The Journal of Pain, Vol 16, No 2 (February), 2015: pp 99-115 Available online at www.jpain.org and www.sciencedirect.com

Critical Reviews Placebo Responses in Long-Standing Complex Regional Pain Syndrome: A Systematic Review and Meta-Analysis Gashirai K. Mbizvo,* Sarah J. Nolan,y Turo J. Nurmikko,*,z and Andreas Goebel*,z *Pain Research Institute, Department of Translational Medicine, University of Liverpool, Liverpool, United Kingdom. y Department of Biostatistics, University of Liverpool, Liverpool,United Kingdom. z Department of Pain Medicine, Walton Centre NHS Foundation Trust, Liverpool, United Kingdom.

Abstract: The typical placebo response (ie, the nonspecific effects in the placebo group including benign natural course, regression to the mean, expectation/conditioning effects, and others) in randomized trials in complex regional pain syndrome (CRPS) is unknown. We recently observed a surprising near-absence of placebo response in a randomized controlled trial we conducted on patients with long-standing ($6 months) CRPS. To investigate the idea that there may be an absence of placebo response in long-standing CRPS further, we conducted a systematic review and meta-analysis of placebo responses in randomized controlled trials conducted in patients with CRPS of $6 months. We systematically identified suitable randomized controlled trials published between 1966 and September 2013. We calculated the mean difference and standard error of the mean difference for placebo responses and synthesized individual effect sizes at 4 specified time periods of interest (15–30 minutes, 1 week, 3–4 weeks, and 6 weeks or more) via meta-analysis using the method of inverse-variance. Heterogeneity was assessed according to the I2 statistic. For primary analysis, we pooled trial-specific effect sizes over the 4 time points. We analyzed data from 340 participants from 18 trials out of a possible 361 participants from 20 trials (94% of participants analyzed). Significant heterogeneity was present between trials; therefore, we interpreted trends from visual inspection of individual trials and pooled estimates. Placebo response was significant at the earliest time period (15–30 minutes). There was no significant evidence of placebo response at any of the other time periods. These results inform the design of future trials, and they caution against the ‘‘therapeutic’’ use of placebo in long-standing CRPS. Perspective: In this meta-analysis of placebo responses in randomized controlled trials in long-standing CRPS, published during 1966 to 2013, we found no evidence for placebo analgesia, except at very early time points. Results inform the design of future placebo analgesia research in long-standing CRPS. ª 2015 by the American Pain Society Key words: Complex regional pain syndrome, reflex sympathetic dystrophy, placebo response, randomized controlled trials, meta-analysis.

T

he term placebo response or placebo effect in clinical trials usually describes the combination of all effects that are not specific to the active trial inter-

vention. In analgesia trials, the placebo response is the sum of pain relief due to a pain condition’s benign natural course, pain regression to the mean,65 and a ‘‘true

G.K.M. and S.J.N. are joint first authors. G.K.M. reports no disclosures. S.J.N. is funded as part of a 3-year research program ‘‘Clinical and Cost Effectiveness of Interventions for Epilepsy in the National Health Service (NHS),’’ which receives financial support from the National Institute for Health Research . T.J.N. serves/has served on scientific advisory boards for Grunenthal, Astellas, Janssen, and Pharmleads; has received speaker honoraria from Pfizer, Astellas and Medtronic; serves on the editorial boards of Pain, Clinical Journal of Pain, Current Pain and Headache Reports, and Neuromodulation; and receives research support from National Institute for Health Research (United Kingdom), Pain Relief Foundation (United Kingdom), and Pfizer Ltd. A.G. has received research support from CSL-Behring, BPL, Biotest, Talecris, Baxter and Grifols, Pfizer, travel support from CSL-Behring and Baxter, speaker honoraria from

Baxter, and consultancy fees from Biotest and Axsome. The authors report no conflicts of interest. All authors contributed substantially to this work. Supplementary data accompanying this article are available online at www.jpain.org and www.sciencedirect.com. Address reprint requests to: Gashirai K. Mbizvo, MBChB, MRes, Pain Research Institute, Department of Translational Medicine, University of Liverpool, Lower Lane, Fazakerley, Liverpool L9 7AL, United Kingdom. E-mail: [email protected] 1526-5900/$36.00 ª 2015 by the American Pain Society http://dx.doi.org/10.1016/j.jpain.2014.11.008

99

100

The Journal of Pain

placebo mechanism.’’ The latter is of context-related magnitude32,64 and caused by mixed factors, including expectation and conditioning.16 Since Beecher’s pivotal studies in the 1950s, placebo responses in pain conditions have generally been seen as particularly large.2,3,60 However, these studies were based predominantly on the review of results from acute pain trials, and often the definition of a placebo response was based on the percentage of patients who had any pain relief with placebo, rather than on the percentage of pain relief in the placebo-treated group.2,3,60 More recent investigations on placebo responses in chronic pain trials have suggested that specific chronic pain conditions have specific expected placebo responses that are generally much smaller than those reported in the earlier trials.15,28,32,36,52 Understanding the sizes of condition-specific placebo responses is important because such information can influence the choice of condition for new compound testing and the design of clinical trials, where, for example, smaller sample sizes will be required when conducting power calculations if smaller placebo responses are expected.36 Understanding the size of placebo response also helps determine whether a placebo could be used to achieve therapeutic effects.35 Finally, placebo response sizes may also convey information about disease-associated neurobiological processes.16 Complex regional pain syndrome (CRPS) is a severe chronic pain, predominately affecting distal limbs after trauma.21,42 Most patients with CRPS improve within the first few months after disease onset, so their response to placebo in a clinical trial would be expected to be large because of the benign natural disease course.11 We recently observed a surprising near-absence of placebo response in our randomized controlled trial (RCT) in patients with CRPS of more than 6 months’ duration.23 Other authors have made similar observations in their trials,44,56 with, for example, Munts et al44 reporting an overall placebo response of 0% in their trial. We wondered whether it might be the case that long-standing CRPS does not demonstrate placebo response, and therefore we have conducted a systematic review and meta-analysis of placebo responses across RCTs in patients with longstanding ($6 months) CRPS. To our knowledge, this is the first review of placebo response in CRPS.

