Cost-effectiveness of ondansetron for postoperative nausea and vomiting

Transcription

1 Cost-effectiveness of ondansetron for postoperative nausea and vomiting M. R. Tramèr, 1 * C. Phillips, 2 D. J. M. Reynolds, 3 H. J. McQuay 1 and R. A. Moore 1 1 Pain Research, Nuffield Department of Anaesthetics, University of Oxford, The Churchill Hospital, Oxford, UK 2 School of Health Science, University of Wales, Swansea, UK 3 Department of Clinical Pharmacology, Oxford Radcliffe Hospital, Oxford, UK Summary The decision as to whether prophylaxis against postoperative nausea and vomiting is better than treatment of established postoperative nausea and vomiting could be made on the basis of costeffectiveness. The cost-effectiveness of ondansetron was calculated using data from published quantitative systematic reviews of randomised trials. Milligrams of ondansetron required to achieve a desired endpoint were chosen as a cost unit. Modelling was based on a cohort of 1000 patients, and examined control event rates (i.e. incidence of postoperative nausea and vomiting without prophylaxis) of between 10 and 90%. In a sensitivity analysis, cost-effectiveness of recommended intravenous doses (4 mg for treatment and prophylaxis) was compared with minimal effective doses as shown by meta-analysis (1 mg for treatment, 8 mg for prophylaxis). Fewer patients experience any postoperative nausea and vomiting symptoms with prophylaxis compared with treatment. But prophylaxis is only marginally more effective than treatment, and treatment of established postoperative nausea and vomiting with effective doses (i.e. 1 or 4 mg) is more cost-effective and safer than prophylaxis with effective doses (i.e. 4 or 8 mg). Reasons for this are the selective treatment of patients who actually need treatment, the high success rate with the lowest dose tested (1 mg) in established postoperative nausea and vomiting, and the disappointing antinausea effect of prophylactic ondansetron even at an eight-fold higher dose. Keywords Complications; postoperative vomiting. Vomiting, anti-emetics; ondansetron.... Correspondence to: Dr M. R. Tramèr. *Present address: Division d Anesthésiologie, Departement APSIC, Hôpital Cantonal Universitaire de Genève, CH-1211 Genève 14, Switzerland Accepted: 30 July 1998 Healthcare providers and purchasers are increasingly confronted by pressures on resources against a background of increasing demands and expectations. This has also been recognised in anaesthesia [1]. Economic evaluations of interventions have been undertaken to ensure that the best use is made of limited resources. For example, the cost-effectiveness of ondansetron has been compared with metoclopramide in the control of cisplatin-induced sickness [2]. Similar analyses have been undertaken to assess the cost-effectiveness of different prophylactic anti-emetic interventions in the surgical setting [3, 4]. Watcha & Smith [4], for instance, performed a cost-effectiveness analysis with data extracted from published studies and a survey of current practice in two hospitals. They concluded that for ondansetron, prophylaxis of postoperative nausea and vomiting (PONV) was more costeffective than therapy only if the overall frequency of PONV was at least 33% [4]. The aim of this study was to assess the relative costeffectiveness of prophylaxis and treatment strategies for dealing with PONV [5]. The intention was not only to establish cost-effectiveness relationships but to base the investigation on the strongest evidence currently available, i.e. data from systematically searched, randomised controlled trials. We chose ondansetron as an anti-emetic, because, for this drug, data on efficacy and harm from two published quantitative systematic reviews of valid randomised controlled trials are available [6, 7] Blackwell Science Ltd

