Time Course of Placebo Effect of Acupuncture on Pain: A Systematic Review

Objectives. Our objective was to investigate the time course of the placebo effect of acupuncture on pain and the factors affecting the placebo effect. Methods. Previously we retrieved three-armed randomized acupuncture trials including sham and no-treatment groups which were published until October 2009. We searched electronic databases again to identify additional trials from October 2009 to December 2011. After a screening of trials, fifteen three-armed acupuncture trials for pain were included in the analysis. Standardized mean differences between the sham and no-treatment groups were calculated for placebo effect. We then plotted the magnitude of the placebo effect over time. Results. The placebo effect gradually has increased for 12 weeks with a standardized mean difference of 0.74 (95% CI: 0.54 to 0.94). Then it decreased after 12 weeks as time passed. When the placebo effects were compared for factors including methodological qualities, they were not affected by all factors, except patient blinding. Trials with sufficient patient blinding showed a larger placebo effect at 8 weeks than those with insufficient patient blinding (P = 0.0009). Conclusion. The placebo effect of acupuncture showed a unique pattern, which was affected by insufficient patient blinding.


Introduction
Randomized trials are necessary to evaluate the effectiveness of acupuncture because acupuncture provides inconsistent bene�ts to patients who are randomly allocated to acupuncture treatment compared to those who directly choose [1]. Randomized trials can provide unbiased results about the efficacy of acupuncture if acupuncture is controlled with a sham acupuncture. For this aim, many efforts have been made to develop optimal sham acupuncture. In fact, sham acupunctures have changed from normal needling of nonacupoints to minimal needling of nonacupoints and nonpenetrating stimulation [2].
Nevertheless, it has been reported that all types of sham acupuncture may not be physiologically inert [3][4][5]. Moreover, according to a series of recent studies [6,7], there is a high possibility of publication bias, where trials with negative results in a treatment group would be less likely to be published, in the three-armed randomized trials including sham and no-treatment groups. When the bias was considered, the magnitude of the placebo effect was calculated to be large [6].
However, all these �ndings [3][4][5][6][7] should be interpreted with caution because they were based on the data measured at one time point, as speci�ed by authors of trials. To date, it has not been studied how the placebo effect changes over time. erefore, we attempted to investigate a one-year time course of the placebo effect of acupuncture. Because bias could be introduced by combining results of trials from different conditions [8], we focused on trials that treated pain.

Study Selection.
Previously, 32 three-armed randomized acupuncture trials were identi�ed using MEDLINE, CINAHL, EMBASE, and the Cochrane Registered Trials from their launch through October 2009 [6,7]. From October 2009 to December 2011, the �rst author searched MEDLINE, SCOPUS, and the Cochrane Registered Trials using the same terms "acupuncture, " "electroacupuncture, " and "electro-acupuncture. " en the �rst author selected randomized clinical trials that met the following conditions: (1) reported data on pain, (2) included an acupuncture group where dry needles were inserted in traditional or painful points, (3) included a sham group where an intervention was considered a sham or a placebo acupuncture in the text, (4) included a no-treatment group where no treatment was applied, and (5) compared the above three groups under identical conditions in one trial. Trials conducted for only one day were excluded.

Data Extraction.
Since the trial duration and assessment time points varied across trials, the assessment time points were grouped into 6 time windows: 0 week, 4 weeks (>0 and ≤4), 8 weeks (>4 and ≤8), 12 weeks (>8 and ≤12), 16 weeks (>12 and ≤16), and 52 weeks (>16 and ≤52). e �rst author then extracted end-point data on pain intensity (e.g., visual analogue scales or other ranking scales) for sham and no-treatment groups reported in those time windows. When more than one pain outcome measure was reported, visual analogue scales were preferred. If end-point data were not available, data on changes from baseline were used. If no data were available, previous studies [9,10] were referenced. Information on trial characteristics, including the methodological qualities, was also extracted. All data were veri�ed by the second author.