Methods Systematic Search Strategy and Selection of Trials RCTs for the treatment of CRPS were identified from the reference lists of trials from 2 systematic reviews of treatments for CRPS. A review by Forouzanfar et al19 identified 28 RCTs published from 1966 to June 2000, and a subsequent review by our group10 identified 43 RCTs from July 2000 to February 2012 using similar methodology; see Supplementary Appendix A for the detailed search methodology for both reviews.

Placebo Responses in Long-Standing CRPS We also repeated the search outlined in Cossins et al10 for the time period from March 2012 to September 2013, searched the reference lists of included trials, and searched trial registries for any details of ongoing trials or unpublished work. Two authors (G.K.M. and S.J.N.) independently screened all eligible trials according to the inclusion criteria below. Any disagreements were resolved by mutual discussion or through adjudication via the senior author (A.G.).

Inclusion Criteria  RCTs for the treatment of CRPS type I or II in adults B We included trials in which a diagnosis CRPS was made using the original International Association for the Study of Pain (IASP) criteria,3 the more recent Budapest criteria (ie, the ‘‘new IASP criteria’’),27 or criteria for the diagnosis of reflex sympathetic dystrophy.17 10,19 B Trials in the reviews conducted in mixed populations of CRPS and non-CRPS patients were included if results were available for CRPS patients as a subgroup. B Trials of all sample sizes were included; we did not specify a minimum sample size for inclusion.  Single-blinded (participants) or double-blinded (participants and investigators) trials B Inadequate blinding of trials may introduce performance and/or detection bias into results; therefore, we made a judgment on the adequacy of blinding in each of the trials (high, low, or unclear risk of bias of blinding; see Supplementary Appendix B for details).  Trials of a parallel or crossover design B In trials of a crossover design, we made a judgment on the presence of a possible carryover effect (see Supplementary Appendix B for details). If for any such trial we considered a carryover effect to be present, we included only data from the group that received placebo treatment first. If data were not reported by treatment period, we excluded the trial.  Trials with at least 1 active treatment arm and a placebo arm B We recorded the route of intervention of active and placebo treatment (intravenous [i.v.], oral, percutaneous, etc) and made a judgment on invasiveness of the intervention (high, medium, or low level of invasiveness; see Supplementary Appendix B for details). B We scrutinized all placebo treatments for the possibility of either active or nocebo (negative or harmful) effects.  Trials that reported an outcome of participantreported pain intensity on the visual analog scale (VAS) or numeric rating scale (NRS) before and after intervention, or change in pain intensity score from baseline by treatment arm B We excluded trials for which we were unable to obtain pain intensity scores in the placebo arm.

Mbizvo et al B If a trial reported an outcome of pain intensity on a validated scale other than VAS or NRS, we examined whether a conversion to the VAS or NRS was possible. We excluded trials that recorded pain outcomes on nonvalidated scales.  Trials of ‘‘long-standing’’ CRPS B We excluded trials in which the participants’ disease duration was not reported. B We excluded patients with less than 6 months’ disease duration a priori, in keeping with the methods used in the underlying systematic review.10 B For all other trials, we compared the participants’ average disease duration to the average duration in trials, which had explicitly excluded patients with less than 6 months’ CRPS duration. We included trials with an average disease duration equal to or longer than the duration in these trials, and we excluded all other trials (see Supplementary Appendix D for details).

Data Extraction Two authors (G.K.M. and S.J.N.) independently extracted the following data from each included trial report. Any disagreements were resolved by mutual discussion or through adjudication via the senior author (A.G.): B Trial authors and year B CPRS subtype (type I, type II, other subtype [eg, post stroke]) B Diagnosis criteria of CRPS (IASP, Budapest criteria, or reflex sympathetic dystrophy) B Parallel or crossover trial design (and an assessment of carryover effect for crossover trials) B Method of blinding of participants and outcome assessors and assessment of adequacy of blinding (high, low, or unclear risk of bias) B Route of intervention (i.v., percutaneous, oral, etc) B Treatment administration details (single or continuous administration and length of treatment time) B Active treatment (including dose) B Placebo treatment B Number of placebo participants B Outcome reported for pain intensity B VAS (0–10 cm), VAS (0–100 mm), or NRS (0–10) B Baseline pain intensity score B Posttreatment pain intensity score(s) B Measurement times during and after treatment (after baseline) B Duration of CRPS in months B Details of any participants withdrawing from the trial, lost to follow-up, or missing data B Intention-to-treat or per-protocol approach taken to analysis We extracted all pain intensity data as reported in the trial report. For statistical analysis, we required a mean pain score and a level of precision (standard deviation, standard error, or 95% confidence interval) both at baseline and posttreatment or a mean change in pain score from baseline and level of precision. If a trial did not report these summary measures, we contacted trial authors for these data. If trial authors could not provide this informa-

The Journal of Pain

101

tion, we excluded the trial from statistical analysis but retained the trial in the narrative section of this review.