2 M. R. Tramèr et al. Cost-effectiveness of ondansetron Table 1 Total estimates of efficiency and harm Definition Value Ref. Success rate with 1 mg treatment 40% [6] Success rate with 4 mg treatment 45% [6] Success rate with 8 mg treatment 44% [6] NNT of 1 mg to prevent nausea 21 [7] NNT of 1 mg to prevent vomiting 15 [7] NNT of 4 mg to prevent nausea 16 [7] NNT of 4 mg to prevent vomiting 6.4 [7] NNT of 8 mg to prevent nausea 6.4 [7] NNT of 8 mg to prevent vomiting 5 [7] NNH for headache with any dose 36 [7] NNH for elevated liver enzymes with any dose 31 [7] Figure 1 Strategies for dealing with PONV. Methods The two strategies to deal with PONV, treatment versus prophylaxis, were displayed graphically as a decision tree (Fig. 1). T1 T3 are the outcomes of treatment and P1 P3 are the outcomes of prophylaxis. By definition T1 and P1 are the same value. Treatment strategy (Fig. 1) Some patients will have no PONV symptoms at any time (T1). A success with treatment (T2) was defined as a nauseated or vomiting patient who had no further episode of nausea or vomiting after one dose of ondansetron. These patients vomited or felt nauseated at least once before they received ondansetron. A treatment failure (T3) was a vomiting or nauseated patient who continued to vomit or to feel nauseated despite treatment with ondansetron. The anti-emetic efficacy of ondansetron 1 mg for treatment of established PONV was shown not to be significantly different from 4 mg or 8 mg [6]. One milligram can therefore be regarded as the minimal effective, or optimal, intravenous dose for treatment of established PONV. Four milligrams, however, was recommended by the manufacturer as the optimal dose. For the purpose of this study therefore both 1-mg and 4-mg doses were chosen for costeffectiveness analyses of treatment of established PONV. Data from systematic review showed that the success rate (i.e. the percentage of nauseated or vomiting patients who would not experience any more nausea or vomiting after administration of ondansetron) was 40% for the 1 mg dose and 45% for the 4 mg dose (Table 1) [6]. Two assumptions were made. First, the success rate with either treatment dose was independent of both clinical setting and patient factors (i.e. independent of type of surgery, anaesthetic technique, gender and age). Secondly, the success rate with placebo would not influence ondansetron s efficacy, and Success rate, Percentage of vomiting or nauseated patients who are treated with the respective dose of ondansetron and who do not continue to vomit or to be nauseated. The success rate with placebo was 20%. NNT/H, Number-needed-to-treat/harm for one outcome in one patient (i.e. prevention of a PONV symptom, or presence of an adverse drug reaction) who would not have had this outcome had they received placebo. therefore there was no need to take into account the placebo response of about 20% [6]. Prophylaxis strategy (Fig. 1) Some patients who receive prophylactic ondansetron would not have vomited anyway (P1). A success with prophylaxis of PONV is a patient who never experienced any nausea or vomiting because he or she had received prophylactic ondansetron (P2). A failure in the prophylaxis arm was a patient who received prophylactic ondansetron but nevertheless vomited or felt nauseated after surgery (P3). For prophylaxis, the appropriate estimate of efficacy is the number-needed-to-treat [8]. This number indicates how many patients have to be treated prophylactically with ondansetron in order to prevent PONV in one of them who would have vomited or been nauseated had they received placebo. Therefore, in this instance, the placebo response is taken into account. Four milligrams is the optimal prophylactic dose recommended by the manufacturer. Meta-analysis, however, showed 8 mg to be the optimal prophylactic dose [7]. Therefore, cost-effectiveness analyses were performed with both the 4 mg and 8 mg dose. Also, meta-analysis has shown that ondansetron s antivomiting effect was consistently more pronounced than its antinausea effect [7]. However, because prophylaxis of nausea is probably as important as prophylaxis of vomiting, we chose the number-neededto-treat to prevent nausea as the appropriate estimate of efficacy for prophylactic doses (Table 1): number-neededto-treat ¼ 16 for 4 mg, and number-needed-to-treat ¼ 6.4 for 8 mg [7]. It was assumed that ondansetron s 1999 Blackwell Science Ltd 227