Data Analysis.
For each time window, the standardized mean differences (SMDs) were calculated using the sham and no-treatment groups to assess the placebo effect. To analyze data from all time windows similarly, the random effects model was used to present summary estimates [11]. e 2 test was conducted to measure the heterogeneity within each time window [12]. 2 values of 25, 50, and 75% are referred to as low, moderate, and high inconsistencies between the trials, respectively. e multivariate analysis was also performed to con�rm that the results were robust [13]. Because withinstudy correlations are unknown, the "riley" option was used.
In the secondary analysis, the following items were examined: allocation concealment, patient blinding, intention-totreat analysis, standardized cointerventions, additional medical help (e.g., the use of rescue analgesics), sham acupuncture type, patient-therapist interaction, the number of treatment sessions, and the number of needles per session. Allocation concealment was considered adequate if researchers screening patients could not predict the next treatment for a patient. Patient blinding was considered adequate if patients could not notice the treatment they receive. Speci�cally, to determine the status of patient blinding, we focused on the patients' real guesses of treatment credibility. Because analytical methods varied across trials and the original data were insufficiently detailed in trials, we extracted the number of patients who perceived to have received true acupuncture in the acupuncture and sham groups and calculated the signi�cance using the two sample proportion test.
Intention-to-treat analysis was considered adequate if all patients assessed at baseline were included in the analysis. All items were analyzed on a dichotomous basis. For the number of treatment sessions and needles per session, a median value was used as a cutoff point. STATA 11.0 was used for the analyses. e data are presented as the SMDs with a 95% con�dence interval, where a positive SMD indicates that sham acupuncture was more effective than no-treatment. e signi�cance was assessed at the level of 0.05. For the secondary analyses, the SMDs at each time point were compared at the level of 0.008 using the interaction test [14] followed by the Bonferroni correction.

Time Course of Placebo Effect.
When the placebo effects were compared for factors including methodological qualities, they were not affected by all factors, except patient blinding. For patient blinding, the patterns were signi�cantly different at 8 weeks ( Figure 3). SMDs for sufficient patient blinding and insufficient patient blinding were 0.71 (0.56 to 0.86) and 0.30 (0.12 to 0.49) at 8 weeks ( 0 000 ), respectively.

