Does anaesthesia with nitrous oxide affect mortality or cardiovascular morbidity? A systematic review with meta-analysis and trial sequential analysis

Transcription

1 British Journal of Anaesthesia 112 (3): (2014) Advance Access publication 9 January doi: /bja/aet416 REVIEW ARTICLES Does anaesthesia with nitrous oxide affect mortality or cardiovascular morbidity? A systematic review with meta-analysis and trial sequential analysis G. Imberger 1 *,A.Orr 2, K. Thorlund 3, J. Wetterslev 1, P. Myles 2,4 and A. M. Møller 5,6 1 Copenhagen Trial Unit, Centre for Clinical Intervention Research, Rigshospitalet, Blegdamsvej 9, Copenhagen Ø DK-2100, Denmark 2 Department of Anaesthesia and Perioperative Medicine, Alfred Hospital, 55 Commercial Road, Melbourne 3004, Australia 3 Department of Clinical Epidemiology and Biostatistics, McMaster University, 1280 Main Street West, Hamilton L8S4L8, Canada 4 Academic Board of Anaesthesia and Perioperative Medicine, Monash University, 55 Commercial Road, Melbourne 3004, Australia 5 Cochrane Anaesthesia Review Group, Rigshospitalet, Blegdamsvej 9, Copenhagen Ø DK-2100, Denmark 6 Department of Anaesthesia, Herlev Hospital, Herlev DK-2730, Denmark * Corresponding author. gimberger@gmail.com Editor s key-points The authors reviewed publications exploring cardiovascular outcomes after nitrous oxide anaesthesia. The evidence-base was insufficient to draw robust conclusions regarding the effect of nitrous oxide anaesthesia on cardiovascular outcomes. Background. The role of nitrous oxide in modern anaesthetic practice is contentious. One concern is that exposure to nitrous oxide may increase the risk of cardiovascular complications. ENIGMA II is a large randomized clinical trial currently underway which is investigating nitrous oxide and cardiovascular complications. Before the completion of this trial, we performed a systematic review and meta-analysis, using Cochrane methodology, on the outcomes that make up the composite primary outcome. Methods. We used conventional meta-analysis and trial sequential analysis (TSA). We reviewed 8282 abstracts and selected 138 that fulfilled our criteria for study type, population, and intervention. We attempted to contact the authors of all the selected publications to check for unpublished outcome data. Results. Thirteen trials had outcome data eligible for our outcomes. We assessed three of these trials as having a low riskof bias. Using conventional meta-analysis, the relative riskof short-term mortality in the nitrous oxide group was 1.38 [95% confidence interval (CI) ] and the relative risk of long-term mortality in the nitrous oxide group was 0.94 (95% CI ). In both cases, TSA demonstrated that the data were far too sparse to make any conclusions. There were insufficient data to perform meta-analysis for stroke, myocardial infarct, pulmonary embolus, or cardiac arrest. Conclusion. This systematic review demonstrated that we currently do not have robust evidence for how nitrous oxide used as part of general anaesthesia affects mortality and cardiovascular complications. Keywords: meta-analysis; nitrous oxide; review, systematic Accepted for publication: 17 July 2013 Nitrous oxide has been used as a general anaesthetic for more than 160 years. Collective anecdotal experience with this drug must be larger than with any other drug used in anaesthesia. Despite this experience, opinions about the role of nitrous oxide in modern-day practice continue to diverge. 1 2 One concern is that exposure to nitrous oxide may increase the risk of cardiovascular complications. Nitrous oxide oxidizes the cobalt atom in vitamin B12, inactivating methionine synthase, causing a decrease in folate metabolism and an increase in homocysteine. 3 Homocysteinaemia after exposure to nitrous oxide has been well demonstrated in vivo 4 6 and long-term homocysteinaemia is known to be associated with an increased risk of ischaemic heart disease. 7 It remains unclear, however, whether this information about this surrogate outcome translates into real clinical risk. To investigate the possible causal association between mortality, cardiac morbidity, and nitrous oxide, the ENIGMA (Evaluation of Nitrous oxide In the Gas Mixture for Anaesthesia) trial group has designed a large, multicentre randomized clinical trial: ENIGMA II is enrolling at-risk patients and is powered to investigate a composite primary outcome of mortality, non-fatal acute myocardial infarction, cardiac arrest, pulmonary embolism, and stroke. 8 & The Author [2014]. Published by Oxford University Press on behalf of the British Journal of Anaesthesia. All rights reserved. For Permissions, please journals.permissions@oup.com

2 Nitrous oxide and cardiovascular outcomes BJA While this trial is underway, several observational studies have been published looking at similar outcomes. Using data from the Intraoperative Hypothermia in Aneurysm Surgery Trial, 9 a secondary analysis was performed, finding an association between nitrous oxide exposure and delayed ischaemic neurological deficits. 10 Sanders and colleagues 11 also performed a secondary subgroup analysis on 1615 participants from the General Anaesthesia comparedwith Local Anaesthesia for Carotid Surgery Trial, finding no evidence that nitrous oxide increases the risk of perioperative vascular adverse events. 12 Turan and colleagues 13 published a retrospective cohort study of 49,016 patients and found that patients who had received nitrous oxide had decreased odds for 30-day mortality. Leslie and colleagues 14 performed a post hoc analysis on 5133 of the patients from the Perioperative Ischemic Evaluation trial, and found no association between nitrous oxide and cardiovascular adverse events. These studies herald a focusing interest on the association between nitrous oxide and cardiovascular adverse events and they certainly contribute to the debate. Being post hoc observational studies, however, potential confounders, especially those confounding by indication, increase the risk of bias and prevent any conclusions from being definitive. The association between nitrous oxide and cardiovascular complications remains unclear, and the results from ENIGMA II are keenly awaited. We thus performed a systematic review and meta-analysis of the five outcomes used as the composite primary outcome in ENIGMA II. We plan to update this meta-analysis when the results from ENIGMA II, and any other future trials, are available. Repeated updates in meta-analysis are analogous to those in interim analyses in a clinical trial. In clinical trials where interim analyses are performed, the concern about increased risk of type 1 error because of repetitive testing and sparse data is well known and sequential hypothesis testing designs and procedures are used to control this increased risk In systematic reviews with meta-analysis, similar sequential hypothesis