Clinical review: Evidence-based perioperative medicine?

The present article outlines the basic principles of Evidence Based Medicine (EBM) and how they should guide clinical practice. The evidence supporting a selection of perioperative interventions is assessed against objective criteria. Many of the perioperative interventions that have been widely adopted into clinical practice are supported by very limited evidence. Conversely a high level of evidence supports other interventions that have not been so widely adopted. This may be due to concerns about limitations in the design and conduct of some of the clinical trials.


Introduction
The present article outlines the basic principles of evidencebased medicine (EBM) and how they should guide clinical practice. Criteria by which items of evidence are judged and the hierarchy of levels of evidence will be reviewed. The evidence supporting a selection of commonly discussed 'perioperative medicine' interventions will be presented and will then be assessed using these objective criteria. The physiological rationale and scientific basis for these interventions will not be discussed in detail.

Evidence-based medicine
EBM is "the integration of best research evidence with clinical expertise and patient values to optimise clinical outcomes and quality of life" [1]. There is often tension between empiricism and EBM: the 'knowledge' of practitioners gained from 'experience' may not agree with the results of an EBM approach to clinical practice [1]. As physicians we recognise that there is an element of art in the practice of medicine. We must also recognise, however, that data derived from rigorous clinical experiments should be more compelling than personal anecdote.
When considering preventative and therapeutic interventions a hierarchy of levels of evidence exists from the highest standard (systematic reviews and large randomised controlled trials [RCTs]), to the lowest level (consensus statements and expert opinion) (see Table 1 for levels of evidence in relation to therapeutic interventions). [2,3]. Clinical practice should, where possible, be guided by Level 1 evidence. In some cases, however, it is not possible to obtain this level of evidence for particular interventions; for example, when conducting the relevant trial would require excessive resources or would be considered unethical. In this situation the highest available standard of evidence should be used (see Table 1).
The well-conducted RCT is the most robust and reliable experimental methodology for comparing preventative and therapeutic clinical interventions. RCTs are designed to demonstrate that any observed differences in outcome between patients allocated to different interventions (A versus B, or A versus control) occur not simply as a result of chance, and that bias is minimised. To this end, RCT design frequently involves blinding of some or all of the involved parties (investigators, patients, carers) to allocation of treatment until after the study is complete. The aim of blinding is to reduce the risk of unrecognised bias being introduced by participants when they are aware of intervention allocation. A RCT is commonly held to be 'large' if it includes at least 1000 patients [4]; few studies in the perioperative setting come close to meeting this criterion. Systematic reviews or overviews are an attempt to avoid the subjective approach of classical narrative review where authors use self-selected references to support an established argument. A well-conducted systematic review should have the following elements: a clearly framed question or objective, the use of an appropriate methodology to search for all relevant literature, and a systematic approach to study selection, assessment of study quality, data extraction and data analysis. Meta-analysis is the use of quantitative methods to summarise the results of a systematic review [5]. Systematic reviews have a useful role both in quantitative pooling of data using the techniques of meta-analysis and by providing qualitative summaries of original reports in a specific area.
Objective methods exist for quality assessment of studies [6,7], and this is an important element of the systematic review process. The old adage 'rubbish in, rubbish out' succinctly summarises the futility of performing a metaanalysis of poor quality studies and then imbuing the summary statistic with any significance. The inclusion of 'poor quality' studies may bias the results of systematic reviews [6].
Heterogeneity reduces the validity of the result of a systematic review. This heterogeneity can be clinical (e.g. variations in the intervention or setting) or can be statistical (e.g. variations in the outcomes suggesting that the studies may not having been addressing the same question) [5]. Evidence of heterogeneity suggests that it may be inappropriate to group together the selected studies.
When assessing preventative or therapeutic interventions the best evidence may be provided by a systematic review of all relevant RCTs. However a systematic review of older, smaller or weaker studies, or demonstrating significant heterogeneity, may be inferior to a large well conducted and generisable RCT.
Founded in 1993, the Cochrane Collaboration is an international, independent, nonprofit organisation that produces and disseminates systematic reviews of healthcare interventions, and promotes the search for evidence in the form of clinical trials and other studies of interventions (http://www.cochrane.org/index0.htm). The major product of the collaboration is the Cochrane Database of Systematic Reviews, published quarterly as part of the Cochrane Library (http://www.cochrane.org/reviews/clibintro.htm). The Cochrane Anaesthesia Review Group (http://www.cochraneanaesthesia.suite.dk/) is one of the many subject-specific collaborative review groups affiliated to the Cochrane Collaboration, and includes within its scope anaesthesia, perioperative medicine, intensive care medicine, prehospital medicine, and resuscitation and emergency medicine. Reviews conducted under the auspices of the Cochrane Collaboration use a clearly defined methodology, and external peer review is integral to each stage of the review process.
Meta-analyses may be subject to bias in the same way as individual RCTs. Publication bias, English-language bias, citation bias, multiple publication bias, database bias and study inclusion bias are recognised and are covered in detail elsewhere [8]. Specific tests for assessing bias are available (e.g. Funnel plot for publication bias) [9,10]. Reviews published by the Cochrane Collaboration seem to show less evidence of bias (13% versus 38% of reviews studied) than systematic reviews selected at random from major journals [9].
Large RCTs and systematic reviews may disagree, and this occurs at a rate greater than chance alone would predict. When both the magnitude and uncertainty of treatment effects were considered, large trials disagreed with metaanalyses in 10-23% of cases [4].

