Effects of a Self-directed Web-Based Strengthening Exercise and Physical Activity Program Supported by Automated Text Messages for People With Knee Osteoarthritis: A Randomized Clinical Trial

Key Points

Question  How does adding a 24-week self-directed strengthening exercise regimen and physical activity guidance supported by automated behavior-change text messages to web-based osteoarthritis (OA) information affect pain and function in people with knee OA?

Findings  This randomized clinical trial of 206 adults with a clinical diagnosis of knee OA found that a web-based exercise intervention supported by text messaging improved knee pain and function at 24 weeks compared with web-based information alone.

Meaning  This freely available digital intervention is a useful and effective option for improving access to recommended OA exercise and/or supporting clinicians in providing exercise management to people with knee OA at scale across the population.

Importance  Exercise therapies are advocated in osteoarthritis (OA) clinical guidelines. However, challenges to accessing exercise may be limiting widespread uptake.

Objective  To evaluate the effects of a self-directed web-based strengthening exercise and physical activity program supported by automated behavior-change text messages on knee pain and function for people with knee OA.

Design, Setting, and Participants  The participant-blinded and assessor-blinded randomized clinical trial enrolled 206 people who met clinical criteria for knee OA in communities across Australia from July 2018 to August 2019, with follow-up taking place at 24 weeks.

Interventions  The control group was given access to a custom-built website with information on OA and the importance of exercise and physical activity. The intervention group was given access to the same information plus a prescription for a 24-week self-directed strengthening regimen and guidance to increase physical activity, supported by automated behavior-change text messages encouraging exercise adherence.

Main Outcomes and Measures  Primary outcomes were change in overall knee pain (numeric rating scale, 0-10) and difficulty with physical function (Western Ontario and McMaster Universities Osteoarthritis Index, 0-68) over 24 weeks. Secondary outcomes were another knee pain measure, sport and recreation function, quality of life, physical activity, self-efficacy, overall improvement, and treatment satisfaction.

Results  Of 206 participants, 180 (87%; mean [SD] age, 60 [8.4] years; 109 [61%] women) completed both 24-week primary outcomes. The intervention group showed greater improvements in overall knee pain (mean difference, 1.6 units; 95% CI, 0.9-2.2 units; P < .001) and physical function (mean difference, 5.2 units; 95% CI, 1.9-8.5 units; P = .002) compared with the control. There was evidence of differences in the proportion of participants exceeding the minimal clinically important improvement in pain (intervention group, 72.1%, vs control, 42.0%; risk difference, 0.30 [95% CI, 0.16-0.44]; P <. 001) and function (intervention group, 68%, vs control, 40.8%; risk difference, 0.27 [95% CI, 0.13-0.41]; P < .001) favoring the intervention. Between-group differences for all secondary outcomes favored the intervention except for physical activity, self-efficacy for function, and self-efficacy for exercise, for which there was no evidence of differences.

Conclusions and Relevance  This randomized clinical trial found that a self-directed web-based strengthening exercise regimen and physical activity guidance supported by automated behavior-change text messages to encourage exercise adherence improved knee pain and function at 24 weeks. This unsupervised, free-to-access digital intervention is an effective option to improve patient access to recommended OA exercise and/or to support clinicians in providing exercise management for people with knee OA at scale across the population.

Trial Registration  Australian New Zealand Clinical Trials Registry Identifier: ACTRN12618001167257

Knee osteoarthritis (OA) is highly prevalent and has no known cure.¹ Recommended first-line management includes education and strengthening exercise.^2-6 However, exercise remains underutilized,^7-10 partly because of limited access to appropriately trained health professionals to prescribe and support exercise.^11-14 Access challenges will continue to worsen given the aging population and rising obesity, and the burden of knee OA is forecast to overwhelm health care systems by 2030.¹⁵ Therefore, there is an urgent need to innovate how exercise is prescribed.¹⁶

Digital technologies may be a feasible solution given that people with knee OA are increasingly seeking information about their condition from internet sources.¹⁷ This information is of variable quality, inconsistently evidence based, and potentially difficult for the general public to comprehend.¹⁸ Some high-quality online OA self-management platforms are freely available through consumer organizations (eg, MyJointPain.org.au). However, they provide general exercise information only. This is problematic as a lack of clear instructions prevents people with knee OA from exercising.^13,19 Adherence to exercise is typically poor in people with knee OA,^20,21 which may explain why the clinical benefits of exercise are not sustained.^21-23 Barriers include participation costs and lack of support.¹³ Cell phone text messages positively influence chronic disease self-management and physical activity behavior.^24,25 Thus, text messages may help support exercise participation without the need for health professional involvement.