testing procedures, such as Trial Sequential Analysis (TSA), can also be applied Given our plan to update our meta-analysis in the future, we performed TSA as part of our current analysis. Methods Protocol and registration The protocol for the review was published on the website of the Copenhagen Trial Unit ( in January The protocol was registered with the PROSPERO database in December 2011 (registration no CRD ). There were two small deviations from this protocol. First, there were errors in the search strategies published in the protocol. These errors were corrected before the searches were run. The corrected searches are provided in Appendix 1. Secondly, we clarified the definition of patients at increased risk of cardiovascular complications slightly, by stating that we considered patients undergoing day surgery as being at low risk of cardiovascular complications. Systematic literature search We conducted a sensitive systematic literature search of Medline, EMBASE, the Cochrane controlled trial register, CINAHL, and ISI Web of Science, updated until May There were no date or language restrictions. We also checked the references of included studies. Two authors (G.I. and A.O.) independently screened all of the abstracts produced by the search to identify eligible studies. Study selection, data extraction, and quality assessment We included all randomized clinical trials, irrespective of language or publication status. We included trials with human adult patients receiving general anaesthesia, for any surgery. We defined a general anaesthetic as any procedure where inhalation agents, systemic agents, or both are given as part of general anaesthesia for the purposes of undertaking a medical procedure. We did not include studies where nitrous oxide was given for the purposes of sedation. We included trials where patients receiving nitrous oxide were compared with patients receiving no nitrous oxide. We included trials only when it was clear that the control group received no nitrous oxide throughout the perioperative period. We excluded trials where participants were randomized to different anaesthetic techniques (apart from the administration of nitrous oxide). We included studies that were able to provide data on any of the following outcomes: (1) Mortality (all cause) including all trials where mortality data were available for the same follow-up period in both the exposure and control groups. We included two comparisons for the mortality outcome: term, follow-up ranging from discharge from the perioperative care unit (PACU) until 30 days after operation (or discharge from hospital). The end point needed to be consistent in the two intervention groups. Mortality long term, including the longest follow-up starting from 30 days after operation. (2) Stroke. (3) Myocardial infarction (MI). (4) Pulmonary embolus (PE). (5) Cardiac arrest. For short-term mortality, stroke, MI, PE, and cardiac arrest, we included data from any trials where the patients were at increased risk of cardiovascular complications. See Appendix 2 for our definition of this increased risk. We accepted all clear and reasonable definitions of these outcomes in individual trials, as long as they were used consistently in both groups and reported explicitly. We selected possible inclusions on the basis of the trial type, participants and interventions described in the abstract and retrieved full copies of these publications. For trials that fulfilled these parameters, but did not report relevant outcomes, we wrote to the authors to ask whether any unpublished data were available. When no address was available on the publication, we searched for previous contact addresses for authors and we contacted departments, co-authors of other publications, or both, in order to find contact details. In 411

3 BJA Imberger et al. the case of mortality, when an author specified that all patients were followed up after leaving PACU, either until a specific time point or until discharge, and that they could confirm that there was no mortality in either group for this period, then we selected these trials for inclusion. In the case of stroke, MI, PE, and cardiac arrest, we only included trials for inclusion if the authors confirmed that these outcomes had been formally registered in each group. Two authors (G.I. and A.O.) independently extracted data from the included trials, including information about the methodology and the data for our outcomes. All duplicate publications were examined to ensure consistency. For the studies with relevant unpublished outcome data, extraction came from the correspondence with authors. Two authors (G.I. and A.O.) reviewed all included trials with regard to their quality. We rated the risk of bias using the guide provided in The Cochrane Handbook of Systematic Reviews of Interventions. 22 We made the assessments with regard to each outcome being included in our review, so these assessments are not necessarily reflective of the quality of the trials for their intended primary outcomes. We assessed each outcome for sequence generation, allocation concealment, blinding, incomplete outcome data, selective outcome reporting, and other sources of bias. When all of the five assessment categories were assessed as adequate for a given outcome, we assigned a low risk of bias for the findings for that outcome. When we assessed one or more categories as inadequate, we assigned an increased risk of bias. Statistical analysis All of our outcomes were categorical and were expressed as risk ratios (RRs) with 95% confidence intervals (CIs). We performed meta-analysis when more than one trial with low risk of bias was eligible for inclusion for a given outcome. All testing was two-sided. To control the increased risk of random error because of sparse data and repetitive testing, we also performed TSA. Conventional meta-analysis was performed using the Review Manager software (RevMan 5.0). TSA was performed using the TSA software ( Trial sequential analysis Repeated updates (sequential multiplicity) and sparse data increase the risk of random error. 23 TSA is a method for metaanalysis that aims to correct for this increased risk In a way similar to monitoring boundaries for interim analyses in single trials, TSA combines an estimation of required information size (RIS) for meta-analysis with monitoring boundaries used as thresholds for statistical significance. The less data that have accumulated, the more conservative the TSA boundaries, making it less likely to declare statistical significance before the RIS has been reached. Similar to a sample size calculation for a single trial, estimating RIS involves a calculation that includes type 1 error, type 2 error, the control event proportion, and the effect size. The calculation for RIS also requires an estimate of heterogeneity; if more heterogeneity is present, RIS increases. 