3 M. R. Tramèr et al. Cost-effectiveness of ondansetron Anaesthesia, 1999, 54, pages prophylactic anti-emetic efficacy would be equal in a patient at high- or low risk of PONV. This assumption is conservative, and might bias the data in favour of prophylaxis (because prophylaxis is likely to be less efficacious in low-risk patients compared with high-risk patients). In addition, the control event rate (i.e. the incidence of PONV without anti-emetic prophylaxis) was taken as an indicator of underlying risk. Estimate of harm The number-needed-to-harm was regarded as the appropriate estimate of the likelihood for drug-related adverse effects (Table 1). The number-needed-to-harm was assumed to be 30 for both prophylaxis and treatment with ondansetron [7]. The adverse drug reaction could be a headache or elevated liver enzymes. This means that one in 30 patients receiving any dose of ondansetron for prophylaxis or treatment of PONV will have such an adverse drug reaction who would not have had it had they received a placebo. This estimate of harm was derived from pooled data with different doses of ondansetron and therefore is likely to overestimate the risk with lower doses (i.e. with 1 mg treatment). Endpoints Two endpoints were considered to be of particular importance: the number of patients who suffered no more than a single episode of nausea or vomiting, and the number of patients who never suffered any nausea or vomiting. In the treatment arm, the number of patients who experienced at most one episode of PONV were patients who did not vomit or feel nauseated at all (i.e. who did not need any treatment) or, when having symptoms, responded to treatment (T1 þ T2). Thus, this more pragmatic approach would accept that a patient vomits once or feels nauseated briefly before an effective treatment is administered. The second, stricter endpoint (i.e. PONVfree patients) was the number of patients in the treatment arm who did not need any treatment (T1). In the prophylaxis arm both endpoints were the number of patients who never experienced any symptoms of PONV, either because they would have had none anyway, or because prophylaxis was successful (P1 þ P2). Modelling The total cost of arriving at each of the endpoints (P1 P3; T1 T3) was calculated for both strategies, treatment and prophylaxis, based on a cohort of 1000 patients. This was done with increasing hypothetical risks of PONV (i.e. control event rates ranging between 10 and 90%), and different doses of ondansetron. One analysis was of 4 mg treatment versus 4 mg prophylaxis (i.e. the doses recommended by the manufacturer), and one was of 1 mg treatment versus 8 mg prophylaxis (i.e. the optimal doses identified by meta-analysis). Relevant cost-effectiveness ratios were: 1 cost per patient independent of outcome (T1 þ T2 þ T3 versus P1 þ P2 þ P3); 2 cost per patient who experiences at most one episode of PONV (T1 þ T2 versus P1 þ P2). As mentioned above, this comparison is between unequal outcomes since those who receive treatment (T2) will by definition have suffered at least one episode of PONV. It is, we believe, ethically acceptable and clinically most relevant; 3 cost per patient who never suffered any PONV symptoms at any time (T1 versus P1 þ P2). Such ratios represent the average cost-effectiveness ratios, which are simply the costs of generating the desired endpoint (a PONV-free patient, for instance) divided by the number of patients involved. Incremental costeffectiveness ratios were calculated to indicate how much it costs to produce an additional unit of effect [9]. Starting from the least effective strategy, i.e. the strategy which generates the lowest number of successes, the difference between the costs of two strategies was divided by the difference in their success rates. Drug-acquisition costs differ widely between countries and even within countries. For instance, hospitals may have different purchasing policies. Therefore, the main cost parameter was the number of mg of ondansetron required rather than the actual price in dollars or pounds. Thus, mg outcomes as reported in this paper may be multiplied by the local price per mg, to generate the actual drug costs