We developed a 24-week self-directed intervention consisting of a website and automated text messages. The website contains educational information on OA and exercise, plus guidance to increase physical activity and a prescription for a structured 24-week self-directed strengthening regimen.²⁶ The text messaging system was developed using behavior change theory²⁷ to support home exercise and has been shown to increase physiotherapist-prescribed home exercise adherence.²⁸ We hypothesized that use of the website and cell phone–based intervention would lead to greater improvements in knee pain and function in people with knee OA compared with a control website offering only educational information on OA and the importance of exercise and physical activity. The control website content was similar to content currently available through high-quality online platforms.

We conducted a parallel, 2-arm, superiority randomized clinical trial that was prospectively registered in the Australian New Zealand Clinical Trials Registry (ACTRN12618001167257). Reporting aligned with the Consolidated Standards of Reporting Trials (CONSORT) guideline,²⁹ relevant Consolidated Standards of Reporting Trials Extension (CONSORT Extension) guideline,^30-32 and the template for intervention description and replication (TIDieR) guideline.³³ The study protocol is available in Supplement 1 and has been published elsewhere.²⁶ Approval was obtained from The University of Melbourne Human Research Ethics Committee (No. 1851085). Digital informed consent was obtained using an online form prior to baseline assessments.

Study participants were recruited nationwide in Australia from July 2018 to August 2019 via online advertisements and the Centre for Health Exercise and Sports Medicine’s volunteer database. Screening was via an electronic survey developed using REDCap software (Vanderbilt University), and eligibility was confirmed by telephone. Inclusion criteria were: (1) OA clinical criteria (age ≥45 years, activity-related knee pain, and morning knee stiffness ≤30 minutes)⁵; (2) knee pain on most days for 3 months or more; (3) average overall knee pain severity of 4 or greater on an 11-point numeric rating scale (NRS) during the previous week; (4) own a cell phone with text messaging; (5) home internet access; and (6) ability to consent, participate, and complete assessments. Exclusion criteria are presented in eTable 1 in Supplement 2.

Participants were randomized with a 1:1 ratio. Computer-generated randomization was prepared by the biostatistician (J.K.) in permuted blocks of sizes 6 to 12. To ensure concealment, the randomization schedule was accessed via a password-protected computer program by a researcher not involved in participant screening, recruitment scheduling, or assessment (S.S. or P.C.). Limited disclosure was used to blind participants who were also assessors, as all outcomes were participant reported. Participants were informed that the study was investigating a range of digital resources (eg, computer, cell phone) to promote knee pain self-management that might include exercise and email or text messaging but were not given specific details of either the intervention or the control groups nor the hypothesis under investigation. The biostatisticians (J.K. and S.J.C.C.) were blinded.

A detailed description of the intervention (My Knee Exercise website and automated text messages) and its development have been published.^26,27 Intervention design features are summarized in eTable 2 in Supplement 2. The website is accessed via https://mykneeexercise.org.au/, and the text messaging system, My Exercise Messages, was adapted for release as an app and is downloadable via app stores. Access is free of charge for both.

The study coordinator (R.K.N.) sent an email to participants in the intervention group describing the intervention (access to the My Knee Exercise website containing a prescription for a 24-week knee strengthening regimen supported by periodic text messages), website access details (the web address and a unique login username and password), and a request to commence the exercise regimen within 1 week. The website was divided into 4 sections (eTable 3 in Supplement 2).

All participants in the intervention group received the same standardized website and were permitted to access it at will for 24 weeks. Participant details were added to the text messaging system, triggering the 24-week automated messages designed to encourage adherence to the prescribed strengthening exercise. Descriptions of all text message types and their frequencies during the 24 weeks are provided in eTable 4 in Supplement 2. Messages were personalized with first names. No changes were made to the website or text messaging system during the trial.

The study coordinator (R.K.N.) sent an email to the participants in the control group describing the intervention (access to the My Knee Education website containing information on knee pain, knee OA, and the importance of exercise and physical activity), website access details (the web address and a unique login username and password), and a request to access the website within 1 week. At enrollment, control participants received a text message prompting them to access the website.

The control website contained the same textual information as the My Knee Education section of the intervention website, but references to the specific strengthening exercise regimen and physical activity guidance were removed. Only general recommendations were retained, similar to those freely accessible from OA web-based consumer resources (eg, generic information on the importance of strengthening exercises and meeting physical activity guidelines).

