Search strategy

To maximise the sensitivity of the search, three separate search strings were combined, including terms for MSK pain, therapeutic exercise and adherence (see online supplementary appendix 1). The search strategy was adapted from a Cochrane review of interventions to improve adherence to exercise for chronic MSK pain in adults.13 The adherence terms were limited to the title only to restrict the search to studies in which adherence was the primary focus. A pilot search was conducted to refine the focus of the strategy while maintaining sufficient sensitivity to identify key studies already known to the authors. The review was not registered a priori.

Data sources

Seven databases were searched from inception to December 2016 using the OvidSP and EBSCO interfaces: MEDLINE, AMED, EMBASE, PsychINFO, HMIC, CINAHL and SPORTDiscus. The Cochrane Database of Systematic Reviews was also searched and free text searching using Google Scholar was carried out. The reference lists of included studies were checked, and Web of Science was searched for papers that had cited included studies to identify further titles that may have met the inclusion criteria. Identified titles and abstracts, then subsequent full texts were screened by pairs of researchers (DLB, AB, MAH, JGQ) independently, and any conflicts were arbitrated by an additional researcher (NEF).

Studies were eligible if they featured all of the following:

1. A definition of exercise adherence: either an explicit definition such as a quote or citation, or a description of how exercise adherence was or should be measured.

2. Adults with MSK pain: including back, neck, hip, knee, ankle, foot, shoulder, elbow, wrist or hand pain, MSK conditions with wider systemic effects such as fibromyalgia and rheumatoid arthritis, postsurgical pain patients where the surgery was for a MSK condition, individuals with a non-specific diagnosis of MSK pain and those with a specific diagnosis (eg, OA or adhesive capsulitis) with or without supporting imaging or other diagnostic test results.

3. Therapeutic exercise: defined as any form of supervised or unsupervised exercise or physical activity specifically provided to patients to treat their MSK pain condition. Any type of exercise was included, such as strengthening, stretching, aerobic or mixed and exercise delivered by an HCP, trained lay representative, fitness instructor or delivered as part of a multidisciplinary package of care. All study types were eligible, not just those where therapeutic exercise was being assessed as an intervention.

Studies were excluded if they:

1. Were not published in English.

2. Comprised conference proceedings where only the abstract has been published.

3. Involved exercise for a non-MSK pain condition, such as cancer, falls prevention, stroke or cardiac rehabilitation.

4. Included participants under 18 years.

5. Included participants with no MSK pain.

Data extraction

Data were extracted independently by two researchers using customised forms. Extracted data were compared for inconsistencies and any corrections made following discussion by the researchers. A third researcher was consulted if needed. We extracted details of the study (design, country, setting), participant details (age, sample size, population, inclusion/exclusion criteria, MSK pain condition), the TE intervention (type, dosage, adherence measure time-point, exercise prescriber, supervised or unsupervised exercise), definitions of adherence (quotes or citations were extracted verbatim), the parameters of exercise adherence assessed (eg, number of repetitions, attendances or intensity of exercise), methods used for measuring adherence (eg, class register or self-report diary) and quantification or values assigned to adherence (eg, number of exercises that should be completed to be classed as adherent). Systematic reviews and protocols were included if they met the above criteria as it was possible they could include information about how adherence is defined. However, only data on the definition of adherence were extracted from systematic reviews to avoid the over-representation of studies that may have been included in a systematic review and as primary research in this review.

Analysis

Data from included studies were summarised in tables. Terms used for describing parameters and measurement methods were standardised and frequency and percentage counts applied. Values for quantification of adherence were standardised by converting to percentages for ease of comparison and grouped according to commonly observed ranges. We did not formally assess the risk of bias since the methodological quality of included studies would not have influenced the utility or relevance of the data that were extracted for the purposes of this systematic review.

Characteristics of included studies

Included studies were published between 1976 and 2016 and originated from 20 different countries (see online supplementary table 1). Sixteen different study types were included and hospital outpatients was the most common study setting (n=31). Seventeen different categories of MSK pain conditions featured, most commonly back and/or neck pain (n=30). The type of TE could be broadly grouped as: strengthening, stretching, aerobic, postural or mind-body exercise24 (eg, yoga), with strengthening (n=43) and stretching (n=35) being the most common. The time-point at which exercise adherence was measured in the studies varied from 1 week to 5 years.

