FormalPara Key Points

Exercise training does not appear to be more effective than placebo interventions for reducing pain intensity when the placebo intervention consisted of sham electrotherapeutic interventions or oral supplementation (omega-3 plus calcium) in individuals with chronic pain.

There are currently no published randomised controlled trials that have utilised an exercise-based placebo intervention in people with chronic primary musculoskeletal pain conditions. This is important because the relative effect of exercise training, contextual factors and natural history remains unknown.

When considered together, exercise training and the associated clinical encounter was more effective than no treatment and standard medical care for reducing pain intensity in people with chronic primary musculoskeletal pain.

1 Background

Chronic musculoskeletal pain affects approximately 20% of the global population [1], impacting work capacity and employment [2], quality of life [3], and mental and physical health [4, 5], and may cause social disadvantage [6]. Pain is the most disabling characteristic of musculoskeletal disorders [7], and is defined as a distressing experience associated with actual or potential tissue damage with sensory, emotional, cognitive and social components [8]. Chronic (persistent) pain is further defined as constant or recurrent pain for a period of time greater than 3 months [9]. Exercise training is a fundamental modality in various position stands and clinical guidelines to manage musculoskeletal pain and disability [10]; however, one study estimated that half (55.2%) of the effect of exercise on pain is related to the exercise prescription variables [11]. The specific mechanisms explaining the effect of exercise training on long-term reductions in pain is not clear [12].

The multidimensional nature of the chronic pain experience [8] may be influenced by factors other than an exercise training stimulus, such as the placebo effect. Previous evidence shows that the overall effect of exercise training on perception, mental health, cognition, and sports performance can be partly explained by placebo effects [13,14,15]. Placebo effects are elicited broadly via two main mechanisms: learning mechanisms (i.e., classical conditioning) and cognitive expectations [16]. In clinical practice, this is referred to as contextual factors that are associated with the experiences of the patient during the clinical encounter, shaped by positive context linked to the clinician’s characteristics (professional reputation, appearance, beliefs, behaviours); patient characteristics (expectations of treatment, beliefs, treatment preference, previous experience, type of injury, sex, age); treatment (diagnostic clarity, diagnostic method, intensity of therapy, observational learning, patient-centred approach, therapeutic touch); and the healthcare setting (environment, architecture, interior design) [17,18,19]. Positive contextual factors as part of a therapeutic encounter in clinical practice can reduce pain intensity [20]. The therapeutic encounter, and in particular the contextual factors surrounding the clinical encounter, are therefore important considerations in clinical practice. In particular, these factors seem to be extremely important during manual therapy where physiotherapists use context to boost placebo and simultaneously reduce nocebo effects [21]. Indeed, two recent Italian surveys showed that half of the physiotherapists working on musculoskeletal disorders are fully aware of frequently using “contextual factors” in their therapies [22], and more than half of the interviewed patients affected by musculoskeletal pain describe contextual factors as a treatment without a specific effect while still believing in its clinical effectiveness [23].

Previous meta-analyses [24, 25] have shown that the specific effect of treatment is modest when compared to the placebo response (i.e., combined relative effects of placebo effects and other non-specific effects such as Hawthorne effect [26] regression to the mean, natural history and spontaneous improvement [27]). One recent meta-analysis showed the placebo response represented approximately 75% of the overall effect on pain intensity reduction for a range of interventions (i.e., electrotherapeutic, pharmacological, non-pharmacological and surgical treatments) in patients with osteoarthritis [28]. Most notably, no exercise-based interventions were included in this review because of the perceived lack of randomised placebo-controlled exercise trials. The authors speculated that this was likely due to the difficulty in designing an adequate exercise placebo intervention. While this perspective is likely accurate given a subsequent meta-analysis [25] in osteoarthritis (knee) identified just one exercise-based placebo controlled study, it brings into question the scientific rigor surrounding the broad support of exercise training in many position statements and its use in the management of musculoskeletal pain conditions.

