Flowchart of retrieved, excluded, and analyzed trials. *Cochrane Central Register of Controlled Trials. †Regional anesthesia (for instance, intercostal nerve block or wound infiltration with local anesthetics). ‡Fewer than 10 patients per treatment group. §Duration of epidural analgesia only intraoperatively or less than 24 hours postoperatively. ∥Continuous data presented as means but without standard deviation, no dichotomous data. RCT indicates randomized controlled trial.
Funnel plot of natural logarithm of odds ratio vs sample size among all studies reporting on pneumonia. Data are odds ratios plotted against weight (sample size) for each trial included in the meta-analysis on pneumonia. An odds ratio of less than 1 signifies less pneumonia with epidural analgesia. The sizes of the bubbles correspond to the sizes of the trials. The dotted vertical line represents the combined odds ratio (0.54).
Postoperative pneumonia with epidural analgesia compared with systemic analgesia: individual trials and cumulative meta-analysis. *Odds ratio (OR) and 95% confidence interval (CI) in individual trials. †Cumulative OR and 95% CIs. Trials are arranged according to date of publication. Trials with zero events are not shown.
Relationship between the year of publication, the incidence of pneumonia with epidural and systemic analgesia, and the efficacy of epidural analgesia in preventing pneumonia compared with systemic analgesia. Gray symbols indicate trials that used patient-controlled analgesia in controls; white symbols, trials that did not use patient-controlled analgesia in controls. Periods for subgroup analyses were chosen by convenience. Score test for linear trend of odds with year: the odds ratio (OR) estimate is an approximation to the OR for a 1-unit increase per year. Hetero indicates heterogeneity; CI, confidence interval. Trials with zero events are not shown.
Pöpping DM, Elia N, Marret E, Remy C, Tramèr MR. Protective Effects of Epidural Analgesia on Pulmonary Complications After Abdominal and Thoracic SurgeryA Meta-Analysis. Arch Surg. 2008;143(10):990-999. doi:10.1001/archsurg.143.10.990
To review the impact of epidural vs systemic analgesia on postoperative pulmonary complications.
Search of databases (1966 to March 2006) and bibliographies.
Inclusion criteria were randomized comparison of epidural vs systemic analgesia lasting 24 hours or longer postoperatively and reporting of pulmonary complications, lung function, or gas exchange. Fifty-eight trials (5904 patients) were included.
Articles were reviewed and data extracted. Data were combined using fixed-effect and random-effects models.
The odds of pneumonia were decreased with epidural analgesia (odds ratio [OR], 0.54; 95% confidence interval [CI], 0.43-0.68), independent of site of surgery or catheter insertion, duration of analgesia, or regimen. The effect was weaker in trials that used patient-controlled analgesia in controls (OR, 0.64; 95% CI, 0.49-0.83) compared with trials that did not (OR, 0.30; 95% CI, 0.18-0.49) and in larger studies (OR, 0.62; 95% CI, 0.47-0.81) compared with smaller studies (OR, 0.37; 95% CI, 0.23-0.58). From 1971-2006, the incidence of pneumonia with epidural analgesia remained about 8% but decreased from 34% to 12% with systemic analgesia (P < .001); consequently, the relative benefit of epidural analgesia decreased also. Epidural analgesia reduced the need for prolonged ventilation or reintubation, improved lung function and blood oxygenation, and increased the risk of hypotension, urinary retention, and pruritus. Technical failures occurred in 7%.
Epidural analgesia protects against pneumonia following abdominal or thoracic surgery, although this beneficial effect has lessened over the last 35 years because of a decrease in the baseline risk.
Epidural local anesthetics, with or without opioids, provide better postoperative pain relief than systemic opioids.1,2 Epidural local anesthetics reduce central sympathetic stimulation, with subsequent favorable effects on coagulation and homeostasis and on gastrointestinal, metabolic, and immune function.3,4 In patients undergoing surgery for hip replacement, epidural local anesthetics were shown to reduce the incidence of venous thrombosis.5 It was also suggested that in patients undergoing vascular surgery, epidural analgesia may reduce the risk of cardiac events.6
The impact of epidural analgesia on the risk of pneumonia in a high-risk population has never been adequately investigated in a meta-analysis. Previously published meta-analyses either did not report on the risk of pneumonia1,6; reported on composite “pulmonary complications” outcomes, lumping very different end points, such as pulmonary infection, atelectasis, respiratory failure, or reintubation7,8; included populations at low risk of postoperative pneumonia (for instance, patients undergoing surgery for hip fracture)5; or included data from only a very limited number of small trials that tested epidural analgesia.9 The largest meta-analysis so far included data on any neuraxial blockade (intrathecal and/or epidural analgesia with or without general anesthesia) and it remained unclear what the impact of epidural analgesia in patients at high risk of postoperative pneumonia was.10 Moreover, none of these meta-analyses addressed the potential of harm related to the use of epidurals. This uncertainty may be one reason why a recently published report on the behalf of the American College of Physicians concluded that the evidence for epidural analgesia to reduce pulmonary complications after noncardiothoracic surgery was conflicting or insufficient.11
The benefit of epidural analgesia to reduce the risk of postoperative pulmonary complications, specifically pneumonia, remains ambiguous. Our meta-analysis was designed to address this issue.
