Introduction

Placebos are drugs, devices or treatments that are physically and pharmacologically inert [1]. Placebo effects may be associated with release of molecules such as dopamine [2, 3], endogenous opioids [4, 5], endocannabinoids [6], oxytocin [7], and vasopressin [8], resulting in clinical improvements in many medical conditions. A pronounced placebo effect occurs in Parkinson’s disease (PD) and was seen in 8–9% of subjects assigned to placebo in a 24-week, randomized, double-blind, placebo-controlled clinical trial, with improvement of symptoms in the order of bradykinesia (94%), rigidity (76%), gait balance/midline function (59%), and tremor (47%) [9]. These improvements were induced by dopamine release in the striatum [10], which altered neuronal activity in the basal ganglia and thalamus [2, 11, 12].

Placebo effects are particularly important in PD clinical practice because improvements are common and marked, and affect the results of clinical trials [13]. Research designs and adjustment of placebo-related factors have been proposed to minimize placebo effects and increase the success of clinical trials, but the best approach for diminishing the placebo effect remains unclear. In this vein, previous studies have suggested that the placebo effect could be associated with prior drug exposure, placebo administration period, and the severity/stage of PD [9, 14,15,16,17]; however, a systematic meta-analysis to find a way to minimize placebo effects has not been performed. Here, we identify placebo-related factors and propose a design to control these factors using a meta-analysis of randomized studies.

Methods

Search strategy

The Preferred Reporting Items for Systematic Reviews and Meta-Analysis (PRISMA) statement was utilized to guide the methodology of the meta-analysis (Supplementary Table 1) [18]. The inclusion criteria were as follows: (1) randomized, placebo-controlled, double-blind, parallel-group design; (2) diagnosis of PD using international consensus criteria including UK Parkinson’s Disease Society Brain Bank criteria [19], Gibb’s criteria [20], Calne’s criteria [21], Gelb’s criteria [22], or Ward and Gibb’s criteria [23]; (3) oral drug proved to be effective for motor symptoms in phase III clinical trials; (4) evaluation of motor symptoms as the primary endpoint; (5) assessment of change in the “on” state using UPDRS III or MDS-UPDRS III from baseline to endpoint; (6) at least 10 subjects reached the endpoint in each group; and (7) written in English. Withdrawal studies in which participants were randomized to continue the investigational drug or placebo after a defined period of the investigational drug administration were excluded. A comprehensive search of three electronic databases (PubMed, Scopus, and Cochrane Library) was conducted on 31st October, 2021. A search of ClinicalTrials.gov of the reference sections of all included articles was also performed. The search terms were (“Parkinson’s disease” OR “Parkinson disease”) AND (“random” OR “randomly” OR “randomized”) AND “placebo”. Two authors (S.H., N.M.) independently evaluated potentially eligible studies identified in the search, after which discrepancies were resolved by mutual agreement between S.H. and N.M.

Data extraction and outcome measures

Data extraction was completed by S.H. and cross-checked by N.M. Intention-to-treat data were used if possible. Extracted data included the publication year, number of study sites, with or without placebo run-in phase, proportion of patients assigned to placebo, treatment period, with or without motor fluctuation, with or without drug treatment at enrollment, age, sex distribution, UPDRS III or MDS-UPDRS III scores in “on” state from baseline to endpoint, disease duration, Hoehn and Yahr stage, levodopa daily dose, rate of withdrawal in the placebo group, and rate of withdrawal due to adverse effects in the placebo group; however, UPDRS III or MDS-UPDRS III scores at the baseline were assessed in the “Without drug at baseline” during the off period. Placebo run-in phase occurs before randomization and all study-eligible subjects are given the placebo treatment [24]. S.H. assessed the risk of bias using the risk of bias tool 2.0 [25] and N.M. cross-checked the result.

