- Anthony Rodgers, codirector ()a,
- Natalie Walker, research fellowa,
- S Schug, professorb,
- A McKee, consultant anaesthetistc,
- H Kehlet, professord,
- A van Zundert, consultant anaesthetiste,
- D Sage, consultant anaesthetistf,
- M Futter, consultant anaesthetistf,
- G Saville, consultant anaesthetistg,
- T Clark, statisticiana,
- S MacMahon, professorh
- a Clinical Trials Research Unit, Department of Medicine, University of Auckland, Private Bag 92019, Auckland, New Zealand,
- b Division of Anaesthesiology, University of Auckland,
- c Department of Anaesthetics, Green Lane Hospital, Epsom, Auckland 1003, New Zealand,
- d Department of Surgical Gastroenterology, Hvidovre University Hospital, DK-2650 Hvidovre, Denmark,
- e Department of Anesthesiology, Intensive Care and Pain Therapy, Catharina Hospital, Michelangelolaan 2, 5623 EJ Eindhoven, Netherlands,
- f Department of Anaesthesia, Auckland and Starship Hospitals, Private Bag 92024, Auckland, New Zealand,
- g Department of Anaesthesia, Royal Cornwall Hospital, Treliske, Truro TR1 3LJ,
- h Institute for International Health, University of Sydney, PO Box 1225, Crows Nest, Sydney, NSW 1585, Australia
- Correspondence to: A Rodgers
- Accepted 4 September 2000
Objectives: To obtain reliable estimates of the effects of neuraxial blockade with epidural or spinal anaesthesia on postoperative morbidity and mortality.
Design: Systematic review of all trials with randomisation to intraoperative neuraxial blockade or not.
Studies: 141 trials including 9559 patients for which data were available before 1 January 1997. Trials were eligible irrespective of their primary aims, concomitant use of general anaesthesia, publication status, or language. Trials were identified by extensive search methods, and substantial amounts of data were obtained or confirmed by correspondence with trialists.
Main outcome measures: All cause mortality, deep vein thrombosis, pulmonary embolism, myocardial infarction, transfusion requirements, pneumonia, other infections, respiratory depression, and renal failure.
Results: Overall mortality was reduced by about a third in patients allocated to neuraxial blockade (103 deaths/4871 patients versus 144/4688 patients, odds ratio=0.70, 95% confidence interval 0.54 to 0.90, P=0.006). Neuraxial blockade reduced the odds of deep vein thrombosis by 44%, pulmonary embolism by 55%, transfusion requirements by 50%, pneumonia by 39%, and respiratory depression by 59% (all P<0.001). There were also reductions in myocardial infarction and renal failure. Although there was limited power to assess subgroup effects, the proportional reductions in mortality did not clearly differ by surgical group, type of blockade (epidural or spinal), or in those trials in which neuraxial blockade was combined with general anaesthesia compared with trials in which neuraxial blockade was used alone.
Conclusions: Neuraxial blockade reduces postoperative mortality and other serious complications. The size of some of these benefits remains uncertain, and further research is required to determine whether these effects are due solely to benefits of neuraxial blockade or partly to avoidance of general anaesthesia. Nevertheless, these findings support more widespread use of neuraxial blockade.
Anaesthesia is commonly classified into two main techniques: general anaesthesia, in which gaseous or intravenous drugs achieve central neurological depression, and regional anaesthesia, in which drugs are administered directly to the spinal cord or nerves to locally block afferent and efferent nerve input.1 Regional anaesthesia for major thoracic, abdominal, or leg surgery relies on neuraxial blockade by injection of local anaesthetic drugs into either the subarachnoid space (spinal anaesthesia) or into the epidural space surrounding the spinal fluid sac (epidural anaesthesia).
The risks of fatal or life threatening events are increased several fold after major surgery, but there is debate about whether the type of anaesthesia has any substantive effect on these risks. Neuraxial blockade has several physiological effects that provide a rationale for expecting to improve outcome with this technique.2 However, the few clinical trials of epidural or spinal anaesthesia that have focused specifically on fatal or life threatening events have generally been too small to detect effects of plausible size reliably. To provide more reliable estimates of the effects of neuraxial blockade on postoperative morbidity and mortality, we conducted a systematic review of all relevant randomised trials.
