Retracted: Effect of Sham Acupuncture on Chronic Pain: A Bayesian Network Meta-Analysis.

Chronic pain is one of the most disabling health conditions worldwide, according to the 2017 Global Burden of Disease Study [ 1 ]. The top three causes of years lived with disability include two kinds of chronic pain [ 1 ], which cause negative emotion, cognitive impairments [ 2 ], and lost productivity [ 3 ]. Additionally, excessive use of analgesics, especially opioids, has posed critical challenges to pain management in the United States [ 4 ] and United Kingdom [ 5 ]. Definite evidence for the long-term efficacy of medications for chronic pain is still sparse [ 6 ]. Acupuncture, a traditional Chinese therapy, has been widely used to manage chronic pain. However, well-designed randomized controlled trials (RCTs) are needed to assess the efficacy of acupuncture. Nevertheless, the effect and design of sham acupuncture (SA) are ambiguous in acupuncture-related RCTs [ 7–9 ], leading to controversy about the efficacy of acupuncture. Inert SA controls are needed to rigorously distinguish the efficacy of acupuncture from a placebo effect [ 10 ]. However, some researchers have shown that SA acts as a non-inert control with some physiological and psychological responses [ 11 , 12 ], and this can result in false-negative results in RCTs with SA controls [ 13 ]. As a result, it is always difficult to decide whether SA controls should be used in the design of an acupuncture RCT. Additionally, it is difficult to decide the type of SA that should be used. SA is divided into roughly two categories: penetrative SA (PSA; at sham acupoints) and non-penetrative SA (NPSA; at real or sham acupoints). NPSA, with a psychological effect, has long been regarded as a standard form of acupuncture placebo control [ 14 ]. In recent years, PSA has been increasingly used because it additionally produces a physiological effect of piercing the skin. Therefore, identifying the differences between SA and non-acupuncture (NA) or PSA and NPSA could provide useful information during the design of acupuncture-related RCTs. As a result, head-to-head RCTs, with different types of SA and NA groups, are needed. Obviously, such RCTs with so many control groups will be confronted with serious ethics issues [ 14 ], but network meta-analyses (NMA) can solve this problem. In the present study, a quantitative assessment of the differences among PSA, NPSA, and NA was conducted with NMA to provide a comprehensive reference for SA.

