- Email: [email protected]
Systematic review: Placebo response in drug trials of fibromyalgia syndrome and painful peripheral diabetic neuropathy—magnitude and patient-related predictors
Ò
PAIN 152 (2011) 1709–1717
www.elsevier.com/locate/pain
Research papers
Systematic review: Placebo response in drug trials of fibromyalgia syndrome and painful peripheral diabetic neuropathy—magnitude and patient-related predictors Winfried Häuser a,b,⇑, Eva Bartram-Wunn a,b, Claas Bartram a,b, Henriette Reinecke c, Thomas Tölle d a
Department of Internal Medicine I, Klinikum Saarbrücken, D-66119 Saarbrücken, Germany Department of Psychosomatic Medicine, Technische Universität München, D-81675 München, Germany Department of Clinical Psychology and Psychotherapy, Technische Universität Darmstadt, D-64293 Darmstadt, Germany d Department of Neurology, Technische Universität München, D-81675 München, Germany b c
Sponsorships or competing interests that may be relevant to content are disclosed at the end of this article.
a r t i c l e
i n f o
Article history: Received 24 November 2010 Received in revised form 21 January 2011 Accepted 24 January 2011
Keywords: Fibromyalgia syndrome Painful peripheral diabetic neuropathy Placebo Systematic review Meta-analysis
a b s t r a c t The magnitude of placebo response and its predictors in fibromyalgia syndrome (FMS) and painful peripheral diabetic neuropathy (DPN) had not been studied. We performed a systematic review by searching MEDLINE, CENTRAL, SCOPUS, and the databases of the U.S. National Institutes of Health and the Pharmaceutical Research and Manufacturers of America until July 2010. We included randomised controlled trials of any pharmacological therapy compared with pharmacological placebo in patients with FMS and painful DPN. Pain values were converted to a 0 to 100 scale. We computed the pooled weighted mean difference (WMD) between pain baseline and end of treatment scores in placebo and active drug groups using a random effects model. A total of 72 studies (9827 patients) in FMS and of 70 studies in DPN (10,297 patients) were included. The pooled WMD in the FMS-placebo group was 7.69 (95% confidence interval [CI] 6.10 to 9.29) and 17.11 (95% CI 16.41 to 17.90) in painful DPN. The pooled WMD in the FMS-active drug group was 13.96 (95% CI 11.93 to 15.99) and in painful DPN was 22.54 (95% CI 20.49 to 24.58). The correlation between WMD in the placebo and active drug group in FMS was r = 0.69 and painful DPN r = 0.47. Placebo accounted for 45% of the response in the drug groups in FMS and for 62% in painful DPN. The placebo response was higher in painful DPN than in FMS (P < .001). The placebo response was not associated with age, sex, and race, but with year of study initiation, pain baseline, and effect size in active drug groups in both diseases. Ó 2011 International Association for the Study of Pain. Published by Elsevier B.V. All rights reserved.
1. Introduction There is no formal definition of placebo that most clinicians and researchers agree upon. In clinical trials, placebos are generally control treatments with a similar appearance to the study treatments, but without their essential components [18]. The placebo response is defined to be the reduction in a symptom as a result of factors related to a patient’s perception of the placebo intervention [40]. The placebo response is determined by the placebo effect (psychological factors such as expectation of benefit, classical conditioning, verbal suggestions, and behaviours manifested by health care providers) as well as by the natural course of disease and by the study design (eg, regression to the mean, uncontrolled parallel interventions) [8,9,31].
⇑ Corresponding author at: Klinikum Saarbrücken gGmbH, Winterberg 1, D-66119 Saarbrücken, Germany. Tel.: +49 681 9632020; fax: +49 681 9632022. E-mail address: [email protected] (W. Häuser).
There is a growing interest of pain medicine in the placebo response in chronic pain trials [39] for 2 reasons: First, the clinical use of placebo had been advocated in editorials [17] and by influential commentators [4]. In clinical practice, physicians use placebo for pain management [38]. The use of placebos is considered to be justified if there is a chance of a clinical improvement by placebo treatment [12]. Whether placebo induces a clinically relevant and stable improvement in chronic pain syndromes needs to be tested [23]. Second, pharmaceutical companies try to identify modifiable trial characteristics to enhance the ability to detect benefits of pharmacological treatments. Negative results of some placebocontrolled trials in chronic neuropathic pain syndromes had been attributed to unexpectedly large reductions of pain intensity in placebo groups that compromised the ability to show significant greater improvements with active medication [11]. The impact of placebo response on drug response in chronic pain trials had not been quantified until now. Topical reviews on placebo response in painful diabetic peripheral neuropathy (DPN) and postherpetic neuralgia (PZN) trials focussed on study characteristics associated
0304-3959/$36.00 Ó 2011 International Association for the Study of Pain. Published by Elsevier B.V. All rights reserved. doi:10.1016/j.pain.2011.01.050
1710
Ò
W. Häuser et al. / PAIN 152 (2011) 1709–1717
with high placebo response rates or negative study outcomes [7,24,33]. Patient characteristics predicting placebo response in painful DPN had not been studied. Studies in osteoarthritis (OA), migraine, and irritable bowel syndrome came to divergent findings on the impact of patient-related predictors such as age, sex, pain baseline, and continent of the study on placebo response rates [10,27,31,44]. The influence of the quantity and quality of the patient–investigator relationship had only been studied in irritable bowel syndrome (IBS) [5,22]. No studies had been conducted on the magnitude and on any types of predictors of placebo response in fibromyalgia syndrome (FMS). Given these open questions on placebo response in chronic pain syndromes, the aims of this systematic review were to determine the magnitude of placebo response on pain, to determine to what amount the placebo response accounts for the response in the active drug group, and to test for potential characteristics of the patient and the patient–investigator interaction associated with the placebo response, in randomised controlled studies of any pharmacological therapy in patients with FMS and painful DPN.
2. Methods The review was performed according to the PRISMA Statement (Preferred Reporting Items for Systematic Reviews and MetaAnalyses [28], see Online Supplementary Table 1) and the recommendations of the Cochrane Collaboration [16]. 2.1. Hypotheses Based on the literature on psychological mechanisms of placebo response [8,31], studies available on determinants of placebo response on pain, and our experiences with patients in clinical studies, we tested the following assumptions. 2.1.1. Patient characteristics Pain at baseline is associated with placebo response rate as shown in OA trials [44] and lamotrigine studies in chronic neuropathic pain [19]. Because of the limited and contradictory findings in the literature, we had no a priori hypotheses on the impact of age, sex, race, and type of disease. 2.1.2. Study-related characteristics that might influence the selection of patients The year of study initiation is positively associated with placebo response rate, as shown in trials in OA [44]. We assumed a higher placebo response rates in studies conducted in Europe compared with studies conducted in North America, as shown in migraine [27]. We assumed that studies conducted in second- or third-world countries have higher placebo response rates than studies in firstworld countries (North America, Western Europe, Japan) because patients in second- or third-world countries have less access to treatment options than patients in first-world countries and therefore have higher treatment expectations. 2.1.3. Quantity of investigator–patient relationship The number of study visits and treatment duration are associated with placebo response rate as shown in IBS [6]. 2.2. Protocol Methods of analysis and inclusion criteria were specified in advance (review protocol available on request).
2.3. Eligibility criteria 2.3.1. Types of studies Double-blind randomized controlled trials of any size with a parallel or cross-over design were included. Studies without randomization and single-blind studies were excluded. Studies with an enriched enrolment with randomized withdrawal design were excluded because of the potential effects of the study design on placebo effects [36]. No language restrictions were made. 2.3.2. Types of participants Patients with painful DPN and FMS, diagnosed by defined criteria, were included. We excluded studies in which DPN was mixed with other neuropathic pain syndromes if no separate data for DPN were reported because of different placebo response rates in painful DPN and PZN trials [7]. 2.3.3. Types of interventions Randomized controlled trials comparing any type of pharmacological medication with pharmacological placebo were included. Studies with nonpharmacological placebos and with pseudo-placebos (active drug without evidence for effectiveness in the disease of interest) were excluded. Studies that combined pharmacological placebo with any other defined treatment, whose effects on pain were tested for, were excluded. 2.3.4. Types of outcomes measures Studies should assess the patient’s ratings of pain intensity. If more than 1 pain score was used, we preferred the following order for the inclusion into analysis: pain Visual Analogue Scale (VAS) 0 to 100, pain Numeric Rating Scale (NRS) 0 to 100, pain VAS 0 to 10, pain NRS 0 to 10, any other pain VAS or NRS, summary scores including nonpainful symptoms (eg, paresthesia and sleep numbness of the feet). For summary scores, we preferred the following order: Total symptom Score (TTS) [45], any other combined score. If intention-to-treat (ITT) and completer analysis were reported, we used ITT outcomes. 2.4. Data sources and searches 2.4.1. Diabetic DPN We used the search strategy of the guidelines of the National Institute for Clinical Excellence on pharmacological management of neuropathic pain (search of literature until October 2008) [29] for MEDLINE. The search strategy was adopted to SCOPUS, the Cochrane Central Register of Controlled trials (CENTRAL), websites of the U.S. National Institutes of Health (NIH) (www.clinicaltrials.gov), and the Pharmaceutical Research and Manufacturers of America (PhRMA) (www.clinicalstudyresults.org). The search was conducted until July 30, 2010. 2.4.2. FMS We expanded our searches used for the German interdisciplinary guideline on the management of FMS [14] in the electronic bibliography databases detailed above until July 30, 2010 (see Online Supplementary Table 2). For both conditions, we reviewed the reference lists of included articles. For painful DPN, we searched the reference list of a recent topical review on the pharmacological management of neuropathic pain [11]. 2.5. Study selection Two authors independently screened the titles and abstracts of potentially eligible studies identified by the search strategy detailed above. The full text articles were then examined independently by 2 authors to determine whether they met the inclusion criteria. Discrepancies were rechecked and resolved by consensus.
