Anaesthetic Drugs and Survival in Cardiac Surgery
Anaesthetic Drugs and Survival in Cardiac Surgery
To address the question whether the choice of the anaesthetic might influence patients' survival after cardiac surgery, we carried out standard meta-analyses and Bayesian network meta-analyses to compare the effect on mortality of desflurane, isoflurane, sevoflurane, and TIVA.
When head-to-head treatment comparisons are not available or conclusive, network meta-analyses can provide estimates of treatment efficacy of multiple treatment regimens. Different treatments are analysed by statistical inference, rather than simply summing up trials that evaluated the same drug management compared with control, so that the results come from combining both direct and indirect estimates. To model the binomial data, we applied the Bayesian hierarchical model using Markov Chain Monte Carlo (MCMC) approaches.
Pertinent studies were independently searched in BioMedCentral, MEDLINE/PubMed, Embase, and the Cochrane Central Register of clinical trials by two expert investigators. Literature searches were last updated on June 1, 2012. The full PubMed search strategy was developed according to Biondi-Zoccai and colleagues and is available in the Appendix. Further hand or computerized searches involved the recent (2010–2012) conference proceedings from the International Anaesthesia Research Society, American Heart Association, American College of Cardiology, American Society of Anesthesiologists, and European Society of Cardiology congresses.
References obtained from database, literature searches with cross-check of references, experts, and manufacturers were first independently examined at a title/abstract level by two investigators and then, if potentially pertinent, retrieved as complete articles. No language restriction was imposed and non-English articles were translated and included in the analyses. The following inclusion criteria were used for potentially relevant studies: random allocation to treatment and comparison between a TIVA and an anaesthesia plan including administration of isoflurane, desflurane, or sevoflurane or a comparison between volatile agents, performed in cardiac surgical patients with no restriction in dose and time of administration. The exclusion criteria were duplicate publications (in this case, the article reporting the longest follow-up was abstracted), non-human experimental studies, and lack of outcome data. Studies in which epidural analgesia/anaesthesia was given to all patients were included. Studies in which ischaemic pre-conditioning or remote ischaemic pre-conditioning were performed in all patients were excluded because ischaemic pre-conditioning has pathways of cardiac protection that are similar to those of volatile anaesthetics even if the cardiac protective properties of volatile agents are not limited to pre-conditioning. Two investigators independently assessed compliance to selection criteria and selected studies for the final analysis, with divergences finally resolved by consensus.
Year of publication, setting, number of patients, volatile agent, anaesthetic comparator, and length of follow-up were collected (Table 1) together with baseline (age, diabetes, ejection fraction, chronic obstructive pulmonary disease, use of beta-blockers, and management of sulfonylurea, theophylline, or allopurinol) (Supplementary Table S1) and procedural (cardioplegia, time of cross-clamping, and number of coronary artery grafts) (Supplementary Table S2) data. Furthermore, we extracted and pooled data on mechanical ventilation, intensive care unit (ICU) stay, hospital stay, troponin I (ng ml), myocardial infarction (as per author definition), and use of inotropic agent.
'Total Intravenous Anaesthesia' was defined as a group not receiving volatile agents. 'Propofol' was defined as a TIVA group receiving propofol as main hypnotic agent and not receiving volatile agents. 'Volatile' (desflurane, isoflurane, or desflurane) was defined as a group receiving a volatile agent (even if added on top of a TIVA regimen and irrespectively on time of administration).
The endpoint of the present systematic review and meta-analysis of randomized trials was to identify differences in mortality at the longest follow-up available between volatile agents and TIVA and to identify whether one or more anaesthetics were superior or inferior in terms of survival, using standard meta-analyses and Bayesian network meta-analyses. If we found that the study had missing or incomplete data on survival, we contacted all authors by letter, e-mail, or both.
