- Email: [email protected]
Single-patient data meta-analysis of 3453 postoperative patients: oral tramadol versus placebo, codeine and combination analgesics
Pain 69 (1997) 287–294
Single-patient data meta-analysis of 3453 postoperative patients: oral tramadol versus placebo, codeine and combination analgesics
R.A. Moore*, H.J. McQuay Pain Research and Nuffield Department of Anaesthetics, University of Oxford, Oxford Radcliffe Hospital, The Churchill, Headington, Oxford OX3 7LJ, UK
Received 5 July 1996; revised version received 23 September 1996; accepted 6 November 1996
Abstract The analgesic effectiveness and safety of oral tramadol were compared with standard analgesics using a meta-analysis of individual patient data from randomised controlled trials in patients with moderate or severe pain after surgery or dental extraction. Calculation of %maxTOTPAR from individual patient data, and the use of .50%maxTOTPAR defined clinically acceptable pain relief. Numberneeded-to-treat (NNT) for one patient to have .50%maxTOTPAR compared with placebo was used to examine the effectiveness of different single oral doses of tramadol and comparator drugs. Eighteen randomised, double-blind, parallel-group single-dose trials with 3453 patients using categorical pain relief scales allowed the calculation of %maxTOTPAR. The use of .50%maxTOTPAR was a sensitive measure to discriminate between analgesics. Tramadol and comparator drugs gave significantly more analgesia than placebo. In postsurgical pain tramadol 50, 100 and 150 mg had NNTs for .50%maxTOTPAR of 7.1 (95% confidence intervals 4.6–18), 4.8 (3.4–8.2) and 2.4 (2.0–3.1), comparable with aspirin 650 mg plus codeine 60 mg (NNT 3.6 (2.5–6.3)) and acetaminophen 650 mg plus propoxyphene 100 mg (NNT 4.0 (3.0–5.7)). With the same dose of drug postsurgical patients had more pain relief than those having dental surgery. Tramadol showed a dose-response for analgesia in both postsurgical and dental pain patients. With the same dose of drug postsurgical pain patients had fewer adverse events than those having dental surgery. Adverse events (headache, nausea, vomiting, dizziness, somnolence) with tramadol 50 mg and 100 mg had a similar incidence to comparator drugs. There was a dose response with tramadol, tending towards higher incidences at higher doses. Single-patient meta-analysis using more than half pain relief provides a sensitive description of the analgesic properties of a drug, and NNT calculations allow comparisons to be made with standard analgesics. Absolute ranking of analgesic performance should be done separately for postsurgical and dental pain. Keywords: Oral tramadol; Codeine; Combination analgesics
1. Introduction The study of analgesics still poses problems, some 40 years after Beecher first described methods of measuring pain and pain relief (Denton and Beecher, 1949; Houde et al., 1965). Problems can be of different sorts, starting with the obvious but important, like the many possible comparisons of drugs, doses, routes of administration and pain condition which make meaningful comparison difficult.
* Corresponding author. Tel.: +44 1865 226161; fax: +44 1865 225775; e-mail: [email protected]
Many controlled trials have been performed, and many published; some 10 000 randomised controlled trials (over 4000 in pharmacological interventions in acute pain) have been identified (Jadad et al., 1996a). Quantitative systematic reviews pool data from a number of trials; while individual trials may have relatively small numbers of patients receiving a particular treatment, meta-analysis allows the result to be confirmed using data from many patients in many trials, thereby increasing the power to determine the ‘true’ result. It can therefore provide a higher quality of evidence on which to base decisions by prescribers, policy-makers and patients. Choice of randomised, controlled trials for systematic
0304-3959/97/$17.00 1997 International Association for the Study of Pain. Published by Elsevier Science Ireland Ltd. PII S0304-3959 (96 )0 3291-5
288
R.A. Moore, H.J. McQuay / Pain 69 (1997) 287–294
reviews is essential. Randomisation (and concealment) of treatment allocation limits selection bias, and blinding of treatments controls observer bias. Inadequacies of randomisation or blinding exaggerate estimates of treatment effect (Schulz et al., 1995). In randomised controlled trials in pain relief, standard methods of measuring pain relief and trial conduct appear to be effectively blinded (McQuay et al., 1996b). Results of systematic reviews have to be easily understood to be useful and used. The elegant numbers-neededto-treat (NNT) approach (Cook and Sackett, 1995) involves defining a clinical end-point, and comparing the rate of that event in a treatment group with the rate in a comparator group; NNT calculations require dichotomous data. NNTs derived for particular benefits or harm can provide a useful starting point for simple verbal and numerical results accessible to any doctor or patient (Chalmers, 1995). Meta-analysis using individual patient data sometimes produces lower estimates of effect of treatment than does meta-analysis using group descriptions (Stewart and Parmar, 1993), though the generality of this has been challenged (Harvey et al., 1993). Using individual patient information may not always be possible, but where possible it is preferred (Oxman et al., 1995) because it is claimed to have the least bias of any meta-analytical method. Studies of pain relief may present an additional complication. The classic design of single dose oral medication with both placebo and active controls to demonstrate analgesic sensitivity (Houde et al., 1965) is explanatory (Schwartz and Lellouch, 1967). Such trials provide evidence that a drug is an analgesic, rather than about the best way to use the compound in practice. In this context, NNT methods are useful as indicators of relative efficacy (Moore et al., 1996). Tramadol has been used in many European countries since the late 1970s, in many different pain conditions. Since most studies with tramadol in Europe had not been conducted according to US regulatory requirements, a completely new programme of clinical studies for registration in the USA of an oral tramadol formulation took place in the late 1980s and early 1990s. Eighteen singledose studies were conducted, nine in dental pain models and nine in postsurgical pain, and the results of the studies have been summarised (Sunshine, 1994). We performed a single-patient data meta-analysis of these 18 studies, and any others (published or unpublished) which could be found and which had categorical pain relief scales, allowing the calculation of the percentage of maximum pain relief obtained by individual patients. Combining data from many studies will help shed additional light on the debate (Stubhaug et al., 1996; VielvoyeKerkmeer, 1995, 1996) about the conflicting results of tramadol in postoperative pain (Sunshine et al., 1992; Stubhaug et al., 1995).
2. Methods 2.1. Primary trials Individual patient data from eighteen primary trials was made available by Gru¨nenthal GmbH, Aachen, Germany and Robert Wood Johnson Pharmaceutical Research Institute, Spring House, PA. One of these had been published (Sunshine et al., 1992). Other studies which used single doses of oral tramadol with categorical pain relief scoring in acute painful conditions were sought by reference to the in-house data from Gru¨nenthal GmbH, from Searle (UK) Ltd. and by searching MEDLINE (1960–1995) and the Oxford Pain Relief Database (1950–1995) (Jadad et al., 1996a) using tramadol as a free text term. There was a prior hypothesis that analgesic drugs produced different analgesic responses in painful dental procedures (such as third molar extractions) than postsurgical procedures (such as abdominal, orthopaedic or gynaecological operations). The prior intention, therefore, was to analyse these conditions separately. Included reports were scored for inclusion and methodological quality using a 3-item scale (Jadad et al., 1996b). Protocols for the Robert Wood Johnson studies of postsurgical pain and of pain due to the extraction of impacted third molars were essentially identical. Trials were of double-blind, single-dose, parallel-group design; randomisation was by computerised random-number generation, stratified on pretreatment pain intensity. Criteria for patient selection were moderate or severe pain and that the patient’s condition was appropriate for management with a centrally acting analgesic and acetaminophen. The age range was 18–70 years. Patients had to be cooperative, reliable, and motivated, and be able to take oral medication. Exclusion criteria included patients with mild or no pain, those who had taken analgesic drugs within 3 h of study drug administration, those needing sedatives during the observation period and those with known contraindications or medical conditions which might interfere with observations. Drugs were given as single oral doses, of placebo (695 evaluable patients), codeine 60 mg (649), tramadol 50 mg (409), tramadol 75 mg (281), tramadol 100 mg (468), tramadol 150 mg (279), tramadol 200 mg (50), aspirin 650 mg plus codeine 60 mg (305), and acetaminophen 650 mg plus propoxyphene 100 mg (316). Patients were given the study drug if they had moderate or severe pain on a 4-point categorical scale (0 = no pain, 1 = slight, 2 = moderate, 3 = severe). Thereafter observations were made at 30 min, and 1, 2, 3, 4, 5 and 6 h after administration. Pain intensity was measured using the same categorical scale, together with a 5-point categorical scale of pain relief (0 = no relief, 1 = a little, 2 = some, 3 = a lot, 4 = complete). Time of remedication was also recorded, as well as a global assessment of therapy (excellent, very good, good, fair or poor) at the final evaluation.