NMA was conducted in accordance with the Cochrane Handbook for Systematic Reviews of Interventions [ 15 ]. The systematic review and NMA were reported in accordance with the Preferred Reporting Items for Systematic Reviews and Meta-analyses (PRISMA) reporting guidelines [ 16 ]. The present study was registered on PROSPERO (CRD42021230466). Five English-language databases (PubMed, Web of Science, Embase, Scopus, and Cochrane Library) and four Chinese-language databases (Chinese Biomedical Literature Database, China National Knowledge Infrastructure, VIP Database for Chinese Technical Periodicals, and Wan fang) were searched from their inception dates to July 5, 2022. The search strategy consisted of three components: intervention (acupuncture or electroacupuncture compared with SA or NA group); outcome (pain intensity); and study type (RCT). There were no restrictions on languages or diseases. Details of the search strategy are shown in Supplementary Data Table S1 . Studies were independently screened by two groups (RZ and YYY; YJZ and ZZX) using the following criteria: 1) RCTs of patients with chronic pain as defined by the International Classification of Diseases (pain persisting beyond 3 months was defined as chronic pain) [ 17 ]; 2) RCTs assessing the effect of acupuncture or electroacupuncture; 3) RCTs using SA or NA as control; and 4) RCTs measuring pain intensity with the Brief Pain Inventory, a numerical rating scale, a visual analog scale, or a verbal rating scale. NPSA was defined as a sham insertion at real and sham acupoints. Sham insertion was defined as needling without penetrating the skin, normally with the use of an auxiliary device. Sham acupoints referred to both non-acupoints and acupoints without specificity for pain management. PSA was defined as the use of all needling methods penetrating the skin at sham acupoints, including deep insertion, minimal insertion, superficial insertion, and quick needle insertion without retention. NA included groups of waiting list controls, usual care, or no treatment. Exclusion criteria were: 1) trials evaluating the effect of acupuncture/electroacupuncture combined with other treatments that were not used in control groups; 2) trials evaluating the efficacy of dry needling at myofascial trigger points (because the relationship between dry needling and acupuncture is still a subject of debate) [ 18 ]; 3) trials with healthy volunteers; and 4) trials using other sham methods as placebo control (transcutaneous laser or transcutaneous electrical stimulation). Two groups (RZ and YHL; YJZ and LRL) evaluated the quality of the included RCTs according to the Cochrane risk-of-bias criteria. There were seven criteria, including randomization sequence generation, allocation concealment, blinding of participants and personnel, blinding of outcome assessment, incomplete outcome data, selective reporting, and other bias. Each item was assigned as having low, high, or unclear risk of bias. Detection bias was evaluated on the basis of the item “blinding of participants,” as pain intensity is a patient-reported outcome. All data were extracted independently by two researchers (XSC and YDC) and verified by other two authors (RZ and YJZ). Data from the intention-to-treat population were preferred if available. Pain intensity measured at the first time point after the end of treatment was used for data synthesis, as this might be the endpoint at which sham controls have the strongest placebo effect. All outcomes were normalized on a scale from 0 to 100 points. Data from analogous groups were first merged during data extraction to avoid the unit-of-analysis errors [ 19 ], when two or more acupuncture/electroacupuncture arms (those with different acupoint selection methods, treatment frequencies, or treatment durations) were set in one trial. The average pain after the end of treatment and the standard deviation were preferentially selected. Data were extracted from figures with GetData Graph Digitizer software (Version 2.26, http://www.getdata-graph-digitizer.com/ ) if only figures were reported. An algebraic manipulation was used to calculate standard deviation if only the 95% confidence interval (CI) or a standard error was available [ 20 ]. Changes from baseline were extracted if none of the above data were available. A Bayesian multiple-treatment NMA with a random-effects model and uninformative priors was performed. The random-effects model was determined through the use of Markov Chain Monte Carlo methods. The parameters of the models were: initial value, 2.5; number of simulation iterations, 20,000 in each of four chains (5,000 adjustment iterations were performed to eliminate the influence of the initial value); and step size, 1. The effect size was expressed as the mean difference (MD) and 95% CI. The findings were significant when the 95% CI excluded the null value. The R packages “meta” [ 21 ] and “gemtc” [ 22 ] in R 3.6.5 (R Core Team, Vienna, Austria) for Windows (Microsoft Corp., Redmond, WA, USA) were used for all meta-analyses. Key assumptions of an NMA included homogeneity, consistency, and transitivity. Homogeneity was evaluated with the Q test and the statistic inconsistency index ( I 2 ). A meta-regression or sensitivity analysis was used to identify the source of a substantial level of heterogeneity when the I 2 value was greater than 50% (substantial heterogeneity). The inconsistency of the model was calculated with the node-splitting approach, in which direct and indirect evidence were separately compared on a particular comparison. Transitivity was evaluated with descriptive statistics for study and population baselines, such as sample size, age, sex, and pain phenotype. Convergence of iterations was evaluated via visual inspection of the four chains to establish homogenous parameter estimates and in accordance with the Brooks-Gelman-Rubin diagnostic. The predesigned sensitivity analyses were conducted to evaluate the robustness of the model. All analyses were repeated in the sensitivity analyses to take into consideration networks derived from inclusion of only 1) trials with nociceptive pain, neuropathic pain, or nociplastic pain [ 23 ]; 2) trials with low frequency (4 weeks) of acupuncture/electroacupuncture; 4) trials combined with analgesic or without analgesic intake; 5) trials with acupuncture or electroacupuncture; and 6) trials with low risk of detection bias.