Outcomes were participant reported and collected via electronic surveys at baseline and at 24 weeks. The 2 primary outcomes were reliable and valid (1) pain and (2) physical function measures recommended for knee OA clinical trials.^34-36 Average overall knee pain in the past week was assessed using an 11-point NRS with terminal descriptors of no pain (score 0) and extreme pain (score 10).³⁷ Limitations with physical functioning during the past week were measured by the physical function subscale of the Western Ontario and McMaster Universities Osteoarthritis Index (WOMAC, Likert version 3.1),³⁵ extracted from the Knee Injury and Osteoarthritis Outcome Score (KOOS),³⁸ with a total score ranging from 0 (no dysfunction) to 68 (maximum dysfunction).

Secondary outcomes were (1) KOOS pain, function in sport and recreation, and knee-related quality-of-life subscales³⁸; (2) Assessment of Quality of Life (version AQoL-6D)³⁹; (3) Physical Activity Scale for the Elderly⁴⁰ (PASE); (4) Arthritis Self-Efficacy Scale (ASES)⁴¹ pain and physical function subscales; (5) Self-Efficacy for Exercise scale (SEE)⁴²; (6) participant-perceived change overall since baseline (24-week follow-up only)⁴³; and (7) overall satisfaction (24-week follow-up only).

Adverse events were participant-reported at 24 weeks, defined as any problem believed to be caused by the study intervention requiring treatment or medication and/or interfering with function for 2 days or more. Custom-developed surveys collected cointervention use (pain medications and other knee OA treatments) at 24 weeks (retrospective recall over previous 6 months). Exercise adherence was assessed by the number of days that knee exercises were performed during the previous week and the Exercise Adherence Rating Scale section B⁴⁴ at 24 weeks. A range of process measures was collected (eTable 5 in Supplement 2). Another article reports qualitative evaluation.⁴⁵

We aimed to detect a conservative effect size of 0.40 (small to moderate between-group difference)⁴⁶ for the primary outcomes. We believed self-directed unsupervised exercise may have smaller effects than the 0.49 for pain and 0.52 for physical function observed with land-based supervised exercise for knee OA.²³ To obtain 80% power, a 2-sided significance level of .05, and a correlation between baseline and follow-up measurements of 0.35 with a 15% loss to follow-up,^47,48 103 participants were required per arm. Assuming between-participant SDs of 2.3 for pain and 11.7 for WOMAC function⁴⁸ and a pre-post correlation of 0.35, this sample allowed greater than 99% power to detect a minimal clinically important difference (MCID) in pain of 1.8 units⁴⁹ and 95% power to detect a MCID in function of 6 units.⁵⁰

Analyses were performed by biostatisticians (S.J.C.C. and J.K.) using Stata, release 16 (StataCorp LLC), and intention-to-treat with all available data from all randomized participants using their randomized group allocation. Baseline characteristics of participants who provided both primary outcomes and those who did not were compared using t tests or χ² tests. P values were 2-tailed and statistical significance was defined as P < .05. Missing outcomes were imputed using chained equations with predictive mean matching and 5 nearest neighbors for continuous outcomes, and logistic regression imputation models for binary improvement outcomes. Imputation models for continuous outcomes at 24 weeks included all primary and secondary outcomes at both baseline and 24 weeks, along with age, sex, body mass index (calculated as weight in kilograms divided by height in meters squared), education level, geographical location, employment status, duration of symptoms, and laterality. The imputation model for the binary variable of global improvement was similar except that all secondary outcome variables at 24 weeks were omitted owing to the tendency for perfect prediction. Data were imputed for each treatment group separately. Estimates from 15 imputed data sets were combined using Rubin rules.⁵¹

For continuous outcomes, the mean (95% CI) difference in change (baseline minus follow-up) between groups was estimated using linear regression models, adjusted for baseline scores. The proportion of participants with self-perceived improvement overall and the magnitude of improvements that met or exceeded MCIDs for NRS pain and WOMAC function were compared between groups using logistic regression models, with results presented as risk ratios and risk differences (both obtained using marginal standardization).⁵² Validity of model assumptions and imputed data sets was assessed using standard diagnostic plots. Analyses were repeated using complete-case data.

