Abstract
Background: Several recent randomized clinical trials have found that the medications being evaluated for neuropathic pain did not significantly differ from placebo for the primary efficacy endpoint, despite encouraging results from prior preclinical and clinical studies. It is unclear whether these trials were unsuccessful because the medications truly lack efficacy or whether characteristics of the trials compromised the demonstration of treatment benefits.
Objective: To identify factors associated with positive (i.e., favors medication) vs negative outcomes of placebo-controlled neuropathic pain trials.
Methods: We examined study characteristics associated with positive vs negative clinical trial outcomes for neuropathic pain treatments using the information provided in a comprehensive meta-analysis and additional ratings for 106 clinical trials.
Results: Univariate analyses indicated that the results of medication vs placebo comparisons were more likely to be positive when medication response rates were greater, placebo response rates were lower, and studies were published earlier. In a multivariate analysis performed to identify independent contributions of study characteristics to trial outcomes, greater medication response, reduced placebo response, and larger sample sizes were each uniquely associated with positive outcomes. In addition, greater medication response rates and parallel groups designs were each independently associated with greater placebo response rates.
Conclusions: The results suggest that study characteristics may contribute to the outcomes of clinical trials of treatments for neuropathic pain and provide an impetus for investigating strategies for decreasing placebo response rates and thereby possibly increasing the likelihood of positive outcomes in trials of efficacious treatments.
Glossary
- NNT=
- numbers needed to treat;
- NRS=
- numerical rating scale;
- VAS=
- visual analogue scale.
In several recent randomized clinical trials of neuropathic pain, the medications being evaluated did not significantly differ from placebo for the primary efficacy endpoint.1–8 Prior to these trials, there was considerable reason to believe that these antidepressant, anticonvulsant, and other medications would demonstrate efficacy because of encouraging results from previous preclinical and clinical studies. It is therefore difficult to determine whether these trials were unsuccessful because the medications truly lack efficacy in the conditions studied or whether other factors compromised the ability of these trials to demonstrate benefits vs placebo.
The randomized clinical trial is the only accepted method for identifying efficacious treatments for neuropathic pain, but few investigators have considered study characteristics and other factors that might influence whether trials succeed in demonstrating efficacy.9–13 Understanding and attenuating any factors that adversely impact the outcomes of neuropathic pain trials could reduce the risk of failure in demonstrating benefits of efficacious treatments. One factor that has received attention is the response in the placebo group, which reflects true placebo effects, natural history, and other temporal processes, regression to the mean, and nonspecific effects.14 The placebo response in studies of pain can be substantial,15,16 and failure to show treatment benefits in neuropathic pain trials might be a result of the magnitude of response in the placebo group.12 We hypothesized that successful outcomes in neuropathic pain clinical trials are associated with lower rates of placebo response and also examined the relationships between other study characteristics and both trial outcomes and placebo response rates.
METHODS
We examined the randomized, placebo-controlled clinical trials of pharmacologic treatments for neuropathic pain included in a recent comprehensive meta-analysis.17 There were 106 different clinical trials examined in this meta-analysis (see appendix E-1 on the Neurology® Web site at www.neurology.org), 14 of which included more than one medication vs placebo comparison, yielding a total of 123 different treatment group comparisons.
Measures.
Trial outcome and study characteristics hypothesized to be related to trial outcome were selected from the meta-analysis17 and also coded from the original studies. Treatment group comparisons within clinical trials in which patients assigned to medication showed significantly greater pain reduction in the primary outcome measure than those assigned to placebo were considered positive, whereas comparisons were considered negative when the medication vs placebo difference in the primary outcome measure was not significant. The proportion of patients assigned to treatment with placebo and with active medication whose pain decreased by 50% or more was recorded for each medication vs placebo comparison. Baseline pain intensity data were available for 65 of 123 active medication vs placebo comparisons that used either a numerical rating scale (NRS) or a visual analogue scale (VAS); the pain intensity at baseline was recorded as 0 to 10 (studies that used an NRS) or transformed to 0 to 10 (studies that used either a 0 to 100 NRS or a 0 to 100 cm VAS) and was calculated as the mean of the values for the active medication and placebo groups when both of these values were provided (these values were usually very similar because all trials were randomized). The duration of the treatment period was recorded in days. The primary outcome measure was recorded as NRS, VAS, or other measure, and NRS and VAS measures were combined into one category for analysis (the primary outcome measure was missing for one trial). Study design was categorized as either crossover or parallel groups. Neuropathic pain condition was categorized as postherpetic neuralgia, painful polyneuropathy, other peripheral, and central; for some analyses, the first three categories were combined and considered peripheral neuropathic pain. It was not possible to categorize three trials because they included patients with both peripheral and central neuropathic pain conditions18 or did not provide diagnoses of the patients enrolled in the trial.19,20 Study medication was categorized as antidepressants, anticonvulsants, opioids, and other, which included a variety of medications with different mechanisms of action.