Definitions of exercise adherence

Adherence data findings extracted from included studies are shown in online supplementary appendix 3. Sixty per cent of studies (n=52) did not provide a clearly specified definition of adherence. Most studies defined adherence by describing the parameters by which it had been assessed (n=73, 85% of 86).

Thirty-four studies (40%) provided at least one clearly specified definition, the most common of which are shown in table 1. Five studies provided definitions that were the authors’ own or were uncited. Frost et al 25 provided their own definition in addition to that of WHO’s definition. Their definition, ‘the extent to which individuals undertake prescribed behaviour accurately and at the agreed frequency, intensity and duration’ was based on the four parameters of adherence they believed characterised rehabilitation prescriptions, namely: frequency, duration, intensity and accuracy.

Parameters used to assess exercise adherence

Frequency of exercise completion was the most commonly used parameter to assess adherence (n=47, 55%) (table 2). Frequency was measured in different ways, including: exercise repetitions, sets or blocks of exercise time, and over different time frames (eg, exercises per day, week or month).

Behavioural parameters were measured to assess adherence in 16 studies (19% of 86). These included HCP assessed elements such as ‘following guidance’ and ‘receptivity to change’, as featured in the Sports Injury Rehabilitation Scale (SIRAS),19 and self-reported elements such as ‘barriers to exercise’. Session attendance, which required the patient to be present at a supervised exercise session, was assessed in 22 (26%) studies. This was slightly different to ‘session completion’ assessed in 11 (13%) studies, which was a self-reported or HCP observed completion of recommended exercises during an exercise session.

Exercise exertion or intensity was assessed in 13 studies (15%), subjectively in 8 studies (9%) and objectively in 5 studies (6%). Seven studies (8%) assessed a parameter reflecting the quality of the patients’ replication of the recommended exercises, which was determined via HCP observation.

Methods used to measure adherence

Of the 86 included studies, 74 (86%) described at least one method of measuring adherence. The most common measurement methods were self-report exercise logs (n=44, 51%), registers of attendance (n=18, 21%) and an existing measurement scale (n=15, 17%). Thirty-three studies (38%) used more than one type of measurement method. Six studies used an objective measure of exercise adherence (eg, pedometer) (table 3). Within the category of ‘existing measurement scales’, 12 different measures of exercise adherence were identified. These included the SIRAS,19 General Adherence Scale26 and Adherence to Physiotherapists Recommendation Scale.27

Values for adherence

Half of the studies (n=44, 51%) did not provide information about what they considered to be a satisfactory value for TE adherence. Forty-two of the studies (49%) described values indicating ‘satisfactory’ TE adherence (table 3). These were grouped into four ranges as shown in table 4. The most common range of values for satisfactory adherence was between 80% and 99% completion of the prescribed exercise/s. Six studies (7%) provided values specifically describing ‘low adherence’. These varied between 0% and 79% completion of the recommended exercises. It was unclear in many of the studies whether cut-off points for satisfactory adherence were determined a priori or post hoc, as they were not described in the methods. Where cut-off points were stated, no references to required therapeutic dosages or other guidelines were mentioned. Some studies used the distribution of participants’ adherence data to assign a value of satisfactory adherence. For example, Granlund et al 28 used the median value of participants’ adherence results to dichotomise them into adherent or non-adherent groups, whereas van Gool et al 29 ordered the participants according to their adherence results, then divided them into three equally sized groups described as ‘low’, ‘intermediate’ or ‘high’ adherence.

Definitions of adherence

Most studies did not state a definition of adherence. This may be because a definition of adherence that is explicitly related to adherence to TE does not exist. Alternatively, it might be that the meaning of adherence is so clear, having undergone so little development between the 1976 version by Sackett and Haynes to the 2003 version from WHO, that there is an assumption that the term is tacitly understood among research and clinical communities. However, the lack of an agreed communicable definition of adherence is a problem, because without a clear picture of what adherence is, HCPs and patients cannot work together to achieve adherence as a shared goal, nor measure or monitor its variability.