It is well accepted that psychological and social factors can moderate the pain experience for patients with chronic musculoskeletal pain [29]. To our knowledge, there are no meta-analyses to date that report the relative effect size of the placebo effect for exercise-based interventions in chronic musculoskeletal pain disorders. Therefore, the aim of this meta-analysis was to determine the relative contribution of the placebo effect, non-specific effects (i.e., natural history, regression to the mean, spontaneous improvement and the Hawthorne effect [26]), and exercise training for reducing pain in adults with chronic musculoskeletal pain. This is important for clinical practice to inform the relative contribution of exercise training and the placebo effects for enhancing patient outcomes [30, 31].

2 Methods

This systematic review and meta-analysis was completed in accordance with the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) [32]. The review was registered prospectively with PROSPERO (CRD42019141096).

2.1 Search Strategy

Five online databases (MEDLINE, CINAHL, SPORTDiscus, EMBASE and CENTRAL) were electronically searched for research published from database inception to February 2021. The search-term strategy can be found in Supplementary Table S1 (Online Supplementary Material, OSM). The search strategy was developed based on the basis of current guidelines for the design of systematic reviews, our prior experience with systematic reviews, and input from content experts. The search had the following limits: MEDLINE (all adult: 19 + years; human), CINAHL (exclude MEDLINE records; human, randomised controlled trials; journal article; all adult), SPORTDiscus (Academic Journal), EMBASE (RCT; not MEDLINE; adult; article) and CENTRAL (trials). To locate additional references, we searched for previously published systematic reviews identified via the Cochrane Database of Systematic Reviews (search terms: placebo exercise; limits: none) and GoogleScholar (search terms: ‘systematic review’ placebo exercise; limits: previous 10 years). There were no additional restrictions for language or year of publication. All results of the search were screened by PJO to exclude duplicates. Independent screening of the titles and abstracts of the remaining studies was completed by JB, CTM, PJO and CAT against the predetermined inclusion and exclusion criteria. The full-text articles were independently assessed against the inclusion and exclusion criteria by four reviewers (CTM, PJO, JB and CAT). Any disagreements were adjudicated by CTM and discussed with co-authors as necessary. Excluded papers at the full-text stage were reassessed against the inclusion and exclusion criteria by KS.

2.2 Inclusion and Exclusion Criteria

To be included, studies were required to be published in a peer-reviewed journal (i.e., grey literature excluded) and be a randomised controlled trial that compared an exercise training intervention to either a non-intervention control group or to a placebo group. All other inclusion criteria followed the Participants, Interventions, Comparators, Outcomes and Study design (PICOS) framework [33].

2.2.1 Participants

Study participants were required to be adults (≥ 18 years) with any chronic primary musculoskeletal pain. Chronic pain was defined as pain duration at baseline ≥ 3 months [34]. There were no restrictions on sex or race. The exclusion criteria consisted of conditions that are not primary musculoskeletal pain in nature or are a result of structural compromise such as fracture, severe scoliosis, osteoporosis and bone pain secondary to metastases. Primary inflammatory conditions such as gout, rheumatoid arthritis, ankylosing spondylitis, psoriatic arthritis, reactive arthritis, juvenile arthritis or secondary pain complaints associated with visceral pain presenting as musculoskeletal pain such as Crohn’s disease and other gastrointestinal disorders, angina, vascular insufficiency, asthma or other breathing-related conditions were also excluded.

2.2.2 Interventions

Interventions consisting of exercise training alone, without the addition of any other treatments (e.g., massage, ultrasound or hot and cold therapy, or education) were included.

2.2.3 Comparators

Comparator control groups consisted of: true control (wait-list control or no treatment control), usual care (standard medical care excluding physical therapies, education, psychotherapies or surgery), or placebo control. A placebo control intervention was defined as any intervention defined as a placebo or sham intervention by the study authors [35].

2.2.4 Outcomes

Any general or disease-specific measure of pain such as a visual analogue or numeric pain scale for pain intensity or Short Form (SF)-36 bodily pain subscale was included.

2.2.5 Study Design

Studies were included if the design was a parallel-arm (individual- or cluster-designed) randomised controlled trial.