R.A. Moore, H.J. McQuay / Pain 69 (1997) 287–294
Adverse experiences volunteered by the patient after non-directive questioning were recorded regardless of any rescue medication used. 2.2. Calculations For each patient the area under the curve of pain relief (categorical scale) against time was calculated (TOTPAR) for 6 h after the study drug was given. If patients needed remedication, pain relief scores reverted to zero and pain intensity scores to the initial value; adverse event recording, but not pain evaluations, continued after remedication. The percentage of the maximum possible for this summary measure was then calculated (%maxTOTPAR) (Cooper, 1991). The number of patients on each treatment who achieved more than 50% of %maxTOTPAR was determined. Relative risk (which indicates how much more likely is an individual given a particular treatment to have a specific outcome than someone not given the treatment) and its 95% confidence intervals were calculated for individual trials using a fixed-effects model (Gardner and Altman, 1989), and number-needed-to-treat (NNT) using the method of Cook and Sackett (Cook and Sackett, 1995). Relative risk and NNT are given with 95% confidence intervals in text and tables. Significance testing for doseresponse of tramadol was done using the Kruskal-Wallis test unstratified for type of surgery. Statistical analysis was done using Statview v.4.5 and Excel 5.0 on a Power Macintosh 8500/150.
289
system rather than pain relief, and therefore had to be excluded from this analysis. Another study (Collins et al., 1996) of several dose levels of oral tramadol after dental surgery was multiple dose and used only pain intensity scoring. It showed significant differences between all tramadol doses and placebo, but again could not be used. No other relevant studies were identified. 3.2. Analgesic efficacy The relative risks and NNTs for each drug tested are shown in Table 2, for dental and postoperative pain. The proportions achieving .50%maxTOTPAR are shown in Fig. 1. There was a clear dose-response for tramadol (P , 0.0001, Kruskal-Wallis test). 3.3. Dental pain Among the dental studies all treatments showed significantly greater pain relief (greater proportion of patients with .50% of %maxTOTPAR) than with placebo (relative risk lower confidence interval . 1.0) except codeine 60 mg. There was a clear dose-response for tramadol, with higher relative risk and lower NNT values with the higher doses. Tramadol doses of 100 and 150 mg produced NNT values of 4.6 (3.6–6.4) and 4.2 (2.9–7.3), respectively, lower than aspirin/codeine with an NNT of 6.3 (4.5–9.8) and acetaminophen/propoxyphene combinations with an NNT of 5.3 (3.4–11.4). 3.4. Postsurgical pain
3. Results 3.1. Search results Individual patient data for 3453 patients from 18 studies was supplied. It consisted of pain intensity and pain relief scores from start of study to 8 h post dose and aggregate adverse effect information. Studies with their codes, drug treatments and numbers of patients, are detailed in Table 1. Data on pain measurements and adverse effects for these single-dose parallel-group double-blind studies was provided. Of the nine post surgical pain studies, two (TR and TV) followed Caesarean section, and one (TX) was conducted with outpatients. Of the nine dental pain studies, three (TI, TI2 and TO) were conducted with outpatients. Tramadol 200 mg was given in only one study in dental pain, and these data were excluded. Study reports were of high methodological quality, scoring the maximum of 5 points on a validated scale (Jadad et al., 1996b). Literature searches through MEDLINE found two relevant studies of oral tramadol in postoperative pain (Sunshine et al., 1992; Stubhaug et al., 1995). The former formed part of the data set supplied. The latter, which did not show a significant difference between tramadol 50 and 100 mg and placebo, used a pain intensity scoring
All treatments showed statistically significantly superior analgesia to placebo. There was a clear dose response for tramadol; tramadol 100 mg had an NNT of 4.8 (3.4–8.2) and tramadol 150 mg had an NNT of 2.4 (2.0–3.1). This was lower than aspirin/codeine and acetaminophen/propoxyphene combinations with NNT values of 3.6 (2.5– 6.3) and 4.0 (3.0–5.7) respectively. The relative analgesic effectiveness of the different analgesics in postoperative pain compared with placebo is shown by the number-needed-to-treat to obtain one patient with .50% of %maxTOTPAR (Fig. 2). 3.5. Dental and postsurgical pain models compared With the exception of tramadol 100 mg, NNTs were lower in postsurgical pain than in dental pain. When the numbers of patients with more than 50% pain relief were compared for each treatment between postsurgical and dental pain (Fig. 1), some treatments produced significantly more pain relief in postsurgical pain. This was the case for codeine 60 mg (relative risk 2.4 (1.6–3.5)), tramadol 75 mg (2.4 (1.5–3.8)) and 150 mg (1.9 (1.4–2.6)), aspirin plus codeine (1.6 (1.04–2.3)) and acetaminophen plus propoxyphene (1.6 (1.1–2.4)).
290
R.A. Moore, H.J. McQuay / Pain 69 (1997) 287–294
This difference was not due to a greater proportion of postsurgical patients with more moderate than severe initial pain intensity. The ratio of moderate to severe initial (baseline) pain intensity was 931:663 (1.40:1) in postsurgical pain, compared with 1294:505 (2.56:1) in dental pain (significantly less severe initial pain in the dental group, relative risk 0.81 (0.77–0.85)). Initial pain intensity stratification produced no consistent or significant differences in the proportion of patients with .50%maxTOTPAR, or in NNTs (data not shown). For postsurgical pain, 44 of 323 patients (13.6%) given placebo had .50%maxTOTPAR,
compared with 28 of 373 patients (7.5%) for dental pain given placebo (relative risk 1.8 (1.2–2.8)). 3.6. Choice of half pain relief In order to test the effect of choices other than half pain relief, NNTs were calculated using dichotomous data for 20–80%maxTOTPAR (combined data). The results for tramadol 75 mg and 150 mg, codeine 60 mg and acetaminophen 650 mg plus proproxyphene 100 mg are shown in Fig. 3. NNTs for an effective drug, tramadol 150
Table 1 Trials, drug treatments and patient numbers Trial Postsurgical pain TA
TC
TR
TV
TW
TW2
TX
TY
TZA
Drug
Patients
Tramadol 50 mg Tramadol 100 mg Codeine 60 mg Placebo Tramadol 50 mg Tramadol 100 mg ASA650 and C60 Codeine 60 mg Placebo Tramadol 75 mg Tramadol 150 mg APAP650 and P100 Placebo Tramadol 50 mg Tramadol 100 mg ASA650 and C60 Codeine 60 mg Placebo Tramadol 50 mg Tramadol 100 mg APAP650 and P100 Codeine 60 mg Placebo Tramadol 75 mg Tramadol 150 mg APAP650 and P100 Codeine 60 mg Placebo Tramadol 75 mg Tramadol 150 mg APAP650 and P100 Codeine 60 mg Placebo Tramadol 75 mg Tramadol 150 mg APAP650 and P100 Codeine 60 mg Placebo Tramadol 75 mg Tramadol 150 mg APAP650 and P100 Codeine 60 mg Placebo
52 58 26 28 40 39 40 39 40 40 40 41 40 31 31 30 29 28 40 40 39 39 40 41 40 39 41 40 35 36 37 33 36 31 28 31 30 30 39 40 38 38 41
Trial Dental pain TE
TE2
TF
TG
TH
TI
TI2
TO
TQ
Drug
Patients
Tramadol 50 mg Tramadol 100 mg ASA650 and C60 Codeine 60 mg Placebo Tramadol 50 mg Tramadol 100 mg ASA650 and C60 Codeine 60 mg Placebo Tramadol 50 mg Tramadol 100 mg ASA650 and C60 Codeine 60 mg Placebo Tramadol 50 mg Tramadol 100 mg ASA650 and C60 Codeine 60 mg Placebo Tramadol 50 mg Tramadol 100 mg ASA650 and C60 Codeine 60 mg Placebo Tramadol 50 mg Tramadol 100 mg ASA650 and C60 Codeine 60 mg Placebo Tramadol 75 mg Tramadol 150 mg APAP650 and P100 Codeine 60 mg Placebo Tramadol 100 mg Tramadol 200 mg Codeine 60 mg Placebo Tramadol 75 mg Tramadol 150 mg APAP650 and P100 Codeine 60 mg Placebo
28 30 28 28 29 23 21 22 24 21 47 49 45 47 49 50 49 41 33 27 47 51 51 48 50 51 50 48 50 49 45 45 42 47 44 50 50 47 53 50 50 49 50 51
ASA650 and C60, aspirin 650 mg plus codeine 60 mg. APAP650 and P100, acetaminophen 650 mg plus propoxyphene 100 mg.