Statistical analysis

Means and standard deviations not reported in the original articles were estimated from medians, ranges, and interquartile ranges [26]. Summary statistics were calculated using the DerSimonian and Laird random-effects model [27]. The primary outcome was the change in “on” state for UPDRS III or MDS-UPDRS III scores in the placebo arm. The standardized mean change using change score standardization (SMCC), a type of standardized mean difference (SMD), was used to combine each effect (Hedge’s g). Differences were computed by single group pretest–post-test design, using the following equations: [28, 29]

$$yi=c\left(df\right)*\frac{{M}_{\text{post}}-{M}_{\text{pre}}}{{\text{SD}}_{D}}$$
$$vi=\left(\frac{1}{n}\right)+\left(\frac{{yi}^{2}}{2*n}\right)$$
$$c\left(df\right)=1-\frac{3}{4df-1},$$

where yi is the effect size; vi is the variance; Mpre is UPDRS III or MDS-UPDRS part III at baseline; and Mpost is UPDRS III or MDS-UPDRS part III at the endpoint.

Heterogeneity between studies was assessed using Q and I2 statistics, with p < 0.1 or I2 > 50% indicating significant heterogeneity. Subgroup and meta-regression analyses were applied to explore possible sources of heterogeneity. In subgroup analyses, studies were stratified into “Early” or “Advanced” with a cutoff at 50% of recruited patients with motor fluctuations. Studies categorized as “Early” were then classified as “Without drug at baseline” or “With drug at baseline” if participants had not or had received levodopa, a dopamine agonist, amantadine, or a monoamine oxidase B inhibitor at enrollment, respectively. “Advanced” studies were further categorized as “Without a run-in phase” or “With a run-in phase” if a placebo run-in phase was not or was used, respectively.

In the univariate meta-regression analysis, a standard linear mixed effects model was first applied. If this model did not fit the data, a quadratic or cubic polynomial model was used, while in multivariate meta-regression analyses, forced entry was applied to include potential covariates, and a Pearson correlation coefficient > 0.6 was used to check multicollinearity among covariates. The following covariates were prespecified to be included in subgroup and meta-regression analyses: with or without motor fluctuation, with or without drug at baseline, with or without a placebo run-in phase, age, sex distribution, and levodopa daily dose. Publication bias was assessed using visual inspection of funnel plots, and adjusted using the trim-and-fill method [30]. Sensitivity analyses were also performed. p < 0.05 was considered to be significant. All analyses were carried out using the “meta”, “metafor”, “ggplot2”, “regplot”, and “corrplot” packages in the R statistical computing environment ver. 4.0.3 (http://www.r-project.org).

Results

Study characteristics

A total of 38 studies [31,32,33,34,35,36,37,38,39,40,41,42,43,44,45,46,47,48,49,50,51,52,53,54,55,56,57,58,59,60,61,62,63,64,65,66,67,68] with 4,828 subjects were included in the meta-analysis (Fig. 1). Of these studies, 21 had a low risk of bias and 17 had a moderate risk (Supplementary Fig. 1). Sixteen studies were in “Early” PD and 22 in “Advanced” PD patients. Of the “Early” studies, 9 were “Without drug at baseline” and 7 were “With drug at baseline”. The “Advanced” studies included 19 “Without a run-in phase” and 3 “With a run-in phase”. The characteristics of all the studies are shown in Table 1. The pooled mean baseline data were: treatment period (range 4–38.6 weeks), study sites (1–129), number of patients administered placebo (20–595), sex distribution (male, 36.4–80.0%), age (59.5–70.2 years), Hoehn and Yahr stage (1.5–3.0), “on” state using UPDRS III or MDS-UPDRS III at baseline (13.9–32.1), levodopa daily dose (0–948 mg/day), and use of the following drugs: entacapone, istradefylline, nebicapone, opicapone, pardoprunox, piribedil, pramipexole, rasagiline, safinamide, tolcapone, tavapadon, and zonisamide.