Discussion
We investigated the pattern of the placebo effect of acupuncture over time, using three-armed trials for pain. e placebo effect gradually increased from baseline until 12 weeks, and decreased aer 12 weeks as time passed. When the placebo effects were compared for factors including methodological qualities, they were not affected by all factors, except patient blinding. Trials with sufficient patient blinding showed a larger placebo effect at 8 weeks than those with insufficient patient blinding ( ). Because we examined the placebo effect over time by separately combining the data for each time window, our �ndings are highly informative in understanding the dynamic pattern of the placebo effect. However, although this approach has been used in recent studies [30,31], there are several limitations. First, not all of the trials provided data relating to each time window. Because of the sparsity of data available for each time window, possible correlations between results for the different time windows were not addressed. erefore, we could not determine the robustness of our results. Second, other time windows are possible to depict the placebo effect. For example, the placebo effect can be described over three time windows, with SMDs of 0.35 (0.18 to 0.51) in 0 to 6 weeks, 0.66 (0.52 to 0.80) in 6 to 12 weeks, and 0.29 (0.17 to 0.41) in 12 to 52 weeks. e degree of the heterogeneity was ≤27%. Although this example shows a unique pattern, the placebo effect also increased from baseline until 12 weeks, as with the current result.
Numerous studies [6,7,9,32] have investigated the placebo effect of acupuncture compared with a no-treatment group. ey analyzed data at one time point and interpreted �ndings in terms of two aspects: (1) the magnitude of the placebo effect of acupuncture and (2) the efficacy of acupuncture compared with sham acupuncture.
First, two previous studies [9,32] demonstrated that the high degree of variability associated with the placebo effect led to cases where sham acupuncture was effective in some situations and ineffective in others. However, they found in general the large placebo effect on pain, with an SMD of 0.42 (0.23 to 0.60) [9] and of 0.53 (0.39 to 0.67) [32], respectively. Speci�cally, one study [9] argued that this large placebo effect may be associated with response bias, where patients report better outcomes to please researchers. Meanwhile, recent two studies argued that the variable placebo effect may be due to the result of the natural process of publication in journals [6,7]. When this publication bias was considered, the SMD for the placebo effect was estimated to be 0.44 (0.29 to 0.59) in general conditions [6].
However, according to the current �ndings, the placebo effect is the most prominent at 12 weeks, with an SMD of 0.74 (0.54 to 0.94). Although we analyzed the similar set of trials to that of previous study [9], this SMD is much greater than that of previous studies [9,32]. Because previous studies analyzed the data measured at one time point [6,7,9,32], their analyses may be confounded by measurement time. In fact, when we used a median of 6 weeks as a cutoff point and reanalyzed the data of one study [9], the summary estimates were different ( : SMD of 0.32 (0.16 to 0.49) for 0 to 6 weeks and 0.65 (0.52 to 0.78) for 8 to 12 weeks. is was also true when reanalyzing the data from another study [32], in which the summary estimates were different ( : SMD of 0.09 (−0.12 to 0.30) for 0 to 6 weeks and 0.64 (0.49 to 0.79) for 8 to 12 weeks. e current �ndings also show that the sham acupuncture was effective at 52 weeks. It can be argued that because sham acupunctures may not be physiologically inert [3][4][5], they may have affected the long-term pain outcomes. However, considering that they have been developed to reduce the speci�c effect of acupuncture, this is not the case. ere is also evidence that the placebo effect may be maintained over 52 weeks in the more extreme cases (e.g., surgery) [33]. It is unclear at present whether the placebo effect of acupuncture lasts over 52 weeks. erefore, we believe that the substantial magnitude (maybe ≤0.27) of the placebo effect at 52 weeks appears to be related to a response bias. is �nding supports other study [34] arguing that the response bias is a major problem in estimating the exact placebo effect.
Second, previous studies showed that the efficacy of acupuncture compared with sham acupuncture was small in magnitude [6,9,32]. However, this �nding was interpreted differently among studies. One study argued that the efficacy of acupuncture might not be clinically relevant [9]. Meanwhile, other study argued that since the total effect of acupuncture including speci�c and nonspeci�c effect appeared to be moderate in magnitude, acupuncture could be   clinically useful from a pragmatic decision maker's viewpoint [32]. ey all argued that insufficient patient blinding may be problematic in estimating the exact efficacy of acupuncture, but none of these studies presented the evidence.
Meanwhile, we did not focus on the acupuncture and sham acupuncture because there is a great controversy surrounding placebos [35]. Nonetheless, our �nding suggests that measurement time is one of important factors to study the efficacy of acupuncture in future. In a clinical practice, clinicians use acupunctures with a variety of skills. us, all types of acupuncture do not show the greatest efficacy at the same time point. However, there are few studies commenting on why they assessed the efficacy at the designated time point. If we know about the effect pattern for acupunctures tested in trials, we can show the maximized efficacy of the acupunctures.
In addition, our �nding emphasizes the importance of patient blinding. Insufficient patient blinding underrated the magnitude of the placebo effect at 8 weeks. For this reason, the efficacy of acupuncture may be exaggerated at 8 weeks in trials with insufficient patient blinding. ere are two possible reasons of why this �nding was not identi�ed in previous studies [6,7,9,32]. First, previous studies simultaneously analyzed the data at different time points, while we categorized the data based on measurement time. Second, we reanalyzed the trials to check whether patient blinding was sufficient or not. Perhaps, a combination of two reasons might induce this discrepancy.
In conclusion, the placebo effect of acupuncture showed a unique pattern according to time: e placebo effect gradually increased from baseline until 12 weeks and decreased aer 12 weeks as time passed. is pattern was affected by patient blinding. e analgesic effect of sham acupuncture was the most prominent at 12 weeks. We believe that our �ndings can assist researchers in the design and conduct of acupuncture trials for pain. ��n��c� �f �n�eres�s e authors declare that there is no con�ict of interests.