Assessment of Quality of Included Trials Methodologic quality of trials included in the reviews of Forouzanfar et al19 and Cossins et al10 had been assessed using a 15-item quality checklist12 (see Supplementary Appendix E for full checklist). In this checklist, each item is weighted, resulting in a maximum total score of 100 for a given study, where a higher score indicates a better quality of study design. For our included trials, we applied the quality scores assigned by Forouzanfar et al19 and Cossins et al,10 and for trials identified in our updated search or from other sources, 2 authors (S.N. and A.G.) independently applied the same checklist and agreed upon a quality score for each study.

Primary Outcome and Assessment of Potential Confounding Factors Our primary endpoint was ‘‘placebo response,’’ defined as the mean change in pain intensity (measured on the VAS 0–100 mm) from baseline to posttreatment. To standardize to a single scale, we assumed that NRS (0–10) and VAS (0–10 cm) were equivalent, and these scales were converted to VAS (0–100 mm) by multiplying NRS pain scores by 10. Where the mean change from baseline was not reported as a summary measure, we calculated the placebo response from data extracted for baseline and from all posttreatment pain scores reported. Where data were reported only graphically, we contacted the corresponding author for numeric data. If this was unsuccessful, we measured the placebo response from the available published graph by hand, whereby 2 authors (G.K.M. and S.J.N.) performed extraction from graphs and compared results for accuracy (see the Results section and Table 1 for details of data extraction methods). Our a priori aim was to model the placebo response over time, with time as a continuous variable in metaregression. However, following identification of relevant studies, we did not have sufficient studies reporting at the same time point(s) to allow this (see the Results section). Therefore, we felt it inappropriate to combine placebo responses for each trial at a single time point (such as earliest measured time or latest measured time) via meta-analysis. Instead, in our primary analysis, the time points for outcome measures were decided upon post hoc in light of the data, where we pooled trial-specific effect sizes separately over 4 time periods: 15 to 30 minutes, 1 week, 3 to 4 weeks, and 6 weeks or more. These were the time periods at which the data naturally congregated across trials, as we observed that trials had reported posttreatment pain scores at various different times ranging from 30 seconds to 91 days. The chosen measurement time periods ensured that we were selecting time points that were actually reported (eg, no studies reported an outcome at 5 weeks, so we did not include this time period). If a trial reported multiple effect sizes at 1 of these time periods (eg, at both 3 and 4 weeks), we took the mean of the reported effect sizes for analysis. These 4 post hoc time periods allow us to

Characteristics of Included Trials

5

PLACEBO PATIENTS

DIAGNOSIS CRITERIA

CRPS SUBTYPEz

ACTIVE TREATMENT

12

RSD

RSD

Cacchio 20098 Goebel 201022 Groeneweg 200824

8 13 12

Budapest Budapest IASP

Type I, poststroke Type I Type I

Groeneweg 200925 Hanna 198926

12 7

IASP RSD

Type I, cold type RSD

Kettler 198837

4

RSD

RSD

Munts 200944 Munts 201043

18 11

IASP IASP

Type I, with dystonia Type I

Perez 200848 Picarelli 201049 Pleger 200450 Price 199851

19 11 10 7

Budapest IASP IASP IASP

Type I Type I Type I Type I or II

Rauck 199353 Robinson 200454 Schwartzman 200956 Sigtermans 200957 Van de Vusse 200461

26 13 10 30 24

RSD IASP IASP IASP IASP

RSD Type I Type I or II Type I Type I

Van der Plas 201362

22

IASP

Type I, with dystonia

NCT00109772 201345,k

92

IASP

Type I

Normal saline (30–50 mL) 1 reserpine (.5 for UE, 1 mg for LE) or guanethidine (20 mg for UE, 30 mg for LE) Active mirror box therapy/mental imagery training I.v. immunoglobulin .5 g/kg Phosphodiesterase-5 inhibitor tadalafil (10 mg for 4 wk and 20 mg for 8 wk) Transdermal 1% isosorbide dinitrate in Vaseline Ketanserin (10 mg for arm pain, 20 mg for leg pain) IVRSB injection I.v. regional droperidol (2.5 mg) 1 heparin (500 U for UE/1,000 U for LE) 1 saline (30 mL for UE/50 mL for LE) Intrathecal glycine (8 mg–32 mg/24 h) Intrathecal methylprednisolone (60 mg) 10% mannitol i.v. 1 saline Motor cortex rTMS Motor cortex rTMS Sympathetic ganglia block of 15 mL 1% lidocaine/ bupivacaine local anesthetic Epidural clonidine injection (300 mg or 700 mg) I.v. pamidronate injection 60 mg I.v. ketamine (.35 mg/kg/h) 1 saline I.v. ketamine (1.2 mg/kg/min) Gabapentin (600 mg o.d. Days 1–2, b.i.d. Days 3–4, t.i.d. Days 5–21) I.m. magnesium sulphate 100 mg/mL solution (5 mL b.i.d. Week 1, 7.5 mL b.i.d. Week 2, 10 mL b.i.d. Week 3) Oral lenalidomide (10 mg taken as two 5-mg capsules o.d.)