A wide search strategy was used to retrieve all trials that randomized surgical patients to either epidural or systemic analgesia and that reported on pulmonary complications, lung function test results, or gas exchange parameters. Since there was an intention to include only data from patients at high risk of pulmonary complications, we concentrated on abdominal or thoracic surgery.12 The MEDLINE, Cochrane, BIOSIS, and CINAHL databases were searched from 1966 to March 2006 for reports related to epidural analgesia (epidural, peridural, extradural) and abdominal or thoracic surgery using the Boolean meanings of “or” and “and.” Bibliographies of selected articles were checked for additional references. There was no language restriction.
Trials in adults (aged ≥18 years) in which epidural analgesia was started preoperatively, intraoperatively, or immediately postoperatively and lasted at least 24 hours were included. Systemic analgesia was defined as opioids given alone or in combination with nonopioid analgesics. When controls received locoregional analgesia (for instance, intercostal nerve blocks), the trial was not considered. Trials with fewer than 10 patients per group or trials on trauma patients were excluded. Each of the retrieved articles was reviewed by one of us (D.M.P.) for inclusion. Queries were resolved by discussion with 2 coauthors (E.M. and C.R.). The primary investigators of 31 reports were contacted to obtain additional information since data reporting was inadequate; 6 answered and of 3 of those,13- 15 additional data were included in our analyses.
One of us (D.M.P.) extracted information on type of epidural analgesia (level of insertion, duration, regimen), systemic analgesia (regimen, route of administration), number of patients, length of observation period, and surgery. Data on pulmonary complications, lung function, gas exchange, and adverse events were extracted from tables or text; definitions were taken as reported in the original articles. For each included trial, the method of randomization, concealment of treatment allocation, degree of blinding, and reporting of dropouts was assessed. Two of us (C.R. and E.M.) checked all extracted information. Discrepancies were resolved by discussion with a coauthor (M.R.T.).
For continuous data, weighted mean differences with 95% confidence intervals (CIs) were calculated. For dichotomous data, Peto odds ratios (ORs) with 95% CIs were computed since there were many zero cells. Trials that had 2 zero cells for an end point (ie, no event occurred in either group) were excluded from the summary OR. An OR less than 1 indicated a beneficial effect with epidural. When the 95% CI around the OR did not include 1, the result was considered statistically significant.
We performed formal heterogeneity testing (P < .10 was considered heterogeneous). When the data were homogenous, we used a fixed-effect model to combine data. When the data were heterogeneous, there was an intention to use a random-effects model. However, sensitivity analyses were performed to identify sources of heterogeneity.
Since ORs cannot be easily extrapolated into daily clinical practice, we computed the number needed to treat (NNT) for beneficial effects and the number needed to harm (NNH) for harmful effects using the control event rate and the OR. We calculated 95% CIs around the NNT/NNH only when the 95% CI around the OR indicated that the result was statistically significant.16
Statistical analyses were performed using STATA (version 9; StataCorp, College Station, Texas) and ReviewManager software (version 4.2; Cochrane Collaboration, Oxfordshire, England).
We retrieved 789 reports but rejected 731 for a variety of reasons, including 3 duplicate publications17- 22 (Figure 1). A total of 58 studies met all inclusion criteria.13- 15,23- 77 They were published between 1971 and 2006 and reported on data from 5904 patients. Of the 41 epidural studies that were included in a previous similar meta-analysis,9 we excluded 21 because the surgery was not abdominal or thoracic, data on postoperative pulmonary complications or lung function could not be extracted, they were not randomized, or controls did not receive systemic opioids.
The included trials tested thoracic, lumbar, or both thoracic and lumbar epidurals. Epidural analgesia was commenced before or at the end of surgery; median duration was 3 days (range, 2 to 5 days). Epidural regimens were with local anesthetics, opioids, or both. Controls received subcutaneous, intravenous, or intramuscular opioids with or without nonopioid analgesics. Opioids were given on demand, regularly, or via patient-controlled analgesia (PCA).