of each strategy. For the same reason we did not consider expenses involved in administration of the drug, cleaning, extra staff time and materials used, costs for rescue anti-emetic medication and costs of unscheduled admission due to prolonged PONV. The total of these supplementary costs may be regarded as a hospital-specific constant. This constant may then be added to the reported costs. Costs which are relevant include staff time for administering a treatment to a patient with PONV, and materials and second-line drug costs. The largest costs would come from readmission or delayed discharge. Subgroup analyses Subgroup analyses were performed to test the impact of different risks of PONVand of different doses of ondansetron on cost-effectiveness. It was assumed that the control event rate (i.e. what happens without anti-emetic prophylaxis) would accurately reflect the true underlying risk in a study population. Therefore, two arbitrarily defined clinical settings were compared: a low-risk setting (control event rate 30%) and a high-risk setting (control event rate 60%) Blackwell Science Ltd

4 M. R. Tramèr et al. Cost-effectiveness of ondansetron Figure 2 PONV strategy: treatment versus prophylaxis with ondansetron. (A) Number of patients who are PONV-free at any time. (B) Milligrams required per patient. (C) Milligrams required per patient who experiences at most 1 PONV episode. (D) Milligrams required per patient who is PONV-free at any time. (A) Treatment 1 mg. (B) Treatment 4 mg. (W) Prophylaxis 4 mg. (X) Prophylaxis 8 mg. The success rate (i.e. absence of any PONV symptoms after application of ondansetron in a patient with established PONV) is 40% with 1 mg, and 45% with 4 mg. The number-needed-to-treat to prevent nausea with 4 mg is 16 (numberneeded-to-treat to prevent vomiting/retching ¼ 6.4). The number-needed-to-treat to prevent nausea with 8 mg is 6.4 (number-needed-totreat to prevent vomiting/retching ¼ 5.0). Control event rate ¼ incidence of PONV in controls (i.e. without anti-emetic prophylaxis). y-axes in Fig. 2(C, D) are shown with a log scale for graphical purposes. Results Modelling: graphical display Number of patients who are PONV-free at any time (Fig. 2A) For all control event rates between 10 and 90%, prophylaxis yielded more PONV-free patients than treatment. This relationship was linear because of the assumption that ondansetron s anti-emetic efficacy was independent of the control event rate. With a control event rate of 30% ( lowrisk setting), 700 of 1000 patients will be completely PONV-free with both treatment doses (1 mg and 4 mg) compared with 763 patients (þ 9%) with 4 mg prophylaxis and 856 patients (þ 22%) with 8 mg prophylaxis. With a control event rate of 60% ( high-risk setting), 400 of 1000 patients will be completely PONV-free with both treatment doses compared with 463 patients (þ 16%) with 4 mg prophylaxis and 556 patients (þ 39%) with 8 mg prophylaxis. Cost per patient (Fig. 2B) The cost per patient (expressed as mg of ondansetron required per patient) was stable across control event rates with prophylaxis. With treatment, the cost per patient was directly related to control event rate. The highest cost per patient was with the most effective prophylactic dose, 8 mg. Cost per patient who experiences at most one episode of PONV (Fig. 2C) Cost per patient who experiences at most one episode of PONV (i.e. mg required per patient who suffers no more than one episode of PONV) increased with increasing control event rates with all strategies but were lowest with 1 mg and 4 mg treatment, respectively. For this endpoint, treatment with 1 mg and 4 mg remained more costeffective than prophylaxis with 4 mg and 8 mg for all control event rates tested. Cost per patient who is PONV-free at any time (Fig. 2D) Cost per PONV-free patient (i.e. mg required per patient who is PONV-free at any time) was lowest with 1 mg treatment with any control event rate. Treatment with 4 mg also showed lower cost per PONV-free patient than both prophylaxis doses but only below a control event rate of 80%. Subgroup analyses Treatment with 4 mg versus prophylaxis with 4 mg (Table 2a and b) In both the low- (control event rate 30%) and the highrisk setting (control event rate 60%), 63 extra PONV-free patients (9% of all patients) were gained with 4 mg prophylaxis (P1 þ P2) compared with 4 mg treatment (T1). However, in the low-risk setting 3.3 times as many mg (i.e mg versus 1200 mg) were required with prophylaxis than with treatment to achieve this. In the high-risk setting it was 1.7 times more (i.e mg versus 2400 mg). Treatment with 1 mg versus prophylaxis with 8 mg (Table 3a and b) When 1 mg for treatment and 8 mg for prophylaxis were 1999 Blackwell Science Ltd 229