291
R.A. Moore, H.J. McQuay / Pain 69 (1997) 287–294 Table 2
Analgesic effectiveness, relative risk and numbers-needed-to-treat (NNT) with 95% confidence intervals for patients achieving at least 50% of %maxTOTPAR on drug treatment compared with placebo Improved on active Dental Codeine 60 mg Tramadol 50 mg Tramadol 75 mg Tramadol 100 mg Tramadol 150 mg Acetaminophen 650 mg and propoxyphene 100 mg Aspirin 650 mg and codeine 60 mg Postsurgical Codeine 60 mg Tramadol 50 mg Tramadol 75 mg Tramadol 100 mg Tramadol 150 mg Acetaminophen 650 mg and propoxyphene 100 mg Aspirin 650 mg and codeine 60 mg
Improved on control
Relative risk (95% CI)
NNT (95% CI)
36/374 41/246 16/95 89/300 29/95 23/91 52/235
28/373 13/225 6/95 22/278 6/95 6/95 13/225
1.3 2.9 2.7 3.8 4.8 4.0 3.8
(0.8–2.1) (1.6–5.2) (1.1–6.5) (2.4–5.8) (2.1–11.1) (1.7–9.4) (2.2–6.8)
50 (16.3–∞) 9.1 (6.1–18.8) 9.1 (5.1–64.5) 4.6 (3.6–6.4) 4.2 (2.9–7.3) 5.3 (3.4–11.4) 6.3 (4.5–9.8)
63/275 38/163 74/186 51/168 106/184 91/225 24/70
35/283 13/136 31/187 13/136 31/187 34/227 4/68
1.9 2.4 2.4 3.2 3.5 2.7 5.8
(1.3–2.7) (1.4–4.4) (1.7–3.5) (1.8–5.6) (2.5–4.9) (1.9–3.8) (2.1–15.9)
mg, were essentially the same at about 2–3 over a wide range of decision points. Those for a slightly less effective analgesic (acetaminophen 650 mg plus propoxyphene 100 mg) rose slightly with pain relief cut-off values above 50%maxTOTPAR, but for codeine 60 mg NNT values which started at about 10 for .20%max-TOTPAR rose rapidly and were not significantly different from placebo by .60%maxTOTPAR. 3.7. Adverse events The incidence of the more common adverse events reported is shown in Fig. 4 for dental and postsurgical pain. Headache, vomiting, nausea, dizziness and somnolence were the most commonly reported adverse events, though predominantly of mild intensity. For dental but not postsurgical pain the adverse event incidence was generally sufficiently high to achieve a statistical difference from placebo for vomiting, nausea, dizziness and somnolence, but not headache. For tramadol
Fig. 1. Proportion of patients with >50%maxTOTPAR.
9.1 7.1 4.4 4.8 2.4 4.0 3.6
(6–23.4) (4.6–17.9) (3.1–7.0) (3.4–8.2) (2.0–3.1) (3.0–5.7) (2.5–6.3)
there was a distinct dose-response in dental pain, with higher doses producing greater incidence of adverse events; this trend was not present in postsurgical pain. 4. Discussion Tramadol is an effective analgesic in postoperative pain. All doses of tramadol were statistically superior to placebo in both postsurgical and dental pain, and there was a significant dose-response. Single oral doses of tramadol 75– 150 mg had analgesic efficacy equivalent to combinations of acetaminophen plus propoxyphene and aspirin plus codeine. Internal sensitivity was demonstrated by two comparator analgesics being statistically superior to placebo, and by the dose response for tramadol. The study methodology (randomised, double-blind trials) avoided known sources of major bias. Search strategies identified a total of 20 randomised trials of oral tramadol in postoperative acute pain with standardised measurements of pain intensity and pain relief. Only two of these had been published in full (Sunshine et al., 1992; Stubhaug et al., 1996) with one other in press (Collins et al., 1996). Results of the other trials had
Fig. 2. NNT and 95% confidence interval for postsurgical patients.
292
R.A. Moore, H.J. McQuay / Pain 69 (1997) 287–294
Fig. 3. Effect of %maxTOTPAR on NNT.
been published in summary form only (Sunshine, 1994). Meta-analytic tools so far developed have concentrated on patients achieving at least 50% pain relief as a single dichotomous measure of clinical effectiveness (Moore et al., 1996), so the two studies which used pain intensity and not pain relief scales could not be included. Of the two studies excluded, one (Stubhaug et al., 1995) could not distinguish oral tramadol from placebo after orthopaedic surgery; the other (Collins et al., 1996) found oral tramadol effective in dentoalveolar surgical pain. Because the only two studies published in full (Sunshine et al., 1992; Stubhaug et al., 1996) came to contrary conclusions about the efficacy of oral tramadol in postoperative pain, a controversy has arisen about its analgesic properties (Stubhaug et al., 1996; Vielvoye-Kerkmeer, 1995, 1996). Examining all the available information (published and unpublished) has demonstrated clear analgesic efficacy; three standard analgesics were distinguished from placebo, as were tramadol 50, 100 and 150 mg. Larger doses of tramadol produced more analgesia. This was done in a single-patient meta-analysis, a method which is claimed to be more conservative than aggregating mean data (Stewart and Parmar, 1993). Sunshine (1994), Cooper (1991) and others have pointed out the variability that can occur in clinical trials of analgesics even in standard settings. This variability may have a number of causes. One is almost certain to
be due to the random play of chance in clinical trials where group sizes are of the order of 30 patients, though there may also be systematic causes. These causes may be apparent only in the systematic examination of large numbers of clinical trials with common end-points. The other consideration is that analgesic trial designs are explanatory (Schwartz and Lellouch, 1967). They are designed to demonstrate that a particular compound is an analgesic in single doses in acute pain. They cannot in themselves determine the value of the intervention in clinical practice, though meta-analysis of such trials may be helpful in determining relative efficacy (Moore et al., 1996). This unique opportunity to use the individual patient data from eighteen trials conducted to a common protocol allowed several questions to be addressed. The first, the original purpose of the studies, was to compare the efficacy and adverse effects of the novel analgesic with placebo and standard oral analgesics. The data also allowed the confirmation of the usefulness of at least 50% pain relief as an indicator of efficacy, comparison of the various analgesics in patients with either moderate or severe baseline pain, and comparison of dental with postsurgical pain. As a clinical outcome 50% relief of pain has historical provenance over 40 years (Denton and Beecher, 1949), and is more readily clinically interpretable than the summary TOTPAR measure. For comparisons of analgesics the question arises whether the 50% relief is a better cutoff than 20% or 80% relief. The best performing analgesic, tramadol 150 mg, had NNT values of 2–3 across a range of cut-offs, from 30 to 60%maxTOTPAR (Fig. 3). The least analgesic drug, codeine 60 mg, showed a rapid rise in NNT beyond 50%maxTOTPAR, and the acetaminophen/proproxyphene combination showed a gradual rise in NNT with higher cut-offs. This provides some empirical support for the use of 50% as the cut-off. Not only does it have a clinically useful resonance, but it also provides sensible discrimination between the best and worst analgesics. That NNT (the reciprocal of the absolute risk reduction, or risk difference) should be relatively unaffected by choice of cut-off is not unexpected. With placebo the proportion of patients achieving a particular level of pain relief falls quickly as %maxTOTPAR increases. For effective analgesics, this proportion falls slowly until high %maxTOTPAR levels are reached. The difference will remain largely unaltered over a wide range of %maxTOTPAR, generating stable NNTs. The imposition of an arbitrary dichotomous outcome, .50%maxTOTPAR, on continuous data (a spectrum of response between no pain relief and complete pain relief) is justified because it allows analgesics to be compared across many different trials. However, it should not be over-interpreted; patients with less than 50%maxTOTPAR can also obtain useful pain relief; conversely those with at least 50%maxTOTPAR may have near maximal pain relief. The reality, though, is that multiple dosing is the
R.A. Moore, H.J. McQuay / Pain 69 (1997) 287–294
293
Fig. 4. Adverse events - headache, nausea, vomiting, dizziness and somnolence.