5

Blanchard 1990 Cacchio 20098 Goebel 201022

INTERVENTION CATEGORY IVRSB Mirror box I.v.

LEVEL OF INVASIVENESS{ High Low Medium

PAIN SCALE

MEAN

SD

VAS VAS NRS

50.60 61.70 79.00

29.00 NA 11.00

STUDY DESIGN

Normal saline (30–50 mL)

24.8 (27.2)

Crossover

Covered mirror box I.v. saline Placebo capsule

14 (7–21)* 19.3 (7.4) 55.7 (52.3)

Parallel Crossover Parallel

Placebo ointment Saline injection

45.5 (29.6) 25.9 (23.89)

Parallel Crossover

37 (28.8)

Crossover

Heparin (500 U for UE/1,000 U for LE) 1 saline (30 mL for UE/50 mL for LE) Saline (sodium chloride .9%) Saline (1.5 mL sodium chloride .9%) I.v. saline Sham rTMS Sham rTMS Saline (15 mL)

108 (60–204)* 48 (24)

Crossover Parallel

14 (5–30)* 79.3 (32.1) 35 (24–72)* 36.4 (25.1)

Parallel Parallel Crossover Crossover

Saline injection I.v. saline I.v. saline I.v. saline Placebo capsule

45.7 (36) 21.6 (3–72)* 87.7 (85.7) 73.2 (61.2) 43 (36)

Crossover Parallel Parallel Parallel Crossover

I.m. placebo injection sodium chloride (.9%)

138 (72–192)

Crossover

Placebo capsule

Minimum of 1 y*

Parallel

BLINDING{

BASELINE PAIN*,**

STUDY REFERENCE*,y

DURATION OF CRPS (MO)x

PERFORMANCE BIAS

DETECTION BIAS

Unclear High Low

Unclear Unclear Low

ANALYSIS APPROACH{

DATA EXTRACTION

PP ITT ITT

Summary statistics Graphical IPD provided

Placebo Responses in Long-Standing CRPS

Blanchard 1990

PLACEBO

The Journal of Pain

STUDY REFERENCE*,y

102

Table 1.

Continued BLINDING{

BASELINE PAIN*,**

STUDY REFERENCE*,y

INTERVENTION CATEGORY

Groeneweg 200824 Groeneweg 200925 Hanna 198926 Kettler 198837 Munts 200944 Munts 201043 Perez 200848 Picarelli 201049 Pleger 200450 Price 199851 Rauck 199353 Robinson 200454 Schwartzman 200956 Sigtermans 200957 Van de Vusse 200461 Van der Plas 201362 NCT00109772 201345,k

Oral Topical IVRSB IVRSB Percutaneous Percutaneous I.v. rTMS rTMS Percutaneous Percutaneous I.v. I.v. I.v. Oral Percutaneous Oral

LEVEL OF INVASIVENESS{ Low Low High High High High Medium Medium Medium High High Medium Medium Medium Low High Low

PAIN SCALE

MEAN

SD

VAS VAS VAS VAS NRS NRS VAS VAS VAS VAS VAS VAS NRS NRS VAS NRS NRS

57.00 51.70 63.33 51.75 68.67 73.00 48.00 88.10 42.60 49.88 75.20 72 (median) 75.00 68.70 64.20 72.00 NAk

12.10 18.90 15.87 35.07 14.26 17.00 23.60 11.40 24.98 20.06 24.98 52–78.6 (IQR) 19.00 14.30 16.00 15.00 NAk

Mbizvo et al

Table 1.

PERFORMANCE BIAS

DETECTION BIAS

Unclear High Unclear High Low Low High Low

Unclear High Unclear High Low Low High Low

High Unclear Unclear High High High Low Low

High Unclear Unclear High High High Low Low

ANALYSIS APPROACH{

DATA EXTRACTION

ITT ITT ITT PP ITT PP ITT ITT ITT ITT PP ITT PP ITT PP PP ITT

Summary statistics Summary statistics Graphical IPD provided IPD provided Summary statistics Summary statistics Graphical Graphical Graphical Graphical Graphical Summary statistics IPD provided Summary statistics Summary statistics Summary statistics

Abbreviations: RSD, reflex sympathetic dystrophy; UE, upper extremity; LE, lower extremity; IVRSB, intravenous regional sympathetic blockade; rTMS, repetitive transcranial magnetic stimulation; o.d., once per day; b.i.d., twice per day; t.i.d., 3 times per day; i.m., intramuscular; SD, standard deviation; PP, per protocol; ITT, intention to treat; IQR, interquartile range. *We were not able to extract data to calculate a mean difference in pain from baseline and associated level of precision for Cacchio et al8 (no measure of precision reported) and Robinson et al54 (median and IQR only reported); therefore, these studies were excluded from statistical analysis. Mean and SD of baseline pain were not available for NCT00109772.45 However as mean difference in pain score from baseline and SD were available, it was possible to include this study in meta-analysis. yWe found evidence of a substantial carryover effect in Hanna26 and therefore extracted data for participants who received placebo in the first period only. zWe considered a diagnosis of RSD to be largely congruent with CRPS type I. xDuration of CRPS in months given as mean (SD) unless marked with an asterisk (Cacchio et al8: median and range; Munts,44 Perez,48 and Van der Plas62: median and IQR; Pleger50 and Robinson et al54: mean and range; NCT0010977245: minimum duration). {See Appendix B for detailed information on judgments of level of invasiveness, adequacy of blinding, and the analysis approach. kAll trial characteristics were taken from limited information provided in the trial report on ClinicalTrials.gov (http://www.clinicaltrials.gov/ct2/show/study/NCT00109772). The trial was terminated prematurely and no publication was available. **All pain intensity scores were converted to a VAS 0 to 100 mm scale (ie, VAS and NRS scores on 0–10 cm scales were multiplied by 10). It is assumed that VAS and NRS scales are equivalent.