Ò
W. Häuser et al. / PAIN 152 (2011) 1709–1717
2.6. Data collection process and data items A structured coding plan was developed prior to the meta-analysis. Two authors independently extracted the data using standard extraction forms. Discrepancies were rechecked, and consensus was achieved by discussion. If needed, a third author reviewed the data to reach a consensus. The coding plan included the following items. 2.6.1. Outcomes We examined change in pain intensity from baseline to endpoint. For trials with more than 1 dosage group of the active treatment, we selected the group with the highest dosage for comparison. 2.6.2. Study characteristics The following variables were collected from all trials: year of initiation (if unavailable, estimated as 3 years before publication [33]; for analysis we calculated the incremental year of study initiation with earliest study set as 0) and publication of the study, sponsoring by pharmaceutical company (if support was not mentioned, we checked the acknowledgement sections and the affiliations of the authors for study support by pharmaceutical companies), continent in which the study took place (Europe, America, Asia, mixed continent samples), number of study sites (if not reported, we chose the number of clinical institutions of the author list in case of study samples <100 patients), percentage of patients screened/randomised, study design (parallel groups or crossover), placebo run-in period, type of active medication (antidepressants, anticonvulsants; other), application of medication (oral, topical, parenteral), total sample sizes in active treatment and placebo group, treatment duration in weeks, number of study visits (If the number of study visits was not reported in detail, we assumed 1 study visit for baseline assessment and randomisation each and for each other point of assessment reported), type of analysis (ITT analysis), publication status (peer reviewed journal vs only available in database). 2.6.3. Patient characteristics Patient characteristics were age, sex (percentage of females), race (percentage of whites/Caucasians), and pain intensity baseline. 2.7. Validity assessment Two authors independently collected the reported quality assessment of reports of randomised clinical trials in terms of Jadad scores, considering the description and sequence of randomization, the double-blind procedure, the appropriateness of randomisation and double-blinding procedures, and the description of withdrawals and dropouts (range 0 to 5) [20]. Discrepancies were rechecked and resolved by consensus and if needed by a third author. Interrater reliability for study characteristics and validity characteristics was computed. 2.8. Dealing with missing data Where details of study outcomes were missing, attempts were made to obtain these data through contacting 74 trial authors. Additional data were provided by 12 authors. Where final standard deviations (SDs) were not available from trial authors, they were calculated from t-values, confidence intervals, or standard errors, where reported in articles [16]. If these data were not available, the final SD was substituted by the mean of the SDs of studies available that used the same outcome scale [16]. In case of more than 2 studies with the same active medication available, we
1711
substituted by the mean of the SDs of the trials with the same medication, if not by the SDs of all trials available. If the mean and SD of pain final was not reported but mean change with SD was given, the mean final was calculated by subtracting the mean change from pain baseline [16]. We substituted missing SD final by SD mean change, because there is no validated formula available to calculate final SD from SD of mean change. In PhRMA, we found the SDs of pain final and the SDs of mean change of 3 studies with duloxetine and 1 study with pregabalin. The Pearson correlation of SD of pain final and SD of mean change of the 4 placebo arms was r = 0.92. We did not ask for not-reported details of study design (eg, method of randomisation, identity of active drug and placebo) because we were unable to receive these details of studies conducted before 2000 in previous systematic reviews on antidepressants in FMS [15]. 2.9. Risk of bias in individual studies The risk of bias in individual studies was assessed by summarizing the study quality in the Jadad score. 2.10. Statistical analysis Descriptive statistics were used to characterize the features of the included trials. Nonparametrical tests were applied for the comparison of continuous variables, and v2 tests for the comparison of categorical variables. Correlations between continuous variables were calculated by the Pearson correlation coefficients. A 2-sided P value of 6.05 was considered significant. We used SPPS version 17.0 (SPSS, Inc, Chicago, Illinois, United States). Meta-analyses were conducted using RevMan Analyses software (RevMan 5.0.24) of the Cochrane collaboration [37] and metaregression analyses by Comprehensive Meta analysis software [3]. Pain scores from 0 to 10 numerical rating scales and other scales were converted to a 0 to 100 scale. We calculated the pooled weighted mean difference (WMD) between baseline and end-oftreatment scores of pain in placebo and active drug groups using a random effects model. This model is more conservative than the fixed-effects model and incorporates both within-study and between-study variance [26]. We used the benchmarks of the Initiative on Methods, Measurement and Pain Assessment in Clinical Trials (IMMPACT) with <10 points decrease on a scale of 0 to 100 to be no important improvement, 10 to 20 points to be a minimally important improvement, and 20 to 27 points to be a moderately important improvement [6] for interpreting the pooled WMD between baseline and end-of-treatment values of pain in placebo and active drug groups. We calculated pooled standardized mean differences (SMD) between baseline and end-of-treatment values of pain by a random effects model, to compare our results with other systematic reviews on placebo response in chronic pain syndromes [18,44]. The calculation of how much the improvement in the active drug period was attributed to the placebo response was performed according to a systematic review on placebo response in antidepressant trials in depression [35]. The I2 statistic was used to estimate the percentage of total variation across studies because of heterogeneity, rather than by chance (i.e., the percentage of variability of associations across studies that is not due to chance or random error, but rather due to real differences in study patients, design, or outcome definitions). I2 values P50% represent substantial heterogeneity. Tau2 was used to determine how much heterogeneity was explained by subgroup differences [16]. The impact of continuous potential predictors on pain response was tested by metaregression, which was performed using the
1712
Ò
W. Häuser et al. / PAIN 152 (2011) 1709–1717
random-effects model. Tau2 variance was calculated by the method of maximum likelihood [3]. To test hypotheses of subgroup effects in case of categorical variables, a test of interaction with a predetermined 2-tailed P < .05 was used [1]. 2.11. Additional analyses 2.11.1. Subgroup analysis Subgroup analyses were prespecified for the following variables: sponsoring by pharmaceutical company, type of publication (peer reviewed journal vs database), type of active medication (antidepressants, anticonvulsants, other), type of application (oral, topical, parenteral), approval by the U.S. Food and Drug Administration (FDA) for FMS (duloxetine, milnacipran, pregabalin) or painful DPN (duloxetine, pregabalin), study design (cross-over vs parallel; single-blind run-in). These subgroup analyses were also used to examine potential sources of clinical heterogeneity. To test our hypotheses, a test of interaction was performed with studies in America vs Europe and first-world vs second- or third-world countries. 2.11.2. Sensitivity analyses Sensitivity analyses were prespecified for studies with and without imputation of data (SDs estimated by mean of SDs of other studies) and/or data extracted from figures, with and without intention-to-treat-analysis and studies with a low (1 to 2) and moderate (3 to 5) Jadad score. We decided post hoc to perform a sensitivity analysis of studies that assessed pure pain scores and composite neuropathic symptom scores including pain in painful DPN. 2.11.3. Risk of bias across studies Publication bias was controlled for by computing ‘‘safe-n rates’’ [30] and the rank correlation test for funnel plot asymmetry. The test examines the rank correlation between standardized intervention effect and its standard error. An asymmetric funnel plot would give rise to such a correlation and may be indicative of publication bias [2].
3. Results 3.1. Search of literature A total of 1251 studies in FMS and 1034 studies in painful DPN fulfilled the first level of inclusion criteria. After excluding studies based on information presented in study abstracts, 107 complete study reports were considered in more detail in FMS and 112 in painful DPN. The final sample size consisted of 72 studies in FMS and 70 studies in painful DPN (see Online Supplementary Figs. 1 and 2 and Tables 3 through 6).
Table 1 Study and patient characteristics in FMS and DPN trials. FMS Number of trials; N 72 Study initiation year; Mean (SD) 1997 (7.1) Only available in database 4 (5.6) Industrial sponsoring; N (%) 56 (77.8) Continent Europe; N (%) 28 (368.9) America; N (%) 42 (58.3) Asia, N (%) 1 (1.4) Mixed continent; N (%) 1 (1.4) Number of countries; Mean (SD) 1.6 (2.5) Number of study sites; Mean (SD) 12.8 (23.5) Parallel design; N (%) 57 (79.2) Single blind run-in; N (%) 9 (12.5) Treatment duration (weeks); Mean (SD) 10.6 (7.6) Number of study visits; Mean (SD) 6.2 (3.4) Active treatment Antidepressants; N (%) 26 (36.1) Anticonvulsants; N (%) 5 (6.9) Other; N (%) 41 (56.9) Application medication Oral; N (%) 58 (80.6) Parenteral; N (%) 10 (13.9) Local; N (%) 4 (5.6) Total sample size; Mean (SD) 176 (269) % Women; Mean (SD) 93.0 (10.1) % Caucasian; Mean (SD) 90.3 (13.5) Mean age; Mean (SD) 47.6 (3.8) Pain baseline (0–10); Mean (SD) 6.5 (0.8)
Painful DPN P 70 1999 (1.0) 9 (12.9) 53 (75.7) 17 (24.3) 35 (50.0) 9 (12.9) 9 (12.9) 2.3 (3.4) 22.2 (30.1) 60 (85.7) 14 (20) 11.6 (8.6) 6.2 (3.0)
NA 0.25 0.12 0.77 0.005
0.006 0.04 0.35 0.22 0.36 0.34 <0.0001
10 (14.3) 28 (40.0) 32 (45.7) 0.73 56 (80.0) 9 (12.9) 5 (7.1) 200 (195) 45.2 (16.2) 76.5 (29.8) 58.3 (3.8) 6.4 (1.0)
0.03 <0.0001 <0.0001 <0.0001 0.25
FMS, fibromyalgia syndrome; DPN, painful diabetic neuropathy; NA, not applicable, SD, standard deviation.
3.2.2. Outcomes All FMS trials and 93% of the painful DPN trials assessed pain by visual analogue or numeric rating scales. Pain intensity was most often based on ratings of average pain in the past 24 h, and the remainder were almost all ratings of current pain. In 7% of the painful DPN trials, only a summary score including other neuropathic symptoms such as paraesthesia and numbness could be analysed. We were able to detect complete means and SD scores for 78% of the FMS and 84% of the painful DPN trials. Therefore we decided that a substitution of missing values was justified. 3.2.3. Reported study quality The means of the Jadad scores did not differ between FMS (mean[SD] = 2.9 [1.8]) and painful DPN (mean[SD] = 2.8 [1.7]) trials (P = .79). 3.2.4. Validity assessments The interrater reliability for the assessments of the study sample and outcome characteristics ranged from 0.85 to 0.91 and was for study quality r = 0.95.
3.2. Study characteristics
3.3. Overall effect size on pain in placebo groups
3.2.1. Study design and patient characteristics There were no significant differences in the year of study initiation, type of publication, type of study design, treatment duration, number of study visits, type of application of medication, and baseline pain scores between the 2 diseases. Trials with painful DPN significantly included more study sites and countries, were more frequently conducted in mixed continent samples, more frequently used anticonvulsants and less antidepressants, had greater total sample sizes, and included more older, male, and non-Caucasians patients than FMS trials (Table 1 and Online Supplementary Tables 3 and 4).
The pooled WMD between pain baseline and end of treatment in the FMS lacebo groups was 7.69 (95% confidence interval [CI] 6.10 to 9.29; I2 = 71%) (see Online Supplementary Fig. 3) and in the FMS active drug group was 17.11 (95% CI 16.41 to 17.90; I2 = 84%) on a pain scale of 0 to 100. The WMD of placebo on pain indicated no important improvement, and the WMD of active drug on pain indicated minimally important improvement according to IMMPACT benchmarks [6]. The pooled SMD on pain in the FMS placebo groups was 0.42 (95% CI 0.35 to 0.49; I2 = 53%), as compared with 0.82 (95% CI 0.72 to 0.92; I2 = 77%) in the active drug group. There was a high linear correlation between the WMDs (r = 0.69,
Ò
1713
W. Häuser et al. / PAIN 152 (2011) 1709–1717
P < .0001) and the SMDs (r = 0.73, P < .001) of the placebo and active drug groups, indicating that reported improvements in placebo and drug groups were highly interdependent. The ratio of the WMDs suggests that 45% of the improvement in the active drug group was attributable to the placebo response based on the assumption that the active drug effect is additive to the one of placebo. The pooled WMD between pain baseline and end of treatment in painful DPN placebo groups was 13.96 (95% CI 11.93 to 15.99, I2 = 85%) (see Online Supplementary Fig. 4) and in active drug groups was 22.54 (95% CI 20.49 to 24.58, I2 = 85%) on a 0 to 100 pain scale. The WMD of placebo on pain indicated minimally important improvement, and the WMD of active drug on pain indicated moderately important improvement according to IMMPACT benchmarks [6]. The overall SMD on pain in the painful DPN placebo groups was 0.71 (95% CI 0.61 to 0.81; I2 = 81%) as compared with d = 1.14 (95% CI 1.02 to 1.25; I2 = 85%) in the active drug group. There was a moderate linear correlation between the WMDs (r = 0.47, P < .001) and a high correlation of the SMDs (r = 0.52, P < .001) of the placebo and active drug groups, indicating that reported improvements in placebo and active drug groups were moderately interdependent. The ratio of the WMDs of placebo and active drug suggests that 62% of the improvements in the active drug groups were attributable to the placebo response based on the assumption that the active drug effect is additive to the one of placebo. The test of interaction of the effect size of placebo on pain in FMS and painful DPN trials was significant (P < .001).
3.4. Subgroup analysis The effect size of placebo on pain in trials conducted in America and Europe did not differ in FMS (P = .66) and painful DPN trials (P = .23). The effect size of placebo on pain of studies that included second- or third-world countries and studies conducted only in first-world countries did not differ in painful DPN (P = .61), but did differ in FMS trials (P = .01) (Table 2). 3.5. Sensitivity analysis Excluding studies with substituted or extracted data from figures, no intention-to-treat analysis, low Jadad score, and combined neuropathic symptom scales in painful DPN did not change the significant effect of placebo on pain in FMS and painful DPN trials (Table 3). 3.6. Metaregression analysis Simple linear regressions showed that pain baseline (b = 0.28, P = .003), incremental year of study initiation (b = 0.31, P = .008), number of study sites (b = 0.09, P = .0001), and effect of active medication on pain (b = 0.22, P = .003) were associated with the pooled WMD of placebo on pain in FMS (Table 4). Simple linear regressions showed that pain baseline (b = 0.25, P = .007), incremental year of study initiation (b = 0.38, P = .002), and effect of active medication on pain (b = 0.60, P < .0001) were associated with the pooled WMD of placebo on pain in painful DPN (Table 4).
