Effectiveness of telephone-based interventions for managing osteoarthritis and spinal pain: a systematic review and meta-analysis

Background Osteoarthritis and spinal pain are common and burdensome conditions; however, the majority of patients with these conditions do not receive care that is consistent with clinical practice guidelines. Telehealth models of care have the potential to improve care for osteoarthritis and spinal pain patients. The aim of this review was to assess the effectiveness of verbal real-time telehealth interventions, including telephone-based and videoconferencing interventions to reduce pain intensity and disability in patients with osteoarthritis of the knee or hip and spinal pain (back or neck pain). Methods We searched seven electronic databases from inception to May 2018. Randomised controlled trials (RCTs), cluster-RCTs, and non-randomised controlled trials were included. Two review authors independently extracted data for each included study. Primary outcomes were pain intensity and disability. We conducted primary meta-analyses combining all conditions with similar interventions and comparators. Standardised mean difference (SMD) and 95% confidence intervals (CIs) were calculated using random effects models. We used the Cochrane Risk of Bias tool to assess risk of bias, and GRADE to evaluate the quality of evidence. Results We included 23 studies with 56 trial arms and 4,994 participants. All studies utilised telephone-based interventions. Only two studies used a telephone only approach and the remainder included educational materials and/or face-to-face components. We found no studies utilising videoconferencing. Meta-analysis showed telephone-based interventions (with educational materials) for osteoarthritis and spinal pain improved pain intensity (n = 5 trials, n = 1,357 participants, SMD −0.27, 95% CI [−0.53, −0.01], Tau2 = 0.06, I2 = 74%; moderate-quality evidence) and disability (n = 7 trials, n = 1,537 participants, SMD −0.21, 95% CI [−0.40, −0.02], Tau2 = 0.03, I2 = 56%; moderate-quality evidence) compared to usual care. Meta-analyses found telephone with face-to-face interventions does not improve pain and disability compared to usual care or face-to-face care alone. Discussion We are moderately confident that telephone-based interventions reduce pain intensity and disability in patients with osteoarthritis and spinal pain compared to usual care, but telephone plus face-to-face interventions are no more effective than usual care or face-to-face interventions alone.

INTRODUCTION approaches overcome such barriers by providing better access to recommended care (McLean et al., 2013).
Telephone-based and videoconferencing interventions are a subset of telehealth models of care, which offer direct verbal patient-provider contact and have increasing appeal to support patient care via remote delivery. A previous review of telehealth interventions for patients with chronic non-specific low back pain, which included mostly asynchronous delivery via web-based platforms such as websites, email, and peer communication services, found these to be no more effective than minimalist interventions (i.e. health or non-health-related information) in improving pain and disability in these patients (Dario et al., 2017). In contrast, another recent review found utilising telephone-based and videoconferencing interventions improves physical function in post-surgical rehabilitation of musculoskeletal conditions, including spinal pain, rheumatoid arthritis, and osteoarthritis (Cottrell et al., 2017a). These results combined with other data revealing telephone-based models of care are preferred by patients with chronic musculoskeletal conditions Cottrell et al., 2017b), suggest remotely delivered verbal real-time interventions such as by telephone or videoconferencing have promising effects. However, while a number of trials investigating the use of telephone-based interventions for osteoarthritis and spinal pain have been conducted (Weinberger et al., 1989;Allen et al., 2016;Bennell et al., 2017;Gialanella et al., 2017) there has been no comprehensive review of the evidence regarding the effectiveness of telephone-based or videoconferencing interventions for this patient group. As such, the primary objective of this systematic review was to assess the effectiveness of verbal real-time telehealth interventions, including telephone-based and videoconferencing interventions, to reduce pain intensity and disability in patients with osteoarthritis of the knee or hip and spinal pain (back or neck pain), compared to usual care or face-to-face interventions.

