Physical activity and exercise for chronic pain in adults: an overview of Cochrane Reviews

Abstract Background Chronic pain is defined as pain lasting beyond normal tissue healing time, generally taken to be 12 weeks. It contributes to disability, anxiety, depression, sleep disturbances, poor quality of life, and healthcare costs. Chronic pain has a weighted mean prevalence in adults of 20%. For many years, the treatment choice for chronic pain included recommendations for rest and inactivity. However, exercise may have specific benefits in reducing the severity of chronic pain, as well as more general benefits associated with improved overall physical and mental health, and physical functioning. Physical activity and exercise programmes are increasingly being promoted and offered in various healthcare systems, and for a variety of chronic pain conditions. It is therefore important at this stage to establish the efficacy and safety of these programmes, and furthermore to address the critical factors that determine their success or failure. Objectives To provide an overview of Cochrane Reviews of adults with chronic pain to determine (1) the effectiveness of different physical activity and exercise interventions in reducing pain severity and its impact on function, quality of life, and healthcare use; and (2) the evidence for any adverse effects or harm associated with physical activity and exercise interventions. Methods We searched theCochrane Database of Systematic Reviews (CDSR) on the Cochrane Library (CDSR 2016, Issue 1) for systematic reviews of randomised controlled trials (RCTs), after which we tracked any included reviews for updates, and tracked protocols in case of full review publication until an arbitrary cut‐off date of 21 March 2016 (CDSR 2016, Issue 3). We assessed the methodological quality of the reviews using the AMSTAR tool, and also planned to analyse data for each painful condition based on quality of the evidence. We extracted data for (1) self‐reported pain severity, (2) physical function (objectively or subjectively measured), (3) psychological function, (4) quality of life, (5) adherence to the prescribed intervention, (6) healthcare use/attendance, (7) adverse events, and (8) death. Due to the limited data available, we were unable to directly compare and analyse interventions, and have instead reported the evidence qualitatively. Main results We included 21 reviews with 381 included studies and 37,143 participants. Of these, 264 studies (19,642 participants) examined exercise versus no exercise/minimal intervention in adults with chronic pain and were used in the qualitative analysis. Pain conditions included rheumatoid arthritis, osteoarthritis, fibromyalgia, low back pain, intermittent claudication, dysmenorrhoea, mechanical neck disorder, spinal cord injury, postpolio syndrome, and patellofemoral pain. None of the reviews assessed 'chronic pain' or 'chronic widespread pain' as a general term or specific condition. Interventions included aerobic, strength, flexibility, range of motion, and core or balance training programmes, as well as yoga, Pilates, and tai chi. Reviews were well performed and reported (based on AMSTAR), and included studies had acceptable risk of bias (with inadequate reporting of attrition and reporting biases). However the quality of evidence was low due to participant numbers (most included studies had fewer than 50 participants in total), length of intervention and follow‐up (rarely assessed beyond three to six months). We pooled the results from relevant reviews where appropriate, though results should be interpreted with caution due to the low quality evidence. Pain severity: several reviews noted favourable results from exercise: only three reviews that reported pain severity found no statistically significant changes in usual or mean pain from any intervention. However, results were inconsistent across interventions and follow‐up, as exercise did not consistently bring about a change (positive or negative) in self‐reported pain scores at any single point. Physical function: was the most commonly reported outcome measure. Physical function was significantly improved as a result of the intervention in 14 reviews, though even these statistically significant results had only small‐to‐moderate effect sizes (only one review reported large effect sizes). Psychological function and quality of life: had variable results: results were either favourable to exercise (generally small and moderate effect size, with two reviews reporting significant, large effect sizes for quality of life), or showed no difference between groups. There were no negative effects. Adherence to the prescribed intervention: could not be assessed in any review. However, risk of withdrawal/dropout was slightly higher in the exercising group (82.8/1000 participants versus 81/1000 participants), though the group difference was non‐significant. Healthcare use/attendance: was not reported in any review. Adverse events, potential harm, and death: only 25% of included studies (across 18 reviews) actively reported adverse events. Based on the available evidence, most adverse events were increased soreness or muscle pain, which reportedly subsided after a few weeks of the intervention. Only one review reported death separately to other adverse events: the intervention was protective against death (based on the available evidence), though did not reach statistical significance. Authors' conclusions The quality of the evidence examining physical activity and exercise for chronic pain is low. This is largely due to small sample sizes and potentially underpowered studies. A number of studies had adequately long interventions, but planned follow‐up was limited to less than one year in all but six reviews. There were some favourable effects in reduction in pain severity and improved physical function, though these were mostly of small‐to‐moderate effect, and were not consistent across the reviews. There were variable effects for psychological function and quality of life. The available evidence suggests physical activity and exercise is an intervention with few adverse events that may improve pain severity and physical function, and consequent quality of life. However, further research is required and should focus on increasing participant numbers, including participants with a broader spectrum of pain severity, and lengthening both the intervention itself, and the follow‐up period.


A B S T R A C T Background
Chronic pain is defined as pain lasting beyond normal tissue healing time, generally taken to be 12 weeks. It contributes to disability, anxiety, depression, sleep disturbances, poor quality of life, and healthcare costs. Chronic pain has a weighted mean prevalence in adults of 20%.
For many years, the treatment choice for chronic pain included recommendations for rest and inactivity. However, exercise may have specific benefits in reducing the severity of chronic pain, as well as more general benefits associated with improved overall physical and mental health, and physical functioning.
Physical activity and exercise programmes are increasingly being promoted and o ered in various healthcare systems, and for a variety of chronic pain conditions. It is therefore important at this stage to establish the e icacy and safety of these programmes, and furthermore to address the critical factors that determine their success or failure.

Objectives
To provide an overview of Cochrane Reviews of adults with chronic pain to determine (1) the e ectiveness of di erent physical activity and exercise interventions in reducing pain severity and its impact on function, quality of life, and healthcare use; and (2) the evidence for any adverse e ects or harm associated with physical activity and exercise interventions.

Methods
We searched theCochrane Database of Systematic Reviews (CDSR) on the Cochrane Library (CDSR 2016, Issue 1) for systematic reviews of randomised controlled trials (RCTs), a er which we tracked any included reviews for updates, and tracked protocols in case of full review publication until an arbitrary cut-o date of 21 March 2016 (CDSR 2016, Issue 3). We assessed the methodological quality of the reviews using the AMSTAR tool, and also planned to analyse data for each painful condition based on quality of the evidence.

Description of the condition
Chronic pain has been defined as pain lasting beyond normal tissue healing time, generally taken to be 12 weeks (International Association for the Study of Chronic Pain; Merskey 2011). It contributes to disability, anxiety and depression, sleep disturbances, poor quality of life, and healthcare costs (Leadley 2014; Moore 2014a; Park 2012).
Chronic pain has a weighted mean prevalence in adults of 20% (Breivik 2006; Moore 2014a), which increases as the population ages (32% of adults aged 25 to 34 years, 62% of adults over 75 years; Abdulla 2013; Elliott 1999). This is a greater proportion than people with asthma (To 2012) or diabetes (IDF 2012) in the same population (van Hecke 2013a). The World Health Organization (WHO) recognises chronic pain as a public health problem throughout the world, with one systematic review assessing the growing evidence that the prevalence of chronic pain in the general population is high internationally (34% in low-income countries and 30% in high-income countries; Elzahaf 2012). Chronic painful conditions comprise four of the 10 highest ranking conditions for years lived with disability in 2013 (Vos 2015), and are responsible for considerable loss of quality of life and employment, and increased healthcare costs (Moore 2014b). Despite this, the term 'chronic pain' was only added as a MeSH term in MEDLINE in January 2012 (National Library of Medicine), highlighting the relatively small proportion of specific research dedicated to this population.
Certain factors can contribute to an increased risk of chronic pain (female gender, older age, lower socioeconomic status, geographical and cultural background, and genetics; Smith 2007; van Hecke 2013b). Other factors associated with chronic pain conditions are modifiable, such as smoking status, alcohol intake, nutrition, obesity, comorbidities, employment status and occupational factors, and physical activity level (Smith 2007;van Hecke 2013a).
A review of current issues in the treatment of chronic pain strongly suggests that health professionals traditionally focus on biomedical views of pain, utilising pharmacology first and foremost, and sometimes not addressing potential nonpharmacological approaches such as physical activity and changing attitudes towards chronic pain (Schofield 2011). Guidance o en suggests that lifestyle advice is important: for example, the National Institute for Health and Care Excellence (NICE) osteoarthritis guidelines state that "exercise should be a core treatment ... irrespective of age, comorbidity, pain severity and disability. Exercise should include: local muscle strengthening [and] general aerobic fitness" (NICE 2014).
Non-pharmacological treatments have been developed, investigated, and implemented, with Cochrane Reviews and protocols evaluating the available evidence for psychological, physical, and other non-medical interventions (e.g. cognitive behavioural and behavioural therapy, Eccleston 2014; Williams 2012; TENS, Nnoaham 2008; low-impact/intensity movement/ exercise therapy, Wieland 2013; dietary, Straube 2015; and patient education, Engers 2008; Gross 2009). While evidence for the e ectiveness of these interventions is of variable quantity and quality, the 2013 Scottish Intercollegiate Guideline Network (SIGN) guidelines on the management of chronic pain made strong recommendations on the use of exercise, based on evidence drawn from randomised controlled trials (RCTs), stating: "exercise and exercise therapies, regardless of their form, are recommended in the management of patients with chronic pain" (SIGN 2013).