Perioperative interventions and EBM?
We have chosen to consider a subjectively generated list of perioperative interventions commonly considered to be elements of the practice of the anaesthetist or the perioperative physician. For each intervention we describe the extent of the supporting literature (RCTs and systematic reviews), the level of evidence and the outcome supported (mortality, length of stay, morbidity). We have searched PubMed (http://www.ncbi. nlm.nih.gov/entrez/query.fcgi?db=PubMed), the Cochrane Library (http://www.cochrane.org/reviews/clibintro.htm) and the internet based "Evidence-based Perioperative Medicine" resource (http://www.hcuge.ch/anesthesie/anglais/evidence/ arevusyst.htm) for systematic reviews and RCTs. Where there are more than 10 RCTs addressing a particular question we have not listed them separately. Table 2 summarises the evidence (RCTs and systematic reviews) supporting each listed intervention. However due to limitations in the design and conduct of these studies there is considerable controversy amongst professionals that care for patients in the perioperative setting as to whether some of these interventions are as effective as the data suggest.

'Optimisation'
The observation that patients who survive major surgery have elevated cardiac output and oxygen delivery values led to the development of goal-directed therapy, otherwise known as haemodynamic optimisation. Cardiac output and oxygen delivery are augmented by the administration of intravenous fluids, with or without the addition of inotropic agents such as dobutamine or dopexamine. A treatment algorithm guided by haemodynamic monitoring drives therapy. Treatment goals are often supranormal values of blood-flow-related variables (e.g. cardiac index, oxygen delivery, stroke volume).
Two systematic reviews with meta-analyses studying this strategy in the perioperative setting have concluded that mortality is reduced using this approach [11,12]. The first review showed a significant reduction in mortality if therapy was initiated preoperatively (two trials, 195 patients). The second review of 21 trials (13 perioperative trials, > 1008 patients) revealed significant mortality reduction when patients were treated before the development of organ failure, when there were control group mortalities > 20% and when therapy produced differences in oxygen delivery between the control and protocol groups [12]. An additional systematic review of perioperative fluid volume optimisation following proximal femoral fracture (two studies, 130 patients) showed a reduction in the hospital length of stay in the intervention group [13]. However, a recent large perioperative RCT failed to show any benefit [14]. The large RCT has been criticised on a number of design and conduct issues, and the systematic reviews exhibit significant heterogeneity. This is a good example of a large RCT producing a different result from previous systematic reviews.

Beta-adrenergic receptor blockade
Cardiovascular disease is an important cause of perioperative morbidity and mortality. This is thought to be due to sympathetic over-activity increasing heart rate, myocardial workload and the incidence of arrhythmias. Perioperative blockade of beta-adrenergic receptors may be protective against cardiovascular complications such as myocardial infarction and arrhythmias, and therefore may reduce mortality.
Three (small) randomised trials examined the use of betablockers in surgical patients with known or suspected coronary disease [15][16][17]. Two of these trials demonstrated a significant reduction in mortality in the intervention group. However, one study (200 patients undergoing elective major noncardiac surgery) stopped beta-blockers prior to surgery in patients in the control group who were already taking them, and this is inconsistent with current guidelines [15]. In the other study (112 patients presenting for elective major Available online http://ccforum.com/content/9/1/81 , patients were preselected on the basis of positive dobutamine stress echocardiograms from an initial cohort of 1351 patients, which makes it difficult to generalise these data to a population not selected in this manner [16]. In the third study (107 patients undergoing elective knee arthroplasty), a reduced prevalence and duration of postoperative myocardial ischaemia detected with Holter monitoring was demonstrated [17].
There are no large RCTs (> 1000 patients) or systematic reviews that demonstrate the effect of this intervention on mortality. Nonetheless, the results of these small studies suggest that perioperative beta-blockade improves outcome in patients with known or suspected heart disease. The American College of Cardiology and American Heart Association guidelines now recommend this strategy [18]. A large international RCT is currently being conducted to test the hypothesis that perioperative administration of betaadrenergic blockers in patients with risk factors for ischaemic heart disease reduces mortality (the 'POISE' study).