21 For our TSAs, we estimated the RIS using 0.05 for type 1 error, 0.10 for type 2 error, the control event proportions calculated from the no nitrous oxide groups in all the included trials and the effect size estimated from the included trials with a low risk of bias. We repeated the analysis using relative risk reductions of 10 and 20% as effect size estimates. We used the D 2 (Diversity) 24 present in the included trials as the estimate for heterogeneity. The TSA can be interpreted by viewing the boundaries and whether the cumulative meta-analysis has crossed them. Alternatively, the results can be translated into TSA-adjusted CIs. 21 In this review, we presented the TSA-adjusted CIs. Sensitivity analyses When statistical heterogeneity was present, we presented the results of random-effects meta-analysis [DerSimonian and Laird (DL)]. 25 When no statistical heterogeneity was present, we presented the results of fixed-effect meta-analysis. Groups with zero events were adjusted with a constant continuityadjustment of 0.5 in each arm (as per the default adjustment in the Revman software). 26 A sensitivity analysis was performed using both analysis models (random effects and fixed effect), different continuity adjustments (constant and empirical), 27 and using a Peto fixed-effect model. 28 Results Figure 1 shows the study flow diagram. From the 8282 abstracts reviewed, we considered 138 publications eligible on the basis of the trial type, the participants and the intervention. Of note, none of these publications had any of our outcomes as primary outcomes. Table 1 shows the primary outcomes described in these publications. In 21 (14 being duplicates), authors reported our outcomes as secondary findings. Of the 117 publications that did not mention any of our outcomes, authors provided unpublished but eligible information about mortality from nine publications (three being duplicates). In total, 13 trials were selected as eligible to be included in our meta-analyses. The characteristics of the included studies are summarized in Table 2. Risk-of-bias assessments are presented in Table 3. Synthesis of results term Figure 2 shows a forest plot for the conventional meta-analysis for this outcome. Including only the trials with low risk of bias, the relative risk in the nitrous group was 1.38 (95% CI ). For TSA, using a control event proportion of 0.5%, a relative risk increase of 40%, and a constant continuity adjustment of 0.5 events per group, the accrued information size (2648) was only 1.63% of the estimated RIS (162802). Consequently, there was too little information for the TSA software to construct boundaries and calculate TSA-adjusted 95% CI. Sensitivity analyses were performed using both analysis models (random effects and fixed effect), different continuity 412

4 Nitrous oxide and cardiovascular outcomes BJA 8282 titles and abstracts reviewed 146 selected as potentially eligible and full papers reviewed 8 excluded based on study type, participants or intervention 138 selected as eligible based on study type, participants and intervention 117 publications not reporting relevant outcomes 21 publications reporting relevant outcomes 26 unable to find a contact address 37 sent, no reply 45 sent, reply stating that no relevant outcomes were measured 9 sent, reply giving relevant outcomes that were measured but not published 7 eligible trials (+14 duplicates) 6 eligible trials (+3 duplicates) 13 randomized clinical trials included Fig 1 Study flow diagram. adjustments (constant and empirical), and using a Peto fixed-effect model. The point estimates (and corresponding 95% CIs), the calculated heterogeneity (I 2 ) and the consequential estimation of RIS all varied substantially when using different models. This variation is summarized in Table 4. Mortality long-term For both trials included in this meta-analysis, these longterm data were extracted from follow-up publications Using conventional meta-analysis, the relative risk in the nitrous group was 0.94 (95% CI ), as shown in Figure 3. For TSA, the control event proportion used was 25% and the relative risk reduction 6%. The relative risk in the nitrous group was 0.94 (TSA adjusted 95% CI ), the TSA did not cross the futility threshold for rejecting a 6% relative risk reduction, and the accrued information size (1873) was only 6% of the estimated RIS (31 699). Using a relative risk reduction of 10%, the point estimate and CIs were the same, the futility threshold was not crossed and the accrued information was 15% of the estimated RIS (12 179). Using a relative risk reduction of 20%, the point estimate was 0.94 (TSA adjusted 95% CI ) and the TSA did cross the futility threshold. This result indicates that exposure to nitrous oxide does not alter long-term mortality by 20%. See Figure 4 for this analysis. Stroke Figure 5 shows a forest plot for stroke. There was only one trial with low risk of bias. 83 Meta-analysis was not performed. 413

5 BJA Imberger et al. Table 1 Primary outcomes studied in the publications eligible based on study type, population, and intervention Primary outcome(s) n Years of publication (range) Air-filled spaces Anaesthetic depth Cancer recurrence Cognitive function Cost Drug kinetics Folate/homocysteine metabolism General tolerance IVF Length of stay Laryngeal mask related endpoints Long-term morbidity and mortality Myocardial ischaemia Neuropsychology Operating conditions Pain Physiological parameters Cardiology Neurology Respiratory Other PONV Viability of bone marrow cells Wound infection Myocardial infarction Four trials registered data for this outcome There was only one trial with low risk of bias. 83 Meta-analysis was not performed (Fig. 6). Pulmonary embolus Only one trial reported data on PE. 76 There were zero events in each group. Meta-analysis was not performed. Cardiac arrest We found no data for this outcome. Discussion The purpose of this systematic review was to summarize the current evidence from randomized clinical trials associating nitrous oxide with serious cardiovascular complications. Our review presents results using both conventional meta-analysis and TSA. We were able to conduct meta-analyses for shortterm mortality and long-term mortality. For the other four outcomes, there was no more than one trial with low risk of bias, and we therefore did not conduct meta-analyses. In the case of short-term mortality, the conventional meta-analysis (with low risk of bias trials) gives a point estimate indicating increased risk in the nitrous oxide group (RR 1.38) but with very wide CIs (95% CI ). There is huge imprecision in the estimate provided by this evidence. This imprecision, combined with the low control event proportion and the relatively small number of patients included, means that we cannot say that the two groups are different for this outcome. Nor can we say that they are