Description of the interventions
Physical activity has been defined by the WHO as "any bodily movement produced by skeletal muscles that requires energy expenditure, including activities undertaken while working, playing, carrying out household chores, travelling, and engaging in recreational pursuits" (WHO 2015). WHO also states that "exercise ... is a sub-category of physical activity that is planned, structured, repetitive, and aims to improve or maintain one or more components of physical fitness" (WHO 2015).
Physical activity for health can take many di erent forms: it can be structured exercise, such as in classes, gym-based, or a DVD or programme performed at home; or unstructured and involve adding just a few small activities each day (activities of daily living). Physical activity and exercise can also vary in intensity, duration, and type: aerobic (such as walking) or more focused on increasing flexibility, strength, or balance. Physical activity and exercise can also be taught (or led) by another individual such as an exercise professional, or initiated and maintained through the person's own initiative and motivation.
Both physical activity and exercise can be performed on land or in the water, and can range from whole-body to localised (body sitespecific) training. Most forms of exercise can also be modified to be performed where there is restricted movement (e.g. in a chair, a bed, or another assistive device).

How the intervention might work
Physical activity and exercise can be adapted for an individual, and is something people can do to help themselves. It is likely to be associated with minimal adverse e ects, such as interactions with medication and potential for abuse in adults with chronic pain, when compared to pharmaceutical and surgical interventions. It is therefore an attractive option to help manage an individual's pain if the systematic reviews show benefit. However, current evidence suggests that simply giving an individual advice to exercise is insu icient to bring about significant change (SIGN 2013), and a badly prescribed intervention that does not consider the individual's conditions and present state of health and fitness, such as one that does not incorporate pacing or gradual progression, may bring about adverse events such as pain 'flare-ups', or lead to cardiac or respiratory events (American College of Sports Medicine 2007). This suggests that supervised or structured interventions may be more fruitful, though this is currently unconfirmed.
Since the 1980s, primary care physician advice for treating pain has changed, moving away from "rest", to minimising or eliminating bedrest and instead remaining active (back pain, Waddell 1987). Exercise may have specific benefits in reducing the severity of chronic pain, as well as more general benefits associated with improved overall physical and mental health, and physical functioning of people with chronic pain, as depression (Finan 2013), deconditioning (Bousema 2007), and obesity are commonly observed in these people (headache/migraine, Bigal 2012; fibromyalgia, Ursini 2011). For example, studies have revealed that a single bout of exercise increases the production of endogenous opioids, leading to transient anti-nociception in both animals and humans, and repeated exercise produces long-lasting anti-nociception in otherwise untreated animals (Stagg 2011). Aerobic exercise is also strongly linked to weight loss (Messier 2013), which in turn has implications for the management of chronic pain as the pressure on joints is reduced. Alternatively, resistance exercise, or other forms of strength training, can improve the person's capacity to support bone and cartilage through improved musculature supporting movement around a joint, with potential to relieve sti ness (Mayer 2008) and bringing about some pain relief. Resistance training through repetitive full range-ofmotion exercise around the lumbar spine (in chronic low back pain) may a ect disc metabolism itself, with the possibility that the exercise programme could improve metabolic exchange in the lumbar discs and aid in repair (Mooney 2006). Training to improve balance and flexibility also has benefits as it reduces the risk of falls, and the potential for further pain or injury (Harvard 2013).

Why it is important to do this overview
If physical activity and exercise interventions are shown to e ectively and safely reduce pain intensity or frequency (or both), they are likely to be a preferable alternative or adjunct therapy to pharmacological/surgical treatments for chronic pain. The interventions could promote personal involvement of individuals in the management of their pain, thus increasing self-e icacy and the ability to self-manage. In turn this could lead to an increase in overall quality of life and a consequent reduction in healthcare use. In addition, exercise is of great importance for cardiovascular (Vigorito 2014) and bone health (Sakuma 2012). Reduced physical function and consequent lack of mobility in people with chronic pain is associated with increased all-cause and cardiovascular mortality (Nüesch 2011), with other studies linking severe chronic pain to general increased all-cause mortality (Moore 2014a; Torrance 2010).
Physical activity and exercise programmes are increasingly being promoted and o ered in various healthcare systems (American College of Sports Medicine (ACSM) 'Exercise is Medicine' global pledge at the Inaugural World Congress 2010) and for a variety of chronic pain conditions, including arthritis (Fransen 2014; Silva 2010), fibromyalgia (Busch 2013), and dysmenorrhoea (Brown 2010). At this stage it is important to establish the e icacy and safety of these programmes, and furthermore to address the critical factors that determine their success or failure.
It is therefore important to identify whether (and how) exercise interventions can be e ectively and safely applied in people with chronic pain.
With a number of systematic reviews published by Cochrane evaluating the e ectiveness of exercise in various painful conditions, it is timely and important to bring together all relevant published information to evaluate the current evidence, and identify the availability and quality of evidence-based exercise interventions. This overview will determine the extent to which the published systematic reviews have accurately assessed the evidence for exercise in chronic pain conditions/syndromes, which will help to direct future guidelines and identify current research gaps.

O B J E C T I V E S
To provide an overview of Cochrane Reviews of adults with chronic pain to determine (1) the e ectiveness of di erent physical activity and exercise interventions in reducing pain severity and its impact on function, quality of life, and healthcare use; and (2) the evidence for any adverse e ects or harm associated with physical activity and exercise interventions.

M E T H O D S Criteria for considering reviews for inclusion
We included only systematic reviews of RCTs of physical activity and exercise in participants with chronic pain, and published in the Cochrane Database of Systematic Reviews. The included reviews had to fulfil the following criteria:

Participants
Adults (aged 18 years and over) reporting chronic non-cancer pain, including persistent (e.g. chronic back pain, fibromyalgia) and intermittent (e.g. migraine, dysmenorrhoea) pain, for at least three months (12 weeks) in any body site.

Intervention
Reviews of RCTs assessing physical activity or exercise as the intervention (any reviews where that assessed physical activity or exercise as a stand-alone intervention). This included physical activity interventions that could be initially taught by an exercise professional, or involve periodical/ongoing supervision.

Exclusions
Interventions not deemed physical activity or exercise using the WHO definition, such as manipulation, mobilisation, or passive movement. Any multi-modal interventions were excluded if physical activity/exercise could not be assessed for e ect (the e ect of exercise must have been measured distinctly).

Comparison
Usual care, waiting list control, placebo/sham treatment, other treatment, or a combination of treatments (as long as the e ect of exercise could be measured distinctly).
This could be presented and analysed as change on a continuous scale, the proportion of participants who 'responded', or, ideally, in a dichotomised format as the proportion of participants in each group who achieved a predetermined threshold of improvement (e.g. outcome in individual participants of at least 50% pain intensity reduction, or no worse than mild pain, at the end of the trial, with at least 30% pain intensity reduction as a secondary outcome, or recovery; Moore 2013).
• A review included some interventions of interest or reported only some outcomes of interest. In this case we extracted the interventions and outcomes of interest, but we did not include interventions or outcomes outside the scope of this overview. • Reviews occasionally included papers that included children and adults together, but the results for adults were not reported or analysed separately in the included papers or the review. In this case we made a judgement as to whether the review could be included based on the proportion of adults. Our intention was to include only those reviews where more than 80% of participants were adults.

Search methods for identification of reviews
We searched the Cochrane Database of Systematic Reviews (CDSR), 2016, Issue 1, on the Cochrane Library for relevant reviews using the search strategy: (pain or migraine or headache) and (exercise or activity or physical). We did not seek non-Cochrane reviews.

Data collection and analysis
Two overview authors (LG, CC) independently carried out searches and selected reviews for inclusion. Disagreements were resolved through discussion, and a third overview author (RAM) acted as arbitrator where necessary.
Two overview authors (independently carried out assessment of methodological quality (LG, CC), and extracted data (LG, RAM). Any disagreements were resolved through discussion, or involving a third overview author if necessary (DM).
One overview author (LG) tracked results of the search for the most up to date version of each review and protocol that fulfilled the inclusion criteria.

Selection of reviews
Included reviews assessed RCTs of the e ects of exercise for pain management in adults (as defined by individual reviews), compared with any of the listed comparators, and included: • a clearly defined clinical question; • details of inclusion and exclusion criteria; • details of databases searched and relevant search strategies; • participant-reported pain severity (primary outcome measure); • summary results for at least one other desired outcome.

Data extraction and management
Two overview authors (LG, RAM) independently extracted data from the included review using a standardised data extraction form and checked for agreement prior to entry into Microso Excel for Windows. We did not extract data from reports included in the reviews again, neither did we undertake any re-analysis of data from reviews. Data were not entered for analysis into Cochrane's statistical so ware due to the lack of relevant and comparable data (RevMan 2014).
We collected the following information (where available) from the reviews: • number of included studies and participants; • intervention (exercise or activity type) and dose (frequency/ intensity); • comparator; • condition treated; • time of assessment; • duration of follow-up; • relevant outcomes.
Where possible we extracted risk ratio (RR), number needed to treat for an additional beneficial outcome (NNTB), mean di erence (MD), and standardised mean di erence (SMD), and other relevant statistical data for the primary and secondary outcomes. This included: • obtaining 50% pain relief (participant-reported); • obtaining any other measure of 'improvement' (participantreported); • adverse events; • death; • withdrawals.