Fig. 1
figure 1

Flowchart for selection of eligible studies

Full size image
Table 1 Characteristics of the included studies
Full size table

Placebo effect in patients with Parkinson’s disease

Placebo significantly improved UPDRS III or MDS-UPDRS III in heterogenous studies (SMD = − 0.25, 95% CI − 0.35 to − 0.14, p < 0.001, I2 = 92%; Fig. 2). In subgroup analysis, placebo was not significant in “Early” studies (SMD = − 0.13, 95% CI − 0.30 to 0.04, p = 0.15, I2 = 94%; Fig. 2), but was significant in “Advanced” studies (SMD = − 0.33, 95% CI − 0.41 to − 0.25, p < 0.001, I2 = 70%; Fig. 2). The placebo effect in “Advanced” studies was significantly higher than that in “Early” studies (p = 0.04; Fig. 2). Stratification of the “Early” and “Advanced” studies revealed significant differences in the placebo effect among four subgroups (p < 0.01). In “Early” studies, the placebo effect was not significant in those “Without drug at baseline” (SMD = − 0.01, 95% CI − 0.24 to 0.22, p = 0.94, I2 = 93%), but was significant in those “With drug at baseline” (SMD = − 0.24, 95% CI − 0.33 to − 0.15, p < 0.001, I2 = 39%) (group difference: p = 0.06; Fig. 3). In “Advanced” studies, the placebo effect was significantly higher “With a run-in phase” (SMD = − 0.46, 95% CI − 0.57 to − 0.34, p < 0.001, I2 = 0%) than “Without a run-in phase” (SMD = − 0.31, 95% CI − 0.40 to − 0.22, p < 0.001, I2 = 71%) (group difference: p = 0.04; Fig. 3).

Fig. 2
figure 2

Forest plot showing placebo effects in studies stratified into “Early” and “Advanced” PD. Results on the left indicate improvement with placebo. There was a significant difference in placebo effect between “Early” and “Advanced” studies (p = 0.04)

Full size image
Fig. 3
figure 3

Forest plot showing placebo effects in studies “Without drug at baseline”, “With drug at baseline”, “Without a run-in phase”, and “With a run-in phase”. “Early” were subdivided to “Without drug at baseline” and “With drug at baseline”. “Advanced” were stratified to “Without a run-in phase” and “With a run-in phase”. There were significant differences between groups (p < 0.01). The placebo effect was not significant in “Without drug at baseline” studies, but was higher in studies “With a run-in phase” than “Without a run-in phase” (p = 0.04). Run-in phase, which was originally expected to attenuate the placebo effect, did not suppress the placebo effect

Full size image

Meta-regression analysis

The results of univariate and multivariate meta-regression analyses for all studies and for studies “Without drug at baseline” and “Without a run-in phase” are presented in Supplementary Table 2. Bubble plots using univariate meta-regression analysis are shown in Fig. 4. For all studies, univariate meta-regression analyses showed that “Without drug at baseline” (coefficient − 0.33, 95% CI − 0.51 to − 0.16, p < 0.001, R2 = 56.8%), longer placebo administration period (coefficient 0.02, 95% CI 0.01 to 0.03, p < 0.001, R2 = 54.0%), and lower UPDRS III or MDS-UPDRS III scores at baseline (coefficient − 0.03, 95% CI − 0.05 to − 0.01, p < 0.01, R2 = 41.1%) were significantly related to a lower placebo effect (Fig. 4a–c). Disease duration was not significantly associated with a placebo effect using a linear mixed effects model (coefficient − 0.03, 95% CI − 0.05 to 0.00, p = 0.05, R2 = 32.6%; Fig. 4d), but a quadratic polynomial model revealed a significant association (coefficient − 0.16, 95% CI − 0.26 to − 0.07, p < 0.01, R2 = 30.7%; Supplementary Fig. 2). Univariate meta-regression analysis showed that the placebo effect dissipated after 225 days of placebo administration.