similar. For TSA, the accrued information was such a small percentage of the estimated RIS (,2%) that the boundaries were unable to be constructed. That is, the quantity of the information was too small to even be analysed with TSA. We conclude, therefore, that there is currently insufficient evidence to assess whether nitrous oxide increases the risk of short-term mortality in high-risk groups having surgery. For the meta-analysis for short-term mortality, events were sparse and there were zero-event groups in two of the three included trials. In this situation (sparse data and zero-event groups), the choice of continuity adjustment can have a large impact on the estimateof effect (RR) and on the calculated heterogeneity (I 2 ). I 2 affects the decision about which metaanalytic model to use, which in turn can have a large impact on the estimate of effect when the data are sparse. Each combination of effect measure, continuity adjustment, and meta-analytic model represents a different way to model the situation. When data are sparse, it is difficult to judge which model is most appropriate. We, therefore, did sensitivity analyses using each different model. The Cochrane handbook suggests using the Peto odds ratio (with fixed-effect metaanalysis) for zero-event data, 26 and we included this in the sensitivity analysis. An extensive variation in point estimates, calculated heterogeneity (and consequently estimated RIS) was demonstrated. This variation reflects the enormous amount of uncertainty, based on these data, of the effect of nitrous oxide on short-term mortality. Long-term mortality provides an example of how TSA can assist with interpreting the results of meta-analysis. Using conventional techniques, the point estimate was close to 1 (RR 0.94) with rather narrow CIs (95% CI ). One might be tempted to conclude, on the basis of these results, that there is no difference in long-term mortality with or without nitrous oxide for this population. For the TSA for this outcome, we used acontrol event proportion of 25% and a relative risk reduction of 6% for the effect size (derived from included trials with a low risk of bias). To test this model, the RIS was large (24 627). However, the number of randomized patients actually accrued in this meta-analysis was,10% (1873). So concluding no effect here would be similar to stopping a trial after only 10% of the sample size had been recruited, finding narrow CIs that crossed 1.0 and concluded that an intervention had no effect (despite being grossly underpowered). Indeed, the futility boundaries constructed with TSA were not crossed, and we conclude that there is insufficient evidence to assess whether nitrous oxide increases or decreases long-term mortality by an effect estimate of 6%. When we tested this model using effect estimates of relative risk reductions of 10%, there was also insufficient evidence. For a relative risk reduction of 20%, however, the TSA just crossed the futility boundary. The population of patients included in this 414

6 415 Table 2 Characteristics of the included trials. *Published in Fleischmann and colleagues. 162 Published in Leslie and colleagues 148 Trial Antonini and colleagues 58 Chowdhury and colleagues 36 Eger and colleagues 61 El-Galley and colleagues 95 Fleischmann and colleagues 162 Hohner and colleagues 89 Iacopino and colleagues 112 Jensen and colleagues 131 Knüttgen and colleagues 30 Krogh and colleagues 97 Leung and colleagues 37 Other publications reporting the same study Koblin and colleagues 55 Kozmary and colleagues 88 Lampe and colleagues 132 Lampe and colleagues 104 Lampe and colleagues 125 Waldman and colleagues 137 Akca and colleagues 92 Fleischmann and colleagues 35 Jensen and colleagues 122 Jensen and colleagues 66 Kalman and colleagues 67 Participants Randomized (nitrous vs no nitrous) 40 patients having laparoscopic cholecystectomy 90 patients having transphenoidal removal of pituitary tumours 100 elective total hip arthroscopy 70 carotid endarterectomy 100 transphenoidal hypophysectomy 28 patients having laparoscopic donor nephrectomies 418 ASAI IIIpatients having elective colon resection, expected to last longer than 2 h 49 patients with coronary artery disease having abdominal aorta surgery 40 ASA I and II patients having neurosurgical procedures 40 patients having intra-abdominal operations on the colon and rectum 42 patients having intracranial procedures in the sitting position 139 patients having elective major colonic surgery 228 patients aged.65 yr, having any surgical procedure and expected to stay in hospital for.48 h Intervention Control Outcomes for inclusion and number of participants able to be included N 2 O and O 2 FIO2 0.4 O 2 and N 2 O 1:2 Air and O 2 FIO2 0.4 (n¼40) Unpublished 40% O 2 in air (n¼85) Unpublished 60% N 2 O and 40% O 2 100% O 2 (n¼270) MI (n¼70) Published 70% N 2 OinO 2 Air and O 2, mixture (n¼28) Unpublished 65% N 2 O and 35% O 2 35% O 2 in N 2 (n¼408) Mortality long (n¼203)* Published 60% N 2 O and 40% O 2 (and 5 10 mg fentanyl kg 21 ) 40% O 2, in air (and mg fentanyl kg 21 ) (n¼47) MI (n¼47) Published 30% O 2 and 67% N 2 O 30% O 2 and 66% N 2 (n¼40) Unpublished 30% O 2 in N 2 O (and higher doses of propofol and fentanyl) O 2 and N 2 O 1:1 N 2 OinO 2 with an FIO 2 of 0.3 N 2 O and O 2 (% at anaesthetist s discretion) no difference in between 2 groups P 0.48) FIO2 Oxygen in air with an Mortality of 0.3 FIO2 O 2 and N 2 1:1 Air in oxygen with an of 0.3 FIO2 O 2 with/without air (% at anaesthetist s discretion) short (n¼40) Unpublished (n¼42) Unpublished (n¼139) Unpublished (n¼210) Published Primary outcome(s) Multiple Postoperative cognitive defects Multiple postoperative complications Bowel distension Surgical wound infection Bowel distension Haemodynamics Ventricular function Myocardial ischaemia Cerebral autoregulation Bowel function Atelectasis Recovery Incidence of venous air embolism Operating conditions Postoperative course Postoperative delirium or cognitive decline Continued Nitrous oxide and cardiovascular outcomes BJA