Nineteen trials (3504 patients) reported the number of patients who had pneumonia. With epidural analgesia, the odds of pneumonia were significantly decreased (OR, 0.54; 95% CI, 0.43-0.68; NNT, 18 [range, 14-27]) (Table 1). The symmetrical funnel suggested that publication bias was unlikely (Figure 2); however, the data were heterogeneous (Table 1).
Sensitivity analyses were performed to identify potential sources of heterogeneity (Table 1). We looked at the impact of type of analgesia in controls (PCA vs none), site of surgery (thoracic vs abdominal), insertion site of epidurals (thoracic vs lumbar), duration of analgesia (2-5 days), epidural drug regimens (local anesthetics vs opioids vs both), and trial size (<100 vs >100 patients per group). Epidural analgesia was significantly more efficacious compared with systemic analgesia in preventing pneumonia in trials where controls did not receive PCA (OR, 0.30) compared with trials where controls did receive such a device (OR, 0.64) (P < .01). Similarly, epidural analgesia was more efficacious in preventing pneumonia in trials with fewer than 100 patients per group (OR, 0.37) compared with larger trials (OR, 0.62) (P = .08). Site of surgery, insertion site of epidurals, and duration of postoperative analgesia had no significant impact on the efficacy of epidurals. Epidural local anesthetics were as efficacious as combinations of local anesthetics and opioids. The efficacy of epidural opioids alone was tested in one small trial only.
Since the trials were published over a period of 35 years, we tested whether older trials reported on different estimates of efficacy compared with younger trials. First, cumulative meta-analysis was performed using pneumonia as the end point (Figure 3). Over the years, the cumulative OR moved toward unity, suggesting that the efficacy of epidurals, compared with systemic analgesia, had decreased between 1971 and 2006. Second, trials were divided into subgroups representing 4 decades since publication of the first trial in 1971; meta-analyses were performed within each stratum (Figure 4A). In the 1970s, 2 trials were published; the OR for the prevention of pneumonia with epidural compared with systemic analgesia was 0.17 (NNT, 4). In the 1980s, the OR was 0.36 (NNT, 9); in the 1990s, it was 0.46 (NNT, 23); and between 2000 and 2006, it was 0.62 (NNT, 25). Although the OR moved toward unity, the evidence that the effect of epidurals on pneumonia was modified by periods was weak (χ23 = 5.04; P = .17).
To better understand the change in efficacy of epidural analgesia over time, we analyzed the incidences of pneumonia with epidural and systemic analgesia separately (Figure 4B and C). In patients receiving epidural analgesia, the average incidences of pneumonia during the 4 decades were between 3.9% and 8.6%; the score test for linear trend of odds with year was not significant (χ21 = 0.00; P = .99) (Figure 4B). In controls receiving systemic analgesia, the average incidence of pneumonia during the 4 decades decreased from 34.3% to 8.3% and then increased to 12%; the score test for linear trend of odds with year was highly significant (χ21 = 21.02; P < .001) (Figure 4C). The definition of pneumonia, the use of concomitant physiotherapy, and the technique of epidural analgesia have not changed over time. However, there was a change in systemic analgesia techniques toward PCA and multimodal analgesia methods (Table 2).
Since the method of analgesia in controls (PCA) and trial size had an impact on the incidence of pneumonia in controls and the efficacy of epidurals, 2 further sensitivity analyses were performed. In control groups, the score test for linear trend of odds with year remained statistically significant when only trials that did not use PCA in controls were considered (χ21 = 19.58; P < .001) or when only trials with fewer than 100 patients per group were considered (χ21 = 33.83; P < .001).
Epidural analgesia significantly decreased the odds of prolonged (>24 hours) ventilation and reintubation (Table 1). Definitions and diagnosis of atelectasis and respiratory depression varied considerably between trials; meta-analysis was deemed inappropriate. Epidural analgesia significantly increased forced vital capacity at 24 hours, forced expiratory volume in 1 second at 24 hours, and peak expiratory flow rate at 24 hours (Table 3). Epidural analgesia significantly increased arterial oxygen pressure at 24 and 72 hours (Table 3).
Fourteen trials reported on the incidence of postoperative myocardial infarction. Definitions of infarction included an increase in serum concentration of myocardial-specific isoenzyme fractions of creatine kinase or lactic dehydrogenase and typical electrocardiogram changes (elevation/depression of the ST segment and/or new Q waves). The odds of myocardial infarction were significantly decreased with epidural analgesia (OR, 0.55; NNT, 48) (Table 4).
