Systematic review on the recurrence of postoperative nausea and vomiting after a first episode in the recovery room – implications for the treatment of PONV and related clinical trials

Postoperative nausea and vomiting (PONV) are frequent adverse events after anaesthesia. In western countries about 10% of the population undergoes a surgical procedure every year. Assuming that still a quarter of all patients suffer from PONV, more than 18 million patients are nauseated or vomit after their surgery in Europe. Although these symptoms are self-limiting and rarely cause major medical problems they are distressing for the patient and may have negative economic implications, e.g. by causing unanticipated hospital admission in surgical outpatients due to intractable vomiting.

There is some evidence that patients at a considerable high risk for developing PONV can benefit from prophylactic antiemetic treatment [1]. However, applying the same approach for patients at a low or moderate risk is not unequivocally supported and fails to improve patient satisfaction compared to symptomatic treatment in the postoperative recovery room [1]. The main reason for the poor results of any conventional antiemetic strategy is that the number of patients that need to be treated prophylactically highly depends on the baseline risk of the population. For instance, if high risk patients with an expected risk for PONV of 80% are treated with an antiemetic intervention that halves this baseline risk then the achieved absolute risk reduction of 40% translates into a NNT of 2.5 that without doubt is highly efficient. In a population with a low risk for PONV (e.g. 10%), however, the same intervention will lead to a (disappointing) 5% absolute risk reduction and the NNT is 20.

This simple calculation demonstrates that routine antiemetic prophylaxis is not indicated from a medical (potential side effects of antiemetics) and economic point of view (increased costs). However, the optimal threshold (percentage of expected risk) at which prophylaxis is indicated cannot be quantified since several other issues modify the perceived risk. Thus, a recent consensus conference did not give a clear recommendation regarding this point [2].

The situation is further complicated by the lack of a clinically satisfying solution to predict PONV. The accuracy of prediction using one of the available simple scoring systems [3, 4] is at best increased to a discriminating power of about 70% compared to a random guess with a pre-hoc accuracy of 50% [5, 6]. Furthermore, even if there was a perfect prognostic tool, theoretical consideration using computer simulation of the effectiveness and efficacy of various prophylactic antiemetic strategies could not identify one optimal approach since the "most effective" or "most efficient" strategy strongly depends on the distribution of patients in a given institution [7]. As a consequence, clinical strategies to deal with PONV in the postanaesthetic care unit vary widely between different institutions [8].

An alternative to administering prophylactic antiemetics is to treat established nausea and vomiting postoperatively. This strategy has been shown to be more cost-effective than prophylaxis [9]. However, a systematic review evaluating studies on treatment of PONV stated that there is a discrepancy between the plethora of trials on prevention of PONV and the paucity of trials on the treatment of established symptoms [10]. There are several reasons for this obvious disparity. First, manufacturers of antiemetic drugs have more commercial interest in prophylaxis strategies which may be the reason for the complete lack of trials with older classic antiemetics when treatment of established PONV is concerned. Second, in the absence of evidence in favour of a (gold) standard for treatment of established PONV, one will have to utilize placebo control groups. Since ethics committees have the duty to balance between the patients' interest and the scientific benefit, it can be speculated that they will be reluctant to approve such trials since there is good evidence from several surveys of in- and outpatients [11, 12] that postoperative nausea and vomiting is as unpleasant as postoperative pain. The problems related to placebo treatment are supported by thorough criticism of trialists and sponsoring companies performing placebo-controlled trials in the prophylaxis of postoperative nausea and vomiting [13, 14] – trials that all provided sufficient antiemetic rescue medications for all participants! Third, and most importantly, therapeutic trials are more difficult to perform. Of the total number of patients giving informed consent to participate in such a trial, only about 20–30% (at "best") can be expected to be randomized due to the fact that the majority of patients never suffer from PONV. In addition, the number of patients that are willing to participate in a placebo-controlled PONV-trial will probably be rather low, especially in patients with a high risk to develop emetic symptoms, e.g. due to previous PONV. Furthermore, study protocols will have to be very precise concerning the inclusion criteria and the proceedings in case of adverse events (e.g. because of the necessary definitions when to administer a rescue antiemetic, etc.) and these regulations will probably result in artificial circumstances that might not be representative of a "real world" situation and thus such trials might lack external validity.

Besides all these obstacles, there is an obvious need for trials showing efficacy of an antiemetic for the treatment of PONV-symptoms prior to its routine application. In the absence of an accepted (gold) standard treatment [15], one may have to use placebo control treatment arms. Any efficacy consideration in terms of clinically relevant treatment effects, however, implies some knowledge about the incidence of an event with placebo treatment. Confronted with this dilemma we decided to update a recent review on trials on the efficacy of therapeutic antiemetic interventions [10]. This material was amended by unpublished data that were collected in order to validate a variety of different risk scores to predict the occurrence of PONV [5] and to train and validate an artificial neural network for the same purpose [16].