7 BJA Imberger et al. Table 2 Continued Primary outcome(s) Intervention Control Outcomes for inclusion and number of participants able to be included Participants Randomized (nitrous vs no nitrous) Trial Other publications reporting the same study Length of stay (hospital) (n¼2012) Stroke (n¼2012) MI (n¼2012) Mortality long (n¼1670) Published 70% N2O and 30% O2 80% O2 and 20% N patients having a surgical procedure expected to last.2 h and expected to stay in hospital.72 h Myles and Chan and colleagues 83 colleagues 98 Myles and colleagues 5 Myles and colleagues 56 Leslie and colleagues 148 Leslie and colleagues 87 Length of stay (hospital and ICU) (n¼87) Stroke (n¼87) MI (n¼87) PE (n¼87) Published O2 and N2O 0.8:1.2 Air and O2 1.5:0.5 Singh and 116 patients having elective colleagues 76 craniotomies for supratentorial tumours meta-analysis was patients having non-cardiac surgery expected to last longer than 2 h and patients with colon cancer having colon resection expected to last longer than 2 h For this population, we can conclude, with a type II risk of 10%, that exposure to nitrous oxide does not alter longterm mortality by 20% (number needed to harm¼5). Clearly, for mortality we are clinically interested in an effect size much smaller than this for an intervention as commonly used as nitrous oxide. We therefore conclude that, for longterm mortality, for a clinically relevant effect size, we have insufficient evidence. The major strength of this review was its thorough search. We designed our approach to maximize sensitivity. With the amount of published trial work including nitrous oxide as an intervention, this task was large, but we feel confident that our systematic and transparent approach gave us a good chance to collect a complete summary of the evidence derived from current randomized controlled trials. This systematic review did not aim to investigate all the outcomes relevant to the use of nitrous oxide. Rather, it aimed to systematically collect and analyse what we know about nitrous oxide and mortality and major cardiovascular complications, before the completion of ENIGMA II. Modelling the significance of the findings can be challenging. TSA allows the definition of a hypothesis, with regard to control event proportions and effect size estimate, in the same way as is done for a randomized controlled trial. Moreover, heterogeneity is included. As you change the results of these parameters, that is, as you change the assumptions in the model, you may change your conclusions. So the answers in the meta-analyses depend crucially on the questions being asked, and the assumptions being made. In this systematic review, we estimated control event proportions by pooling all the no-nitrous oxide groups included in the relevant meta-analysis, the effect size was estimated from the included trials with a low risk of bias and for an a priori anticipated relative risk reduction of 10 and 20%, and we used the D 2 present in the included studies to adjust for heterogeneity. Consequently, our conclusions answer the questions that are defined by these parameters and depend on these assumptions. As with any attempt to synchronize evidence and to model the significance of the findings, there are limitations. Of particular note, in this review, the potential for clinical heterogeneity was large because the inclusion criteria were broad. Estimates of heterogeneity in meta-analysis can be unreliable when numbers are small. 164 Consequently, we may have under-estimated the heterogeneity, and this fact further reduces the confidence we can have in the existing evidence. Nitrous oxide has been investigated extensively with regard to many different primary outcomes. From our search, we did not find any randomized clinical trials that handled our outcomes as primary outcomes. In most cases, we were extracting data that were reported incidentally. The disadvantage of this is that the study design would not have been intended for our outcomes. Moreover, because our outcomes were incidental for many of these trials, we retrieved some of our data from correspondence with authors. There were

8 417 Table 3 Assessments of risk of bias for included trials. *Blinding: participants and personnel, outcome assessment. Other: evidence of selective outcome reporting, other potential threats to validity. Information sourced from correspondence with investigators Trial Randomization Concealment Blinding* Attrition Other Low risk of bias Fleischmann and colleagues 162 Mortality long Leung and colleagues 37 Myles and colleagues 83 Stroke MI Mortality long Computer generated Sealed envelopes Anaesthetists not blinded, surgeons blinded, outcome assessors blinded Computerized random number list Computer-generated code Randomization sequence concealed Automated telephone service Outcome assessment blinded Anaesthetists not blinded Outcome assessment blinded Surgeon and patient not informed of treatment group 10 (2%) short-term mortality Well explained Data sheet lost for 4 (unknown groups) 2 lost to follow-up from the nitrous group (surgical complications) 4 lost to follow-up from the no nitrous group (1 withdrawn by a physician and 3 found not to meet the inclusion criteria) 11 (5%) 10 from the initial study (see Fleischmann and colleagues 162 ) 1 extra lost to long-term follow-up (from the nitrous group) 18 (8%) for the primary outcome Reasons not described Denominator/percentage not quoted for the mortality data Assume it was the same as for the primary outcome 38 (2%) 23 (2%) from the group who got nitrous 15 (1%) from the group who did not get nitrous All explained deferred surgery or development of exclusion criteria 380 (19%) 38 (2%) during original trial(see Myles and colleagues 83 ) 340 (17%) not contacted or uncontactable (49.1% from nitrous group, 50.9% from the no nitrous group) 2 (0.1%) declined to participate Equal numbers of nitrous/no nitrous in the attrition population Increased risk of bias Antonini and Not described Not described Surgeons blinded to intervention Nil colleagues 58 Chowdhury and colleagues 36 Computer generated Not described Those performing the analyses were blinded 5 (5%) All from the group who received no nitrous Well explained and unlikely to be linked with our outcomes Eger and colleagues 61 MI Not fully described Not described Outcome assessment blinded Nil 40 (57% of those randomized in relevant group) Reason for attrition not described (postoperative MI was not a primary outcome/concern in the paper) Post-hoc analysis Post-hoc analysis Continued Nitrous oxide and cardiovascular outcomes BJA

9 418 Table 3 Continued Trial Randomization Concealment Blinding* Attrition Other El-Galley and colleagues 95 Hohner and colleagues 89 Iacopino and colleagues 112 Jensen and colleagues 131 Knüttgen and colleagues 30 Krogh and colleagues 97 Singh and colleagues 76 Stroke MI PE Card withdrawal technique Not described Surgeon blinded Nil Not fully described Not described Subjective outcome assessment blinded Not described Not described Study conducted in a blinded fashion Set of numbered envelopes 2 (4%) Well described Nil Randomized blindly Surgical team members blinded Nil Multiple publications, each describing different outcomes Not fully described Not described Not described Nil Not described Not described Surgeons blinded Nil Computer-generated Opaque envelopes Surgeons and outcome assessment blinded 29 (25%) 15 (27%) from the group who got nitrous 14 (23%) from the group who did not get nitrous All explained remained intubated postoperative and therefore ineligible for measurement of the primary outcome of the study Quantity/nature of attrition a source for increased risk of bias for our outcomes BJA Imberger et al.