Search strategy
We conducted a systematic review following the PRISMA statement (Moher et al., 2009) and prospectively registered on PROSPERO (CRD42015027626). We searched Medline, Embase, AMED, Medline In-Process, PsycINFO, CINAHL, SportDiscus from inception to May 2018 to identify eligible studies (Table S1. Example search strategy for MEDLINE). We used a combination of relevant keywords based on those used in other systematic reviews to construct a search strategy including search terms for participants, intervention, study design, and comparator. The search strategy was reviewed and performed by an information specialist Debbie Booth (DB), and modified to suit each database. We searched trial registries (ClinicalTrials.gov, the Australian and New Zealand Clinical Trials Registry and the World Health Organisation International Clinical Trials Registry Platform) in May 2018. We also conducted a manual search of the reference lists of all included studies. The corresponding authors of all included studies were contacted via email to request details of any other potentially eligible studies.

Study selection
We included randomised controlled trials (RCTs), cluster RCTs (C-RCTs) and non-randomised controlled trials that had a parallel comparison group as per the a priori trial registration. Trials with non-random assignment of groups were included given Medical Research Council recommendations that non-randomised designs may represent an appropriate evaluation design for some complex health promotion interventions (Craig et al., 2008). Eligible comparison groups included other interventions, no treatment, usual care, wait-list control or attention control. To be eligible, trials had to include participants with osteoarthritis of the knee or hip, or spinal pain (back or neck pain). We included trials that defined osteoarthritis as confirmed by clinical assessment or medical diagnosis, including patient self-report of such diagnosis, with or without diagnostic imaging. For spinal pain, we included any trial that included clinically diagnosed or participant self-reported back, neck, thoracic, or cervical pain. Studies with mixed populations of musculoskeletal conditions were included where separate data were provided for osteoarthritis and spinal pain. We included trials that did not specify the location of osteoarthritis, as we assumed those studies would be representative of patients with knee or hip osteoarthritis as these are the most prevalent types of osteoarthritis (Vos et al., 2016). There were no restrictions on intensity or duration of participant symptoms. Studies that included patients with a serious pathology (e.g. cancer, infection, etc.) or included patients in the postoperative period were excluded. We excluded studies including other chronic pain conditions such as headache, rheumatoid arthritis, and neuropathic pain because they have a clearly different etiology and clinical course. There were no restrictions on the basis of publication language, status or date.
We included trials that involved service delivery by any person (i.e. therapist, health professional or trained operator) by telephone or videoconferencing in which there was a direct person-to-person verbal exchange of information. The service could be used to provide any aspect of care (e.g. delivery of advice, education, behaviour modification treatment, ongoing support). We included studies that specifically aimed to test the effectiveness of a telephone-based or videoconferencing intervention. Complex interventions with one or more delivery component (e.g. face-to-face sessions or educational materials in addition to telephone or videoconferencing) were included if the telephone or videoconferencing component was the main method of intervention delivery, defined as at least 50% of the total number of intervention contacts conducted via telephone or videoconferencing. Trials were included if they reported a valid measure of at least one of the following primary review outcomes: pain intensity or disability (including physical function), the core outcomes recommended to be used in clinical trials (e.g. OMERACT-OARSI, IMMPACT) (Dworkin et al., 2005). Secondary outcomes included psychological symptoms, self-efficacy, behavioural outcomes related to treatment (weight loss, physical activity, healthcare or medication use, treatment adherence), health-related quality of life, recovery, subjective improvement in symptoms, fear avoidance, and adverse events.