5 M. R. Tramèr et al. Cost-effectiveness of ondansetron Anaesthesia, 1999, 54, pages Decision tree Number of Total mg Patients (Fig. 1) patients spent Table 2 Treatment with 4 mg versus prophylaxis with 4 mg (a) Control event rate 30% ( low-risk setting) Treatment of established PONV with 4 mg No PONV (no need for treatment) T Successful treatment (45%) T Failure with treatment T All patients with 1 episode of PONV T1 þ T All patients T1 þ T2 þ T Prophylaxis of PONV with 4 mg No PONV anyway P No PONV because of prophylaxis (NNT 16) P PONV despite prophylaxis P All PONV-free patients P1 þ P All patients P1 þ P2 þ P Difference in mg spent (factor x) 3.3 x (b) Control event rate 60% ( high-risk setting) Treatment of established PONV with 4 mg No PONV (no need for treatment) T Successful treatment (45%) T Failure with treatment T All patients with 1 episode of PONV T1 þ T All patients T1 þ T2 þ T Prophylaxis of PONV with 4 mg No PONV anyway P No PONV because of prophylaxis (NNT 16) P PONV despite prophylaxis P All PONV-free patients P1 þ P All patients P1 þ P2 þ P Difference in mg spent (factor x) 1.7 x NNT, Number-needed-to-treat to prevent PONV. used, 156 extra PONV-free patients (16% of all patients) were gained with prophylaxis (P1 þ P2) compared with treatment (T1) in both low- and high-risk settings. However, in the low-risk setting 27 times as many mg were required with prophylaxis than with treatment to achieve this (i.e mg versus 300 mg). In the high-risk setting it was 13 times more (i.e mg versus 600 mg). Incremental cost-effectiveness analysis Endpoint: no more than one episode of PONV (Table 4a) With a low control event rate (i.e. 30%), prophylaxis with 8 mg yielded the highest number of patients who experienced no more than one episode of PONV while prophylaxis with 4 mg yielded the lowest. Ondansetron 1 mg as treatment of established PONV was the cheapest strategy (0.4 mg were spent per patient who experienced no more than one episode of PONV). Changing from the least successful strategy, prophylaxis with 4 mg, to treatment with 1 mg resulted in a higher success rate and, because many fewer mg were required, resulted in a sparing effect: almost 65 mg were saved for each additional success. Only a slight improvement would be achieved by switching from 1 mg to 4 mg treatment; an additional 15 patients with no more than one episode of PONV would be spared. However, 60 mg would be required to generate one additional patient who had no more than one episode of PONV. Finally, another 324 mg would be required to gain one additional patient who profited when changing the strategy from treatment with 4 mg to prophylaxis with 8 mg (Table 4a). With a high control event rate (i.e. 60%) both treatment doses yielded more patients who had no more than one episode of PONV than either prophylaxis dose (Table 4a). Again, treatment of established PONV with 1 mg ondansetron was the cheapest strategy (0.9 mg were spent per patient who experienced no more than one episode of Blackwell Science Ltd