The methodological details for the internal validity and risk of bias assessment, for the statistical analyses and for the details on the conduction of the Bayesian network meta-analyses are reported as Supplementary data. In summary, the internal validity was evaluated according to the Cochrane Collaboration methods; the overall risk of bias was expressed as low, moderate, or high; the evidence of publication bias was assessed by analytic appraisal based on both Peters' and Begg's test; the heterogeneity assumption among studies within direct contrast was evaluated by means of Cochran Q-test and by I by Higgins and Thompson; the validity and the symmetry of the entire Bayesian network meta-analysis was investigated visually by a graph of the network configuration. The presence of effect-modifiers attributable to heterogeneity was considered acceptable if the χP-value was >0.10. Mortality data from individual studies were analysed in order to compute pooled odds ratio (OR) with pertinent 95% confidence intervals (CIs), by means of the inverse variance method with the fixed effect model or the DerSimonian–Laird method with random effect model; the pairwise association between each treatment was delineated by a graphical representation of the network; the network analysis was carried out modelling the binary outcome mortality with the Bayesian hierarchical model (binomial model with logit link function) using the MCMC approach; the indirect estimate was calculated as the difference from the appropriate direct estimates and the corresponding 95% credibility intervals (CrI) was obtained by normal approximation; we selected the fixed or random effect model calculating the posterior mean of residual deviance (Dres) and the deviance information criterion (DIC) statistics.
To explore the association between log-risk of mortality and both the length of study follow-up and the year of publication, we performed meta-regression analyses using the Bayesian approach. Other sub-analyses on mortality outcome were performed analysing the three volatile agents (isoflurane, desflurane, and sevoflurane) separately, in studies using propofol as TIVA, in studies with >100 patients and stratifying by setting, such as overall coronary artery bypass graft (CABG) patients, off-pump or on-pump with cardiopulmonary bypass (CPB), and non-CABG surgery.
Sensitivity meta-analyses were performed by analysing data from studies with low risk of bias and by sequentially removing each study from the overall dataset. The statistical analysis was performed by STATA (release 11, College Station, TX, USA), winBUGS (release 1.4, freeware available by BUGS project), and SAS 2002–2008 program (release 9.2 by SAS Institute, Inc., Cary, NC, USA). Statistical significance was set at the two-tailed 0.05 level for hypothesis testing. Unadjusted P-values are reported throughout. This study was performed in compliance with the Cochrane Collaboration and Preferred Reporting Items for Systematic Reviews and Meta-Analyses guidelines.
Methods
To address the question whether the choice of the anaesthetic might influence patients' survival after cardiac surgery, we carried out standard meta-analyses and Bayesian network meta-analyses to compare the effect on mortality of desflurane, isoflurane, sevoflurane, and TIVA.
When head-to-head treatment comparisons are not available or conclusive, network meta-analyses can provide estimates of treatment efficacy of multiple treatment regimens. Different treatments are analysed by statistical inference, rather than simply summing up trials that evaluated the same drug management compared with control, so that the results come from combining both direct and indirect estimates. To model the binomial data, we applied the Bayesian hierarchical model using Markov Chain Monte Carlo (MCMC) approaches.
Search Strategy and Study Selection
Pertinent studies were independently searched in BioMedCentral, MEDLINE/PubMed, Embase, and the Cochrane Central Register of clinical trials by two expert investigators. Literature searches were last updated on June 1, 2012. The full PubMed search strategy was developed according to Biondi-Zoccai and colleagues and is available in the Appendix. Further hand or computerized searches involved the recent (2010–2012) conference proceedings from the International Anaesthesia Research Society, American Heart Association, American College of Cardiology, American Society of Anesthesiologists, and European Society of Cardiology congresses.
Study Selection
References obtained from database, literature searches with cross-check of references, experts, and manufacturers were first independently examined at a title/abstract level by two investigators and then, if potentially pertinent, retrieved as complete articles. No language restriction was imposed and non-English articles were translated and included in the analyses. The following inclusion criteria were used for potentially relevant studies: random allocation to treatment and comparison between a TIVA and an anaesthesia plan including administration of isoflurane, desflurane, or sevoflurane or a comparison between volatile agents, performed in cardiac surgical patients with no restriction in dose and time of administration. The exclusion criteria were duplicate publications (in this case, the article reporting the longest follow-up was abstracted), non-human experimental studies, and lack of outcome data. Studies in which epidural analgesia/anaesthesia was given to all patients were included. Studies in which ischaemic pre-conditioning or remote ischaemic pre-conditioning were performed in all patients were excluded because ischaemic pre-conditioning has pathways of cardiac protection that are similar to those of volatile anaesthetics even if the cardiac protective properties of volatile agents are not limited to pre-conditioning. Two investigators independently assessed compliance to selection criteria and selected studies for the final analysis, with divergences finally resolved by consensus.