Data extraction and quality assessment
After removing duplicates, two pairs of independent authors screened the titles, abstracts, and full-texts of all identified studies (KO and AW, RH, LW, SY, SK) (Higgins & Green, 2011). A third reviewer resolved any disagreements (CW). Two authors independently extracted data from eligible studies, using a standardised data extraction tool (KO and AW). Information regarding study characteristics (design, participants, interventions, outcomes) was extracted and a third author resolved any disagreement (CW). Authors were contacted to provide further information if eligibility was unclear or to provide data in the appropriate form. When data were unavailable (e.g. standard deviations) estimations were calculated using recommended methods in the Cochrane Handbook (Higgins & Green, 2011). Included C-RCTs were assessed for unit of analysis error and adjusted as required.
Two pairs of independent authors assessed the risk of bias and overall quality of the evidence using the Cochrane Collaboration's tool (KO and FT, RH) (Higgins & Green, 2011). A third reviewer resolved any disagreement (CW). We also considered sources of other bias, including whether the intervention was delivered as intended, whether groups were comparable at baseline and whether contamination between groups occurred.
The overall quality of the evidence for each pooled analysis was assessed using the GRADE criteria (Guyatt et al., 2011). The quality of evidence was downgraded by one level when appropriate according to the following criteria: study design limitations, inconsistency of results, imprecision, indirectness, and publication bias, resulting in the quality of evidence being judged as 'high quality', 'moderate quality', 'low quality', or 'very low quality'. We used the I 2 and Tau 2 statistic to assess heterogeneity between trials, and Tau 2 > 1 and I 2 > 50% was used to identify high heterogeneity (Higgins et al., 2003). We planned to examine Egger's test to assess publication bias if a sufficient number of studies (n ! 10) were included (Egger et al., 1997).

Data synthesis and analysis
If studies reported data for multiple follow-up points, data from the longest time point was extracted for inclusion in meta-analyses (Kroon et al., 2014). Included studies were synthesised according to the intervention components (e.g. telephone or videoconferencing with educational materials or face-to-face contact) and by type of control: usual care (no treatment, usual care, wait-list control, or an attention control) or, other intervention (intervention with no telephone or videoconferencing component) (KO, CW, RH). To limit clinical heterogeneity we did not pool studies with clearly different intervention focus (e.g. physical activity vs. medication support).
Where different measures were used to measure the same outcome (i.e. disability), the most appropriate measure was selected based on the strength of their measurement properties (i.e. reliability, validity, and responsiveness) and its frequency of use in the included studies (to improve comparability across trials) (Fisher et al., 2014). For outcomes assessed using standard scales (e.g. Western Ontario and McMaster Universities Osteoarthritis Index), we used overall scores when possible. When measures presented did not point in the same direction (i.e. if some measures increase with disease severity whilst others decreased) we multiplied the mean values from one set of studies by -1 to ensure all were in the same direction (Higgins & Green, 2011). For the outcome psychological symptoms, if both anxiety and depression were presented separately, we chose to extract data on depression (Kroon et al., 2014).
Where possible, for each continuous outcome we calculated standardised mean differences (SMDs) (which allowed us to combine different measures of the same outcome) and 95% confidence intervals (CIs) and used the random effects model to pool estimates for each analysis using RevMan version 5.3.5. Generic inverse variance method was used to account for the inclusion of both C-RCTs and RCTs. In all instances where we could not combine data in a meta-analysis, including data from non-randomised trials, we provided a narrative summary of the trial findings according to the review objectives. We conducted primary meta-analyses combining all conditions with similar intervention and comparators. We also present meta-analyses by condition where possible (i.e. osteoarthritis and spinal pain separately). Cohen's d was used to classify the intervention effect sizes as small (d = 0.2), medium (d = 0.5), and large (d ! 0.8) (Carson, 2012).
We planned to perform subgroup analyses to explore the potential effect of modality of intervention delivery (telephone and videoconferencing), intervention type (single (e.g. telephone only) and multicomponent (e.g. telephone-based intervention and educational materials)), and duration of spinal pain (acute and chronic (i.e. pain that lasted longer than 3 months)). Sensitivity analyses were planned to explore the influence of overall high risk of bias on pooled treatment effects. Overall risk of bias was defined as being at high risk of bias for one or more key domains (i.e. selection bias, performance bias, detection bias, attrition bias, reporting bias) (Higgins & Green, 2011). Sensitivity analyses were planned to explore the influence of small trials (sample size <100 per group) on pooled treatment effects, as it has been previously reported that small trials tend to report larger benefits of treatment than larger trials in osteoarthritis research (Nüesch et al., 2010).