The Journal of Pain

103

104

The Journal of Pain

Placebo Responses in Long-Standing CRPS

Figure 1. PRISMA study flow diagram. Stages of eligibility screening of RCTs identified by systematic search. examine a range of time points of clinical interest (ie, early and late measurements) at which data were collected and to investigate any differences in placebo response over time. In general, trials with negative findings may be less likely to be published than trials with positive findings; this is known as publication bias and is a common issue in meta-analysis.59 In the context of a large placebo response, a large active response is needed to achieve statistical significance. As ‘‘placebo’’ response and ‘‘real drug’’ response in the active trial arm may be subadditive, a large placebo response may be associated with a negative trial outcome, and consequently, this metaanalysis may be at increased risk of publication bias31 (however, note that this has not consistently been shown in chronic pain trials14). We intended to investigate whether there was any evidence of the presence of publication bias in the context of CPRS trials by examining both asymmetry of funnel plots (which requires a minimum of 10 trials in analysis59) and individual trial characteristics and findings (see Statistical Analysis section). We performed several subgroup analyses at each specified time period to assess the relationship between placebo response and potential confounding factors of interest: baseline pain, trial publication date, adequacy of blinding, level of invasiveness of the route of trial intervention, trial design, and analysis approach (see Statistical Analysis section and Supplementary Appendix B for details).

Statistical Analysis We calculated a mean difference and standard error of mean difference for change in pain intensity from baseline for each trial (see Supplementary Appendix C for details). We then synthesized the individual effect sizes at the 4 specified time periods of interest (15– 30 minutes, 1 week, 3–4 weeks, and 6 weeks or more) via meta-analysis using the method of inverse variance. Pooled mean differences and 95% confidence intervals were reported for each time period separately. We assessed the level of heterogeneity present between trials by visual inspection of forest plots and formally according to the I2 statistic29,30 (the percentage of variability between trials that is due to statistical heterogeneity). A rough interpretation of this I2 statistic is as follows (see Supplementary Appendix C for further details):  0 to 40% may not be important;  30 to 60% may represent moderate heterogeneity;  50 to 90% may represent substantial heterogeneity;  75 to 100% represents considerable heterogeneity. Because of differences in trial design and participant characteristics, we anticipated that heterogeneity would be present between trials; therefore, we performed random-effects meta-analysis at each specified time period. We performed subgroup analysis to assess for the impact of the adequacy of blinding of participants and outcome assessors (high, low, or unclear risk of bias), the level of invasiveness of the route of

Mbizvo et al Table 2.

The Journal of Pain

105

Quality Assessment: Hierarchical List of Quality Scores* for 20 RCTs Included in the Review PARAMETERS

STUDY REFERENCEy

A

Bz

C

D

E

F

G

H

I

J

K

L

M

N

O

TOTAL

Maximum scores Sigtermans 200957 Goebel 201022 Perez 200848 Munts 200944 Rauck 199353 Groeneweg 200925 Munts 201043 Groeneweg 200824 Van der Plas 201362 Van de Vusse 200461 NCT00109772 201345 Price 199851 Pleger 200450 Robinson 200454 Blanchard 19905 Kettler 198837 Picarelli 201049 Schwartzman 200956 Cacchio 20098 Hanna 198926