Data Abstraction and Study Characteristics
Year of publication, setting, number of patients, volatile agent, anaesthetic comparator, and length of follow-up were collected (Table 1) together with baseline (age, diabetes, ejection fraction, chronic obstructive pulmonary disease, use of beta-blockers, and management of sulfonylurea, theophylline, or allopurinol) (Supplementary Table S1) and procedural (cardioplegia, time of cross-clamping, and number of coronary artery grafts) (Supplementary Table S2) data. Furthermore, we extracted and pooled data on mechanical ventilation, intensive care unit (ICU) stay, hospital stay, troponin I (ng ml), myocardial infarction (as per author definition), and use of inotropic agent.
'Total Intravenous Anaesthesia' was defined as a group not receiving volatile agents. 'Propofol' was defined as a TIVA group receiving propofol as main hypnotic agent and not receiving volatile agents. 'Volatile' (desflurane, isoflurane, or desflurane) was defined as a group receiving a volatile agent (even if added on top of a TIVA regimen and irrespectively on time of administration).
The endpoint of the present systematic review and meta-analysis of randomized trials was to identify differences in mortality at the longest follow-up available between volatile agents and TIVA and to identify whether one or more anaesthetics were superior or inferior in terms of survival, using standard meta-analyses and Bayesian network meta-analyses. If we found that the study had missing or incomplete data on survival, we contacted all authors by letter, e-mail, or both.
The methodological details for the internal validity and risk of bias assessment, for the statistical analyses and for the details on the conduction of the Bayesian network meta-analyses are reported as Supplementary data. In summary, the internal validity was evaluated according to the Cochrane Collaboration methods; the overall risk of bias was expressed as low, moderate, or high; the evidence of publication bias was assessed by analytic appraisal based on both Peters' and Begg's test; the heterogeneity assumption among studies within direct contrast was evaluated by means of Cochran Q-test and by I by Higgins and Thompson; the validity and the symmetry of the entire Bayesian network meta-analysis was investigated visually by a graph of the network configuration. The presence of effect-modifiers attributable to heterogeneity was considered acceptable if the χP-value was >0.10. Mortality data from individual studies were analysed in order to compute pooled odds ratio (OR) with pertinent 95% confidence intervals (CIs), by means of the inverse variance method with the fixed effect model or the DerSimonian–Laird method with random effect model; the pairwise association between each treatment was delineated by a graphical representation of the network; the network analysis was carried out modelling the binary outcome mortality with the Bayesian hierarchical model (binomial model with logit link function) using the MCMC approach; the indirect estimate was calculated as the difference from the appropriate direct estimates and the corresponding 95% credibility intervals (CrI) was obtained by normal approximation; we selected the fixed or random effect model calculating the posterior mean of residual deviance (Dres) and the deviance information criterion (DIC) statistics.
To explore the association between log-risk of mortality and both the length of study follow-up and the year of publication, we performed meta-regression analyses using the Bayesian approach. Other sub-analyses on mortality outcome were performed analysing the three volatile agents (isoflurane, desflurane, and sevoflurane) separately, in studies using propofol as TIVA, in studies with >100 patients and stratifying by setting, such as overall coronary artery bypass graft (CABG) patients, off-pump or on-pump with cardiopulmonary bypass (CPB), and non-CABG surgery.
Sensitivity meta-analyses were performed by analysing data from studies with low risk of bias and by sequentially removing each study from the overall dataset. The statistical analysis was performed by STATA (release 11, College Station, TX, USA), winBUGS (release 1.4, freeware available by BUGS project), and SAS 2002–2008 program (release 9.2 by SAS Institute, Inc., Cary, NC, USA). Statistical significance was set at the two-tailed 0.05 level for hypothesis testing. Unadjusted P-values are reported throughout. This study was performed in compliance with the Cochrane Collaboration and Preferred Reporting Items for Systematic Reviews and Meta-Analyses guidelines.