Table 2 Subgroup analysis of the effect size (WMD baseline – end of treatment) in placebo groups in FMS and DPN. Outcome title
FMS
Painful DPN
Number of studies N (%)
WMD (95% CI)
Hetero geneity (I2 [%]; Tau)
Number of studies
WMD (95% CI)
Hetero geneity (I2; Tau)
Industrial sponsoring Yes No
56 16
8.98 (7.51–10.44) 3.86 ( 0.65–8.37)
59; 13.34 71; 53.59
53 17
15.04 (12.85–17.23) 9.25 (4.01–14.50)
86; 51.15 72; 72.84
Type of publication Peer-reviewed journal Database
68 4
7.41 (5.70–9.12) 10.50 (7.92–13.07)
72; 27.72 0; 0
61 9
13.20 (10.81–15.58) 17.64 (14.51–20.77)
85; 65.52 75; 17.03
7.89 (5.73–10.04) 7.15 (4.57–9.73) 6.00 ( 3.10–15.10) 10.00 (6.22–13.78)
77; 30.08 55; 21.21
35 17 9 9
15.25 10.69 11.13 18.05
32; 95; 85; 64;
12.62 (8.68–16.55) 7.14 (5.44–8.83)
66; 11.39 70; 26.39
13 57
15.30 (8.96–21.65) 13.59 (11.88–15.31)
93; 116.22 70; 25.39
Continent America 42 Europe 28 Asia 1 Mixed continent 1 Patients of second- or third-world countries Yes 5 No 67
(13.84–16.66) (3.36–18.01) (5.13–17.13) (15.09–21.01)
5.04 220.46 62.24 11.86
Type of active medication Antidepressants Anticonvulsants Other
26 5 41
8.32 (6.00–10.64) 11.75 (9.61–13.88) 6.51 (4.31–8.72)
67; 17.76 28; 1.65 59; 24.12
10 28 32
17.45 (13.57–21.33) 14.29 (12.45–16.12) 12.48 (8.04–16.92)
77; 26.09 61; 13.98 90; 125.87
FDA approved Yes No
13 59
12.04 (10.72–13.36) 6.42 (4.23–8.61)
2; 36.00 67; 40.55
14 56
16.89 (14.35–19.44) 12.97 (10.37–15.56)
72; 16.68 86; 71.25
Type of application Oral Parenteral Topical
58 10 4
7.92 (6.23–9.62) 8.08 (1.35–14.82) 5.67 ( 0.87–12.22)
72; 23.16 70; 71.15 57; 23.91
56 9 5
14.74 (13.14–16.34) 9.43 ( 3.06–21.91) 10.57 ( 0.17–21.30)
70; 22.59 97; 325.93 61; 83.70
Study design Parallel Cross-over
57 15
9.09 (7.58–10.60) 2.41 ( 0.33–5.16)
59; 14.62 37; 8.86
60 10
15.00 (12.93–17.07) 4.44 ( 1.46–10.35)
85; 49.88 46; 39.08
Single-blind run-in Yes No
9 63
8.11 (4.35–11.87) 7.59 (5.80–9.38)
84; 25.85 68; 26.39
14 56
15.06 (11.31–18.08) 13.68 (11.31–16.06)
76; 34.74 86; 60.00
WMD, weighted mean difference; CI, confidence interval; FMS, fibromyalgia syndrome; DPN, diabetic peripheral neuropathy; SMD, standardized mean difference; FDA, US Food and Drug Administration.
Ò
1714
W. Häuser et al. / PAIN 152 (2011) 1709–1717
Table 3 Sensitivity analysis of the effect size (WMD baseline – end of treatment) in placebo groups in FMS and DPN. Outcome title
FMS
Painful DPN
Number of studies N (%) Data extracted from figures and/or substituted Yes 16 (22.2) No 56 (77.8)
WMD (95% CI)
Hetero geneity (I2 [%]; Tau)
Number of studies
WMD (95% CI)
Hetero geneity (I2; Tau)
4.89 (1.75–8.03) 8.27 (6.51–10.02)
26; 10.10 74; 25.07
12 58
13.50 (9.18–17.81) 14.06 (11.76–16.36)
79; 39.62 85; 58.60
Pain outcome measure Pain only Mixed outcome measure Intention-to-treat analysis Yes No
72 0
7.69 (6.10–9.29)
71; 25.11
65 5
13.64 (11.94–15.33) 20.15 (6.38–33.93)
74; 30.41 96; 223.76
37 35
9.16 (7.39–10.93) 6.17 (3.58–8.57)
62; 14.47 69; 31.31
49 21
15.10 (12.84–17.35) 9.94 (6.40–13.48)
85; 48.80 67; 35.58
Jadad score Low (1–2) Moderate (3–5)
26 46
10.08 (7.43–12.72) 6.51 (4.62–8.39)
67; 22.76 68; 21.99
28 42
14.13 (11.68–16.94) 14.05 (11.22–16.89)
70; 28.08 88; 66.51
WMD, weighted mean difference; CI, confidence interval; FMS, fibromyalgia syndrome; DPN, diabetic peripheral neuropathy.
3.7. Risks of bias 3.7.1. Heterogeneity There was a substantial statistical heterogeneity of analysis for the effect size on pain in FMS and painful DPN trials. Heterogeneity was reduced <50% in case of studies only available in databases, anticonvulsants as active drug, FDA-approval, and cross-over design in FMS. The effect of placebo on pain remained significant except in trials without industrial sponsoring, studies conducted in Asia, those with topical application, and those with a cross-over design (Table 2). Heterogeneity was reduced <50% in case of studies conducted in America and cross-over design in painful DPN. The effect of placebo on pain remained significant except in trials with a cross-over design and with topical and parenteral application (Table 2). Subgroup analyses were not corrected for associations with other factors. 3.7.2. Reported study quality By metaregression, there was no significant association of the Jadad score and pooled WMD of placebo on pain in painful DPN trials (b = 0.01, P = .64). In FMS trials, there was a significantly negative association between the Jadad score and the pooled WMD on pain (b = 1.39, P = .004), indicating that low study quality such as inadequate randomisation or blinding might be associated with higher placebo response rates. 3.7.3. Publication bias Kendall’s tau of the Begg rank correlation test of the pooled WMD (active drug end and placebo end) was not significant in
FMS (P = .39) and painful DPN (P = .43), suggesting no relevant publication bias. In FMS, 315 unpublished studies reporting no positive placebo response would be needed to reduce the pooled WMD of placebo to 1.0. In painful DPN, 1038 [42] unpublished studies reporting no positive placebo response would be needed to reduce the pooled WMD of placebo to 1.0 (9.9). These rates confirm the stability of improvements in the placebo groups. 4. Discussion 4.1. Summary of main findings We found no important improvement of pain in placebo groups in FMS and minimally important improvement of pain in placebo groups in painful DPN trials according to IMMPACT benchmarks. We found that 45% (62%) of the effect size of active medication groups in FMS (painful DPN) was attributable to the placebo response. The placebo response was associated with year of study initiation, pain baseline, and effect size in active drug groups in both diseases. 4.2. Comparison with other reviews on placebo response in drug therapy in chronic pain 4.2.1. Magnitude and predictors of placebo response rates Our analysis of drug trials in painful DPN and FMS supports the conclusion of a systematic review on placebo in different clinical conditions that the mean effect size for placebos in medicine is small in magnitude [18]. The pooled effect of placebo on pain was 3.2 (95% CI 1.6 to 4.7) points on a 100-point scale in an
Table 4 Predictors of effect size (WMD baseline – end of treatment) in placebo groups in FMS and DPN by metaregression analyses. Predictor
FMS Degrees of freedom
Age Female sex Caucasian race Pain baseline Year of study initiation (incremental) Number of study sites Number of countries Treatment duration Number of study visits Effect of active treatment
66 65 36 71 71 71 71 71 71 71
Painful DPN Coef. (B) 0.0001 0.02 0.005 0.28 0.31 0.09 0.02 0.26 0.005 0.22
P
Degrees of freedom
.88 .83 .29 .003 .008 .0001 .21 .13 .66 .003
69 63 43 64 68 68 69 69 68 69
Coef. (B) 0.005 0.01 0.06 0.25 0.38 0.01 0.40 0.003 0.35 0.60
P .99 .90 .09 .007 .002 .90 .10 .60 .24 <.0001
Significant observations (P < .05) are in boldface type. WMD, weighted mean difference; FMS, fibromyalgia syndrome; DPN, diabetic peripheral neuropathy; Coef (B), regression coefficient of each regression representing the slope of each model.
Ò
W. Häuser et al. / PAIN 152 (2011) 1709–1717
analysis of studies with several acute and chronic pain syndromes [21], and thus lower than the ones we found in FMS and painful DPN. The SMD of placebo on pain compared with untreated controls in OA trials was 0.51 (95% CI 0.46 to 0.55) [44] and thus between the pre- and post-SMD values of placebo on pain in FMS and painful DPN trials. The substantial heterogeneity of the effect size of placebo on pain in both diseases confirms findings from previous analysis of chronic neuropathic pain trials that the extent of response in the placebo group is highly variable [11,19]. In line with systematic reviews on placebo response in OA [44] and 1 pooled analysis of 3 lamotrigine trials in neuropathic pain [19], the pain-relieving effect of placebo was positively associated with pain baseline scores in FMS and painful DPN. In accordance with childhood and adolescence migraine trials, we found no association between age and sex and placebo response rates [11]. In contrast to our results, higher placebo response rates were associated with higher age in irritable bowel syndrome (IBS) [31]. In contrast to studies on placebo response in the prophylaxis of migraine [27], we found no differences in the effect sizes of placebo in studies conducted in North America and in Europe. Higher placebo response rates were associated with a longer duration of the study and a greater number of study visits in 1 IBS study [5]. These associations failed to reach significance in FMS and painful DPN trials. However, it should be noted that the placebo response was stable over time in DPN as in other chronic neuropathic pain syndromes [33]. The strong effect of publication year on placebo response, which was also found in trials in OA [44] and depression [35], is difficult to explain. Year of study initiation, number of patients, study sites, and countries included were significantly correlated due to large industry-sponsored studies on duloxetine and pregabalin in both diseases (details not shown). Waber et al. [42] were able to show that the analgesic placebo effect is higher if people expect the drug to be expensive and valuable. The placebo response rates in trials of drugs licensed by the FDA for FMS were higher than in nonFDA approved medication as well as in studies including secondor third-world countries compared with studies that included only first-world countries in FMS (Table 2). We speculate that large multicenter trials of well-known pharmaceutical companies with expensive medications including second- and third world countries induce high treatment expectations, resulting in higher placebo response rates. 4.2.2. Is the placebo response attributable to a placebo effect? A recent Cochrane review on placebo interventions for all clinical conditions argued that the effects of placebo interventions could be, at least in part, attributed to the natural course of disease and regression to the mean. However, by comparing patients who were randomly allocated to placebo group or no treatment group (natural course of disease), the authors found a statistically significant SMD of placebo on pain that was very variable, even among trials with a low risk of bias (hedge’s g = 0.25; 95% CI 0.18 to 0.35) [18]. The association between high baseline pain scores and high placebo response rates in FMS and painful DPN trials can support the assumption that the placebo effect is partially a regression to the mean [35]. However, the effect of placebo on pain was positively associated with the year of study initiation, too. Because year of study initiation and intensity of pain baseline were not significantly associated (details not shown), we hypothesize that this association can be explained by contextual factors. Therefore, we speculate that the response rates in placebo groups in FMS and painful DPN on pain are attributable to a placebo effect and regression to the mean.