6 M. R. Tramèr et al. Cost-effectiveness of ondansetron Table 3 Treatment with 1 mg versus prophylaxis with 8 mg Decision tree Number of Total mg Patients (Fig. 1) patients spent (a) Control event rate 30% ( low-risk setting) Treatment of established PONV with 1 mg No PONV (no need for treatment) T Successful treatment (40%) T Failure with treatment T All patients with 1 episode of PONV T1 þ T All patients T1 þ T2 þ T Prophylaxis of PONV with 8 mg No PONV anyway P No PONV because of prophylaxis (NNT 6.4) P PONV despite prophylaxis P All PONV-free patients P1 þ P All patients P1 þ P2 þ P Difference in mg spent (factor x) 27 x (b) Control event rate 60% ( high-risk setting) Treatment of established PONV with 1 mg No PONV (no need for treatment) T Successful treatment (40%) T Failure with treatment T All patients with 1 episode of PONV T1 þ T All patients T1 þ T2 þ T Prophylaxis of PONV with 8 mg No PONV anyway P No PONV because of prophylaxis (NNT 6.4) P PONV despite prophylaxis P All PONV-free patients P1 þ P All patients P1 þ P2 þ P Difference in mg spent (factor x) 13 x NNT, Number-needed-to-treat to prevent PONV. PONV). Changing from 4 mg prophylaxis to 8 mg prophylaxis would benefit an additional 93 patients, but this would require 43 mg for each additional patient. Switching to 1 mg treatment increases the number of patients who have no more than one episode of PONV by 84 and, at the same time, would save 88 mg for each additional case (i.e. sparing effect). Another 60 mg would be required per additional successful case when increasing the dose to 4 mg treatment; the gain would be a further 30 patients who experienced no more than one episode of PONV. Endpoint: patients who are PONV-free (Table 4b) With both low- and high control event rates, treatment with 1 mg or 4 mg yielded fewer patients who were PONV-free at any time compared with prophylaxis with 4 mg or 8 mg. Treatment with 1 mg was the cheapest strategy at 0.4 mg per patient who was PONV-free when the control event rate was 30%, and at 1.5 mg per patient when the control event rate was 60%. Switching from 1 mg or 4 mg treatment to prophylaxis with 4 mg, the next more successful strategy, would require 44 mg for each additional PONV-free patient when the control event rate was 30%, and 25 mg for each additional PONV-free patient when the control event rate was 60%. A further change to 8 mg prophylaxis in order to achieve the highest possible number of absolutely PONV-free patients would require an additional 43 mg for each patient who benefited, independent of the control event rate. Drug-related adverse effects With the prophylaxis strategy all patients received the drug. Thus, with 4 mg or 8 mg prophylaxis 33 patients out of 1000 would have an adverse drug reaction, independent of the risk of PONV. When using ondansetron as treatment 1999 Blackwell Science Ltd 231

7 M. R. Tramèr et al. Cost-effectiveness of ondansetron Anaesthesia, 1999, 54, pages Table 4a Incremental costs (i.e. mg of ondansetron) to generate one additional patient who experiences no more than one episode of PONV mg required Total Number of Additional to generate mg spent patients patients Total Additional one additional per patient with spared with mg mg patient with with 1 Strategy 1 PONV 1 PONV spent required 1 PONV Comment PONV Control event rate 30% ( low-risk setting) Prophylaxis 4 mg 763 n/a 4000 n/a Treatment 1 mg ¹3700 ¹65 Sparing effect 0.4 Treatment 4 mg Prophylaxis 8 mg Control event rate 60% ( high-risk setting) Prophylaxis 4 mg 463 n/a 4000 n/a Prophylaxis 8 mg Treatment 1 mg ¹7400 ¹88 Sparing effect 0.9 Treatment 4 mg Table 4b Incremental costs (i.e. mg of ondansetron) to generate one additional patient who is completely PONV free mg required Total mg Number of Additional Total Additional to generate spent per PONV-free PONV-free mg mg one additional PONV-free Strategy patients patients required spent PONV-free patient Comment patient Control event rate 30% ( low-risk setting) Treatment 1 mg 700 n/a 300 n/a Treatment 4 mg > 900 No benefit 1.7 Prophylaxis 4 mg Prophylaxis 8 mg Control event rate 60% ( high-risk setting) Treatment 1 mg 400 n/a 600 n/a Treatment 4 mg > 1800 No benefit 6.0 Prophylaxis 4 mg Prophylaxis 8 mg (1 mg or 4 mg) 10 patients out of 1000 experienced an adverse drug reaction when the control event rate was 30%, and 20 had such a reaction when the control event rate was 60%. Discussion These results show that treatment of established PONV with ondansetron 1 mg is the most cost-effective and safest option, irrespective of the proportion of patients who are likely to experience PONV. A systematic review of randomised trials suggests that