Statistical analysis.
Two-tailed t tests for continuous variables and χ2 tests for categorical variables were calculated to compare the study characteristics associated with positive vs negative outcomes of comparisons between active medications and placebo. Logistic regression analysis21 was performed to identify the unique contributions of study characteristics to positive vs negative outcomes and to evaluate whether placebo group response made an independent contribution to study outcomes after controlling for medication response and other study characteristics. Two-tailed t tests for continuous variables and χ2 tests for categorical variables were also calculated to examine associations between study characteristics and placebo response. Finally, a multiple linear regression analysis was performed to identify which study characteristics were independently associated with placebo response. As recommended,21 only those variables for which associations with study outcome or placebo response in the univariate analyses were p ≤ 0.25 were included in the regression models.
RESULTS
Study characteristics.
As shown in table 1, data were drawn from 106 clinical trials published between 1966 and 2005. Sample sizes varied greatly, with a mean of approximately 80 patients per study and a large range (9 to over 1,200 patients). On average, patients in these trials had pain at baseline that was moderate in intensity. The mean duration of the treatment period was somewhat over 5 weeks, but treatment duration ranged from less than 1 week to 5 months. Approximately two-thirds of the trials used either an NRS or a VAS as the primary outcome measure. Crossover designs were somewhat more common than parallel groups designs, and approximately 60% of the treatment comparisons involved antidepressant or anticonvulsant medications. Approximately half of the patients enrolled in the clinical trials were diagnosed with painful polyneuropathy, and fewer than 10% of the patients had central neuropathic pain.
Table 1 Study characteristics of randomized clinical trials of neuropathic pain treatments
Overall, almost two-thirds of the 123 study comparisons demonstrated significantly greater pain reduction in the primary outcome measure with active medication vs placebo (table 1). Across all comparisons, 46% of patients assigned to active medication reported a 50% or greater decrease in pain whereas only 19% of the patients assigned to placebo reported this level of pain reduction.
Univariate comparisons of positive and negative trials.
Table 2 presents the associations between study characteristics and positive vs negative outcomes for all available data as well as for the 90 medication vs placebo comparisons examined in the multivariate analyses with complete data for all variables. As predicted, the proportion of patients who responded to placebo was significantly greater when the trial outcome was negative than when it was positive. Specifically, for negative outcomes, approximately 27% of patients administered placebo had pain reductions of 50% or more, but when the outcome was positive, only 16% of placebo patients experienced this level of pain relief.
Table 2 Relationships between clinical trial outcomes and study characteristics
As would be expected, the response to active medication was also significantly greater for positive compared with negative study outcomes. In addition, when all available data were analyzed, positive study outcomes were more likely to be found in older trials. There were no significant differences in outcome as a function of sample size, baseline pain intensity, trial duration, primary outcome measure, study design, or whether the neuropathic pain condition was central vs peripheral (the significance of differences in study outcomes between trials examining patients with postherpetic neuralgia, painful polyneuropathy, and other peripheral neuropathic pain syndromes was not tested).
Multivariate comparison of positive and negative trials.
Various study characteristics were significantly associated. For example, compared with crossover designs, parallel groups designs were more recent (mean year of publication, 1999 vs 1993; t104 = −4.20, p < 0.001) and larger (mean sample size, 144 vs 27; t104 = −4.21, p < 0.001). An association between outcome and, for example, year of publication, might therefore be accounted for by changes over time in study designs or sample sizes.
Accordingly, hierarchical logistic regression analysis was performed to identify unique contributions made by the study characteristics to determining positive vs negative study outcomes.21 In the first block of the analysis, medication response, sample size, year of publication, study design, and pain condition were simultaneously entered to evaluate their independent contributions to positive vs negative outcomes. The results indicated that greater response to the study medication and larger sample size were each independently associated with a greater likelihood of a positive outcome (table 3). In addition, positive study outcomes were significantly more common in crossover compared with parallel groups designs.