Identification of trials and data collection
We sought to identify all trials in which patients were randomised to receive intraoperative neuraxial blockade (with epidural or spinal anaesthesia) or not. We did not exclude trials in adult populations in which the group receiving neuraxial blockade group also received general anaesthesia, the general anaesthesia group received postoperative neuraxial blockade, or there was more than one type of neuraxial blockade or general anaesthesia group (in which case similar groups were combined). Eligibility was not based on whether results were published, the language of publication, or the primary aims of the trial —for example, we included a randomised trial designed to assess the effects of neuraxial blockade on cognitive function.3 Trials were ineligible if they were not randomised or were quasi-randomised (such as assignment according to date of birth) or if data were not available before 1 January 1997.
We conducted a computerised search using the electronic databases Current Contents (1995-6), Embase (Excerpta Medica, 1980-96), Medline (1966-96), and the Cochrane Library (1998). We used the key words “regional anaesthesia,” “regional anesthesia,” “spinal,” or “epidural” and the Cochrane Collaboration search terms for randomised trials.4 Once papers were identified, authors' names and study titles were used as search terms. We scrutinised the reference lists of all identified papers and also hand searched selected conference proceedings.
We developed standard data collection sheets to record details of trial design, interventions, patient characteristics, and events. We did not use quality scores because analyses stratified by specific design characteristics are more informative.5 The definitions of events were those used in the original trials, since patients in one trial were directly compared only with those in the same trial. Two reviewers independently recorded the published findings from each study. This process was not blinded. A third reviewer compared the two sets of data collection sheets and any differences were resolved by discussion. We attempted to contact the authors of all trials to verify the data and obtain additional unpublished data. If there was more than one trial report, authors were also asked whether the patient groups overlapped. Lastly, we asked authors if they knew of any other relevant studies (published or unpublished).
Analysis was carried out on an intention to treat basis wherever possible. If no events were reported in the publication or by the authors, we assumed that none occurred. This assumption will generally provide unbiased estimates of proportional effects (the entity typically combined in meta-analysis) but will underestimate absolute effects.6 We calculated odds ratios, 95% confidence intervals, and two sided P values for each outcome of interest using Peto's modification of the Mantel-Haenszel method.7 Homogeneity was assessed by a χ2 test. Whenever possible, we stratified analyses of cause specific outcomes by surgical group and type of anaesthetic to determine whether these factors modified the size or direction of proportional effects. However, there were often too few trials with events for such analyses to be informative, and so subgroup analyses are mostly reported for the crude outcome of total mortality.
We identified 158 potentially eligible trials. Ten studies were excluded because they were quasi-randomised,8-17 and six were excluded because not all participants were randomised and separate information on the randomised patients was not available.18-23 One trial was excluded because the groups differed with respect to heparin treatment as well as anaesthetic technique.24 The remaining 141 trials that met all the inclusion criteria included a total of 9559 patients. 3 25-192 More than one publication was available for 18 studies 46-49 59 60 62-65 72 73 84 85 87-92 94-96 99 100 106 107 124-128 134 135 145 146 156-158 161-163 173 174 187 188 but each study was counted only once. No unpublished eligible studies were identified.