2.3 Data Extraction

After duplicate removal (PJO), data screening was completed using Covidence (https://www.covidence.org/). Data extraction was completed in duplicate by four independent assessors (PJO, CTM, CAT and JB). Extracted information included relevant publication information (i.e., author, title, year, journal), study design, number of participants, participant characteristics (e.g., age and sex), intervention details (e.g., duration, type) and outcome measure (pain). Extracted outcome data were pre- and post-intervention mean and standard deviation (SD) for pain intensity. Data presented as median (interquartile range) or alternate measures of spread/variance were converted to mean and SD using established formulae. Where post-intervention SD was unavailable, pre-intervention SD was utilised based on Cochrane guidelines [36, 37]. In all instances where studies were potentially included, yet where data required for meta-analysis were not available, authors were contacted a minimum of three times over a 4-week period to request the information and final decision for inclusion. Similarity between extracted data from the independent assessors was evaluated through Covidence. Any discrepancies were reviewed by KS and CTM against the original paper. This method was piloted on the first ten studies chosen at random prior to commencing data extraction.

2.4 Risk of Bias Assessment and Quality Assessment

The Cochrane Collaboration Risk of Bias Tool was used to examine potential selection bias (random sequence generation and allocation concealment), performance bias (blinding of patients and personnel), detection bias (blinding of outcome assessment), attrition bias (incomplete outcome data), reporting bias (selective outcome reporting) and other bias [36]. This assessment was completed independently by CAT and JB. Studies were classified as having a low, high or unclear (when reporting was not adequate to rate a specific domain) risk for each type of bias. Any disagreements on the risk of bias were adjudicated by CTM. In addition, to assess the quality of the evidence from the meta-analysis, the Grading of Recommendations Assessment, Development and Evaluation (GRADE) approach was used [38].

2.5 Statistical Analysis

Pairwise random-effects meta-analysis was conducted in Stata 16.0 (Stata Corp, College Station, TX, USA). As all outcomes of interest were continuous, yet possibly subject to small sample bias, Hedges’ g, rather than Cohen’s d, was used as the standardised mean difference effect estimate [39]. In line with Cochrane guidelines, individual study groups were pooled when a study investigated multiple groups defined as exercise training to avoid overlapping samples [40]. The main analysis investigated pooled exercise training versus placebo comparators on pain intensity. Secondary (sub-group) analyses were performed for: (1) exercise training versus any comparator (i.e., placebo, true or usual-care control), (2) exercise training versus true control, and (3) exercise training versus usual-care control. Heterogeneity was assessed for all pairwise comparisons via the I2 statistic [40] and publication bias via visual inspection of funnel plots [see Figs. S1–S4 (OSM)] in addition to calculating the P-value of Egger’s test. Sensitivity analyses included: excluding each individual study from the main analysis [pooled exercise training vs. placebo comparators on pain intensity, Fig. S5 (OSM)] and omitting Mengshoel et al. [41] [Fig. S6 (OSM)] due to conversion of median (minimum, maximum) to mean (SD) in all applicable analyses. An alpha level of 0.05 was taken for statistical significance. The Stata code and data are included in Table S2 (OSM).

3 Results

3.1 Study Selection

A summary of the systematic review process is presented in Fig. 1 according to the PRISMA guidelines. There were 5,263 studies screened against the inclusion and exclusion criteria for title and abstract after removal of duplicates. A total of 426 studies were included in the full-text screening with another 345 subsequently being assessed as ineligible for inclusion [Table S3 (OSM): Detailed reasons for exclusion at full-text stage]. Three studies were identified as potentially eligible, concerning which authors were contacted; however, all were ineligible due to essential pain data being unavailable from the authors [42, 43]. The authors of one study [44] provided raw pain data for analysis and was eligible for inclusion. A total of 79 studies were eligible for meta-analysis. Data were presented as median, minimum, and maximum in one study [41], 95% confidence intervals (CIs) in four studies [45,46,47,48], and standard error of the mean in four studies [49,50,51,52], and were converted to mean (SD).

