Has the Use of Meta-Analysis Enhanced Our Understanding of Therapies for Postoperative Nausea and Vomiting?

Abstract

The MEDLINE database for the years 1995-1998 contains 188 reports of clinical trials involving different therapeutic regimens for the management of postoperative nausea and vomiting (PONV), yet the best strategy for its prevention and therapy remains controversial. Clinicians still debate which patients benefit from antiemetic prophylaxis against PONV, along with the preferred drug, dose, and timing of administration. There have been suggestions that these clinical decisions be evidence-based from systematic reviews that use explicit and reproducible methods to limit bias in identifying and rejecting studies [1,2]. In a quantitative systematic review or meta-analysis, a mathematical synthesis of results of primary studies is performed using complex statistical methods to evaluate the diversity among results and to estimate a common pooled effect with increased precision [3]. These methods have been used widely in epidemiological studies, in evaluation of diagnostic tests, and in establishing guidelines and clinical pathways for the management of specific conditions [4]. Several have recently appeared in the anesthesiology literature [5-8], and in this issue of Anesthesia & Analgesia, three different investigative groups have used this methodology to evaluate the efficacy of both pharmacological [9,10] and nonpharmacological [11] therapies for PONV. The major purported benefit of the meta-analytic approach is that it is a powerful tool for "uncovering statistical significance that was not apparent in small statistically under powered studies" [12]. Small but clinically important differences in therapeutic efficacy have been identified by meta-analysis well before large-scale clinical trials have been completed (e.g., reduced post-myocardial infarction mortality with beta-blockade and fibrinolytic therapy) [13,14]. However, many investigators have reservations about combining the results of studies performed on different patient populations in different places at different times with different doses and for different reasons [15,16]. There are also examples of meta-analysis addressing the same question but reaching different conclusions, depending on the trials that were included or whether a fixed-effects or random-effects model was used [17]. Proponents of meta-analysis received a major setback when large-scale trials of magnesium therapy failed to confirm a finding through meta-analysis that this was an effective, safe, and simple intervention in myocardial infarction [14,18]. Similar discrepancies between evidence derived from meta-analysis and mega-trials were noted with nitrate therapy, and they are reported to occur 10%-23% of the time [13,19,20]. This has led to claims that the large clinical trial should be resurrected as the "gold standard" of best evidence and has led some to recommend that one should "forget about the meta-analysis" [21]. However, others have strongly opposed this suggestion [22]. Given these controversies regarding meta-analysis, we should examine what "new" information is provided by the application of this technique to the management of PONV beyond that provided by adequately designed clinical studies. As elegantly pointed out by Domino et al. [9], meta-analysis has inherent weaknesses that limit its validity. These include publication bias (i.e., underpowered studies with "positive" results are more likely to be published than similarly sized studies with negative results), non-standardization of other confounding factors (e.g., anesthetic technique, type of surgical procedure, patient population, nonuniform methods of data collection), and variable outcome measures (i.e., surrogate end points). Meta-analytic studies of PONV have additional weaknesses based on the quality of the data gathered. Investigators often do not distinguish between nausea and vomiting but combine the symptoms of PONV into a single end point defined as vomiting (emesis), retching, or nausea severe enough for a patient to require "rescue" antiemetic therapy. Although vomiting is an objective end point, nausea severe enough to require therapeutic intervention is highly subjective and would vary among patients as would the threshold for treatment among different postanesthesia care unit nurses caring for patients who experience emetic symptoms. The results of a meta-analysis are only as good as the quality of the collected data, and pooling of results from different studies may lead to intrinsic inconsistencies in the separate incidences of nausea, vomiting (or retching), and combined PONV. The difficulties with using meta-analysis for therapeutic decision-making in PONV are best exemplified by the inconsistent conclusions of separate dose-ranging analyses performed by the same group of investigators for the prevention and treatment of this condition [6,23]. In one study, Tramer et al. [23] recommended that the optimal IV dose of ondansetron for the prophylaxis of PONV was 8 mg. However, in another study [6], the same group concluded that if a