Epidural analgesia significantly increased the odds of intraoperative arterial hypotension (OR, 2.03; NNH, 41) and of postoperative urinary retention (OR, 2.15; NNH, 16) (Table 4). The risk of pruritus depended on the injected opioid (Table 4); with epidural morphine, there was the maximum risk and with sufentanil citrate, the least risk. Ten trials (501 patients) reported on epidural-related technical problems; in 1 trial, the insertion level was lumbar, and in all others, it was thoracic. In 36 patients (7.2%), epidural analgesia had to be abandoned. In 2 trials, the incidence of failures was high, 12% and 15.8%; in the others, it varied between 3.9% and 7.4%. There was no relationship between the publication date of the trials and the rate of technical failures. There was no evidence of any effect of epidural analgesia on renal insufficiency, nausea or vomiting, or in-hospital mortality (Table 4).
In patients undergoing abdominal or thoracic surgery, epidural analgesia decreases the risk of postoperative pneumonia compared with systemic analgesia. Epidural analgesia also decreases the risk of prolonged ventilation or reintubation and improves some lung function parameters and blood oxygenation. The biological basis underlying these associations remains unclear; the improved outcome may be because of a better pain control with epidural analgesia, with subsequently enhanced respiratory function.
Although the beneficial effects of epidurals are numerous and well documented,1,3,5- 7,78,79 the evidence of the usefulness of epidural analgesia for the prevention of postoperative pulmonary complications has been ambiguous. Previous meta-analyses have included limited relevant data9,11 or have tested the effect of epidural analgesia in patients at low risk of pulmonary complications.5 For instance, the recently published systematic review on behalf of the American College of Physicians included only 6 trials compared with 58 in our analyses.11 This may reflect the fact that our literature search was more comprehensive and that our review focused on a specific intervention. Also, in the Urwin et al meta-analysis,5 data from 1200 patients undergoing hip fracture repair were included, and the incidence of pulmonary complications was about 5%. If we assume that epidural analgesia has the power of reducing the odds of pneumonia by 46% (OR, 0.54) (Table 1), then 3400 orthopedic patients are needed to be 95% confident to detect a significant difference in favor of epidural analgesia. We included more trials and more data than previous analyses, and we concentrated on patients at high risk for postoperative pulmonary complications.
Most other pulmonary outcome data may be regarded as surrogate. Atelectasis and respiratory depression were reported, but, as in similar analyses,80,81 definitions varied widely and made it impossible to pool the data. Pneumonia is perhaps the most important pulmonary outcome in this context since it may prolong duration of hospitalization and may lead to death.81- 83
Although cardiac morbidity was not our primary end point, there was evidence that with epidurals, the incidence of myocardial infarction was decreased. The OR was similar to the Beattie et al analysis6 but we analyzed more data, and this may explain why our result was statistically significant. About 50 patients would need to receive an epidural to prevent an infarction in 1 of them. We do not know whether that protective effect is due to the epidural analgesia itself, to improved pain relief or arterial oxygenation, a combination of these factors, or yet another mechanism. We were unable to compare lumbar with thoracic epidurals, since most trials that reported on myocardial infarction tested thoracic epidurals.
Epidural analgesia is not without risk, although severe complications appear to be rare.84- 87 One transient neurological injury in 1700 patients undergoing cardiothoracic surgery with epidural analgesia was estimated.87 In the trials included in our analysis, no severe complications were reported. However, no reporting of adverse events does not mean that none have occurred. Pruritus, although minor harm, was related to specific opioids. Some trials reported on technical failures; the combined estimate was very similar to previous analyses.2 The incidence of technical failures, implying that the attempt to provide postoperative pain relief with an epidural catheter has to be abandoned, has not changed during the last 35 years. This is not surprising since the technique has remained much the same.