Fig. 1
figure 1

Preferred reporting items for systematic reviews and meta-analysis (PRISMA) diagram of the study screening process

Full size image

3.2 Study Characteristics

The details of each included study (n = 79; participants: n = 4843) [41, 44,45,46,47,48,49,50,51,52,53,54,55,56,57,58,59,60,61,62,63,64,65,66,67,68,69,70,71,72,73,74,75,76,77,78,79,80,81,82,83,84,85,86,87,88,89,90,91,92,93,94,95,96,97,98,99,100,101,102,103,104,105,106,107,108,109,110,111,112,113,114,115,116,117,118,119,120,121,122] are shown in Table 1. The sample size of the individual groups within the included studies ranged from seven to 125 participants and mean age ranged from 20 to 76 years. The length of the intervention ranged from 2 to 104 weeks. Of the included studies, 23 investigated adults with fibromyalgia [41, 44, 46, 47, 52,53,54,55,56,57,58,59,60,61,62,63,64,65, 111, 117, 119,120,121]; 18 with knee or hip osteoarthritis [45, 49, 50, 66,67,68,69,70,71,72,73,74,75,76,77,78, 114, 115]; three with chronic patellofemoral pain [79,80,81]; one with chronic hip pain in young adults [82]; one with older adults with chronic lower extremity or low back pain [83]; 23 with chronic low back pain [51, 84,85,86,87,88,89,90, 92,93,94,95,96,97,98,99,100,101, 110, 112, 113, 116, 118]; seven with chronic neck or shoulder pain [48, 102,103,104,105,106,107]; one study with chronic Achilles tendinopathy pain [108]; and one study with chronic osteoarthritis of the hand [109].

Table 1 Characteristics of included studies
Full size table

3.3 Risk of Bias Within Individual Studies

A summary of the risk of bias assessment is shown in Fig. 2, and for each study is shown in Tables S4–S6 (OSM). Overall, no studies were rated as being at low risk of bias, and all were rated as being at high risk of bias (Fig. 2). The primary reason for a high risk of bias was the lack of participant blinding (performance bias; 96.2% high risk) across the majority of studies. Additionally, participants were the outcome assessor for pain in all studies, (detection bias; 100% high risk), which was subjectively reported in the majority of cases via a numeric pain-rating scale or a visual analogue scale. Attrition bias (87.3%), selective reporting bias (88.9%), selection bias: sequence generation (68.4%) and selection bias: allocation concealment (60.8%) were predominately rated as low, with some studies rated as being unclear due to a lack of clear reporting.

Fig. 2
figure 2

Percentage of studies examining the effect of exercise training for reducing pain with low risk, some concerns (unclear) and high risk of bias for each aspect of the Cochrane Risk of Bias Tool (revised version). See Tables S3-S5 (Online Supplementary Material) for the assessment for each individual study. a exercise versus placebo; b exercise training versus all controls; c exercise training versus true control; d exercise training versus usual care

Full size image

3.4 Quantitative Analysis

Four studies were eligible for quantitative analysis comparing the effect of exercise interventions to placebo intervention [60, 69, 80, 85]. There were insufficient placebo-controlled trials from which to perform a quantitative synthesis of the existing evidence to directly determine the effect of natural history, placebo and exercise. A pairwise meta-analysis comparing exercise training to placebo showed that exercise was not more effective for reducing pain when compared to a placebo intervention (g [95% CI]: 0.94 [− 0.17, 2.06], P = 0.098, I2 = 92.46%, studies: n = 4, participants: n = 253; Fig. 3, Table 2). There was no evidence of publication bias within the comparison (Egger’s P = 0.250; Table 2). The overall quality of evidence was rated as very low based on the GRADE criteria. Exploratory meta-regression for the primary outcome data using mean sample baseline pain intensity (studies: n = 4) showed this was unlikely to be a source of heterogeneity [Table S7 (OSM)].