Thus, the aim of this study was twofold. First, it was set up to provide a more reliable incidence estimate for the recurrence of emetic symptoms following an initial emetic episode and after receiving placebo during the early postoperative period (e.g. in the recovery room). Secondly, these data were to propose the choice of clinically relevant effects for future trials studying antiemetic interventions for the treatment of PONV. Based on these data we will also discuss the possibility to perform PONV treatment trials with no placebo comparator group by using a virtual benchmark for such trials.

In order to determine a more reliable incidence estimate for patients with reappearance of nausea or vomiting after a first episode of these symptoms in the recovery room, a systematic review of trials was performed that documented the rate of recurrence of PONV after giving placebo or no treatment. For data retrieval and extraction we updated a previous systematic review of Kazemi-Kjellberg et al. published in 2000 [10].

We used the search strategies of this systematic review and created modified algorithms with the help of a librarian (S.F.). We used the following search terms or roots for inclusion "postoperative OR postsurg*", "nausea OR vomit* OR emesis OR emetic OR retch*", "treatment OR control" and the phrases "chemotherapy OR radiotherapy OR prophylaxis or prevention" for exclusion. There was no restriction to placebo-controlled trials. Given this strategy, we searched Medline, Old Medline (reaching back to 1950 – 1959 Current List of Medical Literature), EMBASE, and the Cochrane Controlled Trials register (CENTRAL) aiming at all trials that provide numbers about the incidence of patients that suffer from any emetic sequelae after a first episode of PONV during the early postoperative period. The date of the last literature search was 16^th December 2005. Additionally we hand-searched the references of all matching papers and review articles for additional reports but did not contact the pharmaceutical industry for further unpublished data. In contrary to the first systematic review [10] we did not exclude articles that reported the incidence of PONV instead of two distinct incidences for nausea and vomiting. For the given study, these PONV incidences were equated with nausea since results from almost all trials suggests that vomiting without prodromal nausea is extremely rare. Thus, "PONV" was defined as any nausea or vomiting or retching (with or without a rescue medication at this stage) in the postoperative period and "vomiting" as retching or vomiting postoperatively. Most of the identified trials did not use an intraoperative antiemetic prophylaxis and do not report risk factors for PONV present in the patients who then suffered from nausea or vomiting in the recovery room. However, studies were not excluded in case such explicit information were lacking. The data from the identified reports were independently extracted by two researchers. The Oxford-score (Jadad-score) was assessed for each study [17]. In the case of disagreement between the two results a referee (L.E.) was consulted to solve the disagreement. The analysis was performed according to the recommendations of the QUORUM-statement [18].

As pointed out in the introduction the treatment of established symptoms will remain a cornerstone of any strategy aiming to decrease the sequelae for patients resulting from postoperative nausea and vomiting despite the increasing use of prediction models and guidelines for PONV-prophylaxis. In short the main reasons are listed as follows:

- Prediction models for PONV for an individual patient that are applicable in the clinical practice have by now not turned out to be satisfactory [5, 6]. Although the accuracy can be increased to some extent using sophisticated statistical methods [16], the main and limiting factor is the absence of strong predictive parameters [22].

- A clinical decision whether or not to administer an antiemetic measure is highly depended on the individual characteristics of the patient and the clinical circumstances. For instance, in the outpatient setting antiemetic prophylaxis is more often warranted than in inpatients. As a consequence, guidelines did not and will probably never define an exact threshold when antiemetic prophylaxis is indicated [2].

- Computer simulations on the effectiveness and efficacy of various antiemetic strategies have yielded no perfect solution. Instead, it is suggested to develop individualized standards at each institution depending on the composition of patient population [23].

The latter analysis further suggests that despite a liberal antiemetic prophylaxis PONV will remain a "10–20%-problem" in the postoperative period. Thus, studies on the treatment of PONV are urgently warranted. However, despite the need for such treatment studies it is very unlikely that such trials – especially placebo controlled trials – will be performed in the near future. Several reasons have been described in the introduction section.

There are only two realistic solutions to deal with this lack of sufficient data on antiemetic treatment.

One is that clinicians rely on the data that is available for the prophylaxis of PONV and extrapolate these results to the demands of treatment.

The other one is to fill this gap of evidence with data from well conducted clinical trials. In the absence of a well established (gold) standard treatment [15], one may stick to the need of running placebo-controlled trials despite their inherent difficulties. For this approach the calculated estimate for the recurring incidence of PONV can be helpful to plan such a placebo-controlled study. For example, using the 0–24 hour estimate of 84% (rate of recurring PONV) and assuming a relative risk reduction by a certain antiemetic treatment then the expected incidence in the active treatment group would be 59% (= 0.84. [1-0.3]). This absolute risk reduction of 25% points is a reasonable margin for a clinically relevant effect and a sample size of 2 × 57 patients can be determined for two parallel groups (placebo/antiemetic), a power of 80% and alpha (one-sided) 2.5% for Fisher's exact test. Note, however, that such a trial will of course be underpowered for smaller effects that might be of some clinical relevance in the absence of other evidence or in case the true incidence of recurrent PONV is lower than predicted.

