, 95% CI 0.45–1.41, P-value for effect=0.4, I=0% with 17 studies included; (iv) sevoflurane [2/300 (0.7%)] vs isoflurane [4/293 (1.4%)], OR=0.71, 95% CI 0.14–3.74, P-value for effect=0.7, I=0% with 4 studies included; and (v) sevoflurane [4/227 (1.8%)] vs desflurane [9/232 (3.9%)], OR=0.70, 95% CI 0.34–1.44, P-value for effect=0.3, I=0% with 3 studies included.
The similarity assumption, within each contrast, was confirmed by I<25%. The network configuration of each contrast analysed by a Bayesian network meta-analysis is reported in Figure 3. As the fixed (Dres=127.5 and DIC=149.4 at the longest follow-up available; Dres=111.5 and DIC=131.4 at short-time mortality) and random (Dres=126.5 and DIC=150.1 at the longest follow-up available; Dres=110.8 and DIC=132.0 at short-time mortality) effects models were indistinguishable in terms of model fit, we selected the first that estimated the effect by slightly increasing the precision. The final results are reported in Table 3. We calculated the indirect estimate as difference from the appropriate direct estimates (probability in favour of inconsistency model equal to 0.03 and to 0.05 at the longest follow-up available and at short-time mortality, respectively) and calculated the indirect 95% CrI by normal approximation.
(Enlarge Image)
Figure 3.
Network configuration. Comparisons between treatments and number of studies for each contrast. Supplementary material references.
The Bayesian network meta-analysis (Table 3) found that the use of sevoflurane (posterior mean of OR=0.31, 95% CrI 0.14–0.64) and desflurane (posterior mean of OR=0.43, 95% CrI 0.21–0.82) were associated with a reduction in mortality when compared with TIVA at the longest follow-up available. When the De Hert study was removed, we found that only the use of desflurane was associated with a significant reduction in mortality with respect to TIVA (posterior mean of OR=0.30, 95% CrI 0.09–0.88).
When the Bayesian network meta-analysis was repeated, including all studies using propofol, we found a significant treatment difference effect between sevoflurane and propofol (posterior mean of OR=0.37, 95% CrI 0.13–0.98). Furthermore, Bayesian meta-regressions of the average follow-up against log-risk of mortality showed no significant effect for time on mortality (regression coefficient=−0.0008, CrI −0.004 to 0.002 and regression coefficient=−0.019, CrI −0.060 to 0.003, including all studies using TIVA or propofol, respectively). The Bayesian meta-regressions of average of the year of publication against mortality log-risk showed no significant effect when including all studies using TIVA (regression coefficient=−0.058, CrI −0.048 to 0.185) and a significant association when analysing only those studies using propofol (regression coefficient=0.259, CrI 0.007–0.545): adjusting for the effect of year of publication, we observed a more intense difference effect between sevoflurane and propofol (posterior mean of OR=0.30, 95% CrI 0.10–0.86).
When repeating all the Bayesian network meta-analyses using short-term mortality (≤30 days after surgery) as an endpoint, we found only a trend towards a reduction in mortality when comparing desflurane vs TIVA (posterior mean of OR=0.41, 95% CrI 0.15–1.04).
Supplementary Table S4 reports, for each anaesthetic agent, the posterior distribution of the probability to be the best and the worst, showing a trend of both TIVA and propofol to be the worst in terms of the long- and short-term survival after cardiac surgery.
Results
Description of Included Trials
Figure 1 shows the flow chart for the selection of randomized trials. Database searches, snowballing, and contacts with experts yielded a total of 2630 citations. Excluding 2518 non-pertinent titles or abstracts, we retrieved in complete form and assessed according to the selection criteria 112 studies. A total of 74 studies were further excluded because of their non-experimental design, including the use of historical controls, or because of duplicate publication. Specifically, we excluded 48 studies because there were no outcome data and further details could not be obtained by the authors, 16 studies because of overlapping populations, 7 observational studies, 3 studies performed in non-cardiac surgery settings. We finally identified 38 eligible randomized clinical trials, which were included in the final analysis (Table 1, Supplementary Tables S1 and S2).
(Enlarge Image)
Figure 1.
Flow diagram for selection of articles.
Study Characteristics
The 38 included trials enrolled 3996 patients, including 1648 (41%) receiving TIVA and 2348 (59%) receiving volatile agents. Specifically, 1086 (27%) patients received propofol, 622 (16%) received isoflurane, 701 (17%) received desflurane, and 1025 (26%) received sevoflurane. The trials had a median sample size of 60 (range 20–414) and were published between 1991 and 2012. Clinical heterogeneity was mostly because of control treatment and the follow-up duration. Most studies [24/38 (63%)] were performed on CABG patients with CPB and only five studies were performed in valve surgery patients (two aortic valve replacement and three mitral surgery). Baseline and procedural features were largely similar across the included studies.
The most common pairwise comparison was sevoflurane vs propofol (11 studies) followed by isoflurane vs propofol (six studies) and desflurane vs propofol (six studies). Three- and four-arm studies represented altogether 16% of the trials. Studies appeared to be of medium quality. Particularly, 19 (50%) of the randomized controlled trials (RCTs) were regarded to have low risk of bias, while the other studies lacked important details on the method used for random sequence generation and allocation (Supplementary Table S3). Many studies did not have blinding of the anaesthesiologists but of the staff collecting the outcome data.