1715
4.3. Mechanisms of placebo response Rapaport et al. [34] and others assume that placebo responders differ from drug responders, therefore indicating that the mechanisms of improvement under placebo might differ from the mechanisms of improvement under active treatment. However, we found a strong correlation between effects in placebo and active drug groups. This strong correlation is a contradiction to the assumption that placebo responders and drug responders are different. We can only speculate on the mechanisms underlying the different placebo response rates in painful DPN and FMS. Dworkin et al. [7] hypothesized that the placebo effect requires intact descending pathways, which are more likely to be present in DPN, a disorder predominantly localized to the peripheral nervous system, than in patients with PZN, who can have lesions in the spinal cord as well. Therefore we speculate that the placebo effect in FMS is limited because of deficient descending pain pathways, even though an experimental study demonstrated that pain relief through expectation superseded descending inhibitory deficits in fibromyalgia patients [13]. 5. Limitations 5.1. Methods Major limitations of meta-analyses are publication bias, quality of the included studies, and insufficient data [35]. We found no indicators for publication bias. We had to exclude some studies because of lacking data. The excluded studies were mainly studies with small sample sizes that had been conducted before 2000. Their results would probably not have substantially changed the results. In contrast to a recent topical review on placebo and treatment group responses in PZN vs painful diabetic peripheral neuropathy clinical trials that did not perform a meta-analysis with random-effect model because of incomplete outcome reporting [8], we were able to extract sufficient data for meta-analysis by substituting missing values. Sensitivity analyses demonstrated that the effect on pain was robust against these procedures. Some of the studies included did not report in detail the characteristics of their study samples, which limited the power of metaregressions of age, sex, and race with effect size on pain. Heterogeneity was substantial for most comparisons and only partially explainable by subgroup comparisons. The reasons for heterogeneity remain unknown. Our method to calculate the percentage of improvement attributable to placebo in the active drug group is not supported by studies and should therefore be interpreted cautiously. Our analysis might underestimate the potential effects of placebo for 2 reasons. First, we performed a secondary analysis of comparison arms of drug treatment. The patients in placebo and real drug groups were aware of the possibility of being treated with placebo. The effects in the placebo groups might have been higher if they would have been designed as treatment interventions (without possibility of a placebo treatment). Second, the average placebo effect cannot be used to deduce the magnitude of individual placebo responses. Individuals who experience a placebo response (eg, at least 30% pain reduction) are likely to have a much higher placebo response than the average effect reported here [41]. 5.2. Variables analysed Meta-analyses are limited in providing understanding of factors that contribute to placebo analgesia because they do not systematically vary factors that affect its magnitude [32]. We could not
1716
Ò
W. Häuser et al. / PAIN 152 (2011) 1709–1717
analyse the impact of rescue medication on placebo response rates because the amount was not reported in the trials. We could not assess patient-related predictors of placebo response rates such as personality characteristics and treatment expectations that had influenced pain response rates in IBS [25] and low back pain trials [43] because these variables had not been assessed in the studies. Moreover, we could only analyse the impact of the quantity, but not the quality, of the patient–investigator relationship. A recent study in IBS demonstrated that the effects of placebo acupuncture could be augmented compared with placebo acupuncture alone by a patient–practitioner relationship augmented by warmth, attention, and confidence [22]. Finally, verbal suggestions and behaviours manifested by health care providers and written information of the patient are likely to vary greatly across research contexts, and consequently generate considerable variability in the placebo effect itself [32]. 6. Conclusions A minimally important improvement in pain according to clinical benchmarks was only found in painful DPN trials. Therefore, placebo cannot be recommended for the management of chronic pain. Situational factors were important for the placebo response rather than individual factors (eg, age, race, and gender). These contextual factors, such as positive expectations and suggestions, should be used in clinical practice to empower the effects of active drug therapy. Conflict of interest statement Dr. Häuser has received honoraria for educational lectures from Eli-Lilly, Janssen-Cilag, Mundipharma, and Pfizer, and congress travel support from Eli Lilly. Dr. Toelle has worked as a consultant and/or speaker for Astellas, Gruenenthal, Eli Lilly, Boehringer Ingelheim Pharmaceuticals, Pfizer, UCB Pharma, and Mundipharma. The other authors have no conflict of interest to declare. Appendix A. Supplementary data Supplementary data associated with this article can be found, in the online version, at doi:10.1016/j.pain.2011.01.050. References [1] Altman DG, Bland JM. Interaction revisited: the difference between two estimates. BMJ 2003;326:219. [2] Begg CB, Mazumdar M. Operating characteristics of a rank correlation test for publication bias. Biometrics 1994;50:1088–101. [3] Comprehensive metaanalysis software. Version 2.0. Englewood, NJ: Biostat; 2010. [4] Cochrane AL. Effectiveness and efficiency. Random reflections on health services. Cambridge: Cambridge University Press for British Medical Journal & The Nuffield Provincial Hospitals Trust, 1989. p. 31. [5] Dorn SD, Kaptchuk TJ, Park JB, Nguyen LT, Canenguez K, Nam BH, Woods KB, Conboy LA, Stason WB, Lembo AJ. A meta-analysis of the placebo response in complementary and alternative medicine trials of irritable bowel syndrome. Neurogastroenterol Motil 2007;19:630–7. [6] Dworkin RH, Turk DC, Wyrwich KW, Beaton D, Cleeland CS, Farrar JT, Haythornthwaite JA, Jensen MP, Kerns RD, Ader DN, Brandenburg N, Burke LB, Cella D, Chandler J, Cowan P, Dimitrova R, Dionne R, Hertz S, Jadad AR, Katz NP, Kehlet H, Kramer LD, Manning DC, McCormick C, McDermott MP, McQuay HJ, Patel S, Porter L, Quessy S, Rappaport BA, Rauschkolb C, Revicki DA, Rothman M, Schmader KE, Stacey BR, Stauffer JW, von Stein T, White RE, Witter J, Zavisic S. Interpreting the clinical importance of treatment outcomes in chronic pain clinical trials: IMMPACT recommendations. J Pain 2008;9:105–21. [7] Dworkin RH, Turk DC, Peirce-Sandner S, McDermott MP, Farrar JT, Hertz S, Katz NP, Raja SN, Rappaport BA. Placebo and treatment group responses in postherpetic neuralgia vs painful diabetic peripheral neuropathy clinical trials in the REPORT database. Pain 2010;150:12–6. [8] Enck P, Benedetti F, Schedlowski M. New insights into the placebo and nocebo responses. Neuron 2008;59:195–206.
[9] Ernst E, Resch KL. Concept of true and perceived placebo effects. BMJ 1995;311:551–3. [10] Evers S, Marziniak M, Frese A, Gralow I. Placebo efficacy in childhood and adolescence migraine: an analysis of double-blind and placebo-controlled studies. Cephalalgia 2009;29:436–44. [11] Finnerup NB, Sindrup SH, Jensen TS. The evidence for pharmacological treatment of neuropathic pain. Pain 2010;150:573–81. [12] Finniss DG, Kaptchuk TJ, Miller F, Benedetti F. Biological, clinical, and ethical advances of placebo effects. Lancet 2010;375:686–95. [13] Goffaux P, de Souza JB, Potvin S, Marchand S. Pain relief through expectation supersedes descending inhibitory deficits in fibromyalgia patients. Pain 2009;145:18–23. [14] Häuser W, Eich W, Herrmann M, Nutzinger DO, Schiltenwolf M, Henningsen P. Fibromyalgia syndrome: classification, diagnosis, and treatment. Dtsch Arztebl Int 2009;106:383–91. [15] Häuser W, Petzke F, Üceyler N, Sommer C. Comparative efficacy and acceptability of amitriptyline, duloxetine and milnacipran in fibromyalgia syndrome—a systematic review with meta-analysis. Rheumatology 2011;50:532-43. [16] Higgins JPT, Green S. Cochrane handbook for systematic reviews of intervention. Version 5.01. Available at: www.cochrane.org/resources/ handbook/handbook.pdf. [accessed 30.07.10]. [17] Ho VMS. The placebo effect: can we use it better? BMJ 1994;309:69–70. [18] Hróbjartsson A, Gøtzsche PC. Placebo interventions for all clinical conditions. Cochrane Database Syst Rev 2010: 20; CD003974. [19] Irizarry MC, Webb DJ, Ali Z, Chizh BA, Gold M, Kinrade FJ, Meisner PD, Blum D, Silver MT, Weil JG. Predictors of placebo response in pooled lamotrigine neuropathic pain clinical trials. Clin J Pain 2009;25:469–76. [20] Jadad AR, Moore RA, Carroll D, Jenkinson C, Reynolds JM, Gavaghan DJ, McQuay HJ. Assessing the quality of reports of randomized clinical trials: is blinding necessary? Control Clin Trials 1996;17:1–12. [21] Kamper SJ, Machado LA, Herbert RD, Maher CG, McAuley JH. Trial methodology and patient characteristics did not influence the size of placebo effects on pain. J Clin Epidemiol 2008;61:256–60. [22] Kaptchuk TJ, Kelley JM, Conboy LA, Davis RB, Kerr CE, Jacobson EE, Kirsch I, Schyner RN, Nam BH, Nguyen LT, Park M, Rivers AL, McManus C, Kokkotou E, Drossman DA, Goldman P, Lembo AJ. Components of placebo effect: randomised controlled trial in patients with irritable bowel syndrome. BMJ 2008;336:999–1003. [23] Kaptchuk TJ, Kelley JM, Deykin A, Wayne PM, Lasagna LC, Epstein IO, Kirsch I, Wechsler ME. Do ‘‘placebo responders’’ exist? Contemp Clin Trials 2008;29:587–95. [24] Katz J, Finnerup NB, Dworkin RH. Clinical trial outcome in neuropathic pain: relationship to study characteristics. Neurology 2008;70:263–72. [25] Kelley JM, Lembo AJ, Ablon JS, Villanueva JJ, Conboy LA, Levy R, Marci CD, Kerr CE, Kirsch I, Jacobson EE, Riess H, Kaptchuk TJ. Patient and practitioner influences on the placebo effect in irritable bowel syndrome. Psychosom Med 2009;71:789–97. [26] Laird NM, Mostellier F. Some statistical methods for combining experimental results. Int J Technol Assess Health Care 1990;6:5–30. [27] Macedo A, Baños JE, Farré M. Placebo response in the prophylaxis of migraine: a meta-analysis. Eur J Pain 2008;12:68–75. [28] Moher D, Liberati A, Teztlaff J, Altman G. And the PRISMA Group. Preferred reporting items for systematic reviews and meta-analyses: the PRISMA statement. Ann Int Med 2009;51:1–7. [29] National Institute for Health and Clinical Excellence. Neuropathic pain. The pharmacological management of neuropathic pain in adults in non-specialist settings. Available at:, 2010. Accessed xxxx. [30] Orwin GR. A fail-safe N for effect size in meta-analysis. J Educ Stat 1983;8:157–9. [31] Pitz M, Cheang M, Bernstein CN. Defining the predictors of the placebo response in irritable bowel syndrome. Clin Gastroenterol Hepatol 2005;3:237–47. [32] Price DD, Finiss DG, Benedetti F. A comprehensive review of the placebo effect: recent advances and current thoughts. Annu Res Psychol 2008;59: 565–90. [33] Quessy SN, Rowbotham MC. Placebo response in neuropathic pain trials. Pain 2008;138:479–83. [34] Rapaport MH, Pollack M, Wolkow R, Mardekian J, Clary C. Is placebo response the same as drug response in panic disorder? Am J Psychiatry 2000;157:1014–6. [35] Rief W, Nestoriuc Y, Weiss S, Welzel E, Barsky AJ, Hofmann SG. Meta-analysis of the placebo response in antidepressant trials. J Affect Disord 2009;118:1–8. [36] Staud R, Price DD. Importance of measuring placebo factors in complex clinical trials. Pain 2008;138:474. [37] The Nordic Cochrane Centre. Review Manager (RevMan) [Computer program]. Version 5.0 for Windows. Copenhagen: The Cochrane Collaboration, 2009. [38] Tilburt JC, Emanuel EJ, Kaptchuk TJ, Curlin FA, Miller FG. Prescribing ‘‘placebo treatments’’: results of national survey of US internists and rheumatologists. BMJ 2008;337:a1938. [39] Turner JA, Deyo RA, Loeser JD, Von Korff M, Fordyce WE. The importance of placebo effects in pain treatment and research. JAMA 1994;271: 1609–14. [40] Vase L, Riley 3rd JL, Price DD. A comparison of placebo effects in clinical analgesic trials versus studies of placebo analgesia. Pain 2002;99:443–52.