The original definition by Sackett and Haynes22 used the term ‘compliance’ not ‘adherence’, and concerns therapeutic regimens generally, not specifically MSK pain or TE. While the more recent WHO iteration of this definition (2003)23 uses the term ‘adherence’, to better reflect the autonomy of the patient and his/her role as an ‘active partner’ in the treatment process, it was developed by groups of experts focussing on adherence issues relating to chronic health conditions other than MSK pain (including hypertension, epilepsy, HIV and tuberculosis). Consequently, pharmacological intervention was the main treatment of concern for both these definitions, but it is not clear whether adherence to TE can be defined in the same way. These possible differences may explain why the WHO definition (2003)23 has not been universally applied to TE for MSK pain. The WHO definition (2003) lacks specificity, providing no indication of the parameters of treatment adherence that should be assessed, how they should be measured or what levels of adherence should be expected. This ambiguity means definitions can be interpreted differently, as shown by this review, where several interpretations of a single definition were associated with authors using different parameters, measurement methods and cut-offs.

It has been proposed that the term compliance should be used in efficacy trials where the intervention has been determined according to the desired outcome and target population, but not necessarily the individual patient or recipient.30 Adherence on the other hand is proposed to reflect the effectiveness of TE in trials or real-life contexts.30 Adherence considers the individual’s role in interpreting the appropriateness of the advised intervention and the influences on their behaviour and motivation, such as the environment, society, previous experience, knowledge, symptoms and resources. Adherence is a more complex measurement than simply comparing with a reference standard intervention dose.30 Until the message we are trying to convey with these terms is clarified by all those to whom it pertains, including patients, the uptake of existing definitions may remain inconsistent in MSK pain and TE literature.

The definition of adherence provided by Frost et al 25 is specific to TE for MSK pain and identified the parameters of TE adherence that should be measured: frequency, duration, intensity and accuracy. However, its development was based on stroke rehabilitation,31 public health,32 falls prevention33 and the use of exercise sheets in physiotherapy,34 not literature focussing specifically on TE prescription for MSK pain. Therefore, it is possible that the definition does not reflect the perceptions of patients with MSK pain or HCPs responsible for prescribing TE. While Frost et al 25 offer the first operational definition of TE adherence, its four parameters are different to those identified as most commonly used in this review. Therefore, the concept of adherence to TE for MSK pain may require further refinement before the development of a new measure of adherence.

Parameters of adherence

We identified eight categories of TE adherence parameters (measurable aspects of adherence). The most common was exercise frequency, a finding consistent with other reviews, suggesting it is a relevant dimension of TE adherence, or simply an easy parameter to measure. However, even exercise frequency was assessed in various ways (eg, repetitions or blocks of time, per day or week), suggesting it is more complex than a simple representation of the total TE undertaken.

The accuracy with which patients replicated their exercises, or the quality of their exercise performance was included as a parameter in seven studies. The scant attention paid to accuracy or quality of performance may be because it is not an important dimension of TE adherence, or that it is too challenging to incorporate into research, despite being a common parameter of adherence to assess in clinical practice.15 HCP observation has been recommended in the assessment of TE adherence35 and is a feature of one of the more commonly used measures: the SIRAS.36 For research purposes, the constraint of requiring an observer and its associated cost implications may be why these parameters have not been regularly assessed. While technology may be able to objectively measure certain parameters of TE,37 the expense of such equipment, the expertise required to operate it and the number of sensors required to measure more complex TE interventions, may limit its implementation.38 The acceptability of the equipment by patients may also limit its effectiveness as a measurement tool, as patients may not adhere to using the measurement device, despite adhering to the TE intervention. Natural adherence behaviour may also be affected by overt monitoring, such as wearing a measurement device, although similar changes may also be associated with subjective measurement methods.

Behavioural components featured in 16 studies, 2 of which, used self-reported barriers to exercise completion as a parameter of adherence.39 40 This is an uncommon approach, as barriers are normally seen as modifiers of adherence levels. However, poor treatment adherence has been associated with a greater perceived number of barriers41 indicating a potential role for barriers in the assessment of adherence. Some studies identified single parameters of TE adherence, whereas other studies identified combinations of parameters (eg, exercise frequency and session attendance), without justification. These findings suggest there are multiple relevant parameters of adherence to TE, but no consensus as to their importance or relevance to a specific context. This may relate to non-specific definitions such as WHO’s focus on pharmacological interventions, whereas TE prescription may potentially have more parameters contributing to adherence, such as frequency, intensity, time, accuracy or a behavioural component, which require different methods for their measurement.