Quality of included reviews
Two overview authors (LG, CC) independently assessed each included review to see if it satisfied the criteria specified in the 'assessment of multiple systematic reviews' (AMSTAR) measurement tool (Shea 2007), for rigorous methodological quality. Arbitration by a third overview author (DM) was necessary for some fields.
High quality reviews were required to fulfil each of the established AMSTAR criteria (further criteria to fulfil each field is listed in Table  1).
For each review we also planned to assess the likelihood of publication bias by calculating the number of participants in studies with zero e ect (relative benefit of one) that would be needed to give an NNTB too high to be clinically relevant (Moore 2008). In this case we would have considered an NNTB of 10 or greater for the outcome of participant-reported pain relief of 30% or greater to be the cut-o for clinical relevance. This method is used as statistical tests for the presence of publication bias have been shown to be unhelpful (Thornton 2000). However, assessment of publication bias was not possible due to the lack of specificity of the populations included within the reviews, and so we were unable to extract comparable data.

Quality of evidence in included reviews
We planned to use two main indicators for the quality of evidence: all included reviews must have used only primary studies that were both randomised and double-blind, so minimising the risk of bias from these items; and all included reviews must have included only people with at least moderate pain intensity at baseline (visual • The first tier used data meeting current best standards, where studies reported the outcome of at least 50% pain intensity reduction from baseline (where 50% was the cut-o for a dichotomous (yes/no) outcome: was a 50% reduction in pain observed?), or its equivalent, without using last observation carried forward (LOCF) or other imputation method for dropouts, reported an intention-to-treat (ITT) analysis, lasted eight or more weeks, had a parallel-group design, and had at least 200 participants (preferably at least 400) in the comparison (Moore 2010). These top-tier results were usually reported first. • The second tier used any available data, but where one or more of these conditions were not met, for example reporting at least 30% pain intensity reduction, using LOCF or a completer analysis, lasting four to eight weeks, and where the numbers of participants were at least 200. • A third tier of evidence related to small amounts of data (fewer than 200 participants), or short studies of less than four weeks, or where there was obvious major heterogeneity between studies, or where there were other shortcomings in allocation concealment, considerable attrition, and incomplete outcome data. For this third tier of evidence, no data synthesis was reasonable, and may have been misleading, but an indication of beneficial e ects might be possible.
This overview examined the quality of all included reviews according to current best standards for reporting in pain. These included the attempt and ability of the reviews to identify studies/ interventions with the maximum evidence of e ectiveness, and minimum risk of bias, including the reporting of the following.
• Outcomes in trials of the proportion of participants obtaining at least 50% pain intensity reduction, or no worse than mild pain, at the end of the trial (with at least 30% pain intensity reduction as a secondary outcome). We did not consider the use of mean changes in pain scores as high quality because responses to pain interventions are not Gaussian, and few people have the mean response. • Duration of included studies of eight weeks or longer.
• Imputation method of baseline observation carried forward (BOCF), LOCF, or worst observation carried forward (WOCF) if adverse event withdrawals were similar in active and control groups. • At least 200 participants per treatment group in included studies, with at least two trials, as a minimum criterion for trustworthiness of any analysis. Pooled analysis of small studies may be considered good quality if at least 400 participants were involved, but we regarded these as being potentially subject to bias.
We extracted the 'Risk of bias' as assessed by the original review authors from included reviews. Counts of low risk of bias were extracted from relevant studies in the included reviews and tabulated under the following headings to evaluate the proportion of studies achieving a low risk of bias for each: • random sequence generation (selection bias); • allocation concealment (selection bias); • blinding of participants and personnel (performance bias); • blinding of outcome assessment (detection bias); • incomplete outcome data (attrition bias); • selective reporting (reporting bias); • sample size; • any other biases.

Data synthesis
Additional quantitative analyses were not required, since we only considered results from properly conducted (Cochrane) reviews.
The aim was to concentrate on specific outcomes such as the proportion of participants with at least 50% pain relief, all-cause or adverse event discontinuations, or serious adverse events, and to explore how these can be compared across di erent treatments for the same condition. We planned to compare only like with like (where possible); for example in study duration, which can be an additional source of bias if insu icient in length (Moore 2010).
However due to the limited data available, we were unable to directly compare and analyse interventions, and have instead reported the evidence qualitatively only. We had also planned to employ subgroup analyses assessing age, condition, and intervention type/intensity, though this was not feasible using the available data from included reviews. For this reason we have also been unable to include a 'Summary of findings' table as planned and stated in the protocol.
Importantly, we have tried to highlight issues of low trial quality, inadequate size, and whether trials were truly valid for the particular condition in making between-therapy comparisons.
We approached each review with four main questions/focus, and extracted data accordingly.
• Did they report exercise versus non-exercise studies? • Did the review or studies included in the review (or both) have low risk of bias? • Did they have our main outcome?
• What were the actual intervention/s included in the review?

R E S U L T S
We included 21 reviews with 381 included studies, totalling 37,143 participants. Of these, 264 studies (19,642 participants) examined exercise versus no exercise/minimal intervention in adults with chronic pain (the focus of this overview) and so were used in the qualitative analysis.

Description of included reviews
The search strategy was performed in the Cochrane Library only, and revealed 475 potentially relevant titles, of which 75 were assessed as full papers.
The search was undertaken on 31 January 2016 (CDSR 2016, Issue 1), a er which any included reviews were tracked for updates, and protocols were followed in case of full review publication until 21 March 2016 (CDSR 2016, Issue 3).

Library
Trusted evidence. Informed decisions. Better health.

Cochrane Database of Systematic Reviews
Detailed information about the included reviews is available in Table 3. Trial and participant number, age, and gender distribution is reported in Table 4.

Specificity of chronic pain condition of included reviews
Following abstract and full paper assessment, 21 reviews fulfilled the inclusion criteria: four in rheumatoid arthritis (Cramp 2013;Han 2004;Hurkmans 2009;Silva 2010)

Exercise and physical activity interventions implemented in the included reviews
Interventions assessed included: any specified style of land-based exercise or physical activity such as one designed to improve strength, range of movement, aerobic capacity, or a combination of these (

Aquatic exercise
Any exercise performed in water. This can include swimming, though many studies will be referring to exercises performed vertically in the water (not horizontally), either using the water to support the body through the exercise, or as resistance against the body.

Range of motion and flexibility exercise
Can be performed in water or on land. The intention is to increase the range of motion around a joint through progressive stretching and mobilising of the muscles around and crossing the joint. For the purposes of this overview, we only included active movement where the movement was brought about by the participant, and not passively moved by an external force such as a therapist.

Aerobic exercise
Can be performed in water or on land. Exercise usually performed continuously to raise the heart rate and breathing rate for a prolonged period. Examples include walking, jogging, running, cycling, and swimming. O en presented as a percentage of the participant's heart rate max (HRmax) -the highest heart rate reached when performing at their absolute maximum. Similarly it may be presented as a percentage of VO 2 max or VO 2 peak (a proportion of the maximum amount of oxygen the muscle can take up per minute), or as an absolute value (mL/kg/minute).

Strength/resistance exercise
Can be performed in water or on land. Exercise performed against a progressive resistance with the intention of improving muscle strength, muscle endurance, muscle power, or a combination of these. Resistance can come from fixed or free weights, elastic bands, body weight (against gravity), and water resistance. It may also involve static or isometric strength (holding a position or weight without moving against it). O en presented as a percentage of the participant's one repetition maximum (1-RM) -the maximum weight they can li /move if they only have to do it once.

Motor control exercise
Can be performed in water or on land. Exercise to bring about activation of the deep trunk muscles, targeting the restoration of control and co-ordination of these 'core muscles' (Saragiotto 2016).

Balance (proprioceptive) training
Can be performed in water or on land (water may be used initially for support). Exercise emphasises the maintenance of balance during visual and perturbation challenges with eyes open or closed, range of motion, and maintaining stability over reduced areas of support and unstable surface (Silva 2010), that is improving balance in increasingly unstable situations.

Tai chi
An ancient Chinese discipline developed from martial arts, involving a continuous series of very controlled (and usually slow) movements designed to improve physical and mental wellbeing.

Yoga
Arising out of Hindu philosophy. Exercise includes breath control, simple meditation, and the adoption of specific bodily postures. It is widely practised for health, relaxation, and control (physically and mentally). Incorporates stretching and flexibility training with isometric strength training (holding certain poses, with no movement against a resistance).

Library
Trusted evidence. Informed decisions. Better health.

Pilates
Developed by Joseph Pilates in the 20th Century, it is a system of exercises (o en using special apparatus) designed to improve physical strength, flexibility, and posture, and enhance mental awareness.

Duration and dose (frequency/intensity) of the exercise and physical activity interventions
A detailed breakdown of each review can be seen in Table 5.

Duration of intervention
Interventions assessed by the included reviews varied in length from a single session (Fransen 2015) to 30 months (Fransen 2015). Only five reviews enforced a minimum intervention period to reduce risk of bias, and were able to attribute any e ects to the intervention (Brown 2010; Busch 2013; Gross 2015a; Hurkmans 2009; Silva 2010).

Frequency
There was large variation in the exercise or physical activity intervention being implemented, ranging from just once a week (Bidonde 2014;Busch 2007;Fransen 2014;Fransen 2015;Han 2004;Saragiotto 2016), to twice a day (Boldt 2014), and some performing a short series of exercises (two-minute duration) every 15 minutes during the day (Gross 2015a). However, when reported, most included studies in the reviews implemented the programme twice a week (or stated at least twice a week).