The study authors for 107 (76%) eligible trials, including 8290 (87%) patients, verified the data collection sheets. In almost all cases, we obtained additional unpublished information from contacting the authors, mostly about trial design, but also about events (for example 18 deaths were not reported in original publications). Table 1 shows the patient characteristics and anaesthetic methods and tables 2 and 3 provide summary details of outcome events. We defined a neuraxial blockade group and a non-neuraxial blockade group for each trial, which necessitated collapsing similar groups in 15 trials with more than one randomised comparison. The neuraxial blockade group had no general anaesthesia in 79 (56%) trials and the same general anaesthesia as the non-neuraxial blockade group in 37 (26%) trials. In 22 (15%) trials the neuraxial blockade group received a general anaesthesia different from that in the non-neuraxial blockade group; the systemic opioid varied in seven trials, 28 34 43 84 120 186 192 the use of inhalational anaesthetic varied in two trials, 71 140 the type of inhalational anaesthetic varied in two trials, 148 165 the induction drug varied in one trial,191 and more than one aspect varied in 10 trials. 42 76 80 81 97 151 161 168 169 189 For three (2%) trials details of the general anaesthesia method were unknown.
Among the 56 trials for which follow up data were available, the mean duration of follow up was about 62 days. Only 13 trials provided follow up data beyond 30 days postoperatively. No events were recorded in 80 trials involving 2941 participants, which were mostly designed to assess the physiological, biochemical, and endocrine effects of neuraxial blockade. The mean follow up in the first 30 days these trials was 11 days, compared with 21 days in trials in which events were observed.
A total of 247 deaths within 30 days of randomisation were recorded in 35 trials. Overall mortality was about one third less in the neuraxial blockade group (odds ratio 0.70, 95% confidence interval 0.54 to 0.90, P=0.006; fig 1) with no clear difference between different surgical groups (fig 2). A specific diagnosis was available for 162 of the deaths. Of these, 73 (45%) were due to pulmonary embolism, cardiac events, or stroke, 50 (31%) were due to infective causes, and 39 (24%) were due to other causes. The observed improvement in survival was due to trends towards reductions in deaths from pulmonary embolism, cardiac events, or stroke (0.73, 0.45 to 1.16), deaths from infection (0.68, 0.39 to 1.21), deaths from other causes (0.84, 0.44 to 1.61), and deaths from unknown causes (0.64, 0.41 to 1.01). There was about one fewer death per 100 patients in the 30 days after randomisation in the neuraxial blockade group (103/4871 (2.1%) versus 144/4688 (3.1%)). Only six intraoperative deaths were recorded, one of which was in the neuraxial blockade group (0.28, 0.06 to 1.45). Ten studies, with a total of 1371 patients, recorded 130 deaths between 30 days and six months. All but two of these studies were on orthopaedic patients. Overall, there was no clear effect of neuraxial blockade on deaths during this period (0.89, 0.61 to 1.28).
Mortality results by type of anaesthesia
Seven trials (with 826 participants) directly randomised patients to spinal or epidural anaesthesia. 25 32 77 104 153 181 Only 13 deaths occurred in these trials, four in the spinal group. However, an indirect comparison between trials of spinal and epidural anaesthesia showed no clear difference between their effects on total mortality (0.68, 0.49 to 0.95 for spinal anaesthesia and 0.68, 0.43 to 1.07 for epidural anaesthesia, P for homogeneity=1.0; fig 2). Mortality was reduced overall whether neuraxial blockade was continued postoperatively (0.68, 0.43 to 1.08) or not (0.70, 0.51 to 0.97). The effect on total mortality was not clearly lower in trials in which neuraxial blockade was combined with general anaesthesia (0.87, 0.53 to 1.41) than in trials in which neuraxial blockade was used alone (0.64, 0.47 to 0.87; P for homogeneity=0.3; fig 2). However, the confidence intervals were wide for the trials that used general anaesthesia. Forty four (18%) deaths occurred in the 22 trials in which the neuraxial blockade group had a different general anaesthesia to that used in the group not allocated neuraxial blockade. The overall effect in this group of trials (0.92, 0.49 to 1.71) was not clearly different (P for homogeneity=0.3) from that in other trials (0.66, 0.49 to 0.88).