RESULTS
After duplicate removal, our search identified 3,182 records. Title and abstract screening excluded 3,075 records. Of the 107 full-text records assessed for eligibility, 57 were excluded (see Fig. 1). A total of 50 records (representing 23 studies) fulfilled the eligibility criteria and were included in the review (Fig. 1). Overall 20 RCTs, 2 C-RCT and 1 non-randomised controlled trial with a total of 4,994 participants (range 30-786) across 56 trial arms were included (Table 1). Trials were published between 1989 and 2018. A total of 12 studies were undertaken in the United States, four in Australia, and one each in Sweden, the Netherlands, Italy, Canada, Nigeria, Brazil, and England (Table 1). Intervention duration ranged from 4 weeks to 2 years, with most (n = 14) having a duration of 6 weeks to 6 months (Table 1). No studies experimentally compared interventions of varying number of calls. Five studies reported secondary analyses that examined associations of telephone call dose with patient outcomes (Thomas et al., 2002;Damush et al., 2003;Hughes et al., 2010;Williams et al., 2016;O'Brien et al., 2018).
Three studies found no association between the number of calls and patient outcomes (Damush et al., 2003;O'Brien et al., 2018;Williams et al., 2018) and two studies indicated that completion of a higher number of calls was associated with greater improvements in pain (Thomas et al., 2002;Hughes et al., 2010), physical function (Hughes et al., 2010) and depression (Hughes et al., 2010).
Eight trials included patients with knee osteoarthritis (n = 1,717), five trials included patients with hip and/or knee osteoarthritis (n = 1,965), three did not specify osteoarthritis type (n = 629), two included patients with acute back pain (n = 241), four included patients with chronic back pain (n = 342), and one included patients with chronic neck pain (n = 100) ( Table 1).
All 23 trials utilised telephone for intervention delivery (i.e. no studies utilised videoconferencing). A total of 16 trials compared a telephone-based intervention with  usual care, six trials compared a telephone-based intervention to a face-to-face intervention, and one trial used a three-arm design comparing a telephone-based intervention to usual care and to a face-to-face intervention (Table 1). Only two trials tested telephone alone as the mode of intervention delivery, five tested telephone combined with face-to-face, nine tested telephone combined with educational materials and seven tested telephone combined with face-to-face and educational materials (Table 1). All studies, except one , implemented an intervention designed specifically for osteoarthritis and spinal pain patients. All interventions focused on supporting self-management and providing education in addition to a range of intervention targets, for example physical activity (see Table 1 for details). A total of 17 studies assessed pain intensity (Weinberger et al., 1989;Mazzuca et al., 1997;Thomas et al., 2002;Buhrman et al., 2004;Hughes et al., 2010;Allen et al., 2010Allen et al., , 2016Allen et al., , 2017 The majority of studies were rated high risk for performance bias (n = 22) and detection bias (n = 21) due to the inability to blind treatments and self-reported outcomes (Fig. S1). Two studies were rated low risk for detection bias as the intervention was part of a cohort multiple RCT and the participants were unaware of reciprocal study groups Williams et al., 2018) and one was also rated low risk for performance bias as the personnel delivering the intervention were also unaware of the reciprocal study groups . The majority of studies were rated low for random sequence (n = 18), attrition bias (n = 17), and other bias (n = 16) and unclear for allocation concealment (n = 11) and reporting bias (n = 13). Egger's test was not undertaken to assess publication bias as there was not a sufficient number of studies for any of the comparisons (n < 10) (Egger et al., 1997). Visual inspection of funnel plot asymmetry was not undertaken as there was not a sufficient number of studies to judge asymmetry (n < 10) (Higgins & Green, 2011). I 2 and Tau 2 statistics suggested statistical heterogeneity in the telephone-based interventions (with educational materials) vs. usual care comparison for pain intensity (I 2 = 74%) and disability (I 2 = 56%).