clinician did not administer any prophylactic antiemetic and waited for a patient to have symptoms of PONV, a dose of 1 mg of ondansetron IV was as effective as any larger dose in preventing further symptoms. This research group has also emphasized that duplicate publication is common in PONV studies, occurring in 25% of all relevant reports; inclusion of duplicate publications in the meta-analysis would result in an unwarranted increase in the precision of the conclusion. The comparative studies of ondansetron, metoclopramide, and droperidol used by Domino et al. [9] in their meta-analysis did not always stipulate the time of administration of the drug, and others have suggested that ignoring this effect may result in invalid conclusions [24,25]. Of interest, Domino et al. [9] had initially concluded that ondansetron was superior to both droperidol and metoclopramide as a prophylactic antiemetic [26]. The preliminary ondansetron-droperidol comparison was extended from 850 to 3865 patients in the final report, but 2061 of the additional patients came from a single, multicenter study sponsored by the manufacturers of ondansetron [27]. This mega-trial clearly showed that droperidol 0.625 or 1.25 mg IV was comparable to ondansetron 4 mg IV and did not result in increased drowsiness or delayed discharge compared with the other treatment groups. In the current meta-analysis by Tramer et al. [10], the authors suggest that ultrasmall doses of droperidol may be effective in limiting the emetic side effects of morphine administered using a patient-controlled analgesia delivery system. They reported that, despite the absence of an established dose-response relationship for droperidol's antiemetic activity, side effects as a result of its administration were increased in a dose-related fashion. The major outcome of this meta-analysis was the hypothesis that the "optimal dose of droperidol is likely to be <0.1 mg/mg of morphine (or <4 mg of droperidol/24 h)." In addition, this systematic review could provide preliminary data for subsequent power analysis if these results are confirmed in large, randomized, placebo-controlled (clinical) trials comparing different doses of droperidol. Finally, the article by Lee and Done [11] suggests that nonpharmacologic techniques are more effective than placebo treatment in preventing vomiting after surgery in adults, but not in children. They concluded that nonpharmacologic techniques are equivalent to commonly used antiemetic drugs in preventing vomiting after surgery in adults. The paucity of data regarding the use of nonpharmacologic techniques led the investigators to combine various types of stimulation at the P-6 acupuncture point (i.e., acupressure, electroacupuncture, transcutaneous electrical acupoint stimulation) and analyze the results collectively rather than for each individual type of acupoint stimulation. However, practicing clinicians need to know which particular method of acupoint stimulation is superior. Only direct comparisons in large-scale studies will answer this question. Furthermore, the conclusion that acupuncture is not effective in children may be questioned, as it is based largely on studies in which the therapy was administered while the child's central nervous system was profoundly depressed by general anesthetic drugs. The results may differ if children received acustimulation while awake. It seems appropriate to ask what clinically useful "new" information has emerged from these quantitative analyses of PONV studies. Although large, multicenter studies have replaced smaller, tightly controlled single-site clinical studies, the results are more difficult to interpret because of the inevitable confounding factors that exist among different testing sites. There are similar problems in attempting to combine studies performed at different times in different locations as part of a quantitative systemic review. The major weakness of meta-analysis is that it magnifies the problems of individual studies. Although randomized controlled trials are included in a meta-analysis, comparisons among trials are not randomized comparisons [28]. At best, the hypothesis engendered by a meta-analysis must still be verified in a large trial. A meta-analysis of 30 trials with 60 patients in each trial does not provide evidence equivalent in quality to a well designed single trial of 1800 patients. Although such large randomized controlled trials are difficult to perform, they remain the "gold standard" for determining the best choice among different therapeutic options. An adequate number of subjects should be recruited into the studies so as to have the necessary power to draw meaningful conclusions regarding clinically relevant outcomes (e.g., patient satisfaction, return to work) [25], rather than surrogate end points (related simply to the incidence of PONV) [12]. Unfortunately, the currently available data from meta-analyses of PONV studies have not addressed these more pertinent issues, and therefore, a cautious attitude to accepting clinical practice guidelines based solely on meta-analytic techniques is still justified.