Methods of measuring adherence

Multiple methods of measuring exercise adherence were identified, including self-report exercise logs, class registers and existing measurement scales, mirroring the findings from recent systematic reviews.14–16 However, different measurement methods were used across studies for the same adherence parameter. For example, the parameter of session attendance was measured using a class register, interview and self-report log. Such variation is understandable given the multiple parameters of adherence and the non-specific guidance provided by WHO’s adherence report,23 which suggests, ‘a multimethod approach combining self-reporting and objective measures’. However, the lack of a single valid and reliable measurement tool of TE adherence means that the relative effectiveness of interventions is difficult to compare across studies.13–16

There was large variability in time points at which adherence was measured, varying from 1 week to 5 years. As adherence is likely to change over time,13 the point at which adherence is measured could substantially influence the rates of adherence observed. To improve the comparability of data, it may be useful to establish adherence measurement timescales (eg, short-term, medium-term and long-term). There should also be improved clarity in the reporting of adherence measurement timing (eg, whether measurements took place daily, weekly or monthly), and for studies where there are multiple deliveries of the TE intervention (eg, weekly exercise classes), the time interval between the TE delivery and the adherence measurement.

Values of adherence

Fewer than half of included studies (49%) defined satisfactory adherence. Two approaches were used, either a predetermined cut-off or a distribution method. Distribution-based methods result in the adherence of an individual being judged relative to the performance of other study participants, rather than any predetermined target level of adherence. This explains why satisfactory adherence levels ranged from 14% to 100% completion of prescribed exercises in this review. Only Hammer et al 42 used predetermined adherence categories justified from the literature.43 However, the original source provides no explanation of how these categories were derived. In our review, satisfactory adherence was most commonly valued between 80% and 99% completion of the prescribed exercises (n=15, 17%). Furthermore, several studies described low adherence, providing ranges of prescribed exercise completion between 0% and 79%. This may suggest that 80% could be a reasonable threshold for satisfactory adherence.

If specific TE doses are unknown, it may be inappropriate and unhelpful to set arbitrary adherence cut-off points for TE interventions for MSK pain. A systematic review and meta-analysis of medication adherence showed that good adherence to placebo medication was associated with lower mortality, a so-called ‘healthy adherer’ effect.44 Therefore, it may be beneficial for a patient to identify as adherent, even in the absence of a known therapeutic dosage. Identifying a patient with MSK pain as non-adherent when the therapeutic dosage of TE is unknown, may deny the patient the benefits of the healthy adherer effect.

Research and clinical implications

There is a need to focus on establishing a clear definition of adherence specifically relating to TE for MSK pain, and the best way to measure adherence. Patients must be involved as active, collaborative partners in future research so that the resulting construct of exercise adherence incorporates their views.45 In the absence of a valid, reliable and acceptable measure of TE adherence for MSK pain, the interpretation of results from trials investigating TE interventions or methods for improving TE adherence may be questioned. This underscores the importance of agreeing on a satisfactory measure/s. From the clinician’s perspective, if he or she cannot be sure of how to accurately define and measure adherence, it may be difficult to determine whether a patient’s lack of progress despite treatment is due to inadequate engagement or ineffective treatment.

Limitations

We adopted an inclusive approach using a modified version of an established Cochrane search strategy to include all study types and MSK pain conditions; however, it is possible that some studies relating to adherence to TE for MSK conditions may have been missed. To minimise error and bias, two independent reviewers assessed both title and abstract and full-text papers for inclusion and data extraction. The search terms were limited to title, or title and abstract. Restricting the search in this way ensured that the included studies were specifically focused on adherence in the context of TE for MSK pain and therefore more likely to be appropriate to this review. It is possible that some relevant studies may have been missed that did not feature adherence in their title. However, different search strategies were piloted to achieve results with a focus on adherence while maintaining acceptable sensitivity.

Since a definition of adherence was not provided in the majority of included studies despite their title featuring a term for adherence, it is unlikely that articles not featuring an adherence term in their title would discuss the concept in any detail. Equally, the majority of definitions used were from the same source, and it is unlikely that this would change if more articles were reviewed. Six studies were excluded as they were not published in English. Consequently, our review is at risk of language bias, and the results may under-represent studies that are not published in English. The countries of origin of the excluded studies were still represented in the final 86 included studies. No low-income countries were studied. Therefore, our findings may only be generalisable to high-income countries and their cultures.