Venous thromboembolism, cardiac events, and stroke
A total of 365 deep vein thromboses were reported from 18 trials. Neuraxial blockade reduced the risk of deep vein thrombosis by almost half (0.56, 0.43 to 0.72; fig 3). Since more than 80% of deep vein thromboses were recorded in orthopaedic trials, there was limited power to detect differences between surgical groups. In nine trials all patients were screened for deep vein thromboses by fibrinogen scanning, 59 87 129 venography, 74 114 132 187 or a combination of methods. 62 94 Proportional reductions in deep vein thromboses were similar in the trials with screening (0.56, 0.42 to 0.75) compared with other trials (0.54, 0.30 to 0.96). Therefore, absolute differences were much greater in the trials with screening (121/463 (26%) for neuraxial blockade versus 178/467 (38%) for no neuroaxial blockade) than in other trials (24/4408 (0.5%) versus 42/4221 (1.0%)). Outcome assessments were known to be blinded in only two trials, and deep vein thromboses were also reduced in these studies (0.46, 0.21 to 0.99). 66 98 A total of 96 pulmonary emboli were reported from 23 trials, 21 (22%) of which were fatal. Overall, there were about half as many pulmonary emboli in patients allocated to neuraxial blockade (0.45, 0.29 to 0.69; fig 3).
A total of 104 myocardial infarctions were reported in 30 trials. Overall, there were about one third fewer myocardial infarctions in patients allocated to neuraxial blockade, but the confidence intervals were compatible with both no effect and a halving in risk (0.67, 0.45 to 1.00; fig 3). Only 42 strokes were reported from eight trials, and the confidence intervals were very wide for this outcome (0.85, 0.46 to 1.57; fig 3).
In total, 473 patients from 16 trials required transfusion of two or more units of blood and 100 patients from 12 trials had a postoperative bleed requiring a transfusion. The requirement for a transfusion of two or more units of blood was reduced by about half in patients allocated neuraxial blockade (0.50, 0.39 to 0.66; fig 3). A similar proportional reduction was found for postoperative bleeds requiring a transfusion (0.45, 0.29 to 0.70; fig 3). There was no clear difference in the proportional effects on either outcome across surgical groups.
In total, 62 wound infections were reported from 14 trials. There were fewer wound infections in those allocated to neuraxial blockade, although the confidence intervals were wide (0.79, 0.47 to 1.33; fig 3). Three hundred and eighty seven cases of pneumonia were recorded in 28 trials, of which 38 (10%) were fatal. The risk of developing pneumonia was less in patients randomised to neuraxial blockade (0.61, 0.48 to 0.76; fig 3). There was no clear difference in the proportional effects with the use of concomitant general anaesthesia (neuraxial blockade versus general anaesthesia: 0.63, 0.46 to 0.87; neuraxial blockade plus general anaesthesia versus general anaesthesia: 0.59, 0.42 to 0.81). However, there was some evidence (P for homogeneity=0.05) that the proportional reduction in pneumonia was greater after thoracic epidural anaesthesia (0.48, 0.35 to 0.67) than after lumbar epidural or spinal anaesthesia (0.76, 0.55 to 1.04). Twelve deaths due to an infective cause other than pneumonia were recorded in six trials, of which two occurred in patients allocated to neuraxial blockade (0.33, 0.10 to 1.07; fig 3).
Other postoperative events
A total of 64 cases of respiratory depression were reported from eight trials. The odds of respiratory depression were reduced by 59% in patients allocated to neuraxial blockade (0.41, 0.23 to 0.73; fig 3). The effect was present in trials with and without concomitant general anaesthesia (neuraxial blockade alone versus general anaesthesia 0.37, 0.11 to 1.21; neuraxial blockade plus general anaesthesia versus general anaesthesia 0.43, 0.22 to 0.81). Fifty cases of renal failure were recorded in 10 trials. Although the risk of renal failure was reduced in patients randomised to neuraxial blockade, the confidence intervals were wide and compatible with both no effect and a two thirds reduction (0.57, 0.32 to 1.00; fig 3).
We conducted several analyses to assess whether the effects on total mortality were dependent on trials with methodological problems or affected by the type of anaesthesia. However, all these tests lacked power to detect moderate sized differences.