Table 3 Logistic regression model predicting clinical trial outcomes from study characteristics (n = 90)
Placebo response was then added to this initial model to determine whether it made an independent contribution to outcomes after accounting for the other study characteristics. In the final regression model, greater placebo response significantly predicted lower likelihood of a positive comparison, and both greater medication response and larger sample size remained significantly associated with positive outcomes. Although the significance of the association between baseline pain intensity and study outcome was p ≤ 0.25 in the univariate analysis, baseline pain was omitted from this logistic regression analysis because of the large amount of missing data; however, when a parallel hierarchical logistic regression analysis was performed with baseline pain included (n = 65), baseline pain was not independently associated with study outcome and the variables that were significantly associated with study outcome remained the same.
Factors associated with the placebo response rate.
The relationships between the placebo response rate and the other study characteristics showed that greater placebo response was associated with greater medication response, greater trial duration, and use of parallel groups designs (table 4). Because outliers may account for relationships between active medication and placebo response rates,22 the Spearman rank order correlation coefficient between placebo and medication response was also calculated, and the relationship between greater placebo response and greater medication response remained significant (rho = 0.42). There was a nonsignificant relationship between greater placebo response and larger sample size, but year of publication, baseline pain, primary outcome measure, or pain condition were not significantly associated with placebo response. A linear multiple regression analysis was conducted to identify unique contributions of the study characteristics to placebo response. As shown in table 5, the results of this analysis indicated that greater placebo response was associated with a greater active medication response and use of parallel groups rather than crossover designs.
Table 4 Relationships between placebo response and study characteristics
Table 5 Multiple regression model predicting placebo response from study characteristics (n = 90)
Sensitivity analyses.
It is possible that some of the clinical trials we examined did not show significant benefits for the active medications vs placebo because the medications lack efficacy for neuropathic pain. That is, some trials may have been truly negative trials, whereas truly efficacious medications may have failed to demonstrate benefit because of methodologic problems in other, so-called, failed trials. However, few trials included active comparators with well-established efficacy, which would make it possible to distinguish negative from failed trials; it is therefore unknown which negative trials reflect lack of drug efficacy and which reflect study failure. Because the inclusion of truly negative as well as failed trials in our analyses could have had an impact on the results, we conducted sensitivity analyses in which only trials examining specific medications and classes of medications approved for neuropathic pain by a US or foreign government regulatory agency or considered first-line in published treatment recommendations for neuropathic pain were included.17,23,24 For these analyses, antidepressants, anticonvulsants, opioid analgesics (including tramadol), cannabinoids, topical lidocaine, and capsaicin were included, and NMDA receptor antagonists, mexiletine, and glycine antagonists were excluded. Specific medications within included classes of medications for which the results of clinical trials have been inconsistent (e.g., lamotrigine, oxcarbazepine, topiramate) were included.
The results of regression analyses excluding medications that have not been approved by a regulatory agency or considered first-line indicated that greater medication response, reduced placebo response, and larger sample sizes were each uniquely associated with positive trial outcomes, and that greater medication response rates and parallel groups designs were each independently associated with greater placebo response rates. These multivariate results are identical to those found in the analyses of trial outcomes for all medications.
DISCUSSION
We identified study characteristics, including placebo response rates, that were associated with positive vs negative clinical trial outcomes of neuropathic pain treatments. Univariate analyses indicated that the results of active medication vs placebo comparisons were more likely to be positive when medication response rates were greater, placebo response rates were lower, and studies were published earlier. In a multivariate logistic regression analysis performed to identify the independent contributions of study characteristics to trial outcomes, greater medication response, reduced placebo response, and larger sample sizes were each uniquely associated with positive outcomes. In addition, the results of univariate and multivariate analyses indicated that greater medication response rates and parallel groups designs were each independently associated with greater placebo response rates. Although greater trial duration was associated with greater placebo response rates in the univariate analyses, trial duration did not make an independent contribution to placebo response in the regression analysis. This same pattern of results was found for depression clinical trials25; considered together, these data suggest that although longer trials may provide greater opportunities for nonspecific treatment effects and spontaneous symptom reduction, any impact of trial duration on placebo response rates appears to be modest and explained by other factors.