We randomized 206 participants, of which 180 (87%; mean [SD] age, 60 [8.4] years; 109 [61%] women) completed both primary outcome measures at 24 weeks (Figure). Groups were similar at baseline (Table 1). Participants who did not complete both primary outcomes at 24 weeks reported greater past use of injections, nonsteroidal anti-inflammatory drugs, and oral opioids to manage knee pain and higher rates of foot problems at baseline (eTable 6 in Supplement 2). Website and text message usage are outlined in eTable 7 in Supplement 2. Fourteen (8%) participants reported no website access in the first month (intervention: n = 3 [3%]; control: n = 11 [13%]) and 85 (49%) participants reported no website access in the past month (intervention: n = 35 [39%]; control: n = 50 [60%]) in the past month. The mean (SD) number of website logins per participant was 6.0 (4.9) in the intervention group and 3.2 (2.6) in the control group. In the intervention group, the mean (SD) number of text messages sent to each participant during the 24 weeks was 60.0 (7.5) and the average participant reply rate was 73% (34%).

At 24 weeks, the number of participant-reported knee exercise sessions in the previous week was similar in both groups (eTable 8 in Supplement 2). There was evidence of between-group differences in adherence to knee exercise measured by the Exercise Adherence Rating Scale (mean difference, 2.6 units; 95% CI, 0.8-4.4 units; P = .005), favoring intervention. Few adverse events were reported, and none were serious (eTable 9 in Supplement 2). More intervention participants experienced knee pain than control participants (n = 8 [9.6%] vs n = 1 [1.3%]; P = .019). Use of pain medications and other treatments for the knee during the 24 weeks was similar across groups, except more control participants used massage, heat or cold, and topical anti-inflammatories (eTable 9 in Supplement 2).

There was evidence of greater improvements in overall pain (mean difference, 1.6 units; 95% CI, 0.9-2.2 units; P < .001) and in WOMAC function (mean difference, 5.2 units; 95% CI, 1.9-8.5 units; P = .002) favoring the intervention (Tables 2 and 3). More participants in the intervention group reached MCIDs in pain and in function than in the control group (Table 4). Analyses using complete case data produced similar results (eTables 10 and 11 in Supplement 2).

There was evidence of between-group differences favoring the intervention in most secondary outcomes—all 3 KOOS subscales (pain, sports/recreation, and quality of life), AQoL, ASES pain, participant change since baseline, and overall satisfaction. Changes in PASE, ASES function, and SEE were similar in both groups (Table 3). More participants in the intervention group reported overall improvement than in the control group (Table 4). Analyses using complete case data produced similar results (eTable 10 and eTable 11 in Supplement 2).

This randomized clinical trial provides robust evidence of the effectiveness of adding a 24-week self-directed strengthening exercise regimen and physical activity guidance supported by automated text messages to web-based information similar to what is currently available online through reputable OA consumer organizations. We found significant between-group differences in pain of 1.6 units (95% CI, 0.9-2.2 units) and function of 5.2 units (95% CI, 1.9-8.5 units) favoring the intervention. While these were just below the MCIDs for pain (1.8 units) and function (6 units), the 95% CIs include these differences within the plausible ranges. Additionally, substantially more participants reached the MCIDs in the intervention group than in the control group, while within-group changes in pain and function exceeded MCIDs in the intervention group but not in the control group. Furthermore, effect sizes (pain, 0.68 and function, 0.39) favoring the intervention group were similar to those observed with land-based therapist-supervised exercise.²³

Taken together, the findings of the present study demonstrate that this easily scalable, unsupervised, free-to-access intervention is effective and may be clinically relevant on a population level. Further supporting its effectiveness are the findings of beneficial effects on secondary outcomes. An additional knee pain measure, function in sport and recreation, 2 quality-of-life measures, pain self-efficacy (3 KOOS subscales, AQoL, and ASES pain), perceived improvement overall, and treatment satisfaction all showed the intervention’s significant benefit, although other measures (PASE, ASES function, and SEE) showed similar changes in both groups.

To our knowledge, only 2 other randomized clinical trials^53,54 have evaluated web-based interventions without health professional contact in OA. One trial evaluated a self-directed progressive lower-limb strength, flexibility, and walking program (Help My Knees)⁵³ in people with knee OA compared with a wait-list control group. Unlike the present study findings, no between-group differences were found with the primary outcome of total WOMAC score (combined measure of pain, stiffness, and function) or secondary outcomes of physical function and pain at 4 months. The other trial evaluated a 9-module physical activity program (Join2Move)⁵⁴ in adults with knee and/or hip OA compared with a wait-list control. Similar to the present study findings, there was evidence of between-group differences in physical function (KOOS; mean difference, 6.5 units; 95% CI, 1.8-11.2 units) and in proportion of participants reporting improvements (odds ratio, 10.7; 95% CI, 4.3-26.4), favoring intervention at 3 months. In contrast with the present study findings, there was no evidence of between-group differences in quality of life or function in sport and recreation at 3 months.