A total of 5,761 studies were identified through database searches, and finally 62 RCTs (1.1%) with 6,806 patients were included in the NMA ( Figure 1 ) [ 24–85 ]. The characteristics of the included RCTs are summarized in Table 1 . Most studies (n = 24, 39%) were conducted in China, followed by Germany (n = 6, 10%), then Korea and the United States (n = 5, 8%). The studies included 12 diseases, such as headache, neck pain, shoulder pain, low back pain, gynecological disease–related pain, musculoskeletal pain, facial pain, cancer pain, rheumatic pain, osteoarthritis, diabetic neuropathy, and chronic stable angina pectoris. Forty-nine trials (79%) set two arms, comparing RA with SA or NA. Twenty-six trials with 644 patients used NPSA, 34 trials with 2,153 patients used PSA, and 18 trials with 808 patients used NA ( Figure 2A ). Acupuncture duration and frequency ranged from 3 days to 18 weeks and from five times per week to once per 2 weeks, respectively. Outline of the search-flow diagram using multiple databases. (A) Network diagram of comparisons separating different types of SA. Each circular node represents a type of treatment. Node size represents the total number of patients receiving this treatment (in brackets). Each line represents a type of direct comparison. Width of lines represents the number of trials comparing the connected treatments. (B) Summary of results from assessment of studies with the Cochrane risk-of-bias tool. Characteristics of clinical studies included in the analysis BPI= Brief Pain Inventory; NRS= numerical rating scale; VAS= visual analog scale. The symbol — indicates that no details are available or item is not applicable. Forty-nine trials (79%) described the method of randomized sequence generation, and 44 (71%) reported the details of allocation concealment, and thus had a low risk of selection bias. All trials had a high risk of performance bias because it was difficult to blind the acupuncturist. Seven two-armed RCTs (11%) with NA had a high risk of detection bias, while nine trials (15%) had an unclear detection bias because of the lack of details about participant blinding. Three studies had high attrition bias risk because of unbalanced missing data (n = 2, 3%) or inaccurate descriptions (n = 1, 2%). Four trials (6%) had a high risk of reporting bias because of the incompletely reported outcomes. Nine trials (15%) had a high risk of other bias because of the imbalanced baseline characteristics among groups or lack of details about participant screening. Finally, 27 trials (44%) had a low risk of bias in all domains, except for the high risk of performance bias and unclear risk for reporting bias ( Figure 2B , Supplementary Data Figure S1 ). Both NPSA and PSA were not significantly superior to NA in improving chronic pain in the primary NMA (RA vs NPSA vs PSA vs NA), yet a weak trend favoring NPSA and PSA was detected (NPSA vs NA: MD = –4.77 points, 95% CI –11.09 to 1.52; PSA vs NA: MD = –4.96 points, 95% CI –10.38 to 0.48, Figure 3 ). Notably, after the integration of the NPSA and PSA groups into an SA group in the secondary NMA (RA vs SA. vs NA), the expanded sample size still did not indicate statistical significance (MD= –4.91 points, 95% CI –9.93 to 0.05, Figure 3 ). Additionally, NPSA and PSA had similar effects on pain relief (MD= 0.18, 95% CI –5.45 to 5.81, Figure 3 ). Pooled estimates of the MDs based on the NMA. As expected, RA was associated with significant pain relief, compared with NA (MD= –16.82 points, 95% CI –21.60 to –12.04), NPSA (MD= –12.03 points, 95% CI –16.62 to –7.41), and PSA (MD= –11.85 points, 95% CI –15.48 to –8.23, Figure 3 ). Forest plots of feasible pairwise comparisons with heterogeneity estimates were generated. Substantial heterogeneity was observed in most pairwise comparisons ( Supplementary Data Figure S2 ). Key characteristics of the studies, including sample size, pain phenotype, the origin of trials, analgesic intake, duration, frequency, and intervention, were adjusted in the meta-regression to assess the cause of heterogeneity. None of these factors caused the heterogeneity ( Supplementary Data Table S2 ). Sensitivity analysis suggested that the pairwise comparisons results were robust ( Supplementary Data Figure S3 ). The results suggested that the included trials had favorable transitivity and consistency in the generation of direct and indirect comparisons ( Supplementary Data Figure S4 ). Furthermore, descriptive statistics revealed that the included trials had acceptable transitivity. Additionally, Brooks-Gelman-Rubin diagnostic and history feature indicated that the models had a good convergence, with the stability and replicability of the inferential iterations for each Markov Chain Monte Carlo chain ( Supplementary Data Figure S5 ). The results of predesigned sensitivity analyses were similar to the primary analysis results ( Supplementary Data Table S3 ). The results did not change regardless of whether the type of pain was nociceptive (n = 38) or nociplastic (n = 19), whether the frequency of acupuncture was high (n = 23) or low (n = 39), whether the duration of treatment was long (n = 35) or short (n = 27), whether the patients received acupuncture (n = 42) or electroacupuncture (n = 22), whether the treatment was in combination with analgesics or not (yes, n = 27; no, n = 26; nine studies lacked information on analgesic intake), or when the detection bias was low (n = 47). The sensitivity analysis of trials with neuropathic pain was not conducted because of data sparsity (n = 5).

HBZ and YJZ were responsible for the conception of this study. RZ collected related literature. RZ, YJZ, XC, HCM, YYY, and ZZX selected the studies. RZ, YJZ, YHL, and LRL evaluated the bias of the studies. XSC and YDC extracted the data, and then RZ, YJZ, XC, and HCM verified the extracted data and carried out the data analysis. RZ and YJZ drafted the manuscript. YHL, XSC, YDC, YYY, ZZX, LRL, YL, and HBZ made critical review and commentary on the manuscript. All authors read and approved the final manuscript.