ondansetron 1 mg is the optimal (i.e. the minimal effective) dose for treatment of established PONV; increasing the dose to 4 or even 8 mg is not followed by a significant improvement [6]. Likewise, intravenous ondansetron 8 mg is likely to be the optimal prophylactic dose; 4 mg is more than 20% less efficacious, and further increase to 16 mg is not followed by any improvement [7]. The manufacturer s recommended dose both for treatment and prophylaxis of PONV is 4 mg. Using only the point estimates of effectiveness from systematic reviews, there was little difference in the number of patients who were either PONV-free or who experienced no more than one episode of PONV in any of these strategies. Because of this, the incremental costeffectiveness ratios were highly unfavourable to prophylaxis because prophylaxis will require substantially greater prescribing of anti-emetic, but with few additional patients having a beneficial outcome. Even this may overstate the case, because point estimates have confidence intervals around them. It might be considered legitimate to say that the number of additional patients apparently obtaining a benefit with prophylaxis was within the uncertainties of our estimates, so that no additional patients would be likely to benefit. If this were the case, prophylaxis exposes Blackwell Science Ltd

8 M. R. Tramèr et al. Cost-effectiveness of ondansetron patients to much higher doses of drug for no benefit, but with higher cost and increased likelihood of harm. Our analysis is open to criticism mainly for three reasons. First, to limit such evaluations to the cost of an anti-emetic without considering associated costs related to the outcome and costs related to the management of each episode of PONV may be regarded as an error [4]. Indeed, contrary to others [2, 4] we have chosen a more pragmatic approach to cost-effectiveness analysis, using the actual mg of the drug required to achieve a desired outcome as the cost unit. The real cost of these mg will be dependent on local hospital policy. These drug acquisition costs can be added to a hospital-specific constant which, again depending on local hospital policy, may include costs for cleaning, extra nursing time, drug application, prolonged stay in the postanaesthesia care unit, treatment of drug-related adverse effects and unanticipated overnight admission due to PONV. We believe that our model is more easily applicable to different hospitals within one country, and certainly to different countries. In money terms how much more expensive prophylaxis will be, will depend mainly on the unit cost of drug bought by hospital pharmacies. Hospitals currently using prophylactic ondansetron could realise significant savings if they made a change to only using ondansetron as treatment of established PONV. The second criticism relates to the commercially available drug vials. It may be argued that the 1 mg dose offers only a theoretical advantage in cost-effectiveness because the usual ondansetron vial contains either 4 mg or 8 mg and hospital policy may forbid sharing ampoules between patients. However, if broken but unfinished vials have to be wasted, hospital pharmacies may decide to purchase the 2 mg vial (for half the cost of a 4 mg vial), as available in the UK [10]. Another option would be for the manufacturer to produce 1 mg ampoules. Finally, we have chosen two different endpoints in this cost-effectiveness analysis. With the first endpoint, PONVfree patient at any time, treatment of established PONV with ondansetron 1 mg is more cost-effective than prophylaxis of PONV with any dose of ondansetron (Fig. 2D). For the 4 mg treatment dose this is true only for control event rates below 80%. Increased cost-effectiveness with treatment compared with prophylaxis became apparent when using the second endpoint, patients who vomit or feel nauseated at most once. With this endpoint, however, we are comparing treatment (where we accept that patients vomit or suffer nausea at least once) with prophylaxis where success is defined by no vomiting or nausea at all. It may be argued that this comparison introduces a bias in favour