Fig. 4
figure 4

Bubble plots for studies “Without drug at baseline” and “Without a run-in phase”. In univariate meta-regression analysis, a lower placebo effect was significantly associated with “Without drug at baseline”, longer placebo administration period, and lower UPDRS III or MDS-UPDRS III scores at baseline. There was no significant association between placebo effect and disease duration. The placebo administration period and disease duration explained heterogeneity in “Without drug at baseline” and “Without a run-in phase” studies, respectively

Full size image

Multivariate meta-regression analysis revealed that the four significant factors in univariate analysis explained 84% of the variance in estimates across studies (“Without drug at baseline”: coefficient − 0.42, 95% CI − 0.60 to − 0.23, p < 0.001; placebo administration period: coefficient 0.02, 95% CI 0.01 to 0.03, p < 0.001; UPDRS III or MDS-UPDRS III scores at baseline: coefficient − 0.02, 95% CI − 0.03 to 0.00, p = 0.03; disease duration: coefficient 0.04, 95% CI 0.01 to 0.06, p < 0.01; all moderators: p < 0.001, R2 = 83.7%). No significant correlation was found between these covariates.

Meta-regression analyses of studies “Without drug at baseline” and “Without a run-in phase” were performed to explore the cause of heterogeneity. In those “Without drug at baseline”, a longer placebo administration period was a significant predictor of a lower placebo effect and the magnitude of the effect reached zero after 154 days of administration (coefficient 0.03; 95% CI 0.02 to 0.04; p < 0.001; R2 = 83.8%; Fig. 4e). In “Without a run-in phase” studies, a longer disease duration was significantly associated with a lower placebo effect (coefficient 0.09; 95% CI 0.05 to 0.13; p < 0.001; R2 = 72.1%; Fig. 4f).

Publication bias and sensitivity analysis

Visual inspections of funnel plots showed symmetry for “Without a run-in phase” studies, but asymmetry for the other three groups (Supplementary Fig. 3). The trim-and-fill adjusted results were stable for “Without a run-in phase” (SMD = − 0.31, 95% CI − 0.40 to − 0.22, p < 0.001, I2 = 71.1%; Supplementary Fig. 3D) and attenuated the placebo effect in the other three groups (“Without drug at baseline”, SMD = 0.29, 95% CI 0.05 to 0.52, p = 0.02, I2 = 94.8%; “With drug at baseline”, SMD = − 0.19, 95% CI − 0.28 to − 0.09, p < 0.001, I2 = 52.7%; “With a run-in phase”, SMD = − 0.45, 95% CI − 0.55 to − 0.34, p < 0.001, I2 = 0.0%; Supplementary Fig. 3A, B, C). Jackknife sensitivity analysis showed that all results in the meta-analysis were highly reproducible (Supplementary Figs. 4 and 5). These findings indicate the stability of the results.

Discussion

In this meta-analysis, the placebo effect in PD was analyzed separately for all studies and for studies in early and advanced stages of PD, with the following results. (1) There was no placebo effect in studies without use of a drug at baseline. (2) In the analysis in all studies, the placebo effect decreased as the placebo administration period increased, and the effect disappeared after about 7 months. This trend was particularly pronounced for studies without use of a drug at baseline in early-stage PD. (3) The placebo effect was lower in studies of advanced PD without a placebo run-in phase than with a placebo run-in phase. (4) The placebo effect was lower in all studies and in early-stage PD studies with a drug at baseline as disease duration was shorter, and in advanced-stage PD studies as disease duration was longer.

The absence of a placebo effect for studies without a drug at baseline is due to the effect being related to previous drug exposure and its learning effect [14, 69]. Therefore, the first placebo administration does not induce a placebo effect [14, 70]. The current results are consistent with previous reports, although the patients in studies without a drug at baseline were not strictly drug-naive due to inclusion of some patients with a history of prior drug use. The placebo effect may become even smaller if only drug-naive patients are included. Among the studies analyzed, only two reported the number of drug-naive patients [33, 37]. Given that the history of prior drug use affects the placebo effect, the proportion of drug-naive patients should be presented in future clinical trials.