These results suggest that acupuncture was significantly associated with reduced chronic pain. The two kinds of placebo acupuncture, NPSA and PSA, have similar effects. Both NPSA and PSA, with a weak but not significant effect, were appropriate to be inert placebo controls in RCTs for chronic pain.

This NMA included 62 RCTs with 6,806 patients with chronic pain. NPSA and PSA were not statistically different from NA in reducing chronic pain. Moreover, NPSA and PSA had similar effects. These results were unchanged in subgroups based on pain phenotype, frequency, duration, acupuncture methods, analgesic intake, and detection bias. This study provides a reference for the use of SA in the design of acupuncture-related RCTs. Studies have shown that SA can produce some psychological and physiological effects. Psychologists have found that various positive factors (somatosensory stimulation, enhanced doctor–patient relationships, and treatment context) in the healing ritual can contribute to a significant placebo effect [ 8 ]. Recent studies have suggested that this placebo effect for patient analgesia might be associated with dynamic patient-brain to clinician-brain concordance during patient–clinician interaction [ 86 , 87 ]. Furthermore, SA can also induce a physiological response that is similar to, but weaker than, that produced by RA, probably leading to a specific effect [ 8 ]. Therefore, there have long been dissenting voices challenging SA, arguing that it is a non-inert placebo intervention and could result in false-negative outcomes [ 8 , 9 ]. Prior meta-analyses also showed that SA has a moderate effect on chronic pain as compared with the NA group [ 88 , 89 ]. However, our finding was inconsistent with these studies. In fact, needling at acupoints without electrical stimulation was defined as SA in the study by Madsen et al. [ 88 ], which might overestimate the overall effect of SA. In the present study, neither NPSA nor PSA was significantly superior to NA in improving chronic pain. Even with an increased sample size in the integrated SA group, the weak trend favoring SA was still not significant. These findings indicated that SA, with a weak but not significant effect, could be set up as an inert placebo control in acupuncture-related RCTs. Although from the perspective of methodology, SA is an appropriate inert placebo control, another argument challenged the necessity of SA control arms when the focus is on the clinical significance of acupuncture [ 14 , 90 ]. Trials without SA have also been recommended recently because of their emphasis on practical applicability and validity over treatment efficacy [ 91 ]. In addition, unnecessary transportation costs for the SA group might raise ethics issues. Despite our findings supporting SA as an inert placebo, the use of SA should depend on various of factors, including the purpose, the feasibility, and the cost of resources of a clinical study. The present study also provided a hint for the validity of meridians and acupoints in chronic pain. The effects of acupuncture theoretically consist of: 1) a response to the needling stimulation at defined meridians and acupoints (specific efficacy of acupuncture); 2) a physiological effect of piercing through the skin, also produced even at non-acupoints [ 12 , 13 , 92 ]; and 3) a psychological effect produced by therapist–patient interaction and heightened expectations [ 86 , 87 , 93 ]. In the present study, PSA and NPSA had similar effects, which suggests that the penetration might not be effective. Furthermore, both NPSA and PSA were not statistically different from NA, indicating that the psychological effect of acupuncture might not be significant in relieving chronic pain. The observed significant association between RA and pain relief originated mainly from the needling response of defined meridians and acupoints. Therefore, the analgesic effect can be produced only when the skin is punctured at the right place (acupoints). The main limitation of this study derives from the inclusion of diverse diseases. The various diseases involved could be the leading cause of substantial heterogeneity, lowering the certainty of evidence. Sensitivity analyses for each disease were not conducted because of the limited number of studies on each type of disease. Additionally, some details during the process of acupuncture could also impact the efficacy. For example, Deqi, a sensation of soreness, warmth, tingling, or heaviness, usually after manipulation, has been considered an important parameter for therapeutic effectiveness in acupuncture treatment [ 94 ]. According to the STandards for Reporting Interventions in Clinical Trials of Acupuncture (STRICTA) 2010 extension, Deqi or another response sought after needling should be reported [ 95 ]. Unfortunately, in the present study, the sensitivity analysis for response sought was not available, because only nine studies required achieving Deqi, and the others did not require the sensation or report any details.

All studies included in our work are available from publicly available online databases, and they can also be obtained from the corresponding author by reasonable request.

Click here for additional data file.