4 4 4 4 4 4 4 4 4 4 4 4 4 2 2 4 2 4 4 2 0

15 15 15 15 15 15 15 15 15 15 15 0 5 10 0 0 5 0 0 0 0

8 4 0 0 0 4 0 0 0 0 4 8 0 0 0 0 0 0 0 0 0

8 8 8 8 8 4 7 4 7 8 7 4 8 7 7 7 7 8 8 4 4

12 12 12 12 4 12 12 12 12 4 4 4 12 12 12 12 4 2 0 12 0

6 6 6 6 6 6 4 6 4 6 4 6 4 4 6 4 4 6 6 4 4

6 6 6 6 6 6 4 6 4 6 3 6 4 3 6 4 4 6 6 3 4

5 5 5 5 5 5 5 5 5 5 5 5 5 5 5 5 5 5 5 5 5

6 4 6 4 6 4 4 6 4 4 4 4 4 6 4 4 4 6 4 0 4

6 4 6 6 6 4 4 6 4 4 4 4 4 0 4 4 4 0 4 0 4

6 0 0 0 4 0 4 0 4 4 0 4 4 0 0 0 4 0 0 4 0

5 3 3 5 4 2 3 3 3 4 4 5 1.5 1 3 1.5 2 4 3 3 1.5

2 2 2 2 2 2 2 1 2 2 2 2 2 2 1 1 1 2 2 2 1

5 5 5 5 5 5 5 5 0 5 5 5 5 5 5 5 5 5 5 0 5

6 6 6 2 2 2 2 2 6 2 2 6 2 6 2 2 2 2 2 2 2

100 84 84 80 77 77 75 75 74 73 67 67 64.5 63 57 53.5 53 50 49 41 34.5

*Fifteen-item quality assessment checklist of de Vet et al12 was used (see Supplementary Appendix E for description of the A through O components of the 15-item quality checklist). yFor trials included in the reviews of Forouzanfar et al19 and Cossins et al,10 we used quality scores assigned by the authors. For trials found in our updated search, we applied the same checklist of de Vet et al.12 zA score of 0 indicates that method of randomization and allocation concealment was not stated in the trial paper.

intervention (high, medium, or low), trial design (crossover or parallel), analysis approach (intention-to-treat or per-protocol), baseline pain (mean baseline pain intensity #65, or more than 65 on VAS scale 0–100), and trial publication date (up to year 2000 or after 2000) (see Supplementary Appendix B for more details of choice of categories and thresholds). Again, we performed subgroup analyses with random-effects metaanalysis due to anticipated heterogeneity between trials and performed a test of subgroup differences to examine the influence of different study and participant characteristics (see Supplementary Appendix C for statistical details of test of subgroup differences). To account for the subgroup analysis of multiple trialrelated factors, pooled mean differences and 99% confidence intervals were reported for subgroups over the separate time periods. We would have liked to perform meta-regression to assess the effect of baseline pain and time since trial publication on a continuous scale; however, the numbers of trials included at each time period were insufficient to provide power to detect an effect in meta-regression38; therefore, post hoc dichotomization of these variables into subgroups allowed for these variables to be investigated as confounding factors (see Supplementary Appendix B for more details). For the same reason, we were unable to perform any multivariable regression analyses to investigate possible interactions between confounding factors or to perform a formal statistical investigation of publication bias via examination of asymmetric funnel plots. Alternatively, we judged

publication bias by closely examining individual trial results and characteristics. We performed our systematic review and metaanalysis in accordance to the guidelines of the Preferred Reporting of Items for Systematic Reviews and MetaAnalyses (PRISMA).40 All statistical analyses were performed in Stata statistical software version 11.2 (StataCorp LP, College Station, TX).58

Results Identification of Eligible CRPS Trials Seventy-one RCTs were identified from the reviews of Forouzanfar et al19 and Cossins et al,10 and 6 RCTs1,18,39,45,62,63 were identified in our updated search, 1 of which was an ongoing RCT46 (see below). After duplicates were removed, this yielded a total of 76 RCTs, from which we identified 20 RCTs that met all of the inclusion criteria for our review, with a total of 361 placebo participants.5,8,22,24-26,37,43-45,48-51,53,54,56,57,61,62 See PRISMA flow diagram in Fig 1 for more details on the screening of the potentially eligible trials and Supplementary Appendix D for details of excluded trials. One of the included trials was identified from searching the ClinicalTrials.gov trial registry (www.clinicaltrials. gov/ct2/show/record/NCT00817128). We refer to this trial throughout this review as NCT00109772,45 as specified on the trial registry. We believe this to be the largest RCT conducted in this group of patients. However, to our knowledge, at the time of writing, no full publication of this trial existed. Therefore, we extracted data

106

Placebo Responses in Long-Standing CRPS

The Journal of Pain

from the limited information in the ‘‘Study Results’’ provided on the trial registry. Table 1 provides design and placebo participant characteristics for all included trials. Table 2 provides quality assessment scores for each trial. We identified evidence of carryover effect in 1 crossover trial,26 and therefore we extracted data for participants who received placebo in the first period only. We were able to extract or calculate a mean baseline pain score, a mean posttreatment pain score, and associated levels of precision for 17 of the 20 included trials. Munts et al44 and Robinson et al54 provided median baseline and posttreatment pain scores, and Cacchio et al8 provided mean pain scores at a range of time points but no associated levels of precision. We contacted the corresponding authors of the 3 trials. Munts et al44 were able to provide individual participant data (IPD) for the trial duration; however, this information could not be provided for Cacchio et al8 and Robinson et al54; therefore, we could not include these 2 trials in the meta-analysis. We were also provided IPD for Goebel et al22 and for some of the trial duration of Sigtermans et al,57 and we were able to extract individual participant pain scores from the publications of Kettler et al37 and Price et al51 Seven trials reported results graphically,26,48-51,53,54 and G.K.M. and S.J.N. extracted those results independently by hand from the published graphs, with consistent agreement between results.