An overall reduction in mortality was still evident after we excluded studies for which the total number of patients originally randomised was not available (0.68, 0.51 to 0.91) 26 180; original authors could not be contacted (0.69, 0.53 to 0.90) 36 38 40 82 83 86 103 115 118 131 137-144 147 150 153 155 166 171 172 179 181 185 190 ; more than 5% of all patients were lost to follow up or excluded after randomisation (0.69, 0.51 to 0.91) 3 14 32 38 57 62 71 74 75 94 108 113 114 120 129 130 140 159 164 165 171 173 181 187 ; or more than 5% of the neuraxial blockade group were excluded after randomisation (0.68, 0.51 to 0.91). 28 32 57 75 94 113 120 129 130 140 159 164 165 171 173 The reduction in mortality was also evident after exclusion of two trials that were stopped before scheduled completion (0.70, 0.53 to 0.91) and exclusion of unpublished data (0.67, 0.51 to 0.88). 28 46 94 109 130 165 Finally, there was no clear evidence of publication bias from tests for trend across groups defined by trial size.
Our overview shows improved survival in patients randomised to neuraxial blockade. Additionally, we found reductions in risk of venous thromboembolism, myocardial infarction, bleeding complications, pneumonia, respiratory depression, and renal failure. There was no clear evidence that these effects, in proportional terms, differed by the type of surgical group or the type of neuraxial blockade, although there was limited power to assess subgroup effects reliably. Furthermore, there was no evidence of “catch up” mortality in the neuraxial blockade group between 30 days and 6 months.
The benefits seen for neuraxial blockade may be conferred by multifactorial mechanisms, including altered coagulation, increased blood flow, improved ability to breathe free of pain, and reduction in surgical stress responses.2 In particular, major surgery induces a “stress response” that is substantially altered by neuraxial blockade but not by general anaesthesia.2 This observation, together with the subgroup comparisons shown here, suggests that these benefits are principally due to the use of neuraxial blockade rather than avoidance of general anaesthesia. Thus the key issue seems to be whether neuraxial blockade is used or not, and the way in which this is achieved is less relevant.
Validity of findings
It is unlikely that bias could explain much of the reduction in mortality. We included all randomised trials, irrespective of their initial aims or reported findings. Most trials were not designed to assess major events, but it is unlikely that we missed many deaths or major non-fatal events because we contacted the authors of trials involving 87% of patients and few patients had no outcome data. However, incidence will have been underestimated for non-fatal events that often go undiagnosed, such as deep vein thrombosis. This finding will not bias relative risk estimates6 unless information is selectively available from trials with extreme results. For deep vein thrombosis, at least, the proportional effect of neuraxial blockade in trials designed to assess this outcome was similar to that in other trials. With regard to other potential biases, lack of blinding may have caused some selective misdiagnosis of non-fatal events, but analyses did not indicate publication bias and the overall reduction in mortality was not dependent on inclusion of trials with unconfirmed data or trials for which intention to treat analyses were not possible. Lastly, even though these data represent most of the randomised evidence potentially available, the confidence intervals were wide for many outcomes and relatively little information was available about cause of death.
If the proportional effects of neuraxial blockade are consistent in different patient populations, neuraxial blockade would be expected to result in about one fewer postoperative death and several fewer major complications for every 100 patients at similar risk to those in the studies. However, even though such benefits would be widely regarded as clinically important, the largest individual trial to date180 did not have the power to reliably detect effects of this size. Lack of statistical power may therefore be the principal reason why previous individual trials, editorials,194 and meta-analyses of trials in hip fracture patients 195 196 have concluded that neuraxial blockade had no important effect on mortality.