Fig. 3
figure 3

Forest plot for the meta-analysis investigating the effectiveness of exercise training versus placebo comparators for reducing musculoskeletal pain. Horizontal lines represent standardised mean difference (Hedges’ g) and 95% confidence intervals. The size of the box represents the weight of each study. The diamond represents the overall estimated effect. INT intervention group, CON control group

Full size image
Table 2 Summary of findings for treatment comparisons
Full size table

When pooling all exercise interventions and comparing to all control comparator groups (true control, usual-care control, placebo control), exercise was more effective than control for reducing pain (g [95% CI]: 0.84 [0.64, 1.04], P < 0.001, I2 = 90.02%, studies: n = 79, participants: n = 4843; Fig. 4, Table 2). There was strong evidence of publication bias within the comparison (Egger’s P < 0.001; Table 2). The overall GRADE quality of evidence was considered very low. The authors of two studies [110, 114] reporting larger than anticipated effect sizes (Fig. 4) were contacted to confirm accuracy of reported data, but no response was received. Sensitivity analysis omitting Mengshoel et al. [41] due to conversion of median (minimum, maximum) to mean (SD) did not change the results [Fig. S6 (OSM)].

Fig. 4
figure 4

Forest plot for the meta-analysis investigating the effectiveness of exercise training versus all control comparators for reducing musculoskeletal pain. Horizontal lines represent standardised mean difference (Hedges’ g) and 95% confidence intervals. The size of the box represents the weight of each study. The diamond represents the overall estimated effect. INT intervention group, CON control group

Full size image

Sub-group analysis for the effect of exercise interventions compared to true control (do-nothing control, wait-list control) comparators showed that exercise was more effective than true control for reductions in musculoskeletal pain (g [95% CI]: 0.99 [0.66, 1.32], P < 0.001, I2 = 92.43%, studies: n = 42, participants: n = 2361; Fig. 5, Table 2), however, there was strong evidence of publication bias within the comparison (Egger’s P < 0.001; Table 2). The overall GRADE quality of evidence was considered very low. Sensitivity analysis omitting Mengshoel et al. [41] due to conversion of median (minimum, maximum) to mean (SD) did not change the results [Fig. S6 (OSM)].

Fig. 5
figure 5

Forest plot for the meta-analysis investigating the effectiveness of exercise training versus true control comparators for reducing musculoskeletal pain. Horizontal lines represent standardised mean difference (Hedges’ g) and 95% confidence intervals. The size of the box represents the weight of each study. The diamond represents the overall estimated effect. INT intervention group, CON control group

Full size image

When comparing the effect of exercise against usual-care control comparator (general practitioner standard care but not physical therapies) groups, exercise was more effective than usual care for reducing musculoskeletal pain (g [95% CI]: 0.64 [0.44, 0.83], P < 0.001, I2 = 76.52%, studies: n = 33, participants: n = 2229; Fig. 6, Table 2), and there was strong evidence of publication bias within the comparison (Egger’s P < 0.001; Table 2).

Fig. 6
figure 6

Forest plot for the meta-analysis investigating the effectiveness of exercise training versus usual-care control comparators for reducing musculoskeletal pain. Horizontal lines represent standardised mean difference (Hedges’ g) and 95% confidence intervals. The size of the box represents the weight of each study. The diamond represents the overall estimated effect. INT intervention group, CON control group

Full size image

3.5 Protocol Deviations Compared with PROSPERO Registration

We initially aimed to conduct a network meta-analysis if data permitted, yet the lack of placebo trials precluded such analyses. Among these few trials (n = 4) there was marked heterogeneity (I2 = 92.46%), which we investigated via meta-regression using mean total sample baseline pain intensity [Table S7 (OSM)]. Given these meta-regressions were post hoc and included less than ten trials, we contend results are purely exploratory in nature.

4 Discussion

The primary aim of this systematic review and meta-analysis was to investigate the relative effect of exercise training, placebo effects and non-specific effects on pain in chronic primary musculoskeletal pain conditions. Only four randomised controlled trials that compared an exercise intervention to placebo were identified: one each in osteoarthritis [69], chronic low back pain [85], chronic patellofemoral pain [80] and fibromyalgia [60]. None of these studies used a placebo exercise-training protocol. In light of the lack of exercise-based placebo-controlled trials, the determination of effect size for placebo effects relative to non-specific effects and exercise was not feasible. Pairwise meta-analysis comparing exercise interventions to non-exercise placebo control showed there was no statistically significant evidence for exercise training to be more effective than placebo for reductions in pain intensity; however, the quality of evidence was rated as very low against the GRADE criteria. There was strong evidence of an approximately large effect size for exercise training when compared to all controls (pooled); however, there was also strong evidence of publication bias and the quality of evidence was rated as very low when assessed against the GRADE criteria.