Intensity
Few studies quantified the intensity of each session. Baseline intensity was o en accepted as low/moderate, with the aim to progress over the intervention period to 70% to 85% of HRmax or heart rate reserve (HRR) for aerobic interventions (Brown 2010; Cramp 2013; Hurkmans 2009), 70% to 80% of an individual's 1-RM, or 50% to 70% maximum voluntary contraction (Koopman 2015) in strength/resistance training programmes (Busch 2013; Hurkmans 2009). In other reviews, intensity was described more loosely as "variable" or "low intensity (very light) to maximum e ort (

Duration (per session)
Individual sessions varied in length from two minutes (Gross 2015a), to 90 minutes (Busch 2013;Cramp 2013;Han 2004) or 120 minutes (Boldt 2014), but mostly situated around 45 to 60 minutes. However, it is important to note that the shorter sessions were o en performed more regularly than longer sessions. With more information it would have been possible to calculate total volume of exercise or physical activity (session duration × frequency per week × number of weeks), for a more accurate and detailed analysis.

Intervention specificity for chronic pain in the included reviews
The focus of this overview was exercise versus no-exercise interventions with the intention of answering the original question: is exercise beneficial, detrimental, or ine ective for people with chronic pain when compared to inactivity? Two of the 21 reviews did not include/locate any studies that examined simply exercise versus no exercise (Lauret 2014; Silva 2010). However, many of the included reviews compared varying exercise modality, duration, intensity, and frequency. The "no-exercise" intervention referred to the control group where there was a minimal intervention (such as sham exercise or education) or wait-list control/no treatment (see Table 3 for more information on control group activity).

Time points reported
Four of the 19 reviews that reported data, reported results at a single time point only ('post-intervention': Bidonde 2014;Busch 2007;Cramp 2013;Han 2004). Reviews also analysed outcome measures immediately post-intervention and at one or more follow-up points. Each review defined short-, intermediate-, and long-term follow-up according to their own assessment, so when the time period was not mentioned explicitly, we grouped the reviews according to the review authors' own classification only, and where a time period (weeks, month, years) was explicitly listed but not defined by the authors, we grouped them as shortterm (follow-up as under six months), intermediate-term (six to 12 months), and long-term (longer than 12 months): short-term:

Long-term follow-up
Of the seven reviews claiming to report "long term" follow-up, one classed long-term as longer than six weeks (intermediate term as one to six weeks' follow-up) (Boldt 2014). The remaining six reviews defined long-term follow up as over 12 months ( . The lowest score was 6/11 (Silva 2010) though five categories were not applicable (n/a) due to there being no included studies. Quality assessment results for each individual review are presented in Table 6.
All reviews except one (Bidonde 2014) fulfilled the basic criteria (questions one to three of Table 1); to follow an 'a priori' design as Cochrane implements a system of protocol publication before undertaking the full reviews, where it also specifies dual study selection and data extraction from a comprehensive literature search. One review did not fulfil the 'a priori' design as this was an update and separation from a broader review series, and so the Library Trusted evidence. Informed decisions. Better health.
Cochrane Database of Systematic Reviews criteria had not been explicitly listed prior to publication for this specific title (Bidonde 2014).
Criteria which scored badly using the AMSTAR tool were characteristics of included studies (question six of Table 1), reporting of publication bias (question 10 of Table 1), and conflict of interest declarations (question 11 of Table 1).
• Included study characteristics were limited, o en reporting the "inclusion criteria" used to recruit participants in the study instead of the characteristics of actual included participants, and excluding information such as participants' age, gender split, ethnicity, and disease status. In the remaining reviews, a cursory statement was commonly made regarding the review authors' conflicts of interests, however, fulfilling the AMSTAR criteria also requires a statement to be made regarding any conflict of interest for any of the included studies.

Risk of bias in included reviews
The original review authors assessed risk of bias (see Table 7). The table shows the number of studies assessed as low risk of bias only, and excluded those that were assessed as unclear or high risk of bias.

Selection bias (randomisation and allocation concealment)
Selection bias had the largest proportion of included studies with low risk of bias (63% and 42% of studies adequately undertaking and reporting the methods used).

Performance and detection bias (blinding participants, personnel, outcome assessors)
With any exercise or physical activity intervention it is very di icult to blind both participants and personnel to the allocation, though some studies included in reviews attempted to by o ering sham exercise.
Due to the di iculty of blinding participants to their group allocation, review authors assessed the risk of bias in di erent ways, which may cause confusion: whereas the majority declared this lack of possible blinding to be high risk of bias or unclear, two reviews labelled such cases as low risk of bias in order not to exclude these studies unnecessarily from their analysis (Lane 2014; Lauret 2014). Without these two reviews, only a small percentage (7.8% or 18/229) of the included studies would have scored low risk of performance bias (blinding of participants and personnel), but by including them (all 35 studies from those two reviews assessed as low risk of bias) the overall proportion of studies assessed as having low risk of bias was closer to 20% (53/264).

Attrition (incomplete outcome data, withdrawals/dropouts)
About 55% (144/264) of the studies included in these reviews showed low risk of bias.

Reporting bias (selective reporting)
Reporting bias was classed as low risk in only 46% of included studies. However, it is important to note this was not due to the remainder having high risk of bias, but instead 'unclear', as trial protocols were not always published or accessible to the review authors to accurately assess/interpret.

Study/sample/group size
Sample size was not always included within the risk of bias assessment. It was therefore extracted directly from each review's table of included study characteristics by a single overview author (LG), and assessed as being low risk of bias when there was a minimum of 50 participants per arm, or 100 in total. Numbers were then separated for the proportion of studies with greater than 100 participants per arm (or 200 in total), and 200 participants per arm (or 400 in total), as this could then be considered higher tiered evidence.
Only 26 out of 264 included studies (10%) across the 21 reviews reported over 100 participants in total (or 50 per arm), a further 6% (15/264) included over 200 participants per arm. The remaining 223 studies (84%) had fewer than 50 participants per arm (or sample size was not reported), o en not reaching 50 in total.

Other bias
The format for reporting bias has changed, and therefore some earlier reviews (that are yet to be updated) did not assess bias using the same format. Others reported additional criteria as 'other bias' including the similarity of baseline characteristics, and similarity of timing points.

Interpretation of results/conclusions by original review authors
For conclusions made by the original review authors, see Table  8. We assessed whether these conclusions/interpretations of the results accurately reflected the information provided within the review, and if any further information should have been included. This final assessment of the review is an important stage in determining any author bias within the review process, as many readers, funders, and policy makers will focus on the author conclusions without a full appraisal of the actual presented data.
Eleven of the 21 reviews reported appropriate conclusions based on the data available in the context of the quality of evidence Cochrane Database of Systematic Reviews Fransen 2014), or we were unable to agree with their interpretation due to di iculty in extracting the data (van der Heijden 2015).

E ect of interventions
We have interpreted results using data reported in the reviews, and did not return to the original studies. Where data have been reported as MDs or as an absolute or relative change score we have used the appropriate scales (where possible) to determine whether this was clinically significant. When data have only been presented as SMD, with or without 95% confidence intervals (CI), with or without level of significance (P value), we have cautiously used the interpretation by Cohen 1988 who defined e ect size using the SMD as small (SMD 0.2 to 0.5), moderate (SMD 0.5 to 0.8), or large (SMD greater than 0.8).
For the purposes of clarity, we have used the term 'intervention' to refer to the exercise or physical activity intervention, and 'control' to refer to the included comparison group which did not involve any exercise or physical activity element.

Self-reported pain (severity)
Part of the inclusion criteria for this overview was for pain severity to be listed as an outcome measure.
Two of the 21 reviews did not include/identify any studies that examined intervention versus control (Lauret 2014; Silva 2010).
Of the remaining reviews that did report studies examining intervention versus control (no physical activity or exercise, or minimal intervention), two did not report pain as an absolute or relative score of severity, intensity, or change as a result of the intervention (Brown 2010; Han 2004), and one review assessed pain-free time and distance during exercise (they did not assess pain using a mean/usual pain scale; Lane 2014). We could not extract relevant data for one review as they compared two di erent exercise interventions and a control but did not report the data compared to the control (Regnaux 2015).
The remaining 15 reviews reported a mean or usual pain score for exercise (intervention) and no-exercise (

Reported baseline pain score
Of the 15 reviews that were able to assess pain (Table 9) HAQ: mean of different category scores, 0 or 1 (mild to moderate disability), up to 2 or 3 (severe to very severe disability); WOMAC pain score: 5 items summed to 0 (no pain) to 20 (worst pain ever); WUSPI: 15 items of 0 to 10 VAS scores, summed to form total of 0 (no pain) to 150 (worst pain ever).
This suggests the majority of participants reviewed had mildto-moderate pain (only one review reported a mean of severe pain (aquatic exercise for fibromyalgia, Bidonde 2014) at the commencement of each intervention (less than 30/100 mild pain, 30/100 to 60/100 moderate pain, more than 60/100 severe pain; Collins 1997), though labelling the majority as having only mild-tomoderate pain should be interpreted with caution due to the lack of specific data available -the baseline data of the intervention group would have been preferable to the proxies we have had to use.