Ò
W. Häuser et al. / PAIN 152 (2011) 1709–1717 [41] Vase L, Petersen GL, Riley 3rd JL, Price DD. Factors contributing to large analgesic effects in placebo mechanism studies conducted between 2002 and 2007. Pain 2009;145:36–44. [42] Waber RL, Shiv B, Carmon Z, Ariely D. Commercial features of placebo and therapeutic efficacy. JAMA 2008;299:1016–7. [43] Wasan AD, Kaptchuk TJ, Davar G, Jamison RN. The association between psychopathology and placebo analgesia in patients with discogenic low back pain. Pain Med 2006;7:217–28.
1717
[44] Zhang W, Robertson J, Jones AC, Dieppe PA, Doherty M. The placebo effect and its determinants in osteoarthritis: meta-analysis of randomised controlled trials. Ann Rheum Dis 2008;67:1716–23. [45] Ziegler D, Hanefeld M, Ruhnau KJ, Meißner HP, Lobisch M, Schütte K, Gries FA. The ALADIN Study Group. Treatment of symptomatic diabetic peripheral neuropathy with the anti-oxidant-lipoic acid: a 3-week multicentre randomized controlled trial (ALADIN Study). Diabetologia 1995;38:1425–33.
PAIN 152 (2011) 1709–1717
www.elsevier.com/locate/pain
Research papers
Systematic review: Placebo response in drug trials of fibromyalgia syndrome and painful peripheral diabetic neuropathy—magnitude and patient-related predictors Winfried Häuser a,b,⇑, Eva Bartram-Wunn a,b, Claas Bartram a,b, Henriette Reinecke c, Thomas Tölle d a
Department of Internal Medicine I, Klinikum Saarbrücken, D-66119 Saarbrücken, Germany Department of Psychosomatic Medicine, Technische Universität München, D-81675 München, Germany Department of Clinical Psychology and Psychotherapy, Technische Universität Darmstadt, D-64293 Darmstadt, Germany d Department of Neurology, Technische Universität München, D-81675 München, Germany b c
Sponsorships or competing interests that may be relevant to content are disclosed at the end of this article.
a r t i c l e
i n f o
Article history: Received 24 November 2010 Received in revised form 21 January 2011 Accepted 24 January 2011
Keywords: Fibromyalgia syndrome Painful peripheral diabetic neuropathy Placebo Systematic review Meta-analysis
a b s t r a c t The magnitude of placebo response and its predictors in fibromyalgia syndrome (FMS) and painful peripheral diabetic neuropathy (DPN) had not been studied. We performed a systematic review by searching MEDLINE, CENTRAL, SCOPUS, and the databases of the U.S. National Institutes of Health and the Pharmaceutical Research and Manufacturers of America until July 2010. We included randomised controlled trials of any pharmacological therapy compared with pharmacological placebo in patients with FMS and painful DPN. Pain values were converted to a 0 to 100 scale. We computed the pooled weighted mean difference (WMD) between pain baseline and end of treatment scores in placebo and active drug groups using a random effects model. A total of 72 studies (9827 patients) in FMS and of 70 studies in DPN (10,297 patients) were included. The pooled WMD in the FMS-placebo group was 7.69 (95% confidence interval [CI] 6.10 to 9.29) and 17.11 (95% CI 16.41 to 17.90) in painful DPN. The pooled WMD in the FMS-active drug group was 13.96 (95% CI 11.93 to 15.99) and in painful DPN was 22.54 (95% CI 20.49 to 24.58). The correlation between WMD in the placebo and active drug group in FMS was r = 0.69 and painful DPN r = 0.47. Placebo accounted for 45% of the response in the drug groups in FMS and for 62% in painful DPN. The placebo response was higher in painful DPN than in FMS (P < .001). The placebo response was not associated with age, sex, and race, but with year of study initiation, pain baseline, and effect size in active drug groups in both diseases. Ó 2011 International Association for the Study of Pain. Published by Elsevier B.V. All rights reserved.
1. Introduction There is no formal definition of placebo that most clinicians and researchers agree upon. In clinical trials, placebos are generally control treatments with a similar appearance to the study treatments, but without their essential components [18]. The placebo response is defined to be the reduction in a symptom as a result of factors related to a patient’s perception of the placebo intervention [40]. The placebo response is determined by the placebo effect (psychological factors such as expectation of benefit, classical conditioning, verbal suggestions, and behaviours manifested by health care providers) as well as by the natural course of disease and by the study design (eg, regression to the mean, uncontrolled parallel interventions) [8,9,31].
⇑ Corresponding author at: Klinikum Saarbrücken gGmbH, Winterberg 1, D-66119 Saarbrücken, Germany. Tel.: +49 681 9632020; fax: +49 681 9632022. E-mail address: [email protected] (W. Häuser).
There is a growing interest of pain medicine in the placebo response in chronic pain trials [39] for 2 reasons: First, the clinical use of placebo had been advocated in editorials [17] and by influential commentators [4]. In clinical practice, physicians use placebo for pain management [38]. The use of placebos is considered to be justified if there is a chance of a clinical improvement by placebo treatment [12]. Whether placebo induces a clinically relevant and stable improvement in chronic pain syndromes needs to be tested [23]. Second, pharmaceutical companies try to identify modifiable trial characteristics to enhance the ability to detect benefits of pharmacological treatments. Negative results of some placebocontrolled trials in chronic neuropathic pain syndromes had been attributed to unexpectedly large reductions of pain intensity in placebo groups that compromised the ability to show significant greater improvements with active medication [11]. The impact of placebo response on drug response in chronic pain trials had not been quantified until now. Topical reviews on placebo response in painful diabetic peripheral neuropathy (DPN) and postherpetic neuralgia (PZN) trials focussed on study characteristics associated
0304-3959/$36.00 Ó 2011 International Association for the Study of Pain. Published by Elsevier B.V. All rights reserved. doi:10.1016/j.pain.2011.01.050
1710
Ò
W. Häuser et al. / PAIN 152 (2011) 1709–1717
with high placebo response rates or negative study outcomes [7,24,33]. Patient characteristics predicting placebo response in painful DPN had not been studied. Studies in osteoarthritis (OA), migraine, and irritable bowel syndrome came to divergent findings on the impact of patient-related predictors such as age, sex, pain baseline, and continent of the study on placebo response rates [10,27,31,44]. The influence of the quantity and quality of the patient–investigator relationship had only been studied in irritable bowel syndrome (IBS) [5,22]. No studies had been conducted on the magnitude and on any types of predictors of placebo response in fibromyalgia syndrome (FMS). Given these open questions on placebo response in chronic pain syndromes, the aims of this systematic review were to determine the magnitude of placebo response on pain, to determine to what amount the placebo response accounts for the response in the active drug group, and to test for potential characteristics of the patient and the patient–investigator interaction associated with the placebo response, in randomised controlled studies of any pharmacological therapy in patients with FMS and painful DPN.
2. Methods The review was performed according to the PRISMA Statement (Preferred Reporting Items for Systematic Reviews and MetaAnalyses [28], see Online Supplementary Table 1) and the recommendations of the Cochrane Collaboration [16]. 2.1. Hypotheses Based on the literature on psychological mechanisms of placebo response [8,31], studies available on determinants of placebo response on pain, and our experiences with patients in clinical studies, we tested the following assumptions. 2.1.1. Patient characteristics Pain at baseline is associated with placebo response rate as shown in OA trials [44] and lamotrigine studies in chronic neuropathic pain [19]. Because of the limited and contradictory findings in the literature, we had no a priori hypotheses on the impact of age, sex, race, and type of disease. 2.1.2. Study-related characteristics that might influence the selection of patients The year of study initiation is positively associated with placebo response rate, as shown in trials in OA [44]. We assumed a higher placebo response rates in studies conducted in Europe compared with studies conducted in North America, as shown in migraine [27]. We assumed that studies conducted in second- or third-world countries have higher placebo response rates than studies in firstworld countries (North America, Western Europe, Japan) because patients in second- or third-world countries have less access to treatment options than patients in first-world countries and therefore have higher treatment expectations. 2.1.3. Quantity of investigator–patient relationship The number of study visits and treatment duration are associated with placebo response rate as shown in IBS [6]. 2.2. Protocol Methods of analysis and inclusion criteria were specified in advance (review protocol available on request).
2.3. Eligibility criteria 2.3.1. Types of studies Double-blind randomized controlled trials of any size with a parallel or cross-over design were included. Studies without randomization and single-blind studies were excluded. Studies with an enriched enrolment with randomized withdrawal design were excluded because of the potential effects of the study design on placebo effects [36]. No language restrictions were made. 2.3.2. Types of participants Patients with painful DPN and FMS, diagnosed by defined criteria, were included. We excluded studies in which DPN was mixed with other neuropathic pain syndromes if no separate data for DPN were reported because of different placebo response rates in painful DPN and PZN trials [7]. 2.3.3. Types of interventions Randomized controlled trials comparing any type of pharmacological medication with pharmacological placebo were included. Studies with nonpharmacological placebos and with pseudo-placebos (active drug without evidence for effectiveness in the disease of interest) were excluded. Studies that combined pharmacological placebo with any other defined treatment, whose effects on pain were tested for, were excluded. 2.3.4. Types of outcomes measures Studies should assess the patient’s ratings of pain intensity. If more than 1 pain score was used, we preferred the following order for the inclusion into analysis: pain Visual Analogue Scale (VAS) 0 to 100, pain Numeric Rating Scale (NRS) 0 to 100, pain VAS 0 to 10, pain NRS 0 to 10, any other pain VAS or NRS, summary scores including nonpainful symptoms (eg, paresthesia and sleep numbness of the feet). For summary scores, we preferred the following order: Total symptom Score (TTS) [45], any other combined score. If intention-to-treat (ITT) and completer analysis were reported, we used ITT outcomes. 2.4. Data sources and searches 2.4.1. Diabetic DPN We used the search strategy of the guidelines of the National Institute for Clinical Excellence on pharmacological management of neuropathic pain (search of literature until October 2008) [29] for MEDLINE. The search strategy was adopted to SCOPUS, the Cochrane Central Register of Controlled trials (CENTRAL), websites of the U.S. National Institutes of Health (NIH) (www.clinicaltrials.gov), and the Pharmaceutical Research and Manufacturers of America (PhRMA) (www.clinicalstudyresults.org). The search was conducted until July 30, 2010. 2.4.2. FMS We expanded our searches used for the German interdisciplinary guideline on the management of FMS [14] in the electronic bibliography databases detailed above until July 30, 2010 (see Online Supplementary Table 2). For both conditions, we reviewed the reference lists of included articles. For painful DPN, we searched the reference list of a recent topical review on the pharmacological management of neuropathic pain [11]. 2.5. Study selection Two authors independently screened the titles and abstracts of potentially eligible studies identified by the search strategy detailed above. The full text articles were then examined independently by 2 authors to determine whether they met the inclusion criteria. Discrepancies were rechecked and resolved by consensus.