norm in pain management, where adverse effects may drive practice as much as analgesia. It has been suggested that differences might be seen between analgesics when tested on pain of initial moderate as opposed to severe intensity. This was not supported by these data. Stratification by initial pain intensity revealed no consistent or significant differences in the proportion of patients with at least 50%maxTOTPAR (data not shown). In these trials the analgesics were more effective in postsurgical pain than in dental pain (Table 2), producing lower NNTs despite significantly more patients with severe pain intensity at baseline in postsurgical pain. In postsurgical trials, significantly more patients given placebo (14%) had at least 50% pain relief than was the case with dental models (8%). These average figures for nine
trials in each group are lower than those found by Cooper (1991). The figures found by Cooper were derived from study mean TOTPAR. McQuay et al. (1996b) have pointed out that means are inadequate descriptors of asymmetrically distributed pain measurements, making comparison between estimates derived by single-patient meta-analysis difficult. What Cooper’s data did show was the great between-trial variability of placebo and active responses. This variability is not limited to acute pain studies, and is seen also in chronic pain studies (McQuay et al., 1996c), as well as in studies with more objective outcomes like postoperative vomiting (Trame`r et al., 1995), and in the response of infants to pulmonary surfactant (Soll and McQueen, 1992).
294
R.A. Moore, H.J. McQuay / Pain 69 (1997) 287–294
Despite analysing results on nearly 3500 patients, there were only 18 trials, nine each in dental and postsurgical pain. In order to make definitive statements about differences between pain models information from many more trials would need to be available. Cooper’s 1991 analysis had information from as many as 63 studies. The differences in analgesic efficacy and adverse events seen in this study support the view that dental and postoperative pain should be considered separately in meta-analytical comparisons of analgesic efficacy, at least when opioid analgesics or combinations with opioids are used. Analyses of other analgesics in postsurgical and dental pain models are needed to allow comparisons to be made of relative effectiveness. This will not be easy, partly because few studies report data in ways which allow meta-analysis from the published reports, and partly because many patients are needed to obtain estimates with narrow confidence intervals. Single-patient meta-analysis is the most useful way of generating comparative information, though that will involve much co-operation between investigators and sponsoring pharmaceutical companies. Authors of reports of trials of analgesics can aid future meta-analysis by including dichotomous outcomes as part of their analysis and report. This can easily be done as an addition to, not to the exclusion of, classical pain measures and analysis (McQuay et al., 1996a).
Acknowledgements This work was carried out with financial support from Gru¨nenthal GmbH. We wish to thank Gru¨nenthal GmbH, the Robert Wood Johnson Pharmaceutical Research Institute, and Searle (UK) Ltd for providing unpublished information, and many colleagues around the world for their helpful discussions. References Chalmers, I., Applying overviews in meta-analysis at the bedside, J. Clin. Epidemiol., 48 (1995) 67–70. Collins, M., Young, I., Sweeney, P., Fenn, G.C., Stratford, M.E., Wilson, A. and Harris, M., The effect of tramadol on dento-alveolar surgical pain, Br. J. Oral Maxillofac. Surg. (1996) in press. Cook, R.J. and Sackett, D.L., The number needed to treat: a clinically useful measure of treatment effect, Br. Med. J., 310 (1995) 452–454. Cooper, S.A., Single-dose analgesic studies: the upside and downside of assay sensitivity. In: M.B. Max, R.K. Portenoy and E.M. Laska (Eds.), The Design of Analgesic Clinical Trials: Advances in Pain Research and Therapy, Raven Press, New York, 1991, pp. 117–124.
Denton, J.E. and Beecher, H.K., New analgesics, J. Am. Med. Assoc., 141(1949) 1051–1057. Gardner, M.J. and Altman, D.G., Statistics with confidence, Br. Med. J. (London), 1989. Harvey, I., Peters, T.J. and Toth, B., Meta-analysis, Lancet, 341(1993) 964. Houde, R.W., Wallenstein, S.L. and Beaver, W.T., Clinical measurement of pain. In: G. De Stevens (Ed.), Analgetics, Academic Press, New York and London, 1965, pp. 75–122. Jadad, A.R., Carroll, D., Moore A. and McQuay, H., Developing a database of published reports of randomised clinical trials in pain research, Pain, 66 (1996a) 239–246. Jadad, A.R., Moore, R.A., Carroll, D., Jenkinson, C., Reynolds, D.J.M., Gavaghan, D.J. et al., Assessing the quality of reports of randomized clinical trials: is blinding necessary? Controlled Clin. Trials, 17 (1996b) 1–12. McQuay, H.J., Angell, K., Carroll, D., Moore, R.A. and Juniper, R.P., Ibuprofen compared with ibuprofen plus caffeine after third molar surgery, Pain, 66 (1996a) 247–251. McQuay, H., Carroll, D. and Moore, A., Variation in the placebo effect in randomised controlled trials of analgesics: all is as blind as it seems, Pain, 64 (1996b) 331–335. McQuay, H.J., Trame`r, M., Nye, B.A., Carroll, D., Wiffen, P.J. and Moore, R.A., A systematic review of antidepressants in neuropathic pain, Pain (1996c) in press. Moore, A., McQuay, H. and Gavaghan, D., Deriving dichotomous outcome measures from continuous data in randomised controlled trials of analgesics, Pain, 66 (1996) 229–237. Oxman, A.D., Clarke, M.J. and Stewart, L.A., From science to practice. Meta-analyses using individual patient data are needed, J. Am. Med. Assoc., 274 (1995) 830–836. Schwartz, D. and Lellouch, J., Explanatory and pragmatic attitudes in therapeutic trials, J. Chron. Dis., 20 (1967) 637–648. Schulz, K.F., Chalmers, I., Hayes, R.J. and Altman, D.G., Empirical evidence of bias, J. Am. Med. Assoc., 273 (1995) 408–412. Soll, J.C. and McQueen, M.C., Respiratory distress syndrome. In: J.C. Sinclair and M.E. Bracken (Eds.), Effective Care of the Newborn Infant, Chapter 15, Oxford University Press, Oxford, 1992, p. 333. Stewart, L.A. and Parmar, M.K., Meta-analysis of the literature or of individual patient data: is there a difference?, Lancet, 341 (1993) 418– 422. Stubhaug, A., Grimstad, J. and Breivik, H., Lack of analgesic effect of 50 mg and 100 mg oral tramadol after orthopaedic surgery: a randomized, double-blind, placebo and standard active drug comparison, Pain, 62 (1995) 111–118. Stubhaug, A., Grimstad, J. and Breivik, H., Reply to Vielvoye-Kerkmeer, Pain, 64(1996) 402. Sunshine, A., New clinical experience with tramadol, Drugs, 47 (Suppl. 1 )(1994) 8–18. Sunshine, A., Olson, N.Z., Zieghelboim, I., Decastro, A. and Minn, F.L., Analgesic oral efficacy of tramadol hydrochloride in postoperative pain, Clin. Pharmacol. Ther., 51 (1992) 740–746. Trame`r, M., Moore, A. and McQuay, H., Prevention of vomiting after paediatric strabismus surgery: a systematic review using the numbersneeded-to-treat method, Br. J. Anaes., 75 (1995) 556–561. Vielvoye-Kerkmeer, A.P.E., Comments on Stubhaug, Grimstad and Breivik, Pain, 62 (1995) 111–118. Vielvoye-Kerkmeer, A.P.E., Comments on Stubhaug, Grimstad and Breivik, Pain, 64 (1996) 401–402.