Primary Outcome: Analysis of Placebo Response Table 3 shows the measurement times of posttreatment pain scores in each trial. Five trials including 59 placebo patients measured posttreatment pain at 15 to 30 minutes,5,37,50,51,53 6 trials including 80 placebo participants measured at 1 week,22,26,44,49,51,57 6 trials including 117 placebo participants measured at 3 to 4 weeks,22,44,56,57,61,62 and 8 trials including 197 placebo participants measured at 6 weeks or more. 24,25,43,45,48,49,56,57 Therefore, 11 trials5,24-26,37,43,45,48,50,53,61,62 reported a posttreatment pain score at only 1 of the 4 time points, 5 reported pain scores at 2 time points,22,44,49,51,56 and 1 trial at 3 time points.57 Fig 2 and Table 4 show results of pooled mean difference in pain score from baseline at each of the time periods. A moderate to substantial amount of heterogeneity is present between trials at each of the time points, ranging from 29.7% at 3 to 4 weeks to 92.3% at 1 week (see the Statistical Analysis section for details on interpretation of heterogeneity); therefore, results must be interpreted with caution. For this reason, we will interpret trends from visual inspection of individual trials and pooled estimates rather than placing emphasis on numeric values of pooled results. Statistically significant evidence of a placebo response is present at only the earliest time period of 15 to 30 minutes (2-sided P value of .014, Table 4); however confidence intervals around the pooled mean difference at this time period are very wide because of the imprecision in individ-

Measurement Times of Posttreatment Pain Scores

Table 3.

STUDY REFERENCE 5

Blanchard 1990 Goebel 201022 Groeneweg 200824 Groeneweg 200925 Hanna 198926 Kettler 198837 Munts 200944 Munts 201043 Perez 200848 Picarelli 201049 Pleger 200450 Price 199851 Rauck 199353 Schwartzman 200956 Sigtermans 200957 Van de Vusse 200461 Van der Plas 201362 NCT00109772 201345 Number of studies Number of participants

15–30 MIN*

1 WK

3–4 WKy

x

x

$6 WKz

x x x x x x

x x x x

x x x x

x

x

x x x x

x x

x 5 57

6 76

6 117

8 197

*Within the time interval 15 to 30 minutes: Blanchard et al,5 Kettler et al,37 and Rauck et al53 reported mean pain score at 20 minutes; Pleger et al50 reported at 15 minutes; and Price et al51 at 30 minutes. yWithin the time interval 3 to 4 weeks: Van de Vusse et al61 and Van der Plas et al62 reported a mean pain score at 3 weeks only, and Schwartzman et al56 reported a mean pain score averaged over Weeks 3 and 4. Goebel et al,22 Munts et al,44 and Sigtermans et al57 reported mean pain scores at both 3 and 4 weeks, so for analysis we averaged over these pain scores. zWithin time intervals of 6 weeks or more, Groenweg et al24 and Picarelli et al49 reported a mean pain score at 12 weeks, Groenweg et al25 at 10 weeks, and Munts et al43 at 6 weeks. Perez et al48 reported mean pain scores at 6 and 9 weeks, Schwartzman et al56 at 8 and 12 weeks, and Sigtermans reported weekly mean pain scores from 6 to 12 weeks. NCT0010977245 reported a mean pain score using averaged morning and evening readings at Week 12. For the 3 studies that reported multiple pain scores, we averaged over the scores in each trial for analysis.

ual trial effect sizes stemming from small sample sizes in individual trials. There is no significant evidence of placebo response at any of the other time periods (see Fig 2 and Table 3). Less heterogeneity is present between trials at 3 to 4 weeks; individual trial estimates are relatively precise at this time period, so the nonsignificant pooled effect size is more precise at this time period compared to other time periods. At 1 week and at 6 weeks or more, there is some indication of a trend toward a placebo effect; however, from visual inspection of Fig 2 at both of these time points the pooled estimate seems to be influenced by 2 particular trials. These are Picarelli et al49 and Sigtermans et al,57 which show large statistically significant placebo responses, whereas the other 4 trials show no evidence of placebo response at 1 week. At 6 weeks or more, Picarelli et al49 shows a large statistically significant placebo response, whereas the other 6 trials show either null or small effect sizes.

Assessment of Publication Bias Fig 3 shows the mean pain scores at baseline and all recorded time points for 17 trials included in the analyses.

Mbizvo et al

The Journal of Pain

107

Figure 2. Forest plot of random effects meta-analysis at 15 to 30 minutes, 1 week, 3 to 4 weeks, and 6 weeks or more. Pooled mean difference of pain intensity from baseline on a VAS (0–100 cm). (We were not able to include NCT0010977245 on this figure as we could not extract information on mean baseline pain score.) These graphs show the range of time points at which trials make posttreatment measurements and the extent of between-trial differences in terms of reported pain score. For example, some trials report large differences in pain score, particularly when measured at early time periods,5,37,51 whereas others report similar pain scores over several time periods.44,48,50,56 From visual inspection of Figs 2 and 3, trial characteristics in Table 1,

and results in Table 4, the results from the trials included range from large placebo responses to statistically significant increases in pain from baseline; therefore, we have no evidence to suspect the existence of publication bias in this review. We were not able to perform a more formal statistical assessment of publication bias via funnel plot (see the Statistical Analysis section) because of an insufficient number of trials at each time point; it is recommended that at least 10 trials are included in a meta-analysis to construct a funnel plot.59

Table 4. Primary Analysis, Pooled Mean Difference in Pain Score From Baseline to Posttreatment at 4 Time Points TIME POINT