Both neuropathic pain and depression are chronic conditions for which existing treatments are symptomatic and clinical trial endpoints rely on patient-reported outcomes. We based our approach and hypotheses on the results of clinical trials of antidepressants in patients with depression. In evaluating antidepressants, it is the rule, not the exception, that some trials show significant benefits whereas others do not25,26; meta-analyses of these studies have found that positive clinical trial outcomes are associated with lower response rates in placebo groups,26 use of flexible vs fixed dosage designs,27–29 and enrollment of patients with more severe depression,28,30 and that greater response to placebo is associated with greater response to antidepressants25,26 and more recent year of publication.25 Our results are generally similar to these, especially in demonstrating that clinical trials of neuropathic pain treatments with greater rates of response in the placebo group are less likely to demonstrate that an active medication is superior to placebo.
Several explanations have been proposed for why a greater placebo response rate appears to increase the likelihood that a clinical trial will fail to demonstrate superior benefits of an efficacious treatment.12 For example, when patients treated with placebo demonstrate substantial improvement, it is difficult for patients treated with an active medication to show an even greater benefit.10 In addition, as the magnitude of the response in placebo groups increases, variability might increase and cause a loss of statistical power in a clinical trial.10 Additional research is needed to determine which of these or other explanations accounts for the association between higher rates of response to placebo and a greater likelihood of failure to show benefits of efficacious medications in clinical trials of neuropathic pain.
The most important implication of our results is that reducing the magnitude of the placebo response might enhance the ability of neuropathic pain trials to detect benefits of efficacious treatments and that strategies for decreasing placebo response rates should therefore be investigated. For example, placebo group response rates in neuropathic pain trials might be attenuated by enrolling patients with greater baseline pain severity, using flexible vs fixed dosage designs, minimizing the number of treatment groups, implementing strategies to decrease patient and staff expectations of improvement, and using novel approaches to identify individuals more likely to respond to placebo.12,31–35
However, although it has often been assumed that reducing the magnitude of response in placebo groups would be associated with a greater likelihood that trials will be positive, there have been few successful demonstrations of this. Indeed, it is possible that efforts that reduce placebo response rates may also be associated with decreased responses to active treatment, and so decreasing placebo responses may not necessarily increase the ability of a trial to demonstrate the benefits of treatment.31 Moreover, it has been argued that the great variability in the magnitude of placebo response rates may simply reflect the random play of chance and small sample sizes often used in clinical trials.36,37
Our results indicate that greater placebo response rates are associated with greater response rates to the active medications studied for neuropathic pain,17 as has been shown in studies of conditions as diverse as depression,25,26 Parkinson disease,38 and peptic ulcer disease.39 The explanation for associations between response rates in placebo and active medication groups in clinical trials is unknown, but a variety of factors may play a role.40–42 These include characteristics of patients enrolled in the clinical trials, methodologic aspects of the trials, expectations of patients and investigative personnel, and nonspecific effects associated with trial participation; the results of an analysis of acute pain trials also suggest such relationships may reflect statistical artifact.22 Importantly, that greater placebo and active medication response rates are significantly associated in clinical trials of neuropathic pain provides further support for the caution that any approaches that decrease placebo responses may not increase the likelihood of demonstrating benefits of efficacious treatments because they might also decrease responses to these treatments.
It has been suggested that smaller crossover trials of neuropathic pain treatments tend to demonstrate stronger treatment effects—as reflected in lower numbers needed to treat (NNT) for pain relief—than larger parallel groups designs.17 In contrast, in the present analyses larger sample sizes were significantly associated with positive outcomes and study design was not associated with trial outcome after accounting for placebo response rates. However, crossover designs were significantly associated with lower rates of placebo response, perhaps because patients receive both active and placebo treatment. These lower rates of placebo response might contribute to the stronger effect sizes that have been reported with crossover designs.13,17 Multiple factors must be considered to adequately interpret these relationships, including differences between analyses of effect size vs study outcome, types of medication studied, and statistical approaches (e.g., use of intention-to-treat methods).
The results of a recent analysis of three clinical trials of painful diabetic neuropathy suggested that the difference between active medication and placebo was greater in patients with greater baseline pain intensity.43 These data could be interpreted to suggest that inclusion criteria requiring higher baseline pain (e.g., ≥5 or 6 on a 0 to 10 NRS rather than ≥4) in neuropathic pain trials might make positive outcomes more likely. However, we did not find significant relationships between baseline pain intensity and either study outcome or placebo response rates, and the results of a recent analysis of depression trials indicated that use of inclusion criteria requiring higher levels of baseline depression was not associated with an increased rate of positive outcomes.29 Considered together, these data suggest that it would be premature to modify inclusion criteria to enroll only patients with severe pain in clinical trials; nevertheless, it would be worthwhile to examine whether stratification to ensure adequate representation of patients with different levels of moderate and severe pain would not only increase generalizability of trial results but might also increase the likelihood of positive outcomes.