Instead of using a placebo comparator researchers may think about using an active comparator and to compare the effectiveness of these two verum groups with each other or against a virtual margin based on the calculations presented here. Note, however, that such a proceeding comes at the price of not having demonstrated efficacy of any of the treatments compared to placebo. This approach addresses the special problem of placebo-control groups within this special niche of research and makes use of "historical" data of otherwise small trials that was pooled using meta-analysis. Despite all legitimate criticisms not to perform clinical trials without a placebo comparator, there are some issues within this special area of research that may justify the use of "historical" data. For the great majority of antiemetic drugs efficacy and effectiveness has been demonstrated in numerous prophylaxis trials with PONV setting and other indications where nausea and vomiting occur (e.g. radio- and chemotherapy for cancer). There is no biological rationale why antiemetics should not work postoperatively when their activity has been proven intraoperatively. Furthermore, meta-analyses that have been performed on the (few) placebo-controlled treatment studies available have yielded results very similar to the numerous meta-analyses that have been performed on the prophylaxis trials. Combined with the difficulties related to placebo control groups within this field and the ethical concerns of using placebo, one may justify the use of empirical rate of recurrent PONV after placebo or placebo/no treatment respectively for a single-arm study planning. By summarizing the existing data from an extensive literature search we can be confident that PONV re-occurs in at least 50% of the cases within the next 1–2 hours after a first event. Within the first 6 hours after a first episode PONV occurs in about at least 65% with no obvious further major increase (to 70–75%) when the 24-hour observation interval is used instead.

Since these "best-case" scenario statements are based on the lower 95% confidence interval bounds of the recurrent PONV rates, as a suggestion, any rate under antiemetic treatment should be better than these margins. This could mean that a similar 95%-confidence interval for the recurrent PONV rate under an investigational antiemetic intervention will not include such a "best-case" scenario value.

Tramèr and colleagues argue in favour of the use of placebo-controlled PONV-trials. They performed a systematic review of efficacy trials investigating ondansetron for the prophylaxis of PONV and found an enormous heterogeneity of the chance to develop nausea and vomiting (control event rate). In part this heterogeneity was assumed to be due to the random variation in small trials. We also observed between-trial variability for the recurrence of PONV. A closer look at figure 1 (PONV), however, suggests that very few trials are responsible for the variability. One trial comprises data from a Japanese group whose results have been repeatedly questioned for validity in the past [24, 25]. However, since this study contributes only 20 patients it has no relevant impact on the global result of the analysis (e.g. estimate of PONV-recurrence after 24 hours observation period is 84% [95%-CI: 73%...91%] when the data of Fujii are included and 86% [95%-CI: 76%...92%] when they are removed from the meta-analysis). In the case of recurrent vomiting, again the report of Fujii [26] and a second small trial with 12 patients reporting on the effectiveness of small doses of propofol after gynaecological laparoscopy [27] might be viewed as outlier.

Besides the limited between-study heterogeneity, the inclusion of numerous small trials does not seem to increase the between-heterogeneity which is in contrast to other quantitative systematic reviews. In addition, the absence of an obvious bias induced by a high or low Jadad-score underlines the consistency of the analysis. The same holds true for the inclusion or exclusion of the data of the two observational studies where no placebo was administered but simply "no treatment" was given to the patients. Inclusion of both trials had a minor effect on the estimates (see table 2). Finally, in agreement with the expectations and as such another indicator of consistency the rate estimates increase with enlarged follow-up periods with PONV rate estimates being always at least as frequent as the estimates for vomiting.

The great majority of trials that investigated the rate of recurrence of emetic symptoms after a placebo or placebo/no treatment within the early postoperative period demonstrate a consistently high incidence of emetic sequelae. Conducting a systematic search of the literature and statistical pooling of the available data using meta-analytic techniques, more reliable estimate of these incidences of recurrent symptoms can be determined. These numbers can be used for defining a first guess clinically relevant goal for future placebo controlled trials.

For several reasons, however, it is unlikely that randomized controlled trials will be performed that help to fill the gap between the huge knowledge that has accumulated by a plethora of antiemetic trials investigating the prophylaxis and the little knowledge on treatment of established symptoms due to the paucity trials on the treatment of PONV.

We conclude that the given estimates provide a solid approximation to "real life" and can be used with some confidence as a clue for future studies for instance in order to perform power calculations. As a rule-of-thumb, for the outcome "PONV" a recurrence rate of 65% can be used during a early period 0–2 hours after administering a placebo and 80% for a longer observation interval (0–6 hours or 0–24 hours respectively). For the recurrence of vomiting the expected incidences are 65% for the early period and 70–75% for the extended interval.

Using the lower 95%-confidence intervals of these estimates ("best-case" scenario) might be used as an indicator to guide future study planning. Within this special area of research, our data might even be used to come up with an alternative to a placebo controlled study design. Thus, research on antiemetic treatment which is urgently required might be stimulated.