NUMBER OF TRIALS

NUMBER OF PARTICIPANTS

POOLED MEAN DIFFERENCE*

95% CONFIDENCE INTERVAL

P VALUEy

I2z

15–30 min 1 wk 3–4 wk 6 wk or more

5 6 6 8

59 86 117 197

18.423 6.772 .326 3.870

33.19 to 3.65 14.92 to 1.38 2.329 to 2.981 9.46 to 1.71

.014 .103 .810 .174

82.80% 92.30% 29.70% 90.30%

NOTE. Trials with a posttreatment measurement at 15 to 30 minutes: Blanchard et al,5 Kettler et al,37 Pleger et al,50 Price et al,51 Rauck et al.53 Trials with a posttreatment measurement at 1 week: Goebel et al,22 Hanna et al,26 Munts et al,44 Picarelli et al,49 Price et al,51 Sigtermans et al.57 Trials with a posttreatment measurement at 3 to 4 weeks: Goebel et al,22 Munts et al,44 Schwartzman et al,56 Sigtermans et al,57 Van de Vusse et al,61 Van der Plas et al.62 Trials with a posttreatment measurement at 6 weeks or more: Groenweg et al,24 Groenweg et al,25 Munts et al,43 Perez et al,48 Picarelli et al,49 Schwartzman et al,56 Sigtermans et al,57 NCT00109772.45 *Mean difference calculated as mean pain score posttreatment – mean pain score at baseline. Pooled mean difference calculated in a random effects meta-analysis via the method of inverse variance. yP value for the existence of placebo effect (ie, 2-sided test of the null hypothesis of mean difference = 0). P value 75% is interpreted as considerable heterogeneity (see Statistical Analysis for more details).

108

The Journal of Pain

Placebo Responses in Long-Standing CRPS

Figure 3. All recorded pain intensity scores (baseline and posttreatment) over the trial duration in 17 included trials (we were not able to include trial NCT0010977232 on this figure as we could not extract information on mean baseline pain score). Baseline mean pain score is recorded as mean pain score on 0 days since treatment. Central circle and bars on the graphs represent mean 6 standard deviation (SD).

Subgroup Analysis of Confounders That May Affect Placebo Response Tables 5 to 8 show the results of subgroup analyses at the 4 time points (see also Table 1 for further details of the design and participant characteristics within trials and Supplementary Appendix B for the definitions of subgroups). Lack of sufficient good quality data prevented us from analyzing these factors together in a multivariable model in order to take account of interactions between the factors and the collective effect of the factors on placebo response. In the primary analysis, only the earliest time point of 15 to 30 minutes posttreatment measurement showed statistically significant evidence of a placebo effect. From subgroup analyses (Table 5), the trial characteristics that may be associated with this significant placebo response are a crossover trial design, a per-protocol analysis approach, high invasiveness of intervention, trials published before 2000, and mean baseline score #65 among placebo participants. In the primary analysis, there was no evidence at 1 week, 3 to 4 weeks, or 6 weeks or more of placebo response, and this is also shown in all subgroup analyses with no statistically significant differences between any subgroups in Tables 6 to 8.

As in the primary analysis, a large amount of heterogeneity is present in the majority of subgroup analyses. Therefore, we will not emphasize the numeric results themselves and will instead only describe trends in the subgroup analyses. These subgroup analyses reveal some trends in trial design over time. The earliest posttreatment measurements at 15 to 30 minutes are made only in trials of a crossover design, and conversely, the latest posttreatment measurements at 6 weeks are made only in parallel design trials. Similarly, the earliest measurements are all recorded in trials with medium to highly invasive interventions. Furthermore, later time points of 3 to 4 weeks and after 6 weeks are reported only in trials published after 2000; trials published before 2000 reported up to a latest time point of 2 weeks, and conversely, only a single trial published after 2000 reported outcomes at the early time interval.50 Finally, there is some evidence that trials judged to have high risk of bias for blinding (ie, inadequately blinded) have placebo responses at the earlier time points, and those judged to have low risk of bias (ie, adequately blinded) have placebo responses at later times. The trends described here highlight some of the trial differences that might have contributed to heterogeneity.

Mbizvo et al Table 5.

The Journal of Pain

109

Pooled Mean Difference in Pain Score From Baseline to 15 to 30 Minutes by Subgroups NUMBER OF

15-30 MIN*

TRIALS

Trial design: crossover Trial design: parallel Blinding: low risk of bias Blinding: unclear risk of bias Blinding: high risk of bias Analysis approach: ITT Analysis approach: per protocol Invasiveness: low Invasiveness: medium Invasiveness: high Mean baseline pain: #65 Mean baseline pain: >65 Year of publication: before 2000 Year of publication: 2000 or after

5

NUMBER OF PARTICIPANTS

POOLED MEAN DIFFERENCEy

99% CONFIDENCE INTERVAL

P VALUEz

1 4 4 1 4

18.42 37.83 to .99 .014 NA: all studies of a crossover design 7 31.71 51.34 to 12.09 65 Year of publication: before 2000 Year of publication: 2000 or after

NUMBER OF TRIALS 4 2 4 1 1 6

3 3 2 4 2 4

NUMBER OF PARTICIPANTS

POOLED MEAN DIFFERENCEy

99% CONFIDENCE INTERVAL

P VALUEz

1.91 5.57 to 1.75 .179 18.45 26.86 to 10.04

Placebo responses in long-standing complex regional pain syndrome: a systematic review and meta-analysis.

The typical placebo response (ie, the nonspecific effects in the placebo group including benign natural course, regression to the mean, expectation/co...
2MB Sizes 0 Downloads 6 Views