Quality judgement/ tiered quality (first, second, third tier evidence)
Our assessment criteria stated that we would accept the information as graded evidence when reported as the number of participants achieving a 50% (first tier evidence) or 30% (second tier evidence) reduction in pain, but none of the included reviews reported results in this way, and so instead we used the reported absolute and relative change values.
None of the included reviews fulfilled the requirements for first tier evidence (at least 50% pain reduction from baseline, study duration longer than eight weeks, and more than 200 participants per arm).
Second tier evidence (at least 30% pain reduction from baseline, study duration between four and eight weeks, and more than 200 participants in total or 100 participants per arm) was also lacking in these reviews; three reviews found at least 30% reduction in pain from baseline (Busch 2007; Busch 2013; van der Heijden 2015), one of which also used long enough exercise programmes (eight to 21 weeks' intervention, Busch 2013) but totalled only 81 participants across two studies. The other two reviews did not fulfil the study duration criteria (interventions from 2.5 weeks, Busch 2007; and three weeks, van der Heijden 2015) or study size criteria.
Consequently results from relevant reviews have been pooled (all tier three quality) where appropriate, though results should be interpreted with caution due to the low quality evidence.

Treatment e ect
Data that could be extracted for pain can be seen in Table 9 for all reviews. Only three reviews found no statistically significant changes in usual or mean pain from any intervention (Cramp 2013; Hurkmans 2009; Koopman 2015 (assumed due to lack of presented data)). The remaining reviews reported a statistically significant e ect of the intervention at one or more time points, in at least one subgroup.
Three reviews found at least 30% pain reduction from baseline (post-intervention - . Three reviews found statistically significant improvements as a result of the intervention, but they did not reach clinical significance (post-intervention, P = 0.02, Bartels 2007; "small to moderate" benefit post-intervention and at six-month follow-up, P < 0.001, Fransen 2014; "moderate e ect" of 7% (95% CI 3 to 11) benefit post-intervention, Bidonde 2014).
Overall, results were inconsistent across interventions and followup (see Table 9), as exercise did not consistently bring about a change (positive or negative) in self-reported pain scores at any single point.

Physical function (objectively or subjectively measured)
Measures of physical function were the primary outcome measure in eight out of 21 reviews (

Treatment e ect
Data that could be extracted for physical function are shown in Table 10. Two reviews which reported physical function had no data to extract (Lauret 2014; Silva 2010), and for one review we were unable to extract the relevant data (Regnaux 2015). Two reviews found no significant di erence in physical function between the intervention and control groups ( Only one review reported statistical significance and large e ect size (both short-term and long-term follow-up: SMD 1.10 (95% CI 0.58 to 1.63) and 1.62 (95% CI 0.31 to 2.94), van der Heijden 2015). However, the original review authors highlighted the low to very low quality of the evidence as many studies had high or unclear risk of bias across multiple domains (van der Heijden 2015).

Psychological function
Only five out of 21 reviews assessed psychological function as mental health (

Treatment e ect
Data that could be extracted for psychological function can be seen in Table 11. There were significant e ects in favour of the intervention for mental health (Bartels 2007) and depression (Busch 2013) scores, and "variable e ect" for depression (Cramp 2013). However, there was also no e ect or no di erences between control and intervention groups reported for mental health (Bidonde 2014; Busch 2013), anxiety (Cramp 2013), and depression (Boldt 2014).

Quality of life
A version of quality of life assessment was reported in nine reviews.

Treatment e ect
Data that could be extracted for quality of life can be seen in Table 12. Four reviews found no significant di erence between intervention and control groups in health-related quality of life post-intervention (9 studies, n = 556) (HRQoL: Boldt 2014; Fransen 2014; Gross 2015a, global assessment: Bidonde 2014; Gross 2015a)), three reviews did not or were unable to report any data (

Adherence to the prescribed intervention
Only one review reported adherence to the intervention as an outcome measure (Regnaux 2015), but the authors were unable to perform an analysis on attendance as most studies did not clearly report attendance or compliance (Regnaux 2015). However, five reviews assessed withdrawals or dropouts ( Data that could be extracted for adherence, withdrawals, and attrition can be seen in Table 13. Pooling all available data for withdrawals/dropout/attrition gave an RR of 1.02 (95% CI 0.94 to 1.12) in favour of the control group (6 reviews, 30 studies, n = 2256, control withdrawal 81/1000, intervention withdrawal 82.8/1000).
One clinically controlled trial (CCT) in one review reported statistically significant improvement in enjoyment of exercise/rest (P = 0.0002) and self-reported benefit from exercise/rest (P = 0.006) at both post-intervention (end of therapy, 10 weeks) and follow-up (four months later) (n = 95, Han 2004).

Healthcare use/attendance
None of the reviews reported healthcare use/attendance.

Adverse events (not death)
Eighteen out of 21 reviews reported adverse e ects (three reviews did not report adverse events as an outcome measure due to lack of studies or other undisclosed reasons; Brown 2010; Lauret 2014; Silva 2010). Two reviews only assessed a specific adverse event ("amputation" Lane 2014; "motor unit survival" Koopman 2015), one review observed "safety -pain and radiological damage" (Hurkmans 2009), and another referred to any "sidee ects" (Han 2004).
Data that could be extracted for adverse events (not death) can be seen in

Death
Only one out of 21 reviews reported death separately to other adverse events (Lane 2014). Based on five studies within the review, death had an RR of 0.71 (95% CI 0.28 to 1.78) in favour of exercise as being protective, though was not statistically significant (P = 0.47).

D I S C U S S I O N
Specificity of the condition: despite the heterogeneous nature of chronic pain, in this overview we have combined several painful conditions covering a number of conditions and diagnoses. Regardless of aetiology, the impact of chronic pain is broadly similar across many conditions.

Summary of main results
Pain severity: there were favourable results in a number of reviews as a result of exercise: only three reviews found no statistically significant changes in usual or mean pain from any intervention. However, results were inconsistent across interventions and followup, as the intervention did not consistently bring about a change (positive or negative) in self-reported pain scores at any single point. The exercise or physical activity interventions did not have a negative e ect on the outcome (did not worsen the pain). A factor in the lack of statistical and clinically significant result may be the baseline pain severity of participants. The majority of the included population had an assumed mild-to-moderate pain severity score (assumed only due to lack of exact group data at baseline). This is o en the desired outcome (post-intervention) of many drug therapies for pain, and it may therefore be di icult to show a clinically significant improvement in these people.
Physical function: physical function/disability was the most commonly reported outcome measure, and was the primary measure in eight out of the 21 reviews. Physical function was significantly (statistically) improved as a result of the intervention in 14 reviews, though even these statistically significant results had only small-to-moderate e ect sizes in all but one review.
Psychological function and quality of life: there were variable results for psychological function and quality of life: results were either favourable to exercise (two reviews reporting significantly large e ect sizes for quality of life), or showed no di erence between groups. There were no negative e ects.
Adherence to the prescribed intervention: could not be assessed in any included review. However, risk of withdrawal/dropout was slightly higher in the exercising group (82.8/1000 participants versus 81/1000 participants), though the group di erence was not significant.
Healthcare use/attendance: not reported in any included review.
Adverse events, potential harm, and death: importantly, exercise caused no actual harm, with most adverse events being increased soreness or muscle pain, which reportedly subsided a er several weeks of the intervention. One review reported a non-significant reduction in risk of death as a result of the intervention.

Overall completeness and applicability of evidence
Of the 21 included reviews, seven could be considered out of date as they were most recently assessed as up-to-date prior to 2010 such that any recent controlled trials assessing pain severity have not been included in this overview (Cochrane recommends updating reviews every two years) (Bartels 2007;Brown 2010;Busch 2007;Han 2004;Hayden 2005;Hurkmans 2009;Silva 2010). We included these reviews in the overview, but they may not be as relevant now due to the elapsed time since they were updated. One protocol that had potential to be included was published in 2006 with no full review available yet (Craane 2006).
Available data suggest that participants in the included reviews and studies would generally be characterised as having mild-moderate pain (moderate greater than 30/100 or 3/10) with only one review reporting moderate-severe pain (severe greater than 60/100 or 6/10). Therefore whether the evidence of change or no change seen here as a result of each intervention is applicable to people further along on the pain spectrum (with higher pain scores/worse pain) is debatable. However, it can be argued that those people are more likely to be assigned medical or surgical interventions than physical activity and exercise alone (where available), and as a group they may be less able to engage in exercise, and may therefore be more di icult to recruit into exercise-only studies. Having said this, the labelling of participants as having mild-moderate pain was a cautious one within this overview due to the lack of specific data available at baseline assessment; only three reviews included baseline pain scores in the intervention group, and two further reviews provided control group baseline scores.
There are still gaps in the available literature, and therefore also within this overview. None of the included reviews examined generalised or widespread chronic pain as a global condition, each instead examined specific conditions that included chronic pain as a symptom or result of the ongoing condition (rheumatoid arthritis, osteoarthritis, fibromyalgia, low back pain, intermittent claudication, dysmenorrhoea, mechanical neck disorder, spinal cord injury, postpolio syndrome, and patellofemoral pain). The pain in these cases can occur secondary to other symptoms such as fatigue, muscle sti ness, di iculty sleeping, and depression, all of which could separately (and more e ectively) be influenced by the intervention. Additionally, only 25% of included studies actively reported adverse events. This may a ect the completeness of the evidence as conclusions have been drawn based on the available data. The included reviews did not discuss the possible impact of this non-reporting by the original trials, and this may lead to underestimating possible adverse events from an intervention, or overestimating its safety.

Library
Trusted evidence. Informed decisions. Better health.

Cochrane Database of Systematic Reviews
The exercise interventions examined in the included reviews were broad; including aerobic, strength, flexibility, range of motion, and core or balance training programmes, as well as yoga, Pilates, and tai chi. Many of these interventions can be accessed in the community by the general public and people with chronic pain, either individually or in classes (yoga, Pilates, tai chi). Other exercise intervention programmes, such as the motor control exercise and proprioceptive (balance) training, requires at least initial supervision by a therapist to teach the correct techniques and provide feedback for progression.