10 Nitrous oxide and cardiovascular outcomes BJA Nitrous oxide No nitrous oxide Risk ratio Risk ratio Study or subgroup Low risk of bias Events Total Events Total Weight M-H, random, 95%Cl M-H, random, 95%Cl Fleischmann % 0.14 [0.01, 2.70] Leung % 3.00 [0.12, 72.88] Myles % 2.95 [0.80, 10.85] Subtotal (95% Cl) % 1.38 [0.22, 8.71] Total events 10 6 Heterogeneity: t 2 =1.25; c 2 =3.64, df=2 (P=0.16); l 2 =45% Test for overall effect: Z=0.34 (P=0.73) Increased risk of bias Antonini Not estimable Chowdhury Not estimable Eger % 3.09 [0.13, 75.17] El Gallery Not estimable Hohner % 1.18 [0.08, 17.82] Iacopino Not estimable Jensen Not estimable Knuttgen % 1.00 [0.07, 14.95] Krogh Not estimable Singh Not estimable Subtotal (95% Cl) % 1.43 [0.28, 7.41] Total events 3 2 Heterogeneity: t 2 =0.00; c 2 =0.31, df=2 (P=0.86); l 2 =0% Test for overall effect: Z=0.43 (P=0.67) Total (95% Cl) Total events Heterogeneity: Test for overall effect: % 1.75 [0.69, 4.40] 13 8 t 2 =0.00; c 2 =3.97, df=5 (P=0.55); l 2 =0% Z=1.18 (P=0.24) Test for subgroup differences: c 2 =0.00, df=1 (P=0.97); l 2 =0% Favours Favours nitrous oxide no nitrous oxide Fig 2 Forest plot for short-term mortality. Table 4 Point estimates, calculated heterogeneity and estimated RIS for short-term mortality (low risk of bias) as the meta-analysis model is varied. CIs shown are calculated using conventional meta-analysis. Analysis model Effect measure Random effects Relative risk Fixed effect Relative risk Peto fixed effect Peto odds ratio Continuity adjustment Point estimate (95% CI) I 2 (%) RIS Constant ( ) Constant ( ) Empirical ( ) Empirical ( ) Constant ( ) Constant ( ) Empirical ( ) Empirical ( ) ( ) publications that were eligible based on trial type, population and intervention. Of these, we were unable to contact 63. It is possible that this approach to data collection introduces bias; perhaps those whom we could not contact represent a different population of studies. Our systematic review demonstrated the high number of outcomes both surrogate and clinical that have been considered in randomized controlled trials investigating nitrous oxide. This list does not include outcomes examined in observational studies, such as environmental issues and staff exposure, which would extend its size further. The length of this list reflects how many different outcomes have been considered relevant to nitrous oxide. Assessing the merits of using this drug involves evaluating the strength of evidence for how nitrous oxide affects each relevant outcome. Advantageous outcomes can then be weighed up against the disadvantageous and a judgement can be made about the merits of using the drug. 419

11 BJA Imberger et al. Study or subgroup Fleischmann 2005 Myles 2007 Nitrous oxide No nitrous oxide Risk ratio Risk ratio Events Total Events Total Weight M-H, fixed, 95%Cl M-H, fixed, 95%Cl % 84.1% 1.02 [0.72, 1.46] 0.92 [0.77, 1.10] Total (95% Cl) % 0.94 [0.80, 1.10] Total events Heterogeneity: c 2 =0.26, df=1 (P=0.61); l 2 =0% Test for overall effect; Z=0.78 (P=0.44) Favours Favours nitrous oxide no nitrous oxide Fig 3 Forest plot for long-term mortality. Cumulative Z-score Information size is a two-sided graph Information size=2932 Favours nitrous Z-curve Number of patients (linear scaled) Favours no nitrous Fig 4 TSA for long-term mortality, testing for a relative risk reduction of 20%. Nitrous oxide No nitrous oxide Risk ratio Risk ratio Study or subgroup Events Total Events Total M-H, fixed, 95%Cl M-H, fixed, 95%Cl Myles 2007 Singh [0.06, 15.68] 1.68 [0.30, 9.58] Favours nitrous oxide Favours no nitrous oxide 420 Fig 5 Forest plot for stroke.