Ò
W. Häuser et al. / PAIN 152 (2011) 1709–1717
2.6. Data collection process and data items A structured coding plan was developed prior to the meta-analysis. Two authors independently extracted the data using standard extraction forms. Discrepancies were rechecked, and consensus was achieved by discussion. If needed, a third author reviewed the data to reach a consensus. The coding plan included the following items. 2.6.1. Outcomes We examined change in pain intensity from baseline to endpoint. For trials with more than 1 dosage group of the active treatment, we selected the group with the highest dosage for comparison. 2.6.2. Study characteristics The following variables were collected from all trials: year of initiation (if unavailable, estimated as 3 years before publication [33]; for analysis we calculated the incremental year of study initiation with earliest study set as 0) and publication of the study, sponsoring by pharmaceutical company (if support was not mentioned, we checked the acknowledgement sections and the affiliations of the authors for study support by pharmaceutical companies), continent in which the study took place (Europe, America, Asia, mixed continent samples), number of study sites (if not reported, we chose the number of clinical institutions of the author list in case of study samples <100 patients), percentage of patients screened/randomised, study design (parallel groups or crossover), placebo run-in period, type of active medication (antidepressants, anticonvulsants; other), application of medication (oral, topical, parenteral), total sample sizes in active treatment and placebo group, treatment duration in weeks, number of study visits (If the number of study visits was not reported in detail, we assumed 1 study visit for baseline assessment and randomisation each and for each other point of assessment reported), type of analysis (ITT analysis), publication status (peer reviewed journal vs only available in database). 2.6.3. Patient characteristics Patient characteristics were age, sex (percentage of females), race (percentage of whites/Caucasians), and pain intensity baseline. 2.7. Validity assessment Two authors independently collected the reported quality assessment of reports of randomised clinical trials in terms of Jadad scores, considering the description and sequence of randomization, the double-blind procedure, the appropriateness of randomisation and double-blinding procedures, and the description of withdrawals and dropouts (range 0 to 5) [20]. Discrepancies were rechecked and resolved by consensus and if needed by a third author. Interrater reliability for study characteristics and validity characteristics was computed. 2.8. Dealing with missing data Where details of study outcomes were missing, attempts were made to obtain these data through contacting 74 trial authors. Additional data were provided by 12 authors. Where final standard deviations (SDs) were not available from trial authors, they were calculated from t-values, confidence intervals, or standard errors, where reported in articles [16]. If these data were not available, the final SD was substituted by the mean of the SDs of studies available that used the same outcome scale [16]. In case of more than 2 studies with the same active medication available, we
1711
substituted by the mean of the SDs of the trials with the same medication, if not by the SDs of all trials available. If the mean and SD of pain final was not reported but mean change with SD was given, the mean final was calculated by subtracting the mean change from pain baseline [16]. We substituted missing SD final by SD mean change, because there is no validated formula available to calculate final SD from SD of mean change. In PhRMA, we found the SDs of pain final and the SDs of mean change of 3 studies with duloxetine and 1 study with pregabalin. The Pearson correlation of SD of pain final and SD of mean change of the 4 placebo arms was r = 0.92. We did not ask for not-reported details of study design (eg, method of randomisation, identity of active drug and placebo) because we were unable to receive these details of studies conducted before 2000 in previous systematic reviews on antidepressants in FMS [15]. 2.9. Risk of bias in individual studies The risk of bias in individual studies was assessed by summarizing the study quality in the Jadad score. 2.10. Statistical analysis Descriptive statistics were used to characterize the features of the included trials. Nonparametrical tests were applied for the comparison of continuous variables, and v2 tests for the comparison of categorical variables. Correlations between continuous variables were calculated by the Pearson correlation coefficients. A 2-sided P value of 6.05 was considered significant. We used SPPS version 17.0 (SPSS, Inc, Chicago, Illinois, United States). Meta-analyses were conducted using RevMan Analyses software (RevMan 5.0.24) of the Cochrane collaboration [37] and metaregression analyses by Comprehensive Meta analysis software [3]. Pain scores from 0 to 10 numerical rating scales and other scales were converted to a 0 to 100 scale. We calculated the pooled weighted mean difference (WMD) between baseline and end-oftreatment scores of pain in placebo and active drug groups using a random effects model. This model is more conservative than the fixed-effects model and incorporates both within-study and between-study variance [26]. We used the benchmarks of the Initiative on Methods, Measurement and Pain Assessment in Clinical Trials (IMMPACT) with <10 points decrease on a scale of 0 to 100 to be no important improvement, 10 to 20 points to be a minimally important improvement, and 20 to 27 points to be a moderately important improvement [6] for interpreting the pooled WMD between baseline and end-of-treatment values of pain in placebo and active drug groups. We calculated pooled standardized mean differences (SMD) between baseline and end-of-treatment values of pain by a random effects model, to compare our results with other systematic reviews on placebo response in chronic pain syndromes [18,44]. The calculation of how much the improvement in the active drug period was attributed to the placebo response was performed according to a systematic review on placebo response in antidepressant trials in depression [35]. The I2 statistic was used to estimate the percentage of total variation across studies because of heterogeneity, rather than by chance (i.e., the percentage of variability of associations across studies that is not due to chance or random error, but rather due to real differences in study patients, design, or outcome definitions). I2 values P50% represent substantial heterogeneity. Tau2 was used to determine how much heterogeneity was explained by subgroup differences [16]. The impact of continuous potential predictors on pain response was tested by metaregression, which was performed using the
1712
Ò
W. Häuser et al. / PAIN 152 (2011) 1709–1717
random-effects model. Tau2 variance was calculated by the method of maximum likelihood [3]. To test hypotheses of subgroup effects in case of categorical variables, a test of interaction with a predetermined 2-tailed P < .05 was used [1]. 2.11. Additional analyses 2.11.1. Subgroup analysis Subgroup analyses were prespecified for the following variables: sponsoring by pharmaceutical company, type of publication (peer reviewed journal vs database), type of active medication (antidepressants, anticonvulsants, other), type of application (oral, topical, parenteral), approval by the U.S. Food and Drug Administration (FDA) for FMS (duloxetine, milnacipran, pregabalin) or painful DPN (duloxetine, pregabalin), study design (cross-over vs parallel; single-blind run-in). These subgroup analyses were also used to examine potential sources of clinical heterogeneity. To test our hypotheses, a test of interaction was performed with studies in America vs Europe and first-world vs second- or third-world countries. 2.11.2. Sensitivity analyses Sensitivity analyses were prespecified for studies with and without imputation of data (SDs estimated by mean of SDs of other studies) and/or data extracted from figures, with and without intention-to-treat-analysis and studies with a low (1 to 2) and moderate (3 to 5) Jadad score. We decided post hoc to perform a sensitivity analysis of studies that assessed pure pain scores and composite neuropathic symptom scores including pain in painful DPN. 2.11.3. Risk of bias across studies Publication bias was controlled for by computing ‘‘safe-n rates’’ [30] and the rank correlation test for funnel plot asymmetry. The test examines the rank correlation between standardized intervention effect and its standard error. An asymmetric funnel plot would give rise to such a correlation and may be indicative of publication bias [2].
3. Results 3.1. Search of literature A total of 1251 studies in FMS and 1034 studies in painful DPN fulfilled the first level of inclusion criteria. After excluding studies based on information presented in study abstracts, 107 complete study reports were considered in more detail in FMS and 112 in painful DPN. The final sample size consisted of 72 studies in FMS and 70 studies in painful DPN (see Online Supplementary Figs. 1 and 2 and Tables 3 through 6).
Table 1 Study and patient characteristics in FMS and DPN trials. FMS Number of trials; N 72 Study initiation year; Mean (SD) 1997 (7.1) Only available in database 4 (5.6) Industrial sponsoring; N (%) 56 (77.8) Continent Europe; N (%) 28 (368.9) America; N (%) 42 (58.3) Asia, N (%) 1 (1.4) Mixed continent; N (%) 1 (1.4) Number of countries; Mean (SD) 1.6 (2.5) Number of study sites; Mean (SD) 12.8 (23.5) Parallel design; N (%) 57 (79.2) Single blind run-in; N (%) 9 (12.5) Treatment duration (weeks); Mean (SD) 10.6 (7.6) Number of study visits; Mean (SD) 6.2 (3.4) Active treatment Antidepressants; N (%) 26 (36.1) Anticonvulsants; N (%) 5 (6.9) Other; N (%) 41 (56.9) Application medication Oral; N (%) 58 (80.6) Parenteral; N (%) 10 (13.9) Local; N (%) 4 (5.6) Total sample size; Mean (SD) 176 (269) % Women; Mean (SD) 93.0 (10.1) % Caucasian; Mean (SD) 90.3 (13.5) Mean age; Mean (SD) 47.6 (3.8) Pain baseline (0–10); Mean (SD) 6.5 (0.8)
Painful DPN P 70 1999 (1.0) 9 (12.9) 53 (75.7) 17 (24.3) 35 (50.0) 9 (12.9) 9 (12.9) 2.3 (3.4) 22.2 (30.1) 60 (85.7) 14 (20) 11.6 (8.6) 6.2 (3.0)
NA 0.25 0.12 0.77 0.005
0.006 0.04 0.35 0.22 0.36 0.34 <0.0001
10 (14.3) 28 (40.0) 32 (45.7) 0.73 56 (80.0) 9 (12.9) 5 (7.1) 200 (195) 45.2 (16.2) 76.5 (29.8) 58.3 (3.8) 6.4 (1.0)
0.03 <0.0001 <0.0001 <0.0001 0.25
FMS, fibromyalgia syndrome; DPN, painful diabetic neuropathy; NA, not applicable, SD, standard deviation.
3.2.2. Outcomes All FMS trials and 93% of the painful DPN trials assessed pain by visual analogue or numeric rating scales. Pain intensity was most often based on ratings of average pain in the past 24 h, and the remainder were almost all ratings of current pain. In 7% of the painful DPN trials, only a summary score including other neuropathic symptoms such as paraesthesia and numbness could be analysed. We were able to detect complete means and SD scores for 78% of the FMS and 84% of the painful DPN trials. Therefore we decided that a substitution of missing values was justified. 3.2.3. Reported study quality The means of the Jadad scores did not differ between FMS (mean[SD] = 2.9 [1.8]) and painful DPN (mean[SD] = 2.8 [1.7]) trials (P = .79). 3.2.4. Validity assessments The interrater reliability for the assessments of the study sample and outcome characteristics ranged from 0.85 to 0.91 and was for study quality r = 0.95.
3.2. Study characteristics
3.3. Overall effect size on pain in placebo groups
3.2.1. Study design and patient characteristics There were no significant differences in the year of study initiation, type of publication, type of study design, treatment duration, number of study visits, type of application of medication, and baseline pain scores between the 2 diseases. Trials with painful DPN significantly included more study sites and countries, were more frequently conducted in mixed continent samples, more frequently used anticonvulsants and less antidepressants, had greater total sample sizes, and included more older, male, and non-Caucasians patients than FMS trials (Table 1 and Online Supplementary Tables 3 and 4).
The pooled WMD between pain baseline and end of treatment in the FMS lacebo groups was 7.69 (95% confidence interval [CI] 6.10 to 9.29; I2 = 71%) (see Online Supplementary Fig. 3) and in the FMS active drug group was 17.11 (95% CI 16.41 to 17.90; I2 = 84%) on a pain scale of 0 to 100. The WMD of placebo on pain indicated no important improvement, and the WMD of active drug on pain indicated minimally important improvement according to IMMPACT benchmarks [6]. The pooled SMD on pain in the FMS placebo groups was 0.42 (95% CI 0.35 to 0.49; I2 = 53%), as compared with 0.82 (95% CI 0.72 to 0.92; I2 = 77%) in the active drug group. There was a high linear correlation between the WMDs (r = 0.69,
Ò
1713
W. Häuser et al. / PAIN 152 (2011) 1709–1717
P < .0001) and the SMDs (r = 0.73, P < .001) of the placebo and active drug groups, indicating that reported improvements in placebo and drug groups were highly interdependent. The ratio of the WMDs suggests that 45% of the improvement in the active drug group was attributable to the placebo response based on the assumption that the active drug effect is additive to the one of placebo. The pooled WMD between pain baseline and end of treatment in painful DPN placebo groups was 13.96 (95% CI 11.93 to 15.99, I2 = 85%) (see Online Supplementary Fig. 4) and in active drug groups was 22.54 (95% CI 20.49 to 24.58, I2 = 85%) on a 0 to 100 pain scale. The WMD of placebo on pain indicated minimally important improvement, and the WMD of active drug on pain indicated moderately important improvement according to IMMPACT benchmarks [6]. The overall SMD on pain in the painful DPN placebo groups was 0.71 (95% CI 0.61 to 0.81; I2 = 81%) as compared with d = 1.14 (95% CI 1.02 to 1.25; I2 = 85%) in the active drug group. There was a moderate linear correlation between the WMDs (r = 0.47, P < .001) and a high correlation of the SMDs (r = 0.52, P < .001) of the placebo and active drug groups, indicating that reported improvements in placebo and active drug groups were moderately interdependent. The ratio of the WMDs of placebo and active drug suggests that 62% of the improvements in the active drug groups were attributable to the placebo response based on the assumption that the active drug effect is additive to the one of placebo. The test of interaction of the effect size of placebo on pain in FMS and painful DPN trials was significant (P < .001).