Quality of the evidence
In assessing the quality of the evidence, we employed the AMSTAR tool to examine the reviews, extracted data on risk of bias to examine the available primary evidence, and evaluated the authors' conclusions to ensure that they were appropriate based on the available data.
The AMSTAR tool is useful in assessing the reporting of a systematic review, though it does not inform us of the actual undertaking or conduct of the review process. All 21 included reviews scored well across the AMSTAR assessment, though this is likely due to the stringent reporting guidelines implemented by Cochrane prior to publication. However, it may be necessary or advisable for the Cochrane guidelines to be further expanded and detailed with regards to reporting study characteristics, publication bias, and conflicts of interest, as these areas o en did not meet the requirements laid out in the AMSTAR criteria (Table 1).
Data extracted from the reviews regarding their assessment of bias (risk of bias) showed moderate level scores at best across all included studies within the included reviews. Other than issues surrounding blinding (which are problematic in exercise intervention studies due to the nature of the intervention), the trials did not consistently and adequately report potential attrition and reporting biases, with less than half of studies within these reviews at low risk of bias.
However, the most prominent issue with regards to bias in these exercise and physical activity intervention studies is the sample size used. This subcategory is not used as standard in the assessment of bias in Cochrane Reviews, despite the increasing volume of research available suggesting that small studies of fewer than 100 participants per arm (Moore 2010; Nüesch 2010) are at increased risk of succumbing to the random e ects in estimating both direction and magnitude of treatment e ects (Moore 1998; Turner 2013) due to greater heterogeneity within and between small studies (IntHout 2015).
Studies within the included reviews here were very small (o en fewer than 50 participants in total). For greater quality and a more reliable e ect, at least 100 participants per arm should be analysed for a study to potentially be classed as tier two evidence (200 per arm for tier one); small studies are known to overestimate the treatment e ect by up to 32% in comparison with larger studies (Deschartes 2013).
Assessing studies for risk of bias based on study size (total number or per arm) should be included in any review or meta-analysis in future, to adequately assess the influence of small trials on the estimated treatment e ect (Nüesch 2010). Inclusion in the standard assessment process may in turn influence the design and undertaking of future research trials to increase the sample size, and produce more consistent clinically and statistically accurate results.
Of the 21 included reviews, 12 used a pain measure as their primary outcome ( , and the remaining nine reviews included the measure as a secondary outcome only. Other outcomes were shared, including physical and psychological function, and quality of life. Likewise, each review team will have included studies that did not use their chosen outcome measures as the primary measure, and that were therefore powered according to a di erent primary outcome. On collating the evidence, some studies may appear underpowered for the outcome(s) of interest to us (Turner 2013), yet were adequately powered for the studies' primary measure. To increase the power of the results of this overview, and the intermediary reviews we have included, intervention studies that focus on painful conditions should include pain intensity as the primary outcome, or at least as a prominent secondary outcome; alternatively review authors should seek to include only those studies that were adequately powered for pain intensity as a primary outcome measure.
Intervention length ranged from a single session to regular sessions over a period of 30 months, though the majority were between eight and 12 weeks. Durations of this length are common among exercise and physical activity intervention studies to allow for physiological adaptation and familiarisation. In contrast, the follow-up period was o en inadequate, as many reviews reported only a single follow-up point (immediately post-intervention), or repeated measures over the short-term (less than six months): only six of the 21 reviews planned to assess participants over the long term ( . With chronic conditions, it would be advisable to include longer follow-up periods (beyond 12 months post-randomisation) as long-term solutions may be more relevant to their control or pain management. It is also possible that initial adaptation and potential benefits as a result of an exercise intervention may take longer to manifest in comparison to a 'healthy' person due to the possible limitations in exercise intensity and progression (a training threshold) beyond which any additional physical training may be detrimental to the underlying pathophysiological mechanisms (Daenen 2015) or simply be additional physical stress with no additional physical benefit (Benton 2011).
We grouped outcome measurement points in this overview into short term (less than six months), intermediate term (six to 12 months), and long term (longer than 12 months). The broad time window for 'short term' outcomes (less than six months) is a potential source of heterogeneity as the early period is the one where time of measurement is most likely to result in variable outcomes. These initial problems could be overcome by use of standard reporting periods in exercise intervention studies (suggested four-weekly within the 'short term' period to assess both neural adaptation and other physiological changes). This would allow review authors to use the data gathered closest to the time point they are assessing, for more accurate analyses. Additionally, by extending the follow-up period beyond one year (long-term follow-up), heterogeneity may be reduced further. Cochrane Database of Systematic Reviews Reviews generally did not enforce a minimum exercise requirement for inclusion in their review. Additionally, not all exercise sessions were supervised or baseline fitness/physical ability was assessed subjectively, and consequently it was not reported whether the intervention was fulfilled as described, or whether the dose was enough to elicit a physiological response. Studies o en rely on the self-report of participants as to the actual physical activity and exercise being undertaken, which can lead to a greater risk of bias, and reduced study quality as it is questionable as to whether the e ect can be truly attributed to the intervention. This was examined in a previous review, where it was concluded that non-subjective physical assessment should be performed where possible (Perruchoud 2014), though these still have challenges regarding implementation.
In summary, the quality of the evidence was low (third tier): within this overview we found no tier one or tier two evidence. This is largely due to the small sample sizes and potentially underpowered studies. A number of studies within the reviews had adequately long interventions, but planned follow-up was limited to less than one year (12 months) in all but six reviews.
Interpretation of the available data, and conclusions drawn by the review authors, were appropriate, although the conclusions were sometimes stronger than warranted by the available data.
Occasionally results were not discussed with regards to the quality of the evidence or risk of bias: it is important to discuss the findings in the context of the quality of the evidence, with complete transparency, as this may a ect future research, and implications for patients, funders, and policy makers.

Potential biases in the overview process
While we have attempted to include all relevant reviews in the overview process, we do concede that by only searching the Cochrane Library, and including only current Cochrane Reviews we may have missed some key literature. However previous publications have referred to the higher quality grading (high AMSTAR score) in Cochrane Reviews due to the basic criteria necessary for publication at any stage (protocol or full review) suggesting they may be the most reliable source of evidence (O'Connell 2013).

Agreements and disagreements with other studies or reviews
This is a summary overview of current Cochrane Reviews, we are not aware of any overviews or reviews summarising non-Cochrane reviews.

A U T H O R S ' C O N C L U S I O N S
There is limited evidence of improvement in pain severity as a result of exercise. There is some evidence of improved physical function and a variable e ect on both psychological function and quality of life. However, results are inconsistent and the evidence is low quality (tier three). Promisingly however, none of the physical and activity interventions assessed appeared to cause harm to the participants.

Implications for practice For clinicians and people with chronic pain
The evidence in this overview suggests that the broad spectrum of physical activity and exercise interventions assessed here (aerobic, strength, flexibility, range of motion, and core or balance training programmes, as well as yoga, Pilates, and tai chi) are potentially beneficial, though the evidence for benefit is low quality and inconsistent. The most commonly reported adverse events were increased soreness or muscle pain, which subsided a er several weeks of the intervention.
Physical activity and exercise may improve pain severity as well as physical function and quality of life.

For policy makers
The evidence showed variable results, though in some reviews there was a clinical and statistical benefit in pain relief and physical function (based on low quality evidence). The evidence suggests that physical activity or exercise is an acceptable intervention in people with chronic pain, with minimal negative adverse e ects. However based on this low quality evidence, we cannot provide direction to the content of an exercise programme should clinicians decide to implement one.

Implications for research
There is a clear need for further research into exercise and physical activity for chronic pain in adults.

General implications
• Future research should report baseline values for outcome measures in both intervention and control groups, together with detailed relevant information about the participants. Knowing the baseline value is relevant to interpreting any change observed as a result of the intervention, and understanding the broader value of the intervention. • Where possible, pain results should be reported as the number of people achieving 50%, 30%, and 10% pain relief, and the number who did not meet that point (dichotomous outcome). These are clinically important cut-o s in pain intervention research, and reporting in this way allows readers to observe the clinical e ect more e ectively. • Reporting should include median and range as well as mean and standard deviation (SD) of results. This will allow readers to review the e ects of any outliers that may have skewed the data, which o en goes unnoticed in the reporting of mean and SD alone. Cochrane Database of Systematic Reviews and is a checklist for detailing the programmes using: why (rationale), what (materials and procedures), who, how, where, when, and how much.

Design
• One previous review highlighted the increased bias o en present in questionnaires and other self-report measures of physical activity in people with chronic pain, and as a result made the recommendation to use objective measures instead, such as accelerometers, or the use of direct and indirect calorimetry, where possible (Perruchoud 2014), though these still have challenges regarding implementation. This would allow direct and exact comparison and analyses of actual energy expenditure and treatment e ect.

Population/participants/sample
• There needs to be a focus on participants with generalised and/ or widespread chronic pain, instead of (or as well as) conditionspecific populations. • Studies should include people with higher pain severity (greater than 50/100 on a 100-point visual analogue scale) at baseline. People with mild-moderate pain should still be included, but it would be advisable to separate the results for analysis, ensuring the study is adequately powered to allow this subgroup analysis in advance. This way we could determine if exercise has benefit overall, or a ects one group more than another, and tailor exercise programmes according to the individual needs. • It has been previously suggested that for 20% to 25% of participants undertaking an exercise programme there is little to no favourable response (Timmons 2014), while a small percentage (5% to 10%) have adverse events (Bouchard 2012). It is therefore vitally important that much larger sample sizes are used: ideally more than 200 participants per arm, though even this number in total would increase the quality of the evidence in the first instance. In this way we may be able to learn to identify individuals who will benefit, and those who will require further intervention.