This systematic review highlights that exercise training does not appear to be more effective than placebo interventions when the placebo intervention consisted of sham electrotherapeutic interventions or oral dietary supplementation in individuals with chronic pain. Additionally, there was a lack of placebo-controlled clinical exercise training trials in chronic musculoskeletal pain conditions from which to determine the relative effects of exercise training, placebo effects and non-specific effects. Previous evidence showed that subjective outcomes, such as pain, are particularly influenced by placebo effects [123], and therefore it is critical to understand the relative contribution of exercise and the placebo effects associated with the clinical encounter. A recent meta-analysis evaluating the effect of various interventions on pain for individuals with osteoarthritis showed that the placebo response (placebo effects and non-specific effects such as natural history, regression to the mean, Hawthorne effect) contributed to 75% of the effect on pain [28]. Another meta-analysis showed that placebo interventions were more effective for reducing pain than no treatment for individuals with fibromyalgia [124]. The limitation with our meta-analysis and these previous meta-analyses is that no exercise-based placebo trials were identified in the published literature. This is important to recognise because the effect size of a placebo intervention on pain can be moderated by the type of placebo administered [125,126,127], and that matching (ensuring that the active treatment is indistinguishable from the placebo) is an important component of placebo study design [128]. Although a placebo response is likely involved in all interventions for pain as part of a clinical encounter [129], the magnitude of effect for placebo as part of an exercise-based intervention for individuals experiencing chronic musculoskeletal pain remains unknown. A better understanding of the direct effect of exercise and that of the placebo response is important as it will enable the direction of research either toward optimising exercise prescription variables or to enhance the contextual factors associated with the clinical encounter for individuals experiencing chronic pain. It is therefore critical that future randomised clinical trials consider including an exercise treatment arm, no-treatment control arm and placebo arm into the design to determine the true effect of exercise, placebo effects and non-specific effects for the management of chronic primary musculoskeletal pain. Due to the amount of funding, time, participants and personnel associated with conducting a three-arm clinical trial, it may be more pragmatic for researchers to consider a trial design that includes a treatment arm and placebo arm alone. These trials can be used to calculate and report the overall treatment response, the proportional contextual effect (inclusive of natural history and regression to the mean), and therefore the specific treatment effect [130]. The overall treatment effect (change in baseline in the active intervention) should be reported in addition to the proportional contextual effect by means of the improvement of the placebo group from baseline divided by the improvement in the active treatment group [130]. It must be emphasised that the design of the placebo intervention as well as the active intervention should be considered. As noted earlier, the type of placebo intervention selected should be indistinguishable from the active treatment. [128, 131] Importantly, the therapeutic ritual, environment, treatment provider and contact time should also be identical [131, 132], as far as practicable, and documented as part of a treatment procedural manual, often referred to as manualised treatments in psychotherapy literature [133]. These design elements are easily achievable for pharmacotherapy trials where the active or sham ingredient is encased within an indistinguishable pill, or the use of sham electrotherapeutic devices such as detuned ultrasound or shortwave diathermy where the visual and audible characteristics resemble the active devices. The design of an exercise-based placebo arm requires additional consideration due to the complexity of the treatment and necessary clinical interaction. A clear understanding of the hypothesised mechanisms of action for exercise on pain that is relative to the specific condition is important. For exercise training interventions, this may involve the manipulation of intensity or load, while maintaining the same exercise type, duration, frequency, method of progression and the surrounding contextual factors of the environment, therapeutic ritual, therapeutic alliance, clinician and language used [131].