There are multiple lines of evidence that the placebo effect is largely related to expectancy [71, 72]. The expectancy of this effect fades over time. The decrease in the placebo effect with an increased duration of placebo administration and the disappearance of the placebo effect about 230 days after the start of administration are in line with a systematic review that found that placebo-associated improvements occurred throughout a 6-month study [9]. The positive correlation between the placebo effect and duration of placebo administration was strong in studies without a drug at baseline in early stage PD, suggesting that the duration of placebo administration is one of the main causes of heterogeneity in studies without a drug at baseline.

In advanced PD, the placebo effect was enhanced by a placebo run-in phase. Such a run-in phase was originally expected to attenuate the placebo effect [24], but questions have recently been raised concerning placebo suppression by a placebo run-in phase in other diseases [15]. Time and expense spent on clinical trials can be saved if a placebo run-in phase is not required. However, we note that in the analyzed studies this phase was relatively short (1–4 weeks) and a longer period might result in placebo suppression.

The placebo effect was significantly lower with a longer disease duration in studies of advanced PD, but significantly lower with a shorter disease duration in those in early stage PD with a drug at baseline. Thus, the relationship between disease duration and placebo effect differs between early and advanced stages and may be biphasic depending on the disease stage, which was supported by the quadratic polynomial model (Supplementary Fig. 2). The ventral striatum has been linked to the placebo effect. In pain, placebo treatments induce a functional MRI response and dopamine release in the ventral striatum measured by PET. Activation of the ventral striatum during pain is a predictor of high efficacy of opioid analgesia [73]. Moreover, patients with pathological gambling tend to have a higher placebo effect and upregulate release of dopamine in the ventral striatum [74]. PD is characterized by different degeneration in the substantia nigra pars compacta (SNpc) and ventral tegmental area in the midbrain, resulting in different dopamine levels in the efferent dorsal striatum and ventral striatum. SNpc is more affected than the ventral tegmental area in early PD, leading to dominant dopamine depletion of the dorsal striatum. Therefore, the “overdose hypothesis” proposes that dopaminergic therapy overstimulates the ventral striatum [75]. In early PD, this overstimulation becomes stronger as the disease progresses, whereas in advanced PD, the overstimulation is weakened because of degeneration in the ventral tegmental area. In addition, negative experiences such as dyskinesia and visual hallucinations increase with disease duration in advanced PD, which may attenuate the placebo effect. Thus, different disease stage-dependent degeneration may lead to differences in the placebo effect between early and advanced stages.

The placebo effect has been reported to be larger with a higher UPDRS III score at baseline [17], and placebo administration facilitates more dopamine release in patients with severe symptoms than in those with mild symptoms, suggesting that disease severity determines the size of the placebo effect [10]. Using all studies in this analysis, the placebo effect was positively correlated with UPDRS III and MDS-UPDRS III scores at baseline, in agreement with previous reports [17].

There are several limitations to keep in mind when interpreting the results of this study. A meta-analysis demonstrated a placebo effect in sham surgery during the off state [76], whereas the current study only assessed the placebo effect in the on state, due to lack of sufficient studies using UPDRS in the off state. Non-motor symptoms, such as cognition, mood and psychosis, were also not included in this analysis because effects of these factors were not shown in the included studies. Also, despite correction for publication bias using the trim-and-fill method, this bias could not be excluded.

Conclusion

This meta-analysis identified influences of treatment history and disease duration on the placebo effect in PD treatment. The effect was lower for groups with drugs at baseline in early stage PD as disease duration was shorter, and in advanced-stage PD as disease duration was longer. The placebo effect disappeared about 7 months after administration of placebo. A placebo run-in phase failed to attenuate the placebo effect. Based on these findings, we recommend that future randomized controlled clinical trials for patients with PD in the early and advanced stages match treatment history and disease duration between the placebo and active drug groups, and use a follow-up period of at least 7 months. In addition, a placebo run-in phase is not recommended. Knowledge of the factors involved in the placebo effect will both improve the quality of randomized controlled clinical trials and enhance drug efficacy for patients in clinical practice.