The relative benefit of epidural analgesia has decreased over the last 3 decades. In the early 1970s, of 4 patients who received an epidural, 1 had no pneumonia postoperatively, which would have been the case had they all received conventional systemic analgesia. Thirty-five years later, this ratio has decreased to 1 in 25. The decrease in the protective effect of epidural analgesia compared with systematic analgesia does not seem to be related to a decrease in the efficacy of the epidurals per se but rather to a decrease in the baseline risk of pneumonia in patients who received systemic analgesia. The reason for this phenomenon is not obvious. The definition of pneumonia, although variable, has not changed over the years (Table 2). Other features of patient care, for instance, respiratory physiotherapy, the routine use of nasogastric tubes and prophylactic antibiotics, or early mobilization, may all have favorably influenced the baseline risk of pneumonia, although the impact of routine prophylactic respiratory physiotherapy on pulmonary complications after thoracic and abdominal surgery remains unproven.80,81 Also, any change in standard care would be expected to have an impact on the incidence of pneumonia with both epidural and systemic analgesia. A remarkable observation was the change in analgesia provided to controls. In trials published in the early 1970s, analgesia in controls was exclusively with subcutaneous on-demand opioids. In the 1980s, clinical practice changed to morphine infusions and the first PCA devices appeared. Since the 1990s, PCA and multimodal approaches with concomitant nonopioid analgesics became standard care. Patient-controlled analgesia with strong opioids may decrease the risk of pulmonary complications compared with conventional on-demand opioid administration.88 Indeed, in controls receiving PCA, the baseline risk of pneumonia was decreased, and this may explain why the relative impact of epidural analgesia in reducing the risk of pneumonia was lower (Table 1). However, even in trials that did not use PCA, a significant decrease in the baseline risk of pneumonia over time could be observed. Whether changes in standard-care analgesia had a direct impact on the incidence of postoperative pneumonia, or whether the decrease in the baseline risk reflects yet other changes in patient care, remains to be elucidated. Modification of standard care leading to changes in the baseline risk has been described in other settings.89,90 A change in the baseline risk may explain why, sometimes, meta-analyses and subsequently published large trials reach different conclusions.91 Those performing meta-analyses should be aware of this phenomenon when combining data from trials of different epochs that do not necessarily represent contemporary clinical practice. This does not mean that older trials should be excluded from meta-analysis, but the impact of changes in standard care on the efficacy of an experimental intervention should be explored in appropriate sensitivity analyses.
Our analysis has limitations. Some are related to problems that are inherent to meta-analyses. First, we cannot exclude that we missed trials, and some trials could not be included since data reporting was incomplete. Second, we did not analyze pain data since most trials did not report on such data. It has been shown before that epidural analgesia provided better pain relief than systemic analgesia.1 Third, the periods for sensitivity analyses were chosen arbitrarily and by convenience, although it is unlikely that different subgroups would have changed our findings. Fourth, we have taken definitions of outcomes as provided in the original trials. Finally, we chose the time of publication of a trial as an indicator of the age of that trial, although time of publication does not mean time of performance of a trial.
In patients undergoing abdominal or thoracic surgery, epidural analgesia is associated with a statistically significant and clinically relevant decrease in the risk of postoperative pneumonia, although the degree of efficacy has lessened over the last decades. That phenomenon is probably due to a decrease in the baseline risk of pneumonia. There is also evidence that pulmonary function and arterial oxygenation are improved with epidural analgesia and that the risk of myocardial infarction is reduced. Postoperative pain control with epidural analgesia is time-consuming, specific technical and pharmacological skills are needed, and professional surveillance of the patients must be guaranteed. Clearly, epidural analgesia is not without risk, and failures may occur. Our analyses provide an evidence base for rational decision making to ensure the most beneficial use of epidurals in surgical patients.
Correspondence: Daniel M. Pöpping, MD, Klinik für Anästhesiologie und Intensivmedizin, Universitätsklinikum, Albert-Schweitzer-Strasse 33, D-48149 Münster, Germany (firstname.lastname@example.org).
Accepted for Publication: May 8, 2007.
Author Contributions: Dr Tramèr had full access to all of the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis. Study concept and design: Pöpping, Marret, and Tramèr. Acquisition of data: Pöpping, Marret, and Remy. Analysis and interpretation of data: Pöpping, Elia, and Tramèr. Drafting of the manuscript: Pöpping, Elia, and Tramèr. Critical revision of the manuscript for important intellectual content: Pöpping, Elia, Marret, Remy, and Tramèr. Statistical analysis: Pöpping, Elia, and Tramèr. Obtained funding: Tramèr. Administrative, technical, and material support: Pöpping and Tramèr. Study supervision: Tramèr.
Financial Disclosure: None reported.
Funding/Support: The study was partially funded by a grant from UPSA (France), a Bristol-Myers Squibb company. Dr Elia's salary was paid by the EBCAP foundation (http://www.hcuge.ch/anesthesie/ebcap.htm).
Role of the Sponsor: The sponsor was not involved in the design or conduct of the study; collection, management, analysis, or interpretation of the data; or preparation, review, or approval of the manuscript. The sponsor solely provided funds for travel, meetings, and translations of non-English papers.
Additional Contributions: Special thanks go to Michael F. M. James, PhD, FRCA, Dr Per Flisberg, Christian Jayr, PD, Bernd Wiedemann, PD, Dr Shahnaz Christina Azad, and Dr Adrianus J. de Vries, who responded to our inquiries. 4T Traduction (Levallois-Perret, France) translated 6 Turkish, Japanese, Chinese, or Spanish papers into English.