Interventions
• Di erent forms of exercise should be researched in detail. For the purposes of this overview, we combined all physical activity and exercise interventions under one banner to determine if there was any e ect. However a number of reviews separately analysed resistance (strength) training, aerobic (endurance), and combination programmes. It is important to continue to examine di erent modalities, but currently there is not enough high quality evidence to exclude or prioritise one specific mode (resistance, endurance, stability) or medium (land/water based), or the proportion of a combination programme to be assigned to each, as all may have individual benefits for people with chronic pain. • Intensity of exercise, duration of individual sessions, and frequency should be investigated. It is this dose alongside duration (of the entire intervention) and adherence that may determine the actual e icacy. • More reviews and trials should attempt to minimise intervention heterogeneity by implementing minimum and maximum requirements. Only this way will the research community be able to determine more accurately the direction and magnitude of e ect of a specific programme or intervention. Many of these important restrictions can be implemented as subgroup analyses, though if this is the case it is important to have adequate study numbers (ideally 200 participants per arm or subgroup). • Due to the chronicity and long-term nature of the condition, physiological and psychological changes may take longer to manifest. It is widely accepted that there is a delay in muscular hypertrophy as a result of exercise, and initial gains within the first few weeks of any training programme will be as a result of neural factors (Enoka 1997); this is also in line with the grading of evidence (tier two evidence or higher requires a minimum of a four-week intervention). This suggests that longer interventions may be necessary (eight weeks for tier one evidence), though assessing participants at regular intervals, including at four weeks, would be beneficial to examine the e ect of the neural adaptation alone.

Measurement (end-points)
• Randomised controlled trials with long-term follow-up are needed. Chronic pain is defined by its chronic nature, and therefore long-term follow-up of results is equally important as the initial short-term e ect (if not more so): outcomes should be assessed beyond one year a er randomisation. In turn this will inform the direct e ect of the intervention, as well as the proportion of the population who maintains the programme of exercise employed in the intervention, or something else under the guise of physical activity as a result of participation. • The broad time window for 'short term' outcomes (less than six months) is a potential source of heterogeneity as the early period is the one where time of measurement is most likely to result in variable outcomes. These initial problems could be overcome by use of standard reporting periods in exercise intervention studies (suggested four-weekly assessment within the 'short term' period to assess both neural adaptation and other physiological changes). This would allow review authors to use the data recorded closest to the time point they are assessing, for more accurate and comparable analyses.

Other
• It would be of interest in future research to determine the reasons for non-participation in regular physical activity or noncompliance to a prescribed exercise intervention in people with chronic pain, and how to overcome these barriers. • Future Cochrane Reviews could include: exercise for chronic pain or chronic widespread pain (and not specific conditions such as osteoarthritis, fibromyalgia, etc.), and exercise for neuropathic pain. These areas have not been covered by Cochrane with an exercise or physical activity intervention.

A D D I T I O N A L T A B L E S Criteria
Specific requirements (possible answers: yes, no, cannot answer, not applicable)

Was an 'a priori' design used?
The research question and inclusion criteria should be established before the conduct of the review.
Note: need to refer to a protocol, ethics approval, or predetermined/a priori published research objectives to score a "yes." 2. Was there duplicate study selection and data extraction?
There should be at least 2 independent data extractors and a consensus procedure for disagreements should be in place.
Note: 2 people do study selection, 2 people do data extraction, consensus process or 1 person checks the other person's work.

Was a comprehensive literature search performed?
At least 2 electronic sources should be searched. The report must include years and databases used (e.g. CENTRAL, MEDLINE, and Embase). Keywords or MeSH terms (or both) must be stated and where feasible the search strategy should be provided. All searches should be supplemented by consulting current contents, reviews, textbooks, specialised registers, or experts in the particular field of study, and by reviewing the references in the studies found.

Cochrane Database of Systematic Reviews
Note: if review indicates that there was a search for "grey literature"or "unpublished literature,"indicate "yes."SIGLE database, dissertations, conference proceedings, and trial registries are all considered grey for this purpose. If searching a source that contains both grey and non-grey, must specify that they were searching for grey/unpublished literature.

Was a list of studies (included and excluded) provided?
A list of included and excluded studies should be provided.
Note: acceptable if the excluded studies were referenced. If there was an electronic link to the list but the link is no longer active, select "no." 6. Were the characteristics of the included studies provided?
In an aggregated form such as a table, data from the original studies should be provided on the participants, interventions, and outcomes. The ranges of characteristics in all the studies analysed, e.g. age, race, sex, relevant socioeconomic data, disease status, duration, severity, or other diseases should be reported.
Note: acceptable if not in table format as long as they are described as above.
7. Was the scientific quality of the included studies assessed and documented?
'A priori' methods of assessment should be provided (e.g. for effectiveness studies if the author(s) chose to include only randomised, double-blind, placebo-controlled studies, or allocation concealment as inclusion criteria); for other types of studies alternative items will be relevant.
Note: can include use of a quality scoring tool or checklist, e.g. Jadad scale, risk of bias, sensitivity analysis, etc., or a description of quality items, with some type of result for EACH study ("low"or "high"is acceptable, as long as it is clear which studies scored "low"and which scored "high;"a summary score/range for all studies is not acceptable).
8. Was the scientific quality of the included studies used appropriately in formulating conclusions?
The results of the methodological rigor and scientific quality should be considered in the analysis and the conclusions of the review, and explicitly stated in formulating recommendations.
Note: might say something such as "the results should be interpreted with caution due to poor quality of included studies."Cannot score "yes"for this question if scored "no"for question 7.
9. Were the methods used to combine findings of studies appropriate?
For the pooled results, a test should be done to ensure the studies were combinable, to assess their homogeneity (i.e. Chi 2 test for homogeneity, I 2 statistic). If heterogeneity exists, a random-effects model should be used or the clinical appropriateness of combining should be taken into consideration (i.e. is it sensible to combine?), or both.
Note: indicate "yes"if they mention or describe heterogeneity, i.e. if they explain that they cannot pool because of heterogeneity/variability between interventions.

Was the likelihood of publication bias assessed?
An assessment of publication bias should include a combination of graphical aids (e.g. funnel plot, other available tests) or statistical tests (e.g. Egger regression test), or both.
Note: if no test values or funnel plot included, score "no."Score "yes"if they mention that publication bias could not be assessed because there were fewer than 10 included studies.

Was the conflict of interest stated?
Potential sources of support should be clearly acknowledged in both the systematic review and the included studies.
Note: to get a "yes,"must indicate source of funding or support for the systematic review AND for each of the included studies.

Review Reason for exclusion from overview
Aggarwal 2011 Not exercise/physical activity    (Continued) RCT: randomised controlled trial.    not reported Any exercise programme used in the treatment of intermittent claudication was included, such as walking, skipping and running. Inclusion of trials was not affected by the duration, frequency or intensity of the exercise programme but these issues were taken into account in the metaanalysis

Duration of pain/ diagnosis
Exercise was compared to six different modes of treatment, the most common being usual care or placebo. Two early trials compared exercise with placebo tablets but in more recent studies usual care was used as the control comparator. Exercise was compared with the following drug therapies: antiplatelet agents pentoxifylline, iloprost, and vitamin E. One study compared exercise with pneumatic foot and calf compression.   Hip or knee OA > 6 months High-intensity physical activity or exercise programme.

Cochrane Database of Systematic Reviews
Low-intensity physical activity or exercise programme and control (no-exercise) group in 1 study.
Pain, physical function, quality of life, adverse effects (related to intervention), severe adverse events or withdrawal (due to intervention)        (Continued) 1RM: one repetition maximum; ACSM: American College of Sport Medicine; HRmax: maximum heart rate; HRR: heart rate reserve, IQR: interquartile range; MCE: motor control exercise; MVC: maximum voluntary contraction; RPE: rating of perceived exertion; wk: week.