Regional anaesthesia/analgesia
Improved postoperative pain relief is important for patient comfort and may decrease the hospital stay and lead to reductions in morbidity. Improved blood flow consequent on sympatholysis has additional potential benefits, including a reduction in thromboembolic complications and improved gastrointestinal function.
A large number of RCTs in this area have been summarised in several systematic reviews with meta-analysis. Pooled analysis of 141 randomised trials involving a total of 9559 patients revealed a 30% reduction in mortality associated with neuraxial blockade and significant reductions in thromboembolic and respiratory complications [19]. Pain relief with epidural analgesia with local anaesthetic agents has been demonstrated to be superior to parenteral opioids alone in a meta-analysis of 100 studies [20]. Regional anaesthesia for hip fracture surgery when compared with general anaesthesia produced comparable results for most of the outcomes studied (16 trials, 2191 patients). Regional anaesthesia may have reduced short-term mortality, but no conclusions could be drawn for longer-term mortality [21]. For hip and knee replacement surgery short-term postoperative pain relief was improved with regional anaesthesia but minor complications were more frequent, and there were insufficient data to draw conclusions on the frequency of rare complications from epidural analgesia, postoperative morbidity or mortality, functional outcomes, or length of hospital stay [22]. Following abdominal surgery, the time to return of gastrointestinal function was reduced with epidural anaesthesia but there were insufficient data to comment on mortality [23].

Nutrition
The concept that improving nutritional status and minimising catabolism will improve perioperative outcome has been tested in a large number of RCTs, and two systematic reviews have addressed questions in this area. A systematic review of 11 studies with 837 patients demonstrated that early enteral feeding postoperatively after gastrointestinal surgery reduced infection rates and the length of stay but did not significantly reduce mortality [24]. When compared with enteral nutrition, total parenteral nutrition does not influence the death rate of surgical patients, but may reduce the complication rate, especially in malnourished patients [25].

Supplemental oxygen therapy
The bactericidal activity of neutrophils is mediated by oxidative killing by the production of superoxide radicals, the rate of generation of which is proportional to the partial pressure of oxygen. At surgical wound sites this is likely to be low due to disrupted vascular supply and decreased blood flow perioperatively. Resistance to infection may be enhanced by the administration of supplemental oxygen to increase tissue oxygenation and augment neutrophil function. A moderate-sized randomised trial (500 patients undergoing elective general surgery) in which 30% or 80% oxygen was administered perioperatively demonstrated a reduction in wound infection in the higher FiO 2 group [26]. However, a more recent small study of similar design found an increase in infection in those administered the higher FiO 2 [27]. There are no RCTs or systematic reviews that demonstrate a beneficial effect of this intervention on mortality.

Maintenance of normothermia
Inhibition of physiological temperature control mechanisms, peripheral redistribution of body heat during general anaesthesia and the cool operating room environment contribute to perioperative hypothermia. Immune cell function becomes impaired as a consequence of low tissue oxygenation and hypothermia. Temperature reduction also has an adverse effect on collagen deposition. These factors have an unfavourable effect on surgical wound healing.
Four small to moderate-sized clinical studies (60-300 patients) have compared active warming (normothermia) with the standard of care (resulting in mild hypothermia) [28][29][30][31]. In 200 patients undergoing elective colorectal surgery, those randomised to receive active warming had a reduced incidence of wound infection [29]. In 60 patients undergoing primary hip arthroplasties, the postoperative blood loss and transfusion requirements were significantly greater in the hypothermic patients [31]. In 300 patients undergoing abdominal, thoracic or vascular surgical procedures with documented coronary artery disease, or risk factors for coronary disease, perioperative cardiac events and ventricular tachycardia occurred less frequently in the normothermic group [28]. In 150 patients undergoing elective major abdominal surgery, a decreased duration of postanaesthetic recovery was observed in the normothermic group [30]. There are no RCTs or systematic reviews that demonstrate a beneficial effect of this intervention on mortality.

Critical care provision
Differences in case-mix-adjusted mortality have been noted between countries where levels of critical care provision differ [32].
In a systematic review of 27 studies, high-intensity ICU physician staffing versus low-intensity ICU physician staffing was associated with reduced hospital and ICU mortality and with reduced hospital and ICU length of stay [33]. No RCTs (large or small) or systematic reviews exist to test the hypothesis that critical provision postoperatively reduces mortality.

Conclusions
Many of the perioperative interventions that have been widely adopted into clinical practice are supported by very limited evidence. For a number of interventions the data are either limited in quantity or quality, or are inconsistent. Systematic reviews are no better than the studies that they bring together, and those that include many small studies are often limited by problems of heterogeneity. Systematic reviews should be conducted with the same methodological rigour expected for RCTs. Systematic reviews conducted under the auspices of the Cochrane Collaboration have an established methodology and peer review process, and they may be less prone to bias than non-Cochrane systematic reviews.