Quantitative Data Synthesis
The overall standard meta-analysis (Fig. 2) showed that the use of volatile agents (isoflurane, desflurane, or sevoflurane) was associated with a reduction in mortality when compared with TIVA at the longest follow-up available [25/1994 (1.3%) in the volatile group vs 43/1648 (2.6%) in the TIVA arm, OR=0.51, 95% CI 0.33–0.81, P-value for effect=0.004, number need to treat=74, P-value for heterogeneity=0.9, I=0% with 35 studies included and the three studies comparing a volatile agent vs another volatile agent excluded].
(Enlarge Image)
Figure 2.
Forest plot of volatile agents (isoflurane, desflurane, or sevoflurane) vs TIVA for the risk of mortality at the longest follow-up available. CI, confidence interval; OR, odds ratio.
Visual inspection of the funnel plot did not identify an important skewed or asymmetrical shape (Supplementary Fig. S1). Nonetheless, as quantitative evaluation suggested a possible presence of publication bias, as measured by Peters' test (P=0.02) and Begg's test (P=0.18), we used the trim-and-fill approach (adding the missing studies as suggested by the computer) to confirm the results of our meta-analysis after adjusting for the theoretical presence of unpublished studies (OR=0.43, 95% CI 0.28–0.64, P-value for effect <0.001, P-value for heterogeneity=0.9, I=0%, with 13 studies added).
The results of secondary and sensitivity analysis are reported in Table 2. Volatile agents were associated with a reduced time of mechanical ventilation, and duration of ICU and hospital stay. Furthermore, of 17 studies with troponin I analysis, 7 significantly favoured the volatile regimen, in 6 we observe a trend in favour of volatile agents and in 4 a trend in favour of TIVA.
When comparing TIVA, isoflurane, desflurane, and sevoflurane through direct comparisons, we found non-significant differences in mortality: (i) isoflurane [3/449 (0.7%)] vs TIVA [10/504 (2.0%)], OR=0.71, 95% CI 0.29–1.75, P-value for effect=0.5, I=0% with 13 studies included; (ii) desflurane [12/680 (1.8%)] vs TIVA [31/771 (4.0%)], OR=0.64, 95% CI 0.35–1.18, P-value for effect=0.15, I=0% with 10 studies included; (iii) sevoflurane [10/865 (1.2%)] vs TIVA [25/833 (3.0%)], OR=0.80, 95% CI 0.45–1.41, P-value for effect=0.4, I=0% with 17 studies included; (iv) sevoflurane [2/300 (0.7%)] vs isoflurane [4/293 (1.4%)], OR=0.71, 95% CI 0.14–3.74, P-value for effect=0.7, I=0% with 4 studies included; and (v) sevoflurane [4/227 (1.8%)] vs desflurane [9/232 (3.9%)], OR=0.70, 95% CI 0.34–1.44, P-value for effect=0.3, I=0% with 3 studies included.
The similarity assumption, within each contrast, was confirmed by I<25%. The network configuration of each contrast analysed by a Bayesian network meta-analysis is reported in Figure 3. As the fixed (Dres=127.5 and DIC=149.4 at the longest follow-up available; Dres=111.5 and DIC=131.4 at short-time mortality) and random (Dres=126.5 and DIC=150.1 at the longest follow-up available; Dres=110.8 and DIC=132.0 at short-time mortality) effects models were indistinguishable in terms of model fit, we selected the first that estimated the effect by slightly increasing the precision. The final results are reported in Table 3. We calculated the indirect estimate as difference from the appropriate direct estimates (probability in favour of inconsistency model equal to 0.03 and to 0.05 at the longest follow-up available and at short-time mortality, respectively) and calculated the indirect 95% CrI by normal approximation.
(Enlarge Image)
Figure 3.
Network configuration. Comparisons between treatments and number of studies for each contrast. Supplementary material references.
The Bayesian network meta-analysis (Table 3) found that the use of sevoflurane (posterior mean of OR=0.31, 95% CrI 0.14–0.64) and desflurane (posterior mean of OR=0.43, 95% CrI 0.21–0.82) were associated with a reduction in mortality when compared with TIVA at the longest follow-up available. When the De Hert study was removed, we found that only the use of desflurane was associated with a significant reduction in mortality with respect to TIVA (posterior mean of OR=0.30, 95% CrI 0.09–0.88).
When the Bayesian network meta-analysis was repeated, including all studies using propofol, we found a significant treatment difference effect between sevoflurane and propofol (posterior mean of OR=0.37, 95% CrI 0.13–0.98). Furthermore, Bayesian meta-regressions of the average follow-up against log-risk of mortality showed no significant effect for time on mortality (regression coefficient=−0.0008, CrI −0.004 to 0.002 and regression coefficient=−0.019, CrI −0.060 to 0.003, including all studies using TIVA or propofol, respectively). The Bayesian meta-regressions of average of the year of publication against mortality log-risk showed no significant effect when including all studies using TIVA (regression coefficient=−0.058, CrI −0.048 to 0.185) and a significant association when analysing only those studies using propofol (regression coefficient=0.259, CrI 0.007–0.545): adjusting for the effect of year of publication, we observed a more intense difference effect between sevoflurane and propofol (posterior mean of OR=0.30, 95% CrI 0.10–0.86).
When repeating all the Bayesian network meta-analyses using short-term mortality (≤30 days after surgery) as an endpoint, we found only a trend towards a reduction in mortality when comparing desflurane vs TIVA (posterior mean of OR=0.41, 95% CrI 0.15–1.04).
Supplementary Table S4 reports, for each anaesthetic agent, the posterior distribution of the probability to be the best and the worst, showing a trend of both TIVA and propofol to be the worst in terms of the long- and short-term survival after cardiac surgery.
Source...