Further analyses evaluated effectiveness trials comparing exercise interventions to usual care (i.e., standard medical care) and to no-treatment controls (i.e., true control). We found statistically strong evidence that exercise training was more effective than usual care and no treatment; however, the quality of evidence is very low according to the GRADE criteria. The exact nature and intensity of the usual-care treatment offered or permitted outside of study participation was not well reported in most studies, and therefore may confound observed effects. Most studies reported that standard care consisted of patient-initiated medical management consisting of pharmacotherapy and general education provided by healthcare professionals outside the study research group. Contextual factors as part of a therapeutic encounter in standard clinical practice [20] are consistent with placebo effects reported in clinical trials [126], and therefore a control arm consisting of standard care may provide some insight into the effect of the therapeutic encounter in the absence of a true placebo control. However, none of the studies included in this meta-analysis reported the specific nature and intensity of treatment provided to the usual-care control group and none attempted to standardise this form of treatment. There is inherent variation in treatment received for participants in such effectiveness trials, and this adds to external validity [134]; however, the use of a usual-care control group outside the confines of the study design prevents meaningful interpretation of the specific effects of exercise on pain in this context.

The findings from our meta-analysis are consistent with previous meta-analyses in fibromyalgia [135, 136], osteoarthritis [137, 138], chronic low back pain [139] and chronic neck pain [140], which all report that exercise provides greater reductions in self-reported pain when compared to usual care and true controls. Therefore, the results presented in our current systematic review and meta-analysis support that the experience and act of engaging in an exercise treatment intervention and its associated interactions appears to be more effective than not receiving any care or receiving standard medical care in individuals with chronic musculoskeletal pain. Our meta-analyses suggested a high level of heterogeneity between included studies (i.e., each meta-analysis was graded down via GRADE). We explored this heterogeneity in the primary meta-analysis (placebo) using mean total sample baseline pain intensity (i.e., scale 0–100 points; four studies). For pain intensity, the regression coefficient was 0.064 (95% CI: − 0.002, 0.131; P = 0.057) and R2 was 51.3%; hence, baseline pain intensity did not appear to moderate pain intensity, and approximately half of the between-study variance was explained by pain. The test statistic for residual homogeneity (Qres) was 14.91 (P < 0.001); thus, the null hypothesis of no residual heterogeneity was rejected. The I2 value suggested that approximately 86% of residual variation was due to heterogeneity that may be explained by other covariates. Overall, this means that baseline pain intensity is unlikely to be a source of heterogeneity.

The primary strength of this meta-analysis is that it was completed using wide search criteria for inclusion of studies specifically noting chronicity of primary musculoskeletal pain. This systematic review and meta-analysis has highlighted major limitations in the exercise training literature with regard to adequately reporting study methods and a lack of rigorous study designs that control for placebo effects or include post hoc secondary analysis that considers factors that have been shown to moderate placebo response in pain trials. Additional analyses via causal mediation analysis [141] may help to identify possible mechanisms to explain how complex interventions such as exercise work [142], when well-designed placebo controlled trials are not feasible. Importantly, an approach such as this to estimate the natural direct effect (specific effect of treatment without the contextual factors) requires pre-planning to ensure the hypothesised contextual factors are defined and measured [143]. Although possible, caution should be exercised due to the number of possible contextual factors at play and the inherent challenges when including multiple mediators [144]. The major limitation of our current study was the inability to fully answer the primary research question to determine the relative effects of specific exercise training, placebo effects and non-specific effects (e.g., via determining proportional placebo effects [24]). We anticipated a low number of placebo-controlled trials; however, the decision to include any chronic primary musculoskeletal condition into the analysis was made to increase the number of trials available for analysis. Unfortunately, this a priori decision did not overcome the inherent lack of studies available.

5 Conclusion

There is very low-quality evidence that exercise training is not more effective than non-exercise placebo treatments in chronic pain. Exercise training and the associated clinical encounter are more effective than true control or standard medical care for reductions in pain for adults with chronic musculoskeletal pain, with very low quality of evidence based on GRADE criteria. The development of clinical practice guidelines heavily relies on evidence from randomised clinical trials; however, in our systematic review and meta-analysis there were insufficient placebo-controlled randomised clinical trials to confirm the specific therapeutic effects of exercise training, placebo effects or non-specific effects on pain reduction. To better inform clinical decision making and clinical practice guidelines, future randomised controlled trials should consider the role of placebo effects and the impact of exercise training on chronic musculoskeletal pain.