Total "Y"
Total "N" n/a n/a n/a n/a n/a Y 6 0 5   Table 6. Methodological quality of included reviews using the AMSTAR tool (Continued) N: no; n/a: not applicable; Y: yes; out of maximum summative score of 11. Following arbitration, the authors removed the response "cannot answer" due to no responses as such.  Silva 2010 0 n/a n/a n/a n/a n/a n/a n/a n/a van der Heijden 2015 10 8 6 0 0 6 9 2, total n > 100 10

Review Review authors' conclusions Overview authors' assessment of conclusions
Bartels 2007 "Aquatic exercise has some short-term beneficial effects on the condition of OA patients with hip or knee OA or both. The controlled and randomised studies in this area are still too few to give further recommendations on how to use this therapy... No long-term effects have been found." Appropriate conclusions based on available data. No mention of quality/risk of bias in conclusions, though found to be high quality in results section.
Bidonde 2014 "Low to moderate quality evidence relative to control suggests that aquatic training is beneficial for improving wellness, symptoms, and fitness in adults with fibromyalgia. Very low to low quality evidence suggests that there are benefits of aquatic and land-based exercise, except in muscle strength (very low quality evidence favoring land). No serious adverse effects were reported." Appropriate conclusions based on available data.
Boldt 2014 "Evidence is insufficient to suggest that non-pharmacological treatments are effective in reducing chronic pain in people living with SCI. The benefits and harms of commonly used non-pharmacological pain treatments should be investigated in randomised controlled trials with adequate sample size and study methodology" Appropriate conclusions based on available data.
Brown 2010 "There is a lack of available evidence to support the use of exercise in the alleviation of symptoms associated with dysmenorrhoea. The limited evidence implies that there are no adverse effects associated with exercise." Review authors should not have commented on lack of adverse events as this was not reported in the included study. The comment on lack of adverse events contravened present Cochrane guidance.
Busch 2007 "There is moderate quality evidence that short-term aerobic training (at the intensity recommended for increases in cardiorespiratory fitness) produces important benefits in people with FM in global outcome measures, physical function, and possibly pain and tender points. There is limited evidence that strength training improves a number of outcomes including pain, global wellbeing, physical function, tender points and depression. There is insufficient evidence regarding the effects of flexibility exercise. Adherence to many of the aerobic exercise interventions described in the included studies was poor." Appropriate conclusions based on available data.
Busch 2013 "We have found evidence in outcomes representing wellness, symptoms, and physical fitness favoring resistance training over usual treatment and over flexibility exercise, and favoring aerobic training over resistance training. Despite large effect sizes for many outcomes, the evidence has been decreased to low quality based on small sample sizes, small number of randomized clinical trials (RCTs), and the problems with description of study methods in some of the included studies." Appropriate conclusions based on available data.
Cramp 2013 "There is some evidence that physical activity interventions ... may help to reduce fatigue in RA. However, the optimal parameters and components of these interventions are not yet established." Appropriate conclusions based on available data. However, no mention of quality/risk of bias of studies in conclusion despite low/unclear quality score in results and discussion sections.

Cochrane Database of Systematic Reviews
No conclusions about effect on pain (insufficient data).
Fransen 2014 "There is currently high-level evidence that land-based exercise will reduce hip pain, and improve physical function, among people with symptomatic hip osteoarthritis." Evidence was good quality though sample sizes were often small (i.e. it is debatable if this was high level evidence as claimed by authors). Agree that results demonstrate small but significant benefit from intervention.
Fransen 2015 "High-quality evidence suggests that land-based therapeutic exercise provides benefit in terms of reduced knee pain and quality of life and moderate-quality evidence of improved physical function among people with knee OA… Despite the lack of blinding we did not downgrade the quality of evidence for risk of performance or detection bias." Appropriate conclusions based on available data. May have been generous with quality assessment but this was stated in conclusions for transparency.
Gross 2015a "…there is still no high quality evidence and uncertainty about the effectiveness of exercise for neck pain… Moderate quality evidence supports the use specific strengthening exercises as a part of routine practice … Moderate quality evidence supports the use of strengthening exercises, combined with endurance or stretching exercises may also yield similar beneficial results. However, low quality evidence notes when only stretching or only endurance type exercises … there may be minimal beneficial effects for both neck pain and function." Appropriate conclusions based on available data.

Han 2004
"Tai chi appears to have no detrimental effects on the disease activity of RA in terms of swollen/tender joints and activities of daily living…tai chi appears to be safe, since only 1 participant out of 121 withdrew due to adverse effects and withdrawals were greater in the control groups than the tai chi groups." Appropriate conclusions based on available data. However, no mention of quality/risk of bias in conclusion despite very low quality score in results section.
Hayden 2005 "Evidence from randomized controlled trials demonstrates that exercise therapy is effective at reducing pain and functional limitations in the treatment of chronic low-back pain, though cautious interpretation is required due to limitations in this literature." Appropriate conclusions based on available data. However, no mention of quality/risk of bias of studies in conclusion despite low quality score in results and discussion sections.
Hurkmans 2009 "Short-term, land-based dynamic exercise programs have a positive effect on aerobic capacity (aerobic capacity training whether or not combined with muscle strength training) and muscle strength (aerobic capacity training combined with muscle strength training) immediately after the intervention, but not after a follow-up period. Short-term, water-based dynamic exercise programs have a positive effect on functional ability and aerobic capacity directly after the intervention but it is unknown whether these effects are maintained after follow-up. Long-term, land-based dynamic exercise programs (aerobic capacity and muscle strength training) have a positive effect on functional ability, aerobic capacity, and muscle strength immediately after the intervention but it is unknown whether these effects are maintained after follow-up... Based on the evidence, aerobic capacity training combined with muscle strength training is recommended for routine practice in patients with RA." Appropriate conclusions based on available data. However, no mention of quality/risk of bias of studies in conclusion.
No conclusions regarding pain severity.

Cochrane Database of Systematic Reviews
Koopman 2015 "Data from two single trials suggested that muscle strengthening of thumb muscles (very low-quality evidence) ... are safe and beneficial for improving muscle strength ... with unknown effects on activity limitations." "We found evidence varying from very low quality to high quality that ... rehabilitation in a warm or cold climate are not beneficial in PPS." "Due to a lack of good-quality data and randomised studies, it was impossible to draw definitive conclusions about the effectiveness of interventions in people with PPS." Appropriate conclusions based on available data.
Lane 2014 "… Exercise therapy should play an important part in the care of selected patients with intermittent claudication, to improve walking times and distances. Effects were demonstrated following three months of supervised exercise although some programmes lasted over one year." Appropriate conclusions based on available data. However, no mention of quality/risk of bias of studies in conclusion.
No conclusions regarding pain severity.
Lauret 2014 "There was no clear evidence of differences between supervised walking exercise and alternative exercise modes in improving the maximum and pain-free walking distance of patients with intermittent claudication….
The results indicate that alternative exercise modes may be useful when supervised walking exercise is not an option for the patient." Appropriate conclusions based on available data. However, no mention of quality/risk of bias of studies in conclusion (in discussion).
Regnaux 2015 "We found very low-to low-quality evidence for no important clinical benefit of high-intensity compared to low-intensity exercise programs in improving pain and physical function in the short term.... The included studies did not provide any justification for the levels of intensity of exercise programs. No authors reported evidence for the minimal and maximal intensity that could be delivered." Appropriate conclusions based on available data. This overview has only used one study of the six included as it alone included a control group, for which we could not extract data as the control comparison was not used in the analysis by the review authors.
Saragiotto 2016 "There is very low to moderate quality evidence that MCE has a clinically important effect compared with a minimal intervention for chronic low back pain... As MCE appears to be a safe form of exercise and none of the other types of exercise stands out, the choice of exercise for chronic low back pain should depend on patient or therapist preferences, therapist training, costs and safety." Appropriate conclusions based on available data.
Silva 2010 "We were not able to provide any evidence to support the application of balance exercises (proprioceptive training) alone in patients with RA." Appropriate conclusions based on available data (no included studies).
van der Heijden 2015 "This review has found very low quality but consistent evidence that exercise therapy for patellofemoral pain syndrome (PFPS) may result in clinically important reduction in pain and improvement in functional ability." No subgroup analysis to differentiate between acute, subacute, and chronic pain made it difficult to extract appropriate data for this review.
Yamato 2015 "No definite conclusions or recommendations can be made as we did not find any high quality evidence for any of the treatment comparisons, outcomes or follow-up periods investigated. However, there is low to moderate quality evidence that Pilates is more effective than minimal inter-Appropriate conclusions based on available data. Cochrane Database of Systematic Reviews vention in the short and intermediate term as the benefits were consistent for pain intensity and disability, with most of the effect sizes being considered medium." There was no subgroup analysis to differentiate between acute, subacute, and chronic pain made it difficult to extract appropriate data for this review (one included study had subacute back pain (> 6 weeks), all others were chronic back pain (> 12 weeks)) but results are presented altogether as chronic pain.      Absolute difference -4 (95% CI -8 to -1) NNTB 6 (95% CI 3 to 22) n/a n/a n/a Small difference (improvement) in aquatic exercise group.
Among the major wellness outcomes, none of the outcomes met the threshold for clinically relevant differences (15%). n/a n/a n/a Function was significantly improved from aerobic exercise training, strength training neared significance.
Moderate effect size.
Busch 2013 HAQ and SF-36 for function 3 (107) Change score MD -6.29 (95% CI -10.45 to -2.13) n/a n/a n/a Significantly favourable effect of exercise. Disability 4 (not reported) n/a n/a n/a n/a "Studies investigating hydrotherapy and tai chi demonstrated statistically significant improvements in the intervention arm compared to the control arm between baseline and follow-up. The studies investigating strength training and Ivengar yoga did not demonstrate a statistically significant difference between study arms." The demonstrated effect size for exercise was equivalent to an improvement of physical function of 7 points (95% CI 1 to 12) on a 0 to 100 scale compared with a control group SMD -0.37 (95% CI -0.57 to -0.16) favoured exercise (P < 0.001) n/a n/a Statistically significant, but small effect size only.  Activity limitation: 6-months post-intervention: ADL-index: MD -2.90 (95% CI -4.73 to -1.07) n/a n/a Activity limitation: favoured intervention at both assessment points.
"The baseline imbalance in favour of the usual care group probably biased these results." No relevant studies n/a n/a n/a n/a No relevant studies.
Regnaux 2015 (OA) WOMAC (0 to 68) disability scale, and muscle strength 1 (68) -excluded control (no-exercise data: n = 34) n/a n/a n/a n/a Could not extract exercise vs control data -data presented for high vs low intensity groups only, not compared to control.

HAQ function
No studies found n/a n/a n/a n/a No studies found.