12 Nitrous oxide and cardiovascular outcomes BJA Nitrous oxide No nitrous oxide Risk ratio Risk ratio Study or subgroup Events Total Events Total M-H, fixed, 95%Cl M-H, fixed, 95%Cl Eger 1990 Hohner 1994 Myles 2007 Singh [0.04, 3.26] 0.24 [0.03, 1.87] 1.82 [0.73, 4.55] 3.36 [0.14, 80.20] Favours Favours nitrous oxide no nitrous oxide Fig 6 Forest plot for MI. While many outcomes can alter in their relative importance, depending on circumstances and individual judgement, mortality and cardiovascular complications are always important. In most situations, these outcomes would be considered more important than all others. If nitrous oxide does not affect mortality or cardiovascular complications, then the merits of its use lie with the balance of other outcomes. If nitrous oxide does affect these important outcomes, then this information should profoundly affect whether this drug is used. Our systematic review has demonstrated that we currently do not have robust evidence for how nitrous oxide affects mortality and cardiovascular complications. We can say neither that it does not have an effect nor that it does. This lack of evidence leaves a striking hole. We look forward with interest, therefore, to the results of ENIGMA II and to the continued debate on the role of nitrous oxide in modern anaesthetic practice. Authors contributions Conceiving the review: G.I., K.T., A.O., J.W., P.M., and A.M.M. Coordinating the review: G.I. Undertaking the search: G.I. Screening the abstracts from the search: G.I. and A.O. Retrieving the full papers for review: G.I. Reviewing full papers for inclusions: G.I. and AO. Contacting the authors of review to confirm outcome measurement: G.I. Appraising the quality of included papers: G.I. and A.O. Abstracting data from included papers: G.I. and A.O. Data management for the review: G.I. and J.W. Statistical analysis: G.I. and J.W. Interpretation of data: G.I., J.W., P.M., and A.M.M. Writing the review: G.I. (principal), K.T., A.O., J.W., P.M., and A.M.M. Acknowledgements We thank Eva Grogro and Federico Riezzo for their assistance with translation. There were many authors of nitrous oxide trials who replied to our s of enquiry. The list is too long to publish here, but we thank each of them greatly for their time and their interest in the project. Declaration of interest None declared. Funding G.I. received a PhD study grant from Rigshospitalet, Copenhagen, Denmark, while conducting this project. No other author received funding for this project. References 1 Baum VC, Willschke H, Marciniak B. Is nitrous oxide necessary in the future? Paediatr Anaesth 2012; 22: Myles PS, Leslie K, Silbert B, Paech MJ, Peyton P. A review of the risks and benefits of nitrous oxide in current anaesthetic practice. Anaesth Intensive Care 2004; 32: Weimann J. Toxicity of nitrous oxide. Best Pract Res Clin Anaesthesiol 2003; 17: Badner NH, Drader K, Freeman D, Spence JD. The use of intraoperative nitrous oxide leads to postoperative increases in plasma homocysteine. Anesth Analg 1998; 87: Myles PS, Chan MT, Kaye DM, et al. Effect of nitrous oxide anesthesia on plasma homocysteine and endothelial function. Anesthesiology 2008; 109: Rao LK, Francis AM, Wilcox U, Miller JP, Nagele P. Pre-operative vitamin B infusion and prevention of nitrous oxide-induced homocysteine increase. Anaesthesia 2010; 65: The Homocysteine Studies Collaboration. Homocysteine and risk of ischemic heart disease and stroke: a meta-analysis. JAMA 2002; 288: Myles PS, Leslie K, Peyton P, et al. Nitrous oxide and perioperative cardiac morbidity (ENIGMA-II) trial: rationale and design. Am Heart J 2009; 157: Todd MM, Hindman BJ, Clarke WR, Torner JC. Mild intraoperative hypothermia during surgery for intracranial aneurysm. N Engl J Med 2005; 135: Pasternak JJ, McGregor DG, Lanier WL, et al. Effect of nitrous oxide use on long-term neurologic and neuropsychological outcome in patients who received temporary proximal arteryocclusion during cerebral aneurysm clipping surgery. Anesthesiology 2009; 110: GALA Trial Collaborative Group. General anaesthesia versus local anaesthesiaforcarotid surgery (GALA): a multicentre, randomised controlled trial. Lancet 2008; 372: Sanders RD, Graham C, Lewis SC, Bodenham A, Gough MJ, Warlow C. Nitrous oxide exposure does not seem to be associated with increased mortality, stroke, and myocardial infarction: a nonrandomized subgroup analysis of the general anaesthesia compared with local anaesthesia for carotid surgery (GALA) trial. Br J Anaesth 2012; 109:

13 BJA Imberger et al. 13 Turan A, Mascha EJ, You J, et al. The association between nitrous oxide and postoperative mortality and morbidity after noncardiac surgery. Anesth Analg 2012; 116: Leslie K, MylesP, Devereaux PJ, et al. Nitrous oxide and serious morbidity and mortality in the POISE trial. Anesth Analg 2013; 116: McPherson K. Statistics: The problem of examining accumulating data more than once. N Engl J Med 1974; 290: Freidlin B, Korn EL, George SL. Data monitoring committees and interim monitoring guidelines. Control Clin Trials 1999; 20: Brok J, Thorlund K, Gluud C, Wetterslev J. Trial sequential analysis reveals insufficient information size and potentially false positive results in many meta-analyses. J Clin Epidemiol 2008; 61: Brok J, Thorlund K, Wetterslev J, Gluud C. Apparently conclusive meta-analyses may be inconclusive trial sequential analysis adjustment of random error risk due to repetitive testing of accumulating data in apparently conclusive neonatal meta-analyses. Int J Epidemiol 2009; 38: Wetterslev J, Thorlund K, Brok J, Gluud C. Trial sequential analysis may establish when firm evidence is reached in cumulative meta-analysis. J Clin Epidemiol 2008; 61: Thorlund K, Devereaux PJ, Wetterslev J, et al. Can trial sequential monitoring boundaries reduce spurious inferences from meta-analyses? Int J Epidemiol 2009; 38: Thorlund K, Engstrøm J, Wetterslev J, Brok J, Imberger G, Gluud C. User Manual for Trial Sequential Analysis (TSA). Copenhagen, Denmark: Copenhagen Trial Unit, Centre for Clinical Intervention Research, 2011; Available from (accessed 30 January 2013) 22 Higgins JPT, Altman DG, Sterne JAC.