Our overview indicates that neuraxial blockade reduces major postoperative complications in a wide range of patients. However, uncertainty about the net benefits of neuraxial blockade is likely to remain among some clinicians and for some patient groups. For example, opinion is divided about whether neuraxial blockade is indicated or contraindicated in patients at risk of cardiac complications,197 and it is unclear whether the differences that we observed reflect the benefits of neuraxial blockade alone or are partly due to the avoidance of the adverse effects of general anaesthesia. Such uncertainties provide the rationale for large randomised trials, such as the ongoing multicentre Australian study of epidural anaesthesia and analgesia in major surgery.198 However, since serious complications associated with neuraxial blockade, such as spinal haematoma, are very rare199-201 and more common side effects, such as headache or urinary retention, are not life threatening, our data support recent trends towards increased use of neuraxial blockade. Furthermore, although we focused on intraoperative anaesthetic techniques, postoperative neuraxial blockade has been shown to have additional benefits, at least for pulmonary complications.202 Overall, therefore our data should result in more widespread use of spinal or epidural anaesthesia.
What is already known on this topic
Neuraxial blockade with epidural or spinal anaesthesia reduces the incidence of deep vein thrombosis and one month mortality in hip fracture patients
Insufficient evidence exists for other postoperative outcomes in this surgical group
What this study adds
Mortality was reduced by one third in patients allocated neuraxial blockade
Reductions in mortality did not differ by surgical group, type of blockade, or in trials in which neuraxial blockade was combined with general anaesthesia
Neuraxial blockade also reduced the risk of deep vein thrombosis, pulmonary embolism, transfusion requirements, pneumonia, respiratory depression, myocardial infarction, and renal failure
We thank all trialists who confirmed data and provided extra information for this overview: T K Abboud, A R Aitkenhead, T Asoh, J F Baron, A Bayer, D Berggren, P Berthelsen, D Bigler, P K Bithal, W P Blunnie, R Bode, F Bonnet, N A Borovskikh, M R Brandt, S Bredbacka, M J Breslow, F P Buckley, K S Channer, S P Chin, R Christopherson, F Chung, E Couderc, R J Cuschieri, J B Dahl, F M Davies, M J Davies, M Davis, M De Kock, J Devulder, W Dick, N D Edwards, S M Frank, R L Garnett, S Gelman, S P Gerrish, M M Ghoneim, M S Gold, A Gottlieb, E Hakansson, M Hasenbos, H Hendolin, S W Henneberg, A Holdcroft, A Hole, R Hosoda, P L Houweling, A O Hughes, C Jayr, J Jenkins, N Jia, R D M Jones, L N Jorgensen, J Kanto, H Kehlet, A Lehtinen, M Licker, R A M Mann, P Maurette, S McGowan, P J McKenzie, A D McLaren, G Mellbring, N Melsen, I Milsom, J Modig, S Moiniche, I Murat, J M Murray, J A Odoom, M S J Pathy, J Pedersen, J S Poll, A V Pollock, J P Racle, S Raja, K Reinhart, H Renck, B Rosberg, B A Rosenfeld, H Rutberg, P Ryan, B Scheinin, W Seeling, N Sharrock, I Smilov, T Stathopoulou, R Stenseth, V I Strashnov, J Takala, J Takeda, M V Tseshinsky, H Tsuji, K J Tuman, N Valentin, J M Watters, L G Welborn, A Wessen, I W C White, C Wiessman, P Williams-Russo, M P Yeager, and O N Zabrodin. We thank Iain Chalmers, Rory Collins, Mike Davis, Konrad Jamrozik, John McCall, Tom Pedersen, John Rigg, and Charles Warlow for their helpful comments and Gary Whitlock, Xin-Hua Zhang, Philippa Day, and Valentine Kravtsov for help with translating papers.
Contributors: AR had the original idea for this study. All authors contributed actively to the protocol. NW and AR performed all searching for trials and AM, SS, and GS abstracted the data. NW and TC carried out all data analysis. AR, NW, AM, TC, and SS wrote the first draft of the paper and HK, AvZ, DS, MF, and SM made revisions. AR will act as guarantor for the paper.
Funding Health Research Council of New Zealand and Astra Pain, New Zealand. NW undertook this research during the tenure of a training fellowship from the Health Research Council of New Zealand. AR is a senior research fellow of the National Heart Foundation of New Zealand.
Competing interests HK has received fees for consulting and speaking at meetings from AstraZeneca.