3.4. Subgroup analysis The effect size of placebo on pain in trials conducted in America and Europe did not differ in FMS (P = .66) and painful DPN trials (P = .23). The effect size of placebo on pain of studies that included second- or third-world countries and studies conducted only in first-world countries did not differ in painful DPN (P = .61), but did differ in FMS trials (P = .01) (Table 2). 3.5. Sensitivity analysis Excluding studies with substituted or extracted data from figures, no intention-to-treat analysis, low Jadad score, and combined neuropathic symptom scales in painful DPN did not change the significant effect of placebo on pain in FMS and painful DPN trials (Table 3). 3.6. Metaregression analysis Simple linear regressions showed that pain baseline (b = 0.28, P = .003), incremental year of study initiation (b = 0.31, P = .008), number of study sites (b = 0.09, P = .0001), and effect of active medication on pain (b = 0.22, P = .003) were associated with the pooled WMD of placebo on pain in FMS (Table 4). Simple linear regressions showed that pain baseline (b = 0.25, P = .007), incremental year of study initiation (b = 0.38, P = .002), and effect of active medication on pain (b = 0.60, P < .0001) were associated with the pooled WMD of placebo on pain in painful DPN (Table 4).
Table 2 Subgroup analysis of the effect size (WMD baseline – end of treatment) in placebo groups in FMS and DPN. Outcome title
FMS
Painful DPN
Number of studies N (%)
WMD (95% CI)
Hetero geneity (I2 [%]; Tau)
Number of studies
WMD (95% CI)
Hetero geneity (I2; Tau)
Industrial sponsoring Yes No
56 16
8.98 (7.51–10.44) 3.86 ( 0.65–8.37)
59; 13.34 71; 53.59
53 17
15.04 (12.85–17.23) 9.25 (4.01–14.50)
86; 51.15 72; 72.84
Type of publication Peer-reviewed journal Database
68 4
7.41 (5.70–9.12) 10.50 (7.92–13.07)
72; 27.72 0; 0
61 9
13.20 (10.81–15.58) 17.64 (14.51–20.77)
85; 65.52 75; 17.03
7.89 (5.73–10.04) 7.15 (4.57–9.73) 6.00 ( 3.10–15.10) 10.00 (6.22–13.78)
77; 30.08 55; 21.21
35 17 9 9
15.25 10.69 11.13 18.05
32; 95; 85; 64;
12.62 (8.68–16.55) 7.14 (5.44–8.83)
66; 11.39 70; 26.39
13 57
15.30 (8.96–21.65) 13.59 (11.88–15.31)
93; 116.22 70; 25.39
Continent America 42 Europe 28 Asia 1 Mixed continent 1 Patients of second- or third-world countries Yes 5 No 67
(13.84–16.66) (3.36–18.01) (5.13–17.13) (15.09–21.01)
5.04 220.46 62.24 11.86
Type of active medication Antidepressants Anticonvulsants Other
26 5 41
8.32 (6.00–10.64) 11.75 (9.61–13.88) 6.51 (4.31–8.72)
67; 17.76 28; 1.65 59; 24.12
10 28 32
17.45 (13.57–21.33) 14.29 (12.45–16.12) 12.48 (8.04–16.92)
77; 26.09 61; 13.98 90; 125.87
FDA approved Yes No
13 59
12.04 (10.72–13.36) 6.42 (4.23–8.61)
2; 36.00 67; 40.55
14 56
16.89 (14.35–19.44) 12.97 (10.37–15.56)
72; 16.68 86; 71.25
Type of application Oral Parenteral Topical
58 10 4
7.92 (6.23–9.62) 8.08 (1.35–14.82) 5.67 ( 0.87–12.22)
72; 23.16 70; 71.15 57; 23.91
56 9 5
14.74 (13.14–16.34) 9.43 ( 3.06–21.91) 10.57 ( 0.17–21.30)
70; 22.59 97; 325.93 61; 83.70
Study design Parallel Cross-over
57 15
9.09 (7.58–10.60) 2.41 ( 0.33–5.16)
59; 14.62 37; 8.86
60 10
15.00 (12.93–17.07) 4.44 ( 1.46–10.35)
85; 49.88 46; 39.08
Single-blind run-in Yes No
9 63
8.11 (4.35–11.87) 7.59 (5.80–9.38)
84; 25.85 68; 26.39
14 56
15.06 (11.31–18.08) 13.68 (11.31–16.06)
76; 34.74 86; 60.00
WMD, weighted mean difference; CI, confidence interval; FMS, fibromyalgia syndrome; DPN, diabetic peripheral neuropathy; SMD, standardized mean difference; FDA, US Food and Drug Administration.
Ò
1714
W. Häuser et al. / PAIN 152 (2011) 1709–1717
Table 3 Sensitivity analysis of the effect size (WMD baseline – end of treatment) in placebo groups in FMS and DPN. Outcome title
FMS
Painful DPN
Number of studies N (%) Data extracted from figures and/or substituted Yes 16 (22.2) No 56 (77.8)
WMD (95% CI)
Hetero geneity (I2 [%]; Tau)
Number of studies
WMD (95% CI)
Hetero geneity (I2; Tau)
4.89 (1.75–8.03) 8.27 (6.51–10.02)
26; 10.10 74; 25.07
12 58
13.50 (9.18–17.81) 14.06 (11.76–16.36)
79; 39.62 85; 58.60
Pain outcome measure Pain only Mixed outcome measure Intention-to-treat analysis Yes No
72 0
7.69 (6.10–9.29)
71; 25.11
65 5
13.64 (11.94–15.33) 20.15 (6.38–33.93)
74; 30.41 96; 223.76
37 35
9.16 (7.39–10.93) 6.17 (3.58–8.57)
62; 14.47 69; 31.31
49 21
15.10 (12.84–17.35) 9.94 (6.40–13.48)
85; 48.80 67; 35.58
Jadad score Low (1–2) Moderate (3–5)
26 46
10.08 (7.43–12.72) 6.51 (4.62–8.39)
67; 22.76 68; 21.99
28 42
14.13 (11.68–16.94) 14.05 (11.22–16.89)
70; 28.08 88; 66.51
WMD, weighted mean difference; CI, confidence interval; FMS, fibromyalgia syndrome; DPN, diabetic peripheral neuropathy.
3.7. Risks of bias 3.7.1. Heterogeneity There was a substantial statistical heterogeneity of analysis for the effect size on pain in FMS and painful DPN trials. Heterogeneity was reduced <50% in case of studies only available in databases, anticonvulsants as active drug, FDA-approval, and cross-over design in FMS. The effect of placebo on pain remained significant except in trials without industrial sponsoring, studies conducted in Asia, those with topical application, and those with a cross-over design (Table 2). Heterogeneity was reduced <50% in case of studies conducted in America and cross-over design in painful DPN. The effect of placebo on pain remained significant except in trials with a cross-over design and with topical and parenteral application (Table 2). Subgroup analyses were not corrected for associations with other factors. 3.7.2. Reported study quality By metaregression, there was no significant association of the Jadad score and pooled WMD of placebo on pain in painful DPN trials (b = 0.01, P = .64). In FMS trials, there was a significantly negative association between the Jadad score and the pooled WMD on pain (b = 1.39, P = .004), indicating that low study quality such as inadequate randomisation or blinding might be associated with higher placebo response rates. 3.7.3. Publication bias Kendall’s tau of the Begg rank correlation test of the pooled WMD (active drug end and placebo end) was not significant in
FMS (P = .39) and painful DPN (P = .43), suggesting no relevant publication bias. In FMS, 315 unpublished studies reporting no positive placebo response would be needed to reduce the pooled WMD of placebo to 1.0. In painful DPN, 1038 [42] unpublished studies reporting no positive placebo response would be needed to reduce the pooled WMD of placebo to 1.0 (9.9). These rates confirm the stability of improvements in the placebo groups. 4. Discussion 4.1. Summary of main findings We found no important improvement of pain in placebo groups in FMS and minimally important improvement of pain in placebo groups in painful DPN trials according to IMMPACT benchmarks. We found that 45% (62%) of the effect size of active medication groups in FMS (painful DPN) was attributable to the placebo response. The placebo response was associated with year of study initiation, pain baseline, and effect size in active drug groups in both diseases. 4.2. Comparison with other reviews on placebo response in drug therapy in chronic pain 4.2.1. Magnitude and predictors of placebo response rates Our analysis of drug trials in painful DPN and FMS supports the conclusion of a systematic review on placebo in different clinical conditions that the mean effect size for placebos in medicine is small in magnitude [18]. The pooled effect of placebo on pain was 3.2 (95% CI 1.6 to 4.7) points on a 100-point scale in an
Table 4 Predictors of effect size (WMD baseline – end of treatment) in placebo groups in FMS and DPN by metaregression analyses. Predictor
FMS Degrees of freedom
Age Female sex Caucasian race Pain baseline Year of study initiation (incremental) Number of study sites Number of countries Treatment duration Number of study visits Effect of active treatment
66 65 36 71 71 71 71 71 71 71
Painful DPN Coef. (B) 0.0001 0.02 0.005 0.28 0.31 0.09 0.02 0.26 0.005 0.22
P
Degrees of freedom
.88 .83 .29 .003 .008 .0001 .21 .13 .66 .003
69 63 43 64 68 68 69 69 68 69
Coef. (B) 0.005 0.01 0.06 0.25 0.38 0.01 0.40 0.003 0.35 0.60
P .99 .90 .09 .007 .002 .90 .10 .60 .24 <.0001
Significant observations (P < .05) are in boldface type. WMD, weighted mean difference; FMS, fibromyalgia syndrome; DPN, diabetic peripheral neuropathy; Coef (B), regression coefficient of each regression representing the slope of each model.
Ò
W. Häuser et al. / PAIN 152 (2011) 1709–1717
analysis of studies with several acute and chronic pain syndromes [21], and thus lower than the ones we found in FMS and painful DPN. The SMD of placebo on pain compared with untreated controls in OA trials was 0.51 (95% CI 0.46 to 0.55) [44] and thus between the pre- and post-SMD values of placebo on pain in FMS and painful DPN trials. The substantial heterogeneity of the effect size of placebo on pain in both diseases confirms findings from previous analysis of chronic neuropathic pain trials that the extent of response in the placebo group is highly variable [11,19]. In line with systematic reviews on placebo response in OA [44] and 1 pooled analysis of 3 lamotrigine trials in neuropathic pain [19], the pain-relieving effect of placebo was positively associated with pain baseline scores in FMS and painful DPN. In accordance with childhood and adolescence migraine trials, we found no association between age and sex and placebo response rates [11]. In contrast to our results, higher placebo response rates were associated with higher age in irritable bowel syndrome (IBS) [31]. In contrast to studies on placebo response in the prophylaxis of migraine [27], we found no differences in the effect sizes of placebo in studies conducted in North America and in Europe. Higher placebo response rates were associated with a longer duration of the study and a greater number of study visits in 1 IBS study [5]. These associations failed to reach significance in FMS and painful DPN trials. However, it should be noted that the placebo response was stable over time in DPN as in other chronic neuropathic pain syndromes [33]. The strong effect of publication year on placebo response, which was also found in trials in OA [44] and depression [35], is difficult to explain. Year of study initiation, number of patients, study sites, and countries included were significantly correlated due to large industry-sponsored studies on duloxetine and pregabalin in both diseases (details not shown). Waber et al. [42] were able to show that the analgesic placebo effect is higher if people expect the drug to be expensive and valuable. The placebo response rates in trials of drugs licensed by the FDA for FMS were higher than in nonFDA approved medication as well as in studies including secondor third-world countries compared with studies that included only first-world countries in FMS (Table 2). We speculate that large multicenter trials of well-known pharmaceutical companies with expensive medications including second- and third world countries induce high treatment expectations, resulting in higher placebo response rates. 4.2.2. Is the placebo response attributable to a placebo effect? A recent Cochrane review on placebo interventions for all clinical conditions argued that the effects of placebo interventions could be, at least in part, attributed to the natural course of disease and regression to the mean. However, by comparing patients who were randomly allocated to placebo group or no treatment group (natural course of disease), the authors found a statistically significant SMD of placebo on pain that was very variable, even among trials with a low risk of bias (hedge’s g = 0.25; 95% CI 0.18 to 0.35) [18]. The association between high baseline pain scores and high placebo response rates in FMS and painful DPN trials can support the assumption that the placebo effect is partially a regression to the mean [35]. However, the effect of placebo on pain was positively associated with the year of study initiation, too. Because year of study initiation and intensity of pain baseline were not significantly associated (details not shown), we hypothesize that this association can be explained by contextual factors. Therefore, we speculate that the response rates in placebo groups in FMS and painful DPN on pain are attributable to a placebo effect and regression to the mean.