Important methodologic limitations of the present study must be acknowledged. First, we examined a limited number of study characteristics. In future research on factors associated with outcomes of neuropathic pain trials, various other study characteristics should be assessed, including use of concomitant and rescue analgesics,13 fixed vs flexible dosage research designs,13,27–29 and emotional distress and other psychological characteristics of study patients.44 In addition, the methodologic features of the studies we examined varied greatly, including use of very different inclusion and exclusion criteria and outcome measures. Furthermore, our analyses were based on using pain reductions of ≥50% to identify the percentages of patients who responded to active medication and to placebo. Pain reductions of ≥30% are also known to be considered a clinically meaningful response by patients,45 and it would be important to determine if the same relationships would be found using this criterion to identify the percentages of patients who responded to active medication and placebo; unfortunately, very few of the randomized trials we examined provided this information.
An additional limitation involves the well-known hazards of interpreting results of published clinical trials when many studies are unpublished. If negative trials are more likely to remain unpublished, positive trials would be over-represented in the published literature, analyses of which may then be misleading. For example, if negative trials are more likely to have higher placebo response rates, then the published literature may provide an underestimate of these rates.46 Our conclusions about the contribution of study characteristics to clinical trial outcomes and placebo response rates in clinical trials of treatments for neuropathic pain are therefore limited to the published literature, and must be re-evaluated when more comprehensive databases of clinical trials become available.
A final important limitation of our study involves the difficulty of adequately distinguishing trials that have negative outcomes because the medications lacked efficacy in the condition studied from trials of efficacious medications that failed to demonstrate benefit because of methodologic shortcomings.47 The results of our sensitivity analyses excluding medications that have not been approved by a regulatory agency or considered first line were identical to those found in the analyses of trial outcomes for all medications, which suggests that our findings were not unduly influenced by trials of medications that lack efficacy. Interpretation of the results of future clinical trials of treatments for neuropathic pain would be greatly facilitated by the inclusion of active comparators with well-established efficacy.
Despite these limitations, our results suggest that various study characteristics, including greater magnitude of placebo response, may make it more difficult for a clinical trial to demonstrate efficacy of a medication for neuropathic pain. Improved understanding of such relationships has the potential to ultimately decrease the number of trials that must be conducted to establish efficacy and also reduce the number of patients enrolled in these trials. Such advances will not only increase the efficiency with which efficacious medications can be identified, but will also expose fewer patients to placebo and to medications that lack efficacy or have unacceptable risks.
Footnotes
-
Supplemental data at www.neurology.org
Editorial, page 250
e-Pub ahead of print on October 3, 2007, at www.neurology.org.
Disclosures: The authors shared responsibility for performing the analyses reported in this article, and no support was received from any external source for this research or its publication. N.B.F. has received research support or honoraria in the past year from Neurosearch A/S and UCB Nordic. R.H.D. has received research support, consulting fees, or honoraria in the past year from Allergan, CombinatoRx, Dara, Dov, Eli Lilly, Endo, EpiCept, Fralex, GlaxoSmithKline, GW Pharmaceuticals, Johnson & Johnson, KAI Pharmaceuticals, Merck, NeurogesX (also stock options), Novartis, Pfizer, Schwarz Pharma, Supernus, US Food and Drug Administration, US National Institute of Health, US Veterans Administration, Wyeth, and XTL Biopharmaceuticals. J.K. reports no potential conflicts of interest.
Received January 19, 2007. Accepted in final form May 22, 2007.
REFERENCES
Disputes & Debates: Rapid online correspondence
NOTE: All authors' disclosures must be entered and current in our database before comments can be posted. Enter and update disclosures at http://submit.neurology.org. Exception: replies to comments concerning an article you originally authored do not require updated disclosures.
- Stay timely. Submit only on articles published within the last 8 weeks.
- Do not be redundant. Read any comments already posted on the article prior to submission.
- 200 words maximum.
- 5 references maximum. Reference 1 must be the article on which you are commenting.
- 5 authors maximum. Exception: replies can include all original authors of the article.
- Submitted comments are subject to editing and editor review prior to posting.