Primary outcomes
Telephone-based interventions (with educational materials) vs. usual care

Pain intensity
Meta-analysis of data from five studies (Buhrman et al., 2004;Allen et al., 2010Allen et al., , 2016Allen et al., , 2017Gialanella et al., 2017) (n = 3 knee and/or hip osteoarthritis, n = 2 spinal pain; total n = 1,357 patients) revealed a small positive intervention effect of telephone-based interventions (with educational materials) on pain intensity compared to usual care (SMD -0.27, 95% CI [-0.53, -0.01], Tau 2 = 0.06, I 2 = 74%; moderate-quality evidence) (Table 2; Fig. S2). Positive intervention effects were found for spinal pain (SMD -0.55, 95% CI [-0.92, -0.19]) but not osteoarthritis when synthesised separately (Table 2; Fig. S2). All planned subgroup analyses were not possible due to the limited number of included studies (Table 2). To reduce risk of clinical heterogeneity, one study  which was not disease specific was not included in the main metaanalyses. This study reported no difference in pain intensity for the telephone-based intervention compared to usual care (Table S2).

Telephone plus face-to-face interventions vs. usual care
Pain intensity Meta-analysis of data from three studies (Li et al., 2017;Rutledge et al., 2018;Williams et al., 2018) (n = 1 trial knee osteoarthritis, n = 2 trials spinal pain; total n = 259 patients) showed no difference between telephone plus face-to-face interventions on pain intensity compared to usual care (SMD -0.08, 95%CI [-0.32, 0.16], Tau 2 = 0.00, I 2 = 0%; moderate-quality evidence) (Table 2; Fig. S4). Similar to the main analysis, no intervention effect was found for spinal pain (Table 2; Fig. S4). All planned subgroup analyses were not possible due to the limited number of included studies (Table 2). Three studies (Weinberger et al., 1989;Mazzuca et al., 1997;Thomas et al., 2002) were not included in main meta-analyses due to non-randomised design or did not report sufficient data (e.g. means at follow-up not reported). All three studies reported no difference in pain intensity between the telephone plus face-to-face interventions compared to usual care (Table S2). Disability Meta-analysis of data from four studies (Damush et al., 2003;Li et al., 2017;Rutledge et al., 2018;Williams et al., 2018) (n = 1 trial knee osteoarthritis, n = 3 trials spinal pain; total n = 398 patients) showed no difference between telephone plus face-to-face interventions compared to usual care on disability (SMD -0.08, 95%CI [-0.28, 0.12], Tau 2 = 0.00, I 2 = 0%; moderate-quality evidence) ( Table 2; Fig. S4). Similar to the main analysis, no intervention effect was found for spinal (Table 2; Fig. S4). Subgroup analyses also revealed no intervention effects for interventions including chronic spinal pain patients (Table 2; Fig. S5). Four studies (Weinberger et al., 1989;Mazzuca et al., 1997;Thomas et al., 2002;Goode et al., 2018) were not included in main meta-analyses due to non-randomised design or did not report sufficient data (e.g. means at follow-up not reported) (Table S2). Three studies reported no difference in disability between the telephone plus face-to-face interventions compared to usual care (Weinberger et al., 1989;Mazzuca et al., 1997;Thomas et al., 2002). One study reported telephone plus face-to-face interventions significantly improved disability compared to usual care (Goode et al., 2018) (Table S2).
Telephone plus comprehensive face-to-face interventions vs. face-to-face interventions alone