1715
4.3. Mechanisms of placebo response Rapaport et al. [34] and others assume that placebo responders differ from drug responders, therefore indicating that the mechanisms of improvement under placebo might differ from the mechanisms of improvement under active treatment. However, we found a strong correlation between effects in placebo and active drug groups. This strong correlation is a contradiction to the assumption that placebo responders and drug responders are different. We can only speculate on the mechanisms underlying the different placebo response rates in painful DPN and FMS. Dworkin et al. [7] hypothesized that the placebo effect requires intact descending pathways, which are more likely to be present in DPN, a disorder predominantly localized to the peripheral nervous system, than in patients with PZN, who can have lesions in the spinal cord as well. Therefore we speculate that the placebo effect in FMS is limited because of deficient descending pain pathways, even though an experimental study demonstrated that pain relief through expectation superseded descending inhibitory deficits in fibromyalgia patients [13]. 5. Limitations 5.1. Methods Major limitations of meta-analyses are publication bias, quality of the included studies, and insufficient data [35]. We found no indicators for publication bias. We had to exclude some studies because of lacking data. The excluded studies were mainly studies with small sample sizes that had been conducted before 2000. Their results would probably not have substantially changed the results. In contrast to a recent topical review on placebo and treatment group responses in PZN vs painful diabetic peripheral neuropathy clinical trials that did not perform a meta-analysis with random-effect model because of incomplete outcome reporting [8], we were able to extract sufficient data for meta-analysis by substituting missing values. Sensitivity analyses demonstrated that the effect on pain was robust against these procedures. Some of the studies included did not report in detail the characteristics of their study samples, which limited the power of metaregressions of age, sex, and race with effect size on pain. Heterogeneity was substantial for most comparisons and only partially explainable by subgroup comparisons. The reasons for heterogeneity remain unknown. Our method to calculate the percentage of improvement attributable to placebo in the active drug group is not supported by studies and should therefore be interpreted cautiously. Our analysis might underestimate the potential effects of placebo for 2 reasons. First, we performed a secondary analysis of comparison arms of drug treatment. The patients in placebo and real drug groups were aware of the possibility of being treated with placebo. The effects in the placebo groups might have been higher if they would have been designed as treatment interventions (without possibility of a placebo treatment). Second, the average placebo effect cannot be used to deduce the magnitude of individual placebo responses. Individuals who experience a placebo response (eg, at least 30% pain reduction) are likely to have a much higher placebo response than the average effect reported here [41]. 5.2. Variables analysed Meta-analyses are limited in providing understanding of factors that contribute to placebo analgesia because they do not systematically vary factors that affect its magnitude [32]. We could not
1716
Ò
W. Häuser et al. / PAIN 152 (2011) 1709–1717
analyse the impact of rescue medication on placebo response rates because the amount was not reported in the trials. We could not assess patient-related predictors of placebo response rates such as personality characteristics and treatment expectations that had influenced pain response rates in IBS [25] and low back pain trials [43] because these variables had not been assessed in the studies. Moreover, we could only analyse the impact of the quantity, but not the quality, of the patient–investigator relationship. A recent study in IBS demonstrated that the effects of placebo acupuncture could be augmented compared with placebo acupuncture alone by a patient–practitioner relationship augmented by warmth, attention, and confidence [22]. Finally, verbal suggestions and behaviours manifested by health care providers and written information of the patient are likely to vary greatly across research contexts, and consequently generate considerable variability in the placebo effect itself [32]. 6. Conclusions A minimally important improvement in pain according to clinical benchmarks was only found in painful DPN trials. Therefore, placebo cannot be recommended for the management of chronic pain. Situational factors were important for the placebo response rather than individual factors (eg, age, race, and gender). These contextual factors, such as positive expectations and suggestions, should be used in clinical practice to empower the effects of active drug therapy. Conflict of interest statement Dr. Häuser has received honoraria for educational lectures from Eli-Lilly, Janssen-Cilag, Mundipharma, and Pfizer, and congress travel support from Eli Lilly. Dr. Toelle has worked as a consultant and/or speaker for Astellas, Gruenenthal, Eli Lilly, Boehringer Ingelheim Pharmaceuticals, Pfizer, UCB Pharma, and Mundipharma. The other authors have no conflict of interest to declare. Appendix A. Supplementary data Supplementary data associated with this article can be found, in the online version, at doi:10.1016/j.pain.2011.01.050. References [1] Altman DG, Bland JM. Interaction revisited: the difference between two estimates. BMJ 2003;326:219. [2] Begg CB, Mazumdar M. Operating characteristics of a rank correlation test for publication bias. Biometrics 1994;50:1088–101. [3] Comprehensive metaanalysis software. Version 2.0. Englewood, NJ: Biostat; 2010. [4] Cochrane AL. Effectiveness and efficiency. Random reflections on health services. Cambridge: Cambridge University Press for British Medical Journal & The Nuffield Provincial Hospitals Trust, 1989. p. 31. [5] Dorn SD, Kaptchuk TJ, Park JB, Nguyen LT, Canenguez K, Nam BH, Woods KB, Conboy LA, Stason WB, Lembo AJ. A meta-analysis of the placebo response in complementary and alternative medicine trials of irritable bowel syndrome. Neurogastroenterol Motil 2007;19:630–7. [6] Dworkin RH, Turk DC, Wyrwich KW, Beaton D, Cleeland CS, Farrar JT, Haythornthwaite JA, Jensen MP, Kerns RD, Ader DN, Brandenburg N, Burke LB, Cella D, Chandler J, Cowan P, Dimitrova R, Dionne R, Hertz S, Jadad AR, Katz NP, Kehlet H, Kramer LD, Manning DC, McCormick C, McDermott MP, McQuay HJ, Patel S, Porter L, Quessy S, Rappaport BA, Rauschkolb C, Revicki DA, Rothman M, Schmader KE, Stacey BR, Stauffer JW, von Stein T, White RE, Witter J, Zavisic S. Interpreting the clinical importance of treatment outcomes in chronic pain clinical trials: IMMPACT recommendations. J Pain 2008;9:105–21. [7] Dworkin RH, Turk DC, Peirce-Sandner S, McDermott MP, Farrar JT, Hertz S, Katz NP, Raja SN, Rappaport BA. Placebo and treatment group responses in postherpetic neuralgia vs painful diabetic peripheral neuropathy clinical trials in the REPORT database. Pain 2010;150:12–6. [8] Enck P, Benedetti F, Schedlowski M. New insights into the placebo and nocebo responses. Neuron 2008;59:195–206.
[9] Ernst E, Resch KL. Concept of true and perceived placebo effects. BMJ 1995;311:551–3. [10] Evers S, Marziniak M, Frese A, Gralow I. Placebo efficacy in childhood and adolescence migraine: an analysis of double-blind and placebo-controlled studies. Cephalalgia 2009;29:436–44. [11] Finnerup NB, Sindrup SH, Jensen TS. The evidence for pharmacological treatment of neuropathic pain. Pain 2010;150:573–81. [12] Finniss DG, Kaptchuk TJ, Miller F, Benedetti F. Biological, clinical, and ethical advances of placebo effects. Lancet 2010;375:686–95. [13] Goffaux P, de Souza JB, Potvin S, Marchand S. Pain relief through expectation supersedes descending inhibitory deficits in fibromyalgia patients. Pain 2009;145:18–23. [14] Häuser W, Eich W, Herrmann M, Nutzinger DO, Schiltenwolf M, Henningsen P. Fibromyalgia syndrome: classification, diagnosis, and treatment. Dtsch Arztebl Int 2009;106:383–91. [15] Häuser W, Petzke F, Üceyler N, Sommer C. Comparative efficacy and acceptability of amitriptyline, duloxetine and milnacipran in fibromyalgia syndrome—a systematic review with meta-analysis. Rheumatology 2011;50:532-43. [16] Higgins JPT, Green S. Cochrane handbook for systematic reviews of intervention. Version 5.01. Available at: www.cochrane.org/resources/ handbook/handbook.pdf. [accessed 30.07.10]. [17] Ho VMS. The placebo effect: can we use it better? BMJ 1994;309:69–70. [18] Hróbjartsson A, Gøtzsche PC. Placebo interventions for all clinical conditions. Cochrane Database Syst Rev 2010: 20; CD003974. [19] Irizarry MC, Webb DJ, Ali Z, Chizh BA, Gold M, Kinrade FJ, Meisner PD, Blum D, Silver MT, Weil JG. Predictors of placebo response in pooled lamotrigine neuropathic pain clinical trials. Clin J Pain 2009;25:469–76. [20] Jadad AR, Moore RA, Carroll D, Jenkinson C, Reynolds JM, Gavaghan DJ, McQuay HJ. Assessing the quality of reports of randomized clinical trials: is blinding necessary? Control Clin Trials 1996;17:1–12. [21] Kamper SJ, Machado LA, Herbert RD, Maher CG, McAuley JH. Trial methodology and patient characteristics did not influence the size of placebo effects on pain. J Clin Epidemiol 2008;61:256–60. [22] Kaptchuk TJ, Kelley JM, Conboy LA, Davis RB, Kerr CE, Jacobson EE, Kirsch I, Schyner RN, Nam BH, Nguyen LT, Park M, Rivers AL, McManus C, Kokkotou E, Drossman DA, Goldman P, Lembo AJ. Components of placebo effect: randomised controlled trial in patients with irritable bowel syndrome. BMJ 2008;336:999–1003. [23] Kaptchuk TJ, Kelley JM, Deykin A, Wayne PM, Lasagna LC, Epstein IO, Kirsch I, Wechsler ME. Do ‘‘placebo responders’’ exist? Contemp Clin Trials 2008;29:587–95. [24] Katz J, Finnerup NB, Dworkin RH. Clinical trial outcome in neuropathic pain: relationship to study characteristics. Neurology 2008;70:263–72. [25] Kelley JM, Lembo AJ, Ablon JS, Villanueva JJ, Conboy LA, Levy R, Marci CD, Kerr CE, Kirsch I, Jacobson EE, Riess H, Kaptchuk TJ. Patient and practitioner influences on the placebo effect in irritable bowel syndrome. Psychosom Med 2009;71:789–97. [26] Laird NM, Mostellier F. Some statistical methods for combining experimental results. Int J Technol Assess Health Care 1990;6:5–30. [27] Macedo A, Baños JE, Farré M. Placebo response in the prophylaxis of migraine: a meta-analysis. Eur J Pain 2008;12:68–75. [28] Moher D, Liberati A, Teztlaff J, Altman G. And the PRISMA Group. Preferred reporting items for systematic reviews and meta-analyses: the PRISMA statement. Ann Int Med 2009;51:1–7. [29] National Institute for Health and Clinical Excellence. Neuropathic pain. The pharmacological management of neuropathic pain in adults in non-specialist settings. Available at:
Ò
W. Häuser et al. / PAIN 152 (2011) 1709–1717 [41] Vase L, Petersen GL, Riley 3rd JL, Price DD. Factors contributing to large analgesic effects in placebo mechanism studies conducted between 2002 and 2007. Pain 2009;145:36–44. [42] Waber RL, Shiv B, Carmon Z, Ariely D. Commercial features of placebo and therapeutic efficacy. JAMA 2008;299:1016–7. [43] Wasan AD, Kaptchuk TJ, Davar G, Jamison RN. The association between psychopathology and placebo analgesia in patients with discogenic low back pain. Pain Med 2006;7:217–28.
1717
[44] Zhang W, Robertson J, Jones AC, Dieppe PA, Doherty M. The placebo effect and its determinants in osteoarthritis: meta-analysis of randomised controlled trials. Ann Rheum Dis 2008;67:1716–23. [45] Ziegler D, Hanefeld M, Ruhnau KJ, Meißner HP, Lobisch M, Schütte K, Gries FA. The ALADIN Study Group. Treatment of symptomatic diabetic peripheral neuropathy with the anti-oxidant-lipoic acid: a 3-week multicentre randomized controlled trial (ALADIN Study). Diabetologia 1995;38:1425–33.