Single-patient data meta-analysis of 3453 postoperative patients: oral tramadol versus placebo, codeine and combination analgesics
R.A. Moore*, H.J. McQuay Pain Research and Nuffield Department of Anaesthetics, University of Oxford, Oxford Radcliffe Hospital, The Churchill, Headington, Oxford OX3 7LJ, UK
Received 5 July 1996; revised version received 23 September 1996; accepted 6 November 1996
Abstract The analgesic effectiveness and safety of oral tramadol were compared with standard analgesics using a meta-analysis of individual patient data from randomised controlled trials in patients with moderate or severe pain after surgery or dental extraction. Calculation of %maxTOTPAR from individual patient data, and the use of .50%maxTOTPAR defined clinically acceptable pain relief. Numberneeded-to-treat (NNT) for one patient to have .50%maxTOTPAR compared with placebo was used to examine the effectiveness of different single oral doses of tramadol and comparator drugs. Eighteen randomised, double-blind, parallel-group single-dose trials with 3453 patients using categorical pain relief scales allowed the calculation of %maxTOTPAR. The use of .50%maxTOTPAR was a sensitive measure to discriminate between analgesics. Tramadol and comparator drugs gave significantly more analgesia than placebo. In postsurgical pain tramadol 50, 100 and 150 mg had NNTs for .50%maxTOTPAR of 7.1 (95% confidence intervals 4.6–18), 4.8 (3.4–8.2) and 2.4 (2.0–3.1), comparable with aspirin 650 mg plus codeine 60 mg (NNT 3.6 (2.5–6.3)) and acetaminophen 650 mg plus propoxyphene 100 mg (NNT 4.0 (3.0–5.7)). With the same dose of drug postsurgical patients had more pain relief than those having dental surgery. Tramadol showed a dose-response for analgesia in both postsurgical and dental pain patients. With the same dose of drug postsurgical pain patients had fewer adverse events than those having dental surgery. Adverse events (headache, nausea, vomiting, dizziness, somnolence) with tramadol 50 mg and 100 mg had a similar incidence to comparator drugs. There was a dose response with tramadol, tending towards higher incidences at higher doses. Single-patient meta-analysis using more than half pain relief provides a sensitive description of the analgesic properties of a drug, and NNT calculations allow comparisons to be made with standard analgesics. Absolute ranking of analgesic performance should be done separately for postsurgical and dental pain. Keywords: Oral tramadol; Codeine; Combination analgesics
1. Introduction The study of analgesics still poses problems, some 40 years after Beecher first described methods of measuring pain and pain relief (Denton and Beecher, 1949; Houde et al., 1965). Problems can be of different sorts, starting with the obvious but important, like the many possible comparisons of drugs, doses, routes of administration and pain condition which make meaningful comparison difficult.
* Corresponding author. Tel.: +44 1865 226161; fax: +44 1865 225775; e-mail: [email protected]
Many controlled trials have been performed, and many published; some 10 000 randomised controlled trials (over 4000 in pharmacological interventions in acute pain) have been identified (Jadad et al., 1996a). Quantitative systematic reviews pool data from a number of trials; while individual trials may have relatively small numbers of patients receiving a particular treatment, meta-analysis allows the result to be confirmed using data from many patients in many trials, thereby increasing the power to determine the ‘true’ result. It can therefore provide a higher quality of evidence on which to base decisions by prescribers, policy-makers and patients. Choice of randomised, controlled trials for systematic
0304-3959/97/$17.00 1997 International Association for the Study of Pain. Published by Elsevier Science Ireland Ltd. PII S0304-3959 (96 )0 3291-5
288
R.A. Moore, H.J. McQuay / Pain 69 (1997) 287–294
reviews is essential. Randomisation (and concealment) of treatment allocation limits selection bias, and blinding of treatments controls observer bias. Inadequacies of randomisation or blinding exaggerate estimates of treatment effect (Schulz et al., 1995). In randomised controlled trials in pain relief, standard methods of measuring pain relief and trial conduct appear to be effectively blinded (McQuay et al., 1996b). Results of systematic reviews have to be easily understood to be useful and used. The elegant numbers-neededto-treat (NNT) approach (Cook and Sackett, 1995) involves defining a clinical end-point, and comparing the rate of that event in a treatment group with the rate in a comparator group; NNT calculations require dichotomous data. NNTs derived for particular benefits or harm can provide a useful starting point for simple verbal and numerical results accessible to any doctor or patient (Chalmers, 1995). Meta-analysis using individual patient data sometimes produces lower estimates of effect of treatment than does meta-analysis using group descriptions (Stewart and Parmar, 1993), though the generality of this has been challenged (Harvey et al., 1993). Using individual patient information may not always be possible, but where possible it is preferred (Oxman et al., 1995) because it is claimed to have the least bias of any meta-analytical method. Studies of pain relief may present an additional complication. The classic design of single dose oral medication with both placebo and active controls to demonstrate analgesic sensitivity (Houde et al., 1965) is explanatory (Schwartz and Lellouch, 1967). Such trials provide evidence that a drug is an analgesic, rather than about the best way to use the compound in practice. In this context, NNT methods are useful as indicators of relative efficacy (Moore et al., 1996). Tramadol has been used in many European countries since the late 1970s, in many different pain conditions. Since most studies with tramadol in Europe had not been conducted according to US regulatory requirements, a completely new programme of clinical studies for registration in the USA of an oral tramadol formulation took place in the late 1980s and early 1990s. Eighteen singledose studies were conducted, nine in dental pain models and nine in postsurgical pain, and the results of the studies have been summarised (Sunshine, 1994). We performed a single-patient data meta-analysis of these 18 studies, and any others (published or unpublished) which could be found and which had categorical pain relief scales, allowing the calculation of the percentage of maximum pain relief obtained by individual patients. Combining data from many studies will help shed additional light on the debate (Stubhaug et al., 1996; VielvoyeKerkmeer, 1995, 1996) about the conflicting results of tramadol in postoperative pain (Sunshine et al., 1992; Stubhaug et al., 1995).
2. Methods 2.1. Primary trials Individual patient data from eighteen primary trials was made available by Gru¨nenthal GmbH, Aachen, Germany and Robert Wood Johnson Pharmaceutical Research Institute, Spring House, PA. One of these had been published (Sunshine et al., 1992). Other studies which used single doses of oral tramadol with categorical pain relief scoring in acute painful conditions were sought by reference to the in-house data from Gru¨nenthal GmbH, from Searle (UK) Ltd. and by searching MEDLINE (1960–1995) and the Oxford Pain Relief Database (1950–1995) (Jadad et al., 1996a) using tramadol as a free text term. There was a prior hypothesis that analgesic drugs produced different analgesic responses in painful dental procedures (such as third molar extractions) than postsurgical procedures (such as abdominal, orthopaedic or gynaecological operations). The prior intention, therefore, was to analyse these conditions separately. Included reports were scored for inclusion and methodological quality using a 3-item scale (Jadad et al., 1996b). Protocols for the Robert Wood Johnson studies of postsurgical pain and of pain due to the extraction of impacted third molars were essentially identical. Trials were of double-blind, single-dose, parallel-group design; randomisation was by computerised random-number generation, stratified on pretreatment pain intensity. Criteria for patient selection were moderate or severe pain and that the patient’s condition was appropriate for management with a centrally acting analgesic and acetaminophen. The age range was 18–70 years. Patients had to be cooperative, reliable, and motivated, and be able to take oral medication. Exclusion criteria included patients with mild or no pain, those who had taken analgesic drugs within 3 h of study drug administration, those needing sedatives during the observation period and those with known contraindications or medical conditions which might interfere with observations. Drugs were given as single oral doses, of placebo (695 evaluable patients), codeine 60 mg (649), tramadol 50 mg (409), tramadol 75 mg (281), tramadol 100 mg (468), tramadol 150 mg (279), tramadol 200 mg (50), aspirin 650 mg plus codeine 60 mg (305), and acetaminophen 650 mg plus propoxyphene 100 mg (316). Patients were given the study drug if they had moderate or severe pain on a 4-point categorical scale (0 = no pain, 1 = slight, 2 = moderate, 3 = severe). Thereafter observations were made at 30 min, and 1, 2, 3, 4, 5 and 6 h after administration. Pain intensity was measured using the same categorical scale, together with a 5-point categorical scale of pain relief (0 = no relief, 1 = a little, 2 = some, 3 = a lot, 4 = complete). Time of remedication was also recorded, as well as a global assessment of therapy (excellent, very good, good, fair or poor) at the final evaluation.
R.A. Moore, H.J. McQuay / Pain 69 (1997) 287–294
Adverse experiences volunteered by the patient after non-directive questioning were recorded regardless of any rescue medication used. 2.2. Calculations For each patient the area under the curve of pain relief (categorical scale) against time was calculated (TOTPAR) for 6 h after the study drug was given. If patients needed remedication, pain relief scores reverted to zero and pain intensity scores to the initial value; adverse event recording, but not pain evaluations, continued after remedication. The percentage of the maximum possible for this summary measure was then calculated (%maxTOTPAR) (Cooper, 1991). The number of patients on each treatment who achieved more than 50% of %maxTOTPAR was determined. Relative risk (which indicates how much more likely is an individual given a particular treatment to have a specific outcome than someone not given the treatment) and its 95% confidence intervals were calculated for individual trials using a fixed-effects model (Gardner and Altman, 1989), and number-needed-to-treat (NNT) using the method of Cook and Sackett (Cook and Sackett, 1995). Relative risk and NNT are given with 95% confidence intervals in text and tables. Significance testing for doseresponse of tramadol was done using the Kruskal-Wallis test unstratified for type of surgery. Statistical analysis was done using Statview v.4.5 and Excel 5.0 on a Power Macintosh 8500/150.
289
system rather than pain relief, and therefore had to be excluded from this analysis. Another study (Collins et al., 1996) of several dose levels of oral tramadol after dental surgery was multiple dose and used only pain intensity scoring. It showed significant differences between all tramadol doses and placebo, but again could not be used. No other relevant studies were identified. 3.2. Analgesic efficacy The relative risks and NNTs for each drug tested are shown in Table 2, for dental and postoperative pain. The proportions achieving .50%maxTOTPAR are shown in Fig. 1. There was a clear dose-response for tramadol (P , 0.0001, Kruskal-Wallis test). 3.3. Dental pain Among the dental studies all treatments showed significantly greater pain relief (greater proportion of patients with .50% of %maxTOTPAR) than with placebo (relative risk lower confidence interval . 1.0) except codeine 60 mg. There was a clear dose-response for tramadol, with higher relative risk and lower NNT values with the higher doses. Tramadol doses of 100 and 150 mg produced NNT values of 4.6 (3.6–6.4) and 4.2 (2.9–7.3), respectively, lower than aspirin/codeine with an NNT of 6.3 (4.5–9.8) and acetaminophen/propoxyphene combinations with an NNT of 5.3 (3.4–11.4). 3.4. Postsurgical pain
3. Results 3.1. Search results Individual patient data for 3453 patients from 18 studies was supplied. It consisted of pain intensity and pain relief scores from start of study to 8 h post dose and aggregate adverse effect information. Studies with their codes, drug treatments and numbers of patients, are detailed in Table 1. Data on pain measurements and adverse effects for these single-dose parallel-group double-blind studies was provided. Of the nine post surgical pain studies, two (TR and TV) followed Caesarean section, and one (TX) was conducted with outpatients. Of the nine dental pain studies, three (TI, TI2 and TO) were conducted with outpatients. Tramadol 200 mg was given in only one study in dental pain, and these data were excluded. Study reports were of high methodological quality, scoring the maximum of 5 points on a validated scale (Jadad et al., 1996b). Literature searches through MEDLINE found two relevant studies of oral tramadol in postoperative pain (Sunshine et al., 1992; Stubhaug et al., 1995). The former formed part of the data set supplied. The latter, which did not show a significant difference between tramadol 50 and 100 mg and placebo, used a pain intensity scoring
All treatments showed statistically significantly superior analgesia to placebo. There was a clear dose response for tramadol; tramadol 100 mg had an NNT of 4.8 (3.4–8.2) and tramadol 150 mg had an NNT of 2.4 (2.0–3.1). This was lower than aspirin/codeine and acetaminophen/propoxyphene combinations with NNT values of 3.6 (2.5– 6.3) and 4.0 (3.0–5.7) respectively. The relative analgesic effectiveness of the different analgesics in postoperative pain compared with placebo is shown by the number-needed-to-treat to obtain one patient with .50% of %maxTOTPAR (Fig. 2). 3.5. Dental and postsurgical pain models compared With the exception of tramadol 100 mg, NNTs were lower in postsurgical pain than in dental pain. When the numbers of patients with more than 50% pain relief were compared for each treatment between postsurgical and dental pain (Fig. 1), some treatments produced significantly more pain relief in postsurgical pain. This was the case for codeine 60 mg (relative risk 2.4 (1.6–3.5)), tramadol 75 mg (2.4 (1.5–3.8)) and 150 mg (1.9 (1.4–2.6)), aspirin plus codeine (1.6 (1.04–2.3)) and acetaminophen plus propoxyphene (1.6 (1.1–2.4)).
290
R.A. Moore, H.J. McQuay / Pain 69 (1997) 287–294
This difference was not due to a greater proportion of postsurgical patients with more moderate than severe initial pain intensity. The ratio of moderate to severe initial (baseline) pain intensity was 931:663 (1.40:1) in postsurgical pain, compared with 1294:505 (2.56:1) in dental pain (significantly less severe initial pain in the dental group, relative risk 0.81 (0.77–0.85)). Initial pain intensity stratification produced no consistent or significant differences in the proportion of patients with .50%maxTOTPAR, or in NNTs (data not shown). For postsurgical pain, 44 of 323 patients (13.6%) given placebo had .50%maxTOTPAR,
compared with 28 of 373 patients (7.5%) for dental pain given placebo (relative risk 1.8 (1.2–2.8)). 3.6. Choice of half pain relief In order to test the effect of choices other than half pain relief, NNTs were calculated using dichotomous data for 20–80%maxTOTPAR (combined data). The results for tramadol 75 mg and 150 mg, codeine 60 mg and acetaminophen 650 mg plus proproxyphene 100 mg are shown in Fig. 3. NNTs for an effective drug, tramadol 150
Table 1 Trials, drug treatments and patient numbers Trial Postsurgical pain TA
TC
TR
TV
TW
TW2
TX
TY
TZA
Drug
Patients
Tramadol 50 mg Tramadol 100 mg Codeine 60 mg Placebo Tramadol 50 mg Tramadol 100 mg ASA650 and C60 Codeine 60 mg Placebo Tramadol 75 mg Tramadol 150 mg APAP650 and P100 Placebo Tramadol 50 mg Tramadol 100 mg ASA650 and C60 Codeine 60 mg Placebo Tramadol 50 mg Tramadol 100 mg APAP650 and P100 Codeine 60 mg Placebo Tramadol 75 mg Tramadol 150 mg APAP650 and P100 Codeine 60 mg Placebo Tramadol 75 mg Tramadol 150 mg APAP650 and P100 Codeine 60 mg Placebo Tramadol 75 mg Tramadol 150 mg APAP650 and P100 Codeine 60 mg Placebo Tramadol 75 mg Tramadol 150 mg APAP650 and P100 Codeine 60 mg Placebo
52 58 26 28 40 39 40 39 40 40 40 41 40 31 31 30 29 28 40 40 39 39 40 41 40 39 41 40 35 36 37 33 36 31 28 31 30 30 39 40 38 38 41
Trial Dental pain TE
TE2
TF
TG
TH
TI
TI2
TO
TQ
Drug
Patients
Tramadol 50 mg Tramadol 100 mg ASA650 and C60 Codeine 60 mg Placebo Tramadol 50 mg Tramadol 100 mg ASA650 and C60 Codeine 60 mg Placebo Tramadol 50 mg Tramadol 100 mg ASA650 and C60 Codeine 60 mg Placebo Tramadol 50 mg Tramadol 100 mg ASA650 and C60 Codeine 60 mg Placebo Tramadol 50 mg Tramadol 100 mg ASA650 and C60 Codeine 60 mg Placebo Tramadol 50 mg Tramadol 100 mg ASA650 and C60 Codeine 60 mg Placebo Tramadol 75 mg Tramadol 150 mg APAP650 and P100 Codeine 60 mg Placebo Tramadol 100 mg Tramadol 200 mg Codeine 60 mg Placebo Tramadol 75 mg Tramadol 150 mg APAP650 and P100 Codeine 60 mg Placebo
28 30 28 28 29 23 21 22 24 21 47 49 45 47 49 50 49 41 33 27 47 51 51 48 50 51 50 48 50 49 45 45 42 47 44 50 50 47 53 50 50 49 50 51
ASA650 and C60, aspirin 650 mg plus codeine 60 mg. APAP650 and P100, acetaminophen 650 mg plus propoxyphene 100 mg.
291
R.A. Moore, H.J. McQuay / Pain 69 (1997) 287–294 Table 2
Analgesic effectiveness, relative risk and numbers-needed-to-treat (NNT) with 95% confidence intervals for patients achieving at least 50% of %maxTOTPAR on drug treatment compared with placebo Improved on active Dental Codeine 60 mg Tramadol 50 mg Tramadol 75 mg Tramadol 100 mg Tramadol 150 mg Acetaminophen 650 mg and propoxyphene 100 mg Aspirin 650 mg and codeine 60 mg Postsurgical Codeine 60 mg Tramadol 50 mg Tramadol 75 mg Tramadol 100 mg Tramadol 150 mg Acetaminophen 650 mg and propoxyphene 100 mg Aspirin 650 mg and codeine 60 mg
Improved on control
Relative risk (95% CI)
NNT (95% CI)
36/374 41/246 16/95 89/300 29/95 23/91 52/235
28/373 13/225 6/95 22/278 6/95 6/95 13/225
1.3 2.9 2.7 3.8 4.8 4.0 3.8
(0.8–2.1) (1.6–5.2) (1.1–6.5) (2.4–5.8) (2.1–11.1) (1.7–9.4) (2.2–6.8)
50 (16.3–∞) 9.1 (6.1–18.8) 9.1 (5.1–64.5) 4.6 (3.6–6.4) 4.2 (2.9–7.3) 5.3 (3.4–11.4) 6.3 (4.5–9.8)
63/275 38/163 74/186 51/168 106/184 91/225 24/70
35/283 13/136 31/187 13/136 31/187 34/227 4/68
1.9 2.4 2.4 3.2 3.5 2.7 5.8
(1.3–2.7) (1.4–4.4) (1.7–3.5) (1.8–5.6) (2.5–4.9) (1.9–3.8) (2.1–15.9)
mg, were essentially the same at about 2–3 over a wide range of decision points. Those for a slightly less effective analgesic (acetaminophen 650 mg plus propoxyphene 100 mg) rose slightly with pain relief cut-off values above 50%maxTOTPAR, but for codeine 60 mg NNT values which started at about 10 for .20%max-TOTPAR rose rapidly and were not significantly different from placebo by .60%maxTOTPAR. 3.7. Adverse events The incidence of the more common adverse events reported is shown in Fig. 4 for dental and postsurgical pain. Headache, vomiting, nausea, dizziness and somnolence were the most commonly reported adverse events, though predominantly of mild intensity. For dental but not postsurgical pain the adverse event incidence was generally sufficiently high to achieve a statistical difference from placebo for vomiting, nausea, dizziness and somnolence, but not headache. For tramadol
Fig. 1. Proportion of patients with >50%maxTOTPAR.
9.1 7.1 4.4 4.8 2.4 4.0 3.6
(6–23.4) (4.6–17.9) (3.1–7.0) (3.4–8.2) (2.0–3.1) (3.0–5.7) (2.5–6.3)
there was a distinct dose-response in dental pain, with higher doses producing greater incidence of adverse events; this trend was not present in postsurgical pain. 4. Discussion Tramadol is an effective analgesic in postoperative pain. All doses of tramadol were statistically superior to placebo in both postsurgical and dental pain, and there was a significant dose-response. Single oral doses of tramadol 75– 150 mg had analgesic efficacy equivalent to combinations of acetaminophen plus propoxyphene and aspirin plus codeine. Internal sensitivity was demonstrated by two comparator analgesics being statistically superior to placebo, and by the dose response for tramadol. The study methodology (randomised, double-blind trials) avoided known sources of major bias. Search strategies identified a total of 20 randomised trials of oral tramadol in postoperative acute pain with standardised measurements of pain intensity and pain relief. Only two of these had been published in full (Sunshine et al., 1992; Stubhaug et al., 1996) with one other in press (Collins et al., 1996). Results of the other trials had
Fig. 2. NNT and 95% confidence interval for postsurgical patients.
292
R.A. Moore, H.J. McQuay / Pain 69 (1997) 287–294
Fig. 3. Effect of %maxTOTPAR on NNT.
been published in summary form only (Sunshine, 1994). Meta-analytic tools so far developed have concentrated on patients achieving at least 50% pain relief as a single dichotomous measure of clinical effectiveness (Moore et al., 1996), so the two studies which used pain intensity and not pain relief scales could not be included. Of the two studies excluded, one (Stubhaug et al., 1995) could not distinguish oral tramadol from placebo after orthopaedic surgery; the other (Collins et al., 1996) found oral tramadol effective in dentoalveolar surgical pain. Because the only two studies published in full (Sunshine et al., 1992; Stubhaug et al., 1996) came to contrary conclusions about the efficacy of oral tramadol in postoperative pain, a controversy has arisen about its analgesic properties (Stubhaug et al., 1996; Vielvoye-Kerkmeer, 1995, 1996). Examining all the available information (published and unpublished) has demonstrated clear analgesic efficacy; three standard analgesics were distinguished from placebo, as were tramadol 50, 100 and 150 mg. Larger doses of tramadol produced more analgesia. This was done in a single-patient meta-analysis, a method which is claimed to be more conservative than aggregating mean data (Stewart and Parmar, 1993). Sunshine (1994), Cooper (1991) and others have pointed out the variability that can occur in clinical trials of analgesics even in standard settings. This variability may have a number of causes. One is almost certain to
be due to the random play of chance in clinical trials where group sizes are of the order of 30 patients, though there may also be systematic causes. These causes may be apparent only in the systematic examination of large numbers of clinical trials with common end-points. The other consideration is that analgesic trial designs are explanatory (Schwartz and Lellouch, 1967). They are designed to demonstrate that a particular compound is an analgesic in single doses in acute pain. They cannot in themselves determine the value of the intervention in clinical practice, though meta-analysis of such trials may be helpful in determining relative efficacy (Moore et al., 1996). This unique opportunity to use the individual patient data from eighteen trials conducted to a common protocol allowed several questions to be addressed. The first, the original purpose of the studies, was to compare the efficacy and adverse effects of the novel analgesic with placebo and standard oral analgesics. The data also allowed the confirmation of the usefulness of at least 50% pain relief as an indicator of efficacy, comparison of the various analgesics in patients with either moderate or severe baseline pain, and comparison of dental with postsurgical pain. As a clinical outcome 50% relief of pain has historical provenance over 40 years (Denton and Beecher, 1949), and is more readily clinically interpretable than the summary TOTPAR measure. For comparisons of analgesics the question arises whether the 50% relief is a better cutoff than 20% or 80% relief. The best performing analgesic, tramadol 150 mg, had NNT values of 2–3 across a range of cut-offs, from 30 to 60%maxTOTPAR (Fig. 3). The least analgesic drug, codeine 60 mg, showed a rapid rise in NNT beyond 50%maxTOTPAR, and the acetaminophen/proproxyphene combination showed a gradual rise in NNT with higher cut-offs. This provides some empirical support for the use of 50% as the cut-off. Not only does it have a clinically useful resonance, but it also provides sensible discrimination between the best and worst analgesics. That NNT (the reciprocal of the absolute risk reduction, or risk difference) should be relatively unaffected by choice of cut-off is not unexpected. With placebo the proportion of patients achieving a particular level of pain relief falls quickly as %maxTOTPAR increases. For effective analgesics, this proportion falls slowly until high %maxTOTPAR levels are reached. The difference will remain largely unaltered over a wide range of %maxTOTPAR, generating stable NNTs. The imposition of an arbitrary dichotomous outcome, .50%maxTOTPAR, on continuous data (a spectrum of response between no pain relief and complete pain relief) is justified because it allows analgesics to be compared across many different trials. However, it should not be over-interpreted; patients with less than 50%maxTOTPAR can also obtain useful pain relief; conversely those with at least 50%maxTOTPAR may have near maximal pain relief. The reality, though, is that multiple dosing is the
R.A. Moore, H.J. McQuay / Pain 69 (1997) 287–294
293
Fig. 4. Adverse events - headache, nausea, vomiting, dizziness and somnolence.