Secondary outcomes
Telephone-based interventions (with educational materials) vs. usual care Meta-analysis of data from five studies (Blixen et al., 2004;Buhrman et al., 2004;Allen et al., 2010Allen et al., , 2016Allen et al., , 2017) (n = 3 trials hip and/or knee osteoarthritis, n = 1 trial unspecified osteoarthritis, n = 1 trials spinal pain; total n = 1,293 patients) showed no difference between telephone-based interventions (with educational materials) and usual care in improving psychological symptoms (SMD 0.03, 95% CI [-0.10, 0.16], Tau 2 = 0.00, I 2 = 4%; moderate-quality evidence) ( Table 2; Fig. S2). Similar to the main analysis, results by patient condition showed no difference between telephone-based interventions and usual care in improving psychological symptoms for osteoarthritis (Table 2; Fig. S2). All planned subgroup analyses were not possible due to the limited number of included studies (Table 2).
Four other studies not included in meta-analysis (reasons included not reporting a measure of variance at follow-up and not disease specific), reported no difference between groups for psychological symptoms (Maisiak, Austin & Heck, 1996), weight loss , physical activity O'Brien et al., 2018), healthcare utilisation O'Brien et al., 2018), or subjective improvement   (Table S2). A further three studies reported telephone-based interventions significantly improved psychological symptoms , physical activity , and recovery (Iles et al., 2011) compared to usual care (Table S2). One study also reported higher fear avoidance in the telephone-based intervention group compared to usual care . Two studies reported data on adverse events O'Brien et al., 2018), one stated adverse events were similar between groups  and the other reported none of the adverse events were associated with the intervention . Three other studies stated no study-related adverse events occurred but did not provide any data (Allen et al., 2010Gialanella et al., 2017) (Table S2).
Five other studies not included in meta-analysis (reasons included not reporting a measure of variance at follow-up, non-randomised study design and dissimilar intervention components), reported no difference between groups for psychological symptoms (Weinberger et al., 1989), weight loss , physical activity (Li et al., 2017;Williams et al., 2018), healthcare utilisation , health-related quality of life (Mazzuca et al., 1997), subjective improvement (Rutledge et al., 2018;Williams et al., 2018), and fear avoidance   (Table S2). One study also reported telephone plus face-to-face interventions significantly improved physical activity and fear avoidance compared to usual care (Damush et al., 2003) (Table S2).

Sensitivity analyses
We were unable to explore the influence of overall high risk of bias on pooled treatment effects as all studies were rated as high risk of bias (rated high risk of bias for one or more key domains). Sensitivity analyses to explore the impact of study size on our treatment effects are presented in Table 2. When excluding small trials, there was no longer an intervention effect of telephone-based interventions on pain intensity (Fig. S8), or disability (Fig. S3) compared to usual care. For psychological symptoms intervention effect was not changed by the removal of studies with small sample size (<100) (Fig. S9).