of treatment. However, among prophylaxis failures there will be quite a large proportion who only vomit or feel nauseated once. Indeed, the success rate in vomiting or nauseated patients treated with placebo was shown to be about 20% [6], suggesting that these patients would not have vomited more than once. In other words, it may be assumed that from most patients perspective a failure of prophylaxis may be the same as a treatment success. What might change these views? Unplanned overnight stays in day-case surgery because of PONV are clearly expensive, but there is no reason to expect that this would be rarer with prophylactic ondansetron than with treatment. There is evidence from systematically searched randomised comparisons between ondansetron and other anti-emetics that admission to hospital because of excessive or prolonged PONV happens in 0.2%, and independent of the prophylactic anti-emetic intervention [11]. Adverse events consequent on the much greater use of ondansetron might incur greater costs ¹ either through treatment of headache and unplanned overnight stay because of headache, or through the unnecessary investigation of raised liver enzymes. Again, there is no evidence that these are major concerns. Can these results be extrapolated to other anti-emetics? No direct comparisons have been found to indicate that any of the commonly prescribed anti-emetics are significantly better or worse than ondansetron [11]. Most other anti-emetics have a much lower cost per dose than ondansetron, and so monetary differences between prophylaxis and treatment may not be great. But if other anti-emetics do not differ in efficacy, but are cheaper, then that does point to other possible savings, albeit recognising the need also to examine their safety [4]. What is clear is that the great variability in results shown in individual trials, both in respect to PONV rates without prophylaxis (i.e. the control event rates), and in the apparent effect of treatment [6, 7], makes useful economic evaluation impossible without relying on evidence which has been gathered systematically. Acknowledgments Dr Tramèr held a UK Overseas Research Student Award and is supported by a grant from the Swiss National Research Foundation (grant no ). Supported by Pain Relief Funds, Pain Relief Unit, Oxford. References 1 Shapiro BA. Why must the practice of anesthesiology change? Its economics, doctor! Anesthesiology 1997; 86; Zbrozek AS, Cantor SB, Cardenas MP, Hill Jr DP. Pharmacoeconolnic analysis of ondansetron versus metoclopramide for cisplatin-induced nausea and vomiting. American Journal of Hospital Pharmacy 1994; 51: Cieslak CD, Watcha MF, Phinips MB, Pennant JH. The dose response relation and cost-effectiveness of granisetron 1999 Blackwell Science Ltd 233

9 M. R. Tramèr et al. Cost-effectiveness of ondansetron Anaesthesia, 1999, 54, pages for the prophylaxis of pediatric postoperative emesis. Anesthesiology 1996; 85: Watcha M, Smith I. Cost-effectiveness analysis of antiemetic therapy for ambulatory surgery. Journal of Clinical Anesthesia 1994; 6: Fisher DM. The big little problem of postoperative nausea and vomiting: do we know the answer yet? Anesthesiology 1997; 87: Tramèr MR, Moore RA, Reynolds DJM, McQuay HJ. A quantitative systematic review of ondansetron in treatment of established postoperative nausea and vomiting. British Medical Journal 1997; 314: Tramèr MR, Reynolds DJM, Moore RA, McQuay HJ. Efficacy, dose response, and safety of ondansetron in prevention of postoperative nausea and vomiting: a quantitative systematic review of randomised placebocontrolled trials. Anesthesiology 1997; 87: Laupacis A, Sackett DL, Roberts RS. An assessment of clinically useful measures of the consequences of treatment. New England Journal of Medicine 1988; 318: Karlsson G, Johanneson M. The decision rules of cost-effectiveness analysis. Pharmaco Economics 1996; 9: Drugs used in nausea and vertigo. British National Formulary 1996; 32: Tramèr MR, Reynolds DJM, Moore RA, McQuay HJ. Why placebo controlled trials are essential and equivalence trials are inadequate. British Medical Journal 1998; 317: Blackwell Science Ltd