(editors) Chapter 8: Assessing risk of bias in included studies. In: Higgins JPT, Green S, eds, Cochrane Handbook for Systematic Reviews of Interventions. Chichester: John Wiley & Sons, 2008; Thorlund K, Imberger G, Walsh M, et al. The numberof patients and events required to limit the risk of overestimation of intervention effects in meta-analysis a simulation study. PLoS One 2011; 6: e Wetterslev J, Thorlund K, Brok J, Gluud C. Estimating required information size by quantifying diversity in random-effects model meta-analyses. BMC Med Res Methodol 2009; 9: DerSimonian R, Laird N. Meta-analysis in clinical trials. Control Clin Trials 1986; 7: Deeks JJ, Higgins JPT, Altman DG. Chapter 9: Analysing data and undertaking meta-analyses. In: Higgins JPT, Green S, eds. Cochrane Handbook for Systematic Reviews of Interventions. Chichester: John Wiley & Sons, 2008; Sweeting MJ, Sutton AJ, Lambert PC. What to add to nothing? Use and avoidance of continuity corrections in meta-analysis of sparse data. Stat Med 2004; 23: Yusuf S, Peto R, Lewis J, Collins R, Sleight P. Beta blockade during and after myocardial infarction: an overview of the randomized trials. Prog Cardiovasc Dis 1985; 27: Hernandez-Palazon J, Martinez-Lage JF, De la Rosa-Carrillo VN, Tortosa JA, Lopez F, Poza M. Anesthetic technique and development of pneumocephalus after posterior fossa surgery in the sitting position. Neurocirugia 2003; 14: Knuttgen D, Stolzle U, Koning W, Muller MR, Doehn M. Airembolism in the sitting position. Oxygen/nitrogen versus oxygen/laughing gas. Anaesthesist 1989; 38: Losasso TJ, Muzzi DA, Dietz NM, Cucchiara RF. Fifty percent nitrous oxide does not increase the risk of venous air embolism in neurosurgical patients operated upon in the sitting position. Anesthesiology 1992; 77: Al Zahrani T, Ibraheim O, Turkistani A, Mazen K. Bispectral index profile during general anaesthesia using nitrous oxide for lower segment caesarean delivery. Internet J Anesthesiol 2006; 10: 7p 33 Coenen P, Lauwers M, Vanlersberhge C, Camu F. Contribution of nitrous oxide to anesthetic depth of sevoflurane and desflurane anesthesia. Br J Anaesth 1999; 82(Suppl. 1): Nakayama M, Iwasaki S, Ichinose H, et al. Effect of nitrous oxide on the bispectral index during sevoflurane anesthesia. Masui 2002; 51: Fleischmann E, Marschalek C, Schlemitz K, et al. Nitrous oxide may not increase the risk of cancer recurrence after colorectal surgery: a follow-up of a randomized controlled trial. BMC Anesthesiol 2009; 9: 1 36 Chowdhury T, Prabhakar H, Kumar BP, Dash HH. To assess the effects of two anesthetic techniques (with and without nitrous oxide) on cognitive function in patients undergoing transsphenoidal pituitary surgery. J Neurosurg Anesthesiol 2011; 23: 436 (Conference publication) 37 Leung JM, Sands LP, Vaurio LE, Wang Y. Nitrous oxide does not change the incidence of postoperative delirium or cognitive decline in elderly surgical patients. Br J Anaesth 2006; 96: Adamus M, Koutna J, Hrabalek L, Konupcikova K, Fischerova L, Francakova J. Comparison of the cost of balanced anaesthesia maintained with or without nitrous oxide at a university hospital. Anesteziol Intenzivni Med 2011; 22: Heath KJ, Sadler P, Winn JH, McFadzean WA. Nitrous oxide reduces the cost of intravenous anaesthesia. Eur J Anaesthesiol 1996; 13: Ekbom K, Assareh H, Anderson RE, Jakobsson JG. The effects of fresh gas flow on the amount of sevoflurane vaporized during 1 minimum alveolar concentration anaesthesia for day surgery: a clinical study. Acta Anaesthesiol Scand 2007; 51: Garrido P, Caldera MV, Deudero E, Villagran MJ, Martinez B, Criado A. Effects of nitrous oxide as second gas during desflurane anaesthesia. Br J Anaesth 1998; 80(Suppl. 1): Gin T, Yau G, Chan K, Gregory MA, Oh TE. Disposition of propofol infusions for caesarean section. Can J Anaesth 1991; 38: Harbut P, Wysoczanska-Harbut J, Boroch M, Kubler A. Propofol consumptionand recovery from target-controlled anaesthesia: influence of the nitrous oxide administration. Br J Anaesth 1999; 82:A Hendrickx JF, Coddens J, Callebaut F, et al. Effect of N 2 O on sevoflurane vaporizer settings during minimal- and low-flow anesthesia. Anesthesiology 2002; 97: Hendrickx JFA, Carette R, Lemmens HJM, De Wolf AM. Large volume N 2 O uptake alone does not explain the second gas effect of N 2 O on sevoflurane during constant inspired ventilation. Br J Anaesth 2006; 96: Illman H, Antila H, Olkkola KT. Quantitation of the effect of nitrous oxide on rocuronium infusion requirements using closed-loop feedback control. Anesthesiology 2008; 108: Illman HL, Antila HMJ, Olkkola KT. Effect of nitrous oxide on cisatracurium infusion demands: a randomized controlled trial. BMC Anesthesiol 2010; 10: Jakobsson J, Heidvall M, Davidson S. The sevoflurane-sparing effect of nitrous oxide: A clinical study. Acta Anaesthesiol Scand 1999; 43: Kopman AF, Chin WA, Moe J, Malik R. The effect of nitrous oxide on the dose-response relationship of rocuronium. Anesth Analg 2005; 100: Koukita Y, Ichinohe T, Hosaka G, et al. The effect of nitrous oxide on the vecuronium-induced neuromuscular blockade during propofol anesthesia. J Jpn Dent Soc of Anesthesiol 2004; 32: