A comparison between enriched and nonenriched enrollment randomized withdrawal trials of opioids for chronic noncancer pain

1Institute for Work & Health; 2Toronto Rehabilitation Institute; 3Division of Physiatry, Department of Medicine, University of Toronto; 4Comprehensive Pain Program, Toronto Western Hospital; 5Krembil Neuroscience Centre, Toronto Western Hospital, Toronto, Ontario Correspondence: Dr Andrea D Furlan, Institute for Work & Health, 481 University Avenue, Suite 800, Toronto, Ontario M5G 2E9. Telephone 416-927-2027 ext 2171, fax 416-927-4167, e-mail afurlan@iwh.on.ca An enriched enrollment randomized withdrawal (EERW) study is a type of randomized, controlled trial in which potential participants are assigned to receive the study drug for a period of time in an open-label prerandomization phase. If they benefit from the drug and can tolerate the side effects, they are randomly assigned to continue in the opioid group, or to receive the alternative drug (typically a placebo) (1,2). This study design has been advocated as useful for studying drugs that provide benefit to only a minority of the people who take it (2). In the case of opioids, it is postulated that only a minority of people can tolerate the adverse effects, and some will not experience analgesic effects because they lack the relevant metabolizing enzymes. However, this type of design may be biased by unblinding of the participants or by provoking withdrawal symptoms that may confound the pain assessments. Moreover, the results of EERW-designed trials do not extrapolate well to more general populations (3). In 2006, we published a systematic review of opioids for chronic noncancer pain (CNCP) (4). At that time, there were not many EERW trials to make a comparison possible. The primary objective of the present updated review is to compare the results between EERW and non-EERW trials of opioids for CNCP among the randomized trials included in the Canadian guideline for the safe and effective use of opioids for CNCP (5). Another objective is to assess the efficacy of weak versus strong opioids, in subgroups of various types of pain and type of comparison group used in the trials (placebo or other drugs). review


MEthoDs
Our group published a systematic review with a meta-analysis of randomized trials of opioids for CNCP in 2006 (4).We updated the searches in MEDLINE (1960 to July 2009), EMBASE (1988 to July 2009) and CENTRAL using the OVID interface.We also reviewed the reference lists in the retrieved articles, reviews and textbooks.Search strategies used for MEDLINE and EMBASE were the same as those used in the original meta-analysis (Appendix 1 and 2, respectively) One reviewer (EI) conducted the electronic searches and entered the data into Reference Manager 11 (Thomson Reuters, USA), removing all duplicates.
Two independent reviewers (AF and LC) screened all titles and abstracts for potential studies meeting the following inclusion criteria: • Study characteristics: Randomized controlled trials in humans published in languages that could be read by members of our team, ie, English, French, Portuguese or Spanish.Studies published only as abstracts were excluded.• Population: CNCP defined as pain lasting more than six months including neuropathic pain conditions, nociceptive pain (osteoarthritis, rheumatoid arthritis, back and musculoskeletal pain) and fibromyalgia (considered to be a functional pain syndrome).We excluded migraines, dental pain, ischemic pain due to vascular disease and abdominal pain (ie, chronic pancreatitis, kidney stones, etc) because these entities are usually not classified as CNCP.• Interventions: Any trial in which the opioid was given via the oral, transdermal, transmucosal or rectal route for at least seven days.We excluded head-to-head comparisons between opioids.
In the present review, we classified the potency of opioids as weak (propoxyphene, codeine, tramadol and hydrocodone) or strong (oxycodone, morphine, oxymorphone, fentanyl and buprenorphine) (6) Hard copies of potential studies were retrieved and assessed for inclusion and risk of bias.Two independent reviewers (AF and LC) met to reach consensus on the included studies and risk of bias.When in doubt, a third reviewer was consulted (AMG).When we needed more information that was not reported in a particular trial, we contacted the corresponding author of that study.
The risk of bias in each trial was assessed using the six domains recommended by the Cochrane Collaboration (Appendix 3): sequence generation, allocation concealment, blinding, incomplete outcome data, selective outcome reporting and other sources of biases (7).Under 'other sources of bias', we included information about whether there was a relationship with the pharmaceutical industry.We extracted the following information about the relationship with the pharmaceutical industry: author affiliation with industry, funding of study by industry, industry providing the study drug or statistical analysis performed by an industry-affiliated statistician.In case of affirmative response to any of these questions, we concluded that there was a relationship with the pharmaceutical industry and a potential to introduce bias.
Meta-analyses and meta-regressions were conducted using Comprehensive Meta Analysis (Biostat, USA) software, with the standardized mean difference being the 'effect size' (ES) for pain and functional outcomes.The ES were classified into small (≤0.5), medium (0.5 to <0.8) and large (≥0.8) as suggested by Cohen (8).For side effects, all meta-analyses were conducted using Revman 5 (Cochrane IMS, USA) using risk differences.Statistical heterogeneity was tested by Q test (c 2 ) reported as I 2 (higher values indicate higher heterogeneity).A clinically relevant difference of side effects was defined when the incidence was at least 10% higher in the opioid group compared with the control group (4).All meta-analyses were conducted using a random-effects model, and meta-regression was conducted using a fixed-effects model.
Subgroups were decided a priori to assess the variations in ES: weak versus strong opioids group; etiology of pain group: nociceptive, neuropathic, fibromyalgia and mixed; and comparison group: placebo and other drugs.In the present review, weak opioids were propoxyphene, codeine, tramadol and hydrocodone, and strong opioids were oxycodone, morphine, oxymorphone, fentanyl and buprenorphine (6).

REsuLts
There were 41 randomized trials included in the 2006 meta-analysis (4).The updated literature searches added 21 new trials.We found one case of triplicate publication about the same trial (9)(10)(11).A total of 62 randomized trials were included in this update (9,.Table 1 shows the characteristics of all included trials.Some trials are shown twice because they have multiple arms and comparisons.There were nine different opioids prescribed in these 62 trials: tramadol, codeine, propoxyphene, morphine, oxycodone, oxymorphone, methadone, transdermal buprenorphine and transdermal fentanyl. All included trials were described as randomized; however, only 32 (51.6%) were judged to be adequate.All but two trials (28,33) were described as double-blinded, but only 40 trials were judged as having adequate methods of double-blinding; for example, double-dummy technique, capsule-in-capsule technique, or identical appearance of active and control medications.Two trials (32,34) stated that the outcomes assessors were supposed to be blinded, but they were able to recognize the allocation of the treatments.
A total of 11,927 patients were randomized, but only 7807 participants finished the trials (4120 dropouts).In the opioid group, the average dropout rate was 35%; 15% of the participants dropped out due to inadequate pain relief and 21% due to side effects (some patients dropped out for both reasons).In the control groups, the average dropout rate was 38% (with 30% of the dropouts due to inadequate pain relief and 10% due to side effects).
Forty-one trials had at least one author who was affiliated with the pharmaceutical industry.Ten trials were clearly not funded by pharmaceutical companies, five did not report any information about funding, and the remaining 47 were funded by the industry.In 47 trials, the drug was provided by the industry, in nine trials it was not, and this information was not reported in the remaining five trials.The statistical analysis was performed by an industry-affiliated statistician in nine trials, by someone not affiliated with the industry in 10 trials, and this was not reported in the remaining 43 trials.When the above information was combined, 54 of the 62 trials (87%) had some type of relationship with the pharmaceutical industry.
Regarding the CNCP diagnoses, 87.1% were classified as nociceptive pain (osteoarthritis, rheumatoid arthritis and back pain without radiculopathy), 9.2% as neuropathic pain (diabetic neuropathy, postherpetic neuralgia, phantom limb pain, 'neuropathic pain' and regional cervicobrachial pain syndrome), 3.2% as fibromyalgia and 0.41% as mixed nociceptive and neuropathic pain.The average age of the included population was 58.1 years (range 40 to 71 years); 63% were female and 85% were white.

Efficacy of opioids compared with placebo (table 2)
Opioids were compared with placebo in 47 trials.Pain was assessed in all trials, and function in 31 trials.The results showed a medium ES in favour of opioids for pain (ES=0.58,95% CI 0.48 to 0.67) and a small ES for function (ES=0.34,95% CI 0.25 to 0.43).The mean score was 6.0 mm longer for nabilone than for dihydrocodeine in the available case analysis and 5.6 mm in the per protocol analysis.Dihydrocodeine provided better pain relief than the synthetic cannabinoid nabilone.Nabilone was significantly superior to dihydrocodeine on the SF-36 (role-physical)

Continued on next page
Two trials were excluded because they examined the efficacy of opioids for acute-on-chronic pain (51,63).There were 14 trials with an enrichment design: two were excluded from the present review because the patients in both the opioid and the placebo groups received a short-acting opioid for breakthrough pain (73,74) and 12 were included (18,20,29,37,51,54,56,57,59,60,63,67).
From the randomized trials that reported data suitable for metaanalysis, the treatment effect among EERW trials was 0.62 (95% CI 0.33 to 0.92) and was 0.57 (95% CI 0.47 to 0.67) among the 39 non-EERW trials for pain outcomes.The difference between the treatment effects was not statistically significant (P=0.6).For functional outcomes, there were three EERW (ES=0.24,95% CI 0.08 to 0.41) and 28 non-EERW trials (ES=0.36,95% CI 0.26 to 0.45).The difference between the treatment effects was not statistically significant (P=0.3).
Regarding the subgroups of types of pain, the meta-analyses showed a medium ES for pain relief and small ES for functional outcomes for  both nociceptive and neuropathic pain (Table 2).For fibromyalgia and mixed pain, the meta-analyses showed a small ES for both pain and functional outcomes (Table 2).

Effectiveness of opioids compared with other drugs (table 2)
In patients with nociceptive pain, opioids were compared with nonsteroidal anti-inflammatory drugs (NSAIDs) in eight trials reported in seven publications (14,27,33,47,48,58,66) and compared with acetaminophen in one trial (36).A meta-analysis of eight trials of opioids compared with NSAIDS showed no difference for pain outcome.When we assessed the type of opioid used in these trials, seven used a weak opioid (codeine, propoxyphene or tramadol) and one trial used a strong opioid (titrated dose of oxycodone and morphine) (33).When analyzed separately, the meta-analysis of the seven trials of weak opioids did not show that opioids are more effective than NSAIDs, but the strong opioid was shown to be more effective than 1000 mg of naproxen daily.NSAIDs were significantly better than opioids for function, but with a very small ES of 0.18.There was only one trial for nociceptive pain comparing codeine plus acetaminophen with acetaminophen alone (36).It was found that at seven days, the addition of codeine was better than providing acetaminophen alone.After this period of time, there was no difference.
In patients with neuropathic pain, opioids were compared with tricyclic antidepressants in three trials (28,34,52), anticonvulsants in one (25) and cannabinoids in one (24).Opioids (either weak or strong) were not more effective than tricyclic antidepressants or anticonvulsants for either pain or function.We found one trial comparing dihydrocodeine with nabilone showing that dihydrocodeine provides better pain relief than nabilone, but the latter was significantly superior to dihydrocodeine on the SF-36 (rolephysical) (24).

DIsCussIon
We identified 62 randomized trials of opioids for a variety of CNCP conditions.Opioids were shown to be effective when compared with placebo in improving pain and function, but the ES were only medium and small, respectively.These conclusions did not change when we analyzed subgroups of EERW/non-EERW design, weak/strong opioids, or when the diagnosis was nociceptive/neuropathic pain.
There is an important difference in side effects reported between EERW and non-EERW designs.Six adverse effects were observed more often in the opioid group than in the placebo group in the non-EERW designs.Consumers of opioids should be aware of these effects to make better informed decisions before they are prescribed long-term opioids for CNCP.All included studies were randomized trials, and the majority were judged to be of high methodological quality.However, most of these trials (74%) were of short duration (ie, shorter than six weeks), with ties to the pharmaceutical industry and a sizeable number of dropouts.Our analyses could only be performed on what was reported in the trials.On some occasions, we contacted the authors of the trials and were able to clarify some details and obtain more information.
The comparisons of opioids with other drugs were based on studies that were not designed as equivalence or noninferiority trials.Therefore, the conclusions derived from these studies should be accepted with some reservation.
Our conclusions are similar to other recently published systematic reviews on this topic.A recently published Cochrane review of oral or transdermal opioids for osteoarthritis of the knee or hip included 10 trials and concluded that, overall, opioids were more effective than control interventions in terms of pain relief and improvement of function.There were no substantial differences in effects according to type of opioid, analgesic potency (strong or weak), daily dose, duration of treatment or follow-up, methodological quality of trials or type of funding.Adverse events were more frequent in opioid groups compared with controls.They found, however, only small to moderate beneficial effect (75).
Another recent Cochrane review examined the long-term safety, efficacy and effectiveness of opioids for CNCP (76).They included 26 studies with 27 treatment groups, of which 12 were administered orally, five transdermally and 10 intrathecally.Twenty-five of the studies were case series, and there was one randomized trial comparing two opioids.They found that opioids were associated with a clinically significant reduction in pain; however, quality of life and functional status were inconclusive due to insufficient evidence and statistical findings.
More research is needed to determine the usefulness of EERW designs in the administration of opioids for CNCP.In principle, enrichment is a process of increasing the proportion of likely responders to a treatment, and is an efficient strategy for maximizing differences between drug and placebo effects in a clinical trial.Some investigators defend the use of enrichment design in opioids for CNCP because opioids may underperform in clinically heterogeneous contexts, which means that substantial efficacy in a particular subgroup of patient may be diluted or masked by poor efficacy in other subgroups.In the case of opioids, a low proportion of responders would produce an average response that would be moderate at best, concealing the good reponse in the minority of responders; therefore, a drug of potential clinical usefulness to some patients might be discarded because, on average, it did not appear to be very effective in a trial (2).On the other hand, disadvantages of EERW designs include an influence on the apparent efficacy, the limited external validity (generalizability) because of the sample selection procedure, and the invalidation of drug-placebo comparisons because of carryover effects or withdrawal symptoms (77).
There is also a need to assess whether chronic use of opioids provides more benefits than harms in the long-term.None of these randomized trials followed patients for more than one year to determine whether they continued to benefit from opioids.

ACknoWLEDGEMEnts:
The authors acknowledge the support from the National Opioid Use Guideline Group (NOUGG), as well as Quenby Mahood and Joanna Liu, librarians from the Institute for Work & Health, for their help with the literature searches and article retrieval.Finally, we thank the authors of the randomized trials who responded to our requests for more information while this systematic review was being conducted.
ContRIButoRs: All authors participated in the conception and writing of this review.Andrea Furlan was responsible for writing the protocol, and performed the data extraction, quality assessment, statistical analyses and report writing.Luis Chaparro was responsible for the data extraction and quality assessment.Emma Irvin was responsible for the literature searches and writing the methods.Angela Mailis-Gagnon was responsible for overseeing the project and editing the final manuscript.

Search terms for chronic noncancer pain
Search terms for opioids

Huse et al (32), 2001
respectively.NNT to obtain 50% and 33% decreases in pain intensity with morphine were 5.6 and 4.5, respectivelyContinued on next page

Primary and secondary Results (as reported in the studies) Placebo-controlled (neuropathic pain) Raja et al (52), 2002
Primary: Sum of pain intensity differences (0-10 NRS) in the first 60 min (SPID-60)Secondary: Proportion of breakthrough episodes with 33% and 50% improvement; time to significant pain relief, pain intensity differences, proportion of episodes with meaningful pain relief and proportion of episodes that required supplemental medication SPID-60 was significantly greater for breakthrough pain episodes treated with fentanyl buccal tablets compared with those in which placebo was administeredPlacebo-controlled (nociceptive pain) Roth (54), 1998 Primary: Time to exit from the study due to therapeutic failure Secondary: Severity of pain* (0-3 numeric scale), ability to perform activities Time to exit from the study because of insufficient pain relief was longer in the tramadol group.Pain at rest and severity of pain on motion were lower in the tramadol group.No differences were noted in general severity of current pain and on disability to perform ADLs Primary: Pain intensity* (0-3 numeric scale), pain relief Secondary: SPID, WOMAC* (physical function subscale)The addition of tramadol/acetaminophen to NSAID or COX-2 selective inhibitor therapy was effective in the treatment of osteoarthritis flare pain

Emkey et al (22), 2004
Primary: Pain intensity* (100 mm VAS) Secondary: Pain relief, WOMAC* (physical function subscale), SF-36 survey Mean final VAS scores, mean final pain relief rating scores, WOMAC physical function and SF-36 role-physical measures were all significantly better with tramadol/ acetaminophen than with placebo B. Tramadol starting at 50 mg/d and reaching 200 mg/d on day 4 C. Tramadol starting at 50 mg/d and reaching 200 mg/d on day 10 Duration of treatment: 2 wks

Kosinski et al (9), 2007; Gana et al (10), 2006; Schein et al (11), 2008
Primary: Pain intensity (100 mm VAS* and 5-point Likert scale) Secondary: Pain relief, activity, sleep, overall efficacy Analgesic efficacy was significantly better with codeine/acetaminophen than with placebo for all criteria except the number of awakenings Continued on next page

TABLE 1 -continued Characteristics of the 62 randomized trials included in the present updated systematic review (grouped by type of opioid)
Primary: Pain intensity* (5-point NRS)Secondary: Pain relief, global improvementBoth suprofen and propoxyphene produced a considerable reduction in pain intensity from baseline after only 1 wk of treatment.This beneficial effect did not diminish with continued therapy.Further improvement occurred in both groups by 24 wks

TABLE 3 Adverse effects of opioids Adverse effects of opioids Randomized trial design
Bolded data: The difference between opioid and placebo is statistically significant (P<0.05) and clinically relevant (difference ≥10%).EERW Enriched enrollment randomized withdrawal