Principle results
The trials in this review all used telephone-based interventions as the primary delivery method. All intervention content focused on self-management principles and providing education. Many studies included additional components such as educational materials and/or face-to-face interactions (21 of 23). We found moderate-quality evidence, as assessed using GRADE criteria (Guyatt et al., 2011), that telephone-based interventions (with educational materials) improves pain intensity, disability, and weight loss but not psychological symptoms compared to usual care. We also found high-quality evidence (Guyatt et al., 2011) that telephone-based interventions (with educational materials) improve self-efficacy compared to usual care. There was moderate-quality evidence (Guyatt et al., 2011) that there is no difference between telephone plus face-to-face interventions and usual care or face-to-face interventions alone in improving pain intensity, disability or psychological symptoms. There were limited studies that assessed the effects of telephone interventions on physical activity, health care or medication use or supporting treatment adherence.
Two previous systematic reviews of telehealth care for musculoskeletal conditions have been published (Cottrell et al., 2017a;Dario et al., 2017). A total of 10 of the studies on osteoarthritis and five on spinal pain were not included in these previous reviews. Cottrell et al. (2017a) found positive effects for real-time telephone or videoconferencing delivery of rehabilitation, mostly post-surgical (arthroplasty) rehabilitation for patients with various musculoskeletal conditions including back pain, neck pain, rheumatoid arthritis, and osteoarthritis. Our review considered a more homogenous patient group, the two most burdensome musculoskeletal conditions (osteoarthritis and spinal pain) and we excluded post-surgical rehabilitation. Our results, however, conflict with Dario et al. (2017) who included studies of any telehealth intervention for patients with non-specific low back pain and found no difference between these interventions and minimalist interventions (i.e. health or non-health-related information). The difference in results might be explained by our definition of a telehealth intervention, which was required to involve verbal telecommunication with direct patient-provider contact (i.e. telephone or videoconferencing), whereas Dario et al's. definition of telehealth interventions included asynchronous methods such as email, e-community or web-based content. It is possible that the verbal person-to-person support via telephone explained the positive effects found in our review.
Our results show that telephone-based models of care improve patients outcomes, compared to usual care. Other evidence suggests telephone-based models of care are preferred by patients with chronic musculoskeletal conditions Cottrell et al., 2017b). Despite this, telephone-based services are not widely available for patients with osteoarthritis or spinal pain. For other chronic conditions and health behaviours, such as diet and physical activity, telephone-based models have received significant investment based on additional evidence, including economic feasibility (Graves et al., 2009). While our review suggests that telephone-based interventions for people with osteoarthritis or spinal pain warrants consideration, there are unknown factors, for example, what the intervention should focus on, for example, weight or physical activity, that could impact the success of this in clinical practice. While our meta-analysis findings regarding self-efficacy point towards a possible mechanism in telephone-based interventions supporting capacity for self-management through self-efficacy, this and other possible mechanisms need to be tested formally.
The telephone is often used in clinical practice as an adjunct to face-to-face care (McLean et al., 2013). This is thought to allow more efficient integration of multidisciplinary roles and complex aspects of care (McLean et al., 2013;Dinesen et al., 2016). For osteoarthritis and spinal pain, our review suggests that for patients who already receive comprehensive face-to-face care additional telephone-based support does not improve pain and disability. We also found that telephone and face-to-face care was not better than usual care. However, we contend that the interventions in this meta-analysis are not comprehensive disease specific models, relative to other interventions we included. Overall, our findings do not support the additional use of telephone to clinical care that provides comprehensive management.
Usual care for many patients with osteoarthritis and spinal pain does not typically align with recommendations in clinical practice guidelines (Williams et al., 2010;Hunter, 2011;Brand et al., 2014). In light of the global disability burden of osteoarthritis and spinal pain, developing effective ways to deliver good quality care to the many patients with these conditions is an important future direction. Our results show that telephone services may be one way to provide remotely delivered care to people who cannot access it, or those who may otherwise receive suboptimal usual care. However, as usual care is often not evidence-based, arguably it may be an inferior control comparison to inform health policy and implementation (Williams et al., 2010;Hunter, 2011;Brand et al., 2014). One important research direction that would inform whether telephone-based models should be supported widely for musculoskeletal conditions is to understand if telephone only models are equivalent to good quality face-to-face care. We identified only one study assessing this comparison (Odole & Ojo, 2013). While the results of the study suggest equivalent effectiveness, more trials are needed to validate this finding, which also uses appropriate research designs (i.e. inferiority designs).

Limitations
The majority of studies included in the meta-analysis focused on patients with osteoarthritis (16 of 23), so caution needs to be taken when generalising overall results to the management of patients with spinal pain. However, when reporting the findings separately for the two trials of spinal pain, the intervention effects for pain intensity and disability remained the same. All included studies were telephone-based interventions so we were unable to evaluate the effect of other verbal real-time telehealth interventions (i.e. videoconferencing interventions) as planned. Another limitation is inconsistent outcome reporting across the included studies. Around a third of included studies reported data that could not be synthesised in the meta-analysis for the core outcomes recommended to be used in clinical trials (e.g. OMERACT-OARSI, IMMPACT) (Dworkin et al., 2005); pain intensity (4/15), disability (5/19), psychological symptoms (3/12), and self-efficacy (1/3). Furthermore, only two included studies reported they collected data to assess cost-effectiveness. Future trials examining the effect of telephone-based interventions should undertake cost-effectiveness analyses to determine whether such interventions can improve patients' outcomes at lower costs than usual care. Finally, in terms of GRADE, the overall quality of evidence was assessed as moderate for all meta-analyses but one (self-efficacy; high-quality of evidence). Although the effect we estimated is likely to be robust, there is a possibility that future high-quality research may change the effect estimates.

CONCLUSIONS
Our review is the first to comprehensively synthesise evidence on telephone-based interventions for osteoarthritis and spinal, which are the most common musculoskeletal conditions and the leading causes of disability worldwide. There is moderate-quality evidence that telephone-based interventions, compared to usual care, are effective for pain and disability for osteoarthritis and spinal patients collectively. Telephone-based services offer the potential to support osteoarthritis and spinal pain patients to access better quality care.