norm in pain management, where adverse effects may drive practice as much as analgesia. It has been suggested that differences might be seen between analgesics when tested on pain of initial moderate as opposed to severe intensity. This was not supported by these data. Stratification by initial pain intensity revealed no consistent or significant differences in the proportion of patients with at least 50%maxTOTPAR (data not shown). In these trials the analgesics were more effective in postsurgical pain than in dental pain (Table 2), producing lower NNTs despite significantly more patients with severe pain intensity at baseline in postsurgical pain. In postsurgical trials, significantly more patients given placebo (14%) had at least 50% pain relief than was the case with dental models (8%). These average figures for nine
trials in each group are lower than those found by Cooper (1991). The figures found by Cooper were derived from study mean TOTPAR. McQuay et al. (1996b) have pointed out that means are inadequate descriptors of asymmetrically distributed pain measurements, making comparison between estimates derived by single-patient meta-analysis difficult. What Cooper’s data did show was the great between-trial variability of placebo and active responses. This variability is not limited to acute pain studies, and is seen also in chronic pain studies (McQuay et al., 1996c), as well as in studies with more objective outcomes like postoperative vomiting (Trame`r et al., 1995), and in the response of infants to pulmonary surfactant (Soll and McQueen, 1992).
294
R.A. Moore, H.J. McQuay / Pain 69 (1997) 287–294
Despite analysing results on nearly 3500 patients, there were only 18 trials, nine each in dental and postsurgical pain. In order to make definitive statements about differences between pain models information from many more trials would need to be available. Cooper’s 1991 analysis had information from as many as 63 studies. The differences in analgesic efficacy and adverse events seen in this study support the view that dental and postoperative pain should be considered separately in meta-analytical comparisons of analgesic efficacy, at least when opioid analgesics or combinations with opioids are used. Analyses of other analgesics in postsurgical and dental pain models are needed to allow comparisons to be made of relative effectiveness. This will not be easy, partly because few studies report data in ways which allow meta-analysis from the published reports, and partly because many patients are needed to obtain estimates with narrow confidence intervals. Single-patient meta-analysis is the most useful way of generating comparative information, though that will involve much co-operation between investigators and sponsoring pharmaceutical companies. Authors of reports of trials of analgesics can aid future meta-analysis by including dichotomous outcomes as part of their analysis and report. This can easily be done as an addition to, not to the exclusion of, classical pain measures and analysis (McQuay et al., 1996a).
Acknowledgements This work was carried out with financial support from Gru¨nenthal GmbH. We wish to thank Gru¨nenthal GmbH, the Robert Wood Johnson Pharmaceutical Research Institute, and Searle (UK) Ltd for providing unpublished information, and many colleagues around the world for their helpful discussions. References Chalmers, I., Applying overviews in meta-analysis at the bedside, J. Clin. Epidemiol., 48 (1995) 67–70. Collins, M., Young, I., Sweeney, P., Fenn, G.C., Stratford, M.E., Wilson, A. and Harris, M., The effect of tramadol on dento-alveolar surgical pain, Br. J. Oral Maxillofac. Surg. (1996) in press. Cook, R.J. and Sackett, D.L., The number needed to treat: a clinically useful measure of treatment effect, Br. Med. J., 310 (1995) 452–454. Cooper, S.A., Single-dose analgesic studies: the upside and downside of assay sensitivity. In: M.B. Max, R.K. Portenoy and E.M. Laska (Eds.), The Design of Analgesic Clinical Trials: Advances in Pain Research and Therapy, Raven Press, New York, 1991, pp. 117–124.
Denton, J.E. and Beecher, H.K., New analgesics, J. Am. Med. Assoc., 141(1949) 1051–1057. Gardner, M.J. and Altman, D.G., Statistics with confidence, Br. Med. J. (London), 1989. Harvey, I., Peters, T.J. and Toth, B., Meta-analysis, Lancet, 341(1993) 964. Houde, R.W., Wallenstein, S.L. and Beaver, W.T., Clinical measurement of pain. In: G. De Stevens (Ed.), Analgetics, Academic Press, New York and London, 1965, pp. 75–122. Jadad, A.R., Carroll, D., Moore A. and McQuay, H., Developing a database of published reports of randomised clinical trials in pain research, Pain, 66 (1996a) 239–246. Jadad, A.R., Moore, R.A., Carroll, D., Jenkinson, C., Reynolds, D.J.M., Gavaghan, D.J. et al., Assessing the quality of reports of randomized clinical trials: is blinding necessary? Controlled Clin. Trials, 17 (1996b) 1–12. McQuay, H.J., Angell, K., Carroll, D., Moore, R.A. and Juniper, R.P., Ibuprofen compared with ibuprofen plus caffeine after third molar surgery, Pain, 66 (1996a) 247–251. McQuay, H., Carroll, D. and Moore, A., Variation in the placebo effect in randomised controlled trials of analgesics: all is as blind as it seems, Pain, 64 (1996b) 331–335. McQuay, H.J., Trame`r, M., Nye, B.A., Carroll, D., Wiffen, P.J. and Moore, R.A., A systematic review of antidepressants in neuropathic pain, Pain (1996c) in press. Moore, A., McQuay, H. and Gavaghan, D., Deriving dichotomous outcome measures from continuous data in randomised controlled trials of analgesics, Pain, 66 (1996) 229–237. Oxman, A.D., Clarke, M.J. and Stewart, L.A., From science to practice. Meta-analyses using individual patient data are needed, J. Am. Med. Assoc., 274 (1995) 830–836. Schwartz, D. and Lellouch, J., Explanatory and pragmatic attitudes in therapeutic trials, J. Chron. Dis., 20 (1967) 637–648. Schulz, K.F., Chalmers, I., Hayes, R.J. and Altman, D.G., Empirical evidence of bias, J. Am. Med. Assoc., 273 (1995) 408–412. Soll, J.C. and McQueen, M.C., Respiratory distress syndrome. In: J.C. Sinclair and M.E. Bracken (Eds.), Effective Care of the Newborn Infant, Chapter 15, Oxford University Press, Oxford, 1992, p. 333. Stewart, L.A. and Parmar, M.K., Meta-analysis of the literature or of individual patient data: is there a difference?, Lancet, 341 (1993) 418– 422. Stubhaug, A., Grimstad, J. and Breivik, H., Lack of analgesic effect of 50 mg and 100 mg oral tramadol after orthopaedic surgery: a randomized, double-blind, placebo and standard active drug comparison, Pain, 62 (1995) 111–118. Stubhaug, A., Grimstad, J. and Breivik, H., Reply to Vielvoye-Kerkmeer, Pain, 64(1996) 402. Sunshine, A., New clinical experience with tramadol, Drugs, 47 (Suppl. 1 )(1994) 8–18. Sunshine, A., Olson, N.Z., Zieghelboim, I., Decastro, A. and Minn, F.L., Analgesic oral efficacy of tramadol hydrochloride in postoperative pain, Clin. Pharmacol. Ther., 51 (1992) 740–746. Trame`r, M., Moore, A. and McQuay, H., Prevention of vomiting after paediatric strabismus surgery: a systematic review using the numbersneeded-to-treat method, Br. J. Anaes., 75 (1995) 556–561. Vielvoye-Kerkmeer, A.P.E., Comments on Stubhaug, Grimstad and Breivik, Pain, 62 (1995) 111–118. Vielvoye-Kerkmeer, A.P.E., Comments on Stubhaug, Grimstad and Breivik, Pain, 64 (1996) 401–402.