Conflicting findings across studies may partially be explained by low engagement levels with web-based programs. In the Help My Knees study,⁵³ 20% of participants did not access the program during the 4-month intervention. Similarly, only 55% of participants completed the second of 9 modules of Join2Move.⁵⁴ The present intervention had much higher levels of user engagement—participant-reported website access was 97% in the first month and 61% in the final month. These differences may be due to variation in the type and level of support provided. In Help My Knees and Join2Move, automated support and exercise reminders seemed to occur just once per week. In contrast, the present intervention used automated text messages (average, 2.5 messages sent per participant per week), designed to address common exercise barriers and facilitators in knee OA.¹³ Such communication has been shown to enhance adherence to physiotherapist-prescribed exercise in people with knee OA.²⁸ The text messages provided along with the web-based exercise program may have facilitated better engagement and enhanced its effectiveness. Furthermore, additional qualitative research⁴⁵ highlights that participants valued the text messages as a prompt to exercise, keeping them accountable to the program.

In the present study, 72% of intervention participants experienced clinically important improvements in pain and 68% in function, demonstrating that most participants experienced meaningful improvements in knee OA symptoms without needing health professional contact. Conversely, 30% of participants did not benefit from the unsupervised approach, suggesting that more intensive, personalized management may be required. This gap could be addressed through a stepped-care approach where people move through a hierarchy of evidence-based interventions based on outcomes.⁵⁵

Our free-to-access, unsupervised program could serve as an entry-level intervention, with participants who do not experience clinical benefits progressing to subsequent steps for more intensive, personalized management. Such an approach has the potential to better distribute limited health care resources and reduce demand for contact with health professionals, thus improving access for those requiring it. The present intervention may also be valuable in low-income to middle-income countries with large unmet needs for physiotherapy care.⁵⁶ Further research is required to evaluate the effectiveness of this program when stepped care and/or face-to-face care is incorporated.

There are several study strengths. The robust randomized clinical trial design with reliable and valid outcomes, participant blinding and assessor blinding, and excellent participant retention enhances internal validity. Generalizability is maximized by broad inclusion criteria, recruitment of participants nationwide, and no restrictions placed on cointervention and medication use. Another strength is the rigorous intervention design, which was evidence informed, incorporated behavior change theory, and included input from clinicians and people with knee OA.

Several limitations are acknowledged. There was potential for bias with participants self-selecting to volunteer for a study investigating different digital resources to support knee OA management, which may have led to inclusion of people with more favorable views of technology. Participants had completed high school or higher education and reported moderate to high self-efficacy at baseline, meaning the study findings may not generalize to people with lower levels of education or self-efficacy. At baseline, participants not completing 24-week outcomes used more analgesics and injections, although symptom severity was similar between groups; therefore, the intervention may not meet the needs of those who prefer pharmacological treatments. We did not include a long-term follow-up; therefore, whether the intervention effects are maintained beyond 24 weeks is unknown. Finally, because the comparator was web-based education, it remains unknown how the intervention compares with clinician-delivered interventions (1-on-1 or group based).

A web-based intervention of self-directed strengthening exercise and physical activity guidance supported by automated behavior-change text messages improved knee pain and function at 24 weeks. The benefits may be clinically relevant and indicate that this unsupervised, free-to-access intervention could be an effective option for improving patient access to recommended OA exercise and supporting clinicians in providing exercise management to people with knee OA at scale across the population.

Accepted for Publication: February 19, 2021.

Published Online: April 12, 2021. doi:10.1001/jamainternmed.2021.0991

Corresponding Author: Rachel K. Nelligan, PGCertPhysio, Centre for Health, Exercise and Sports Medicine, Department of Physiotherapy, School of Health Sciences, The University of Melbourne, Parkville, Victoria, 3010, Australia (rachel.nelligan@unimelb.edu.au).

Author Contributions: Ms Nelligan had full access to all of the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis. Ms Nelligan is a PhD candidate and this trial is contributing to her doctoral dissertation.

Concept and design: Nelligan, Hinman, Bennell.

Acquisition, analysis, or interpretation of data: All authors.

Drafting of the manuscript: Nelligan, Kasza, Crofts, Bennell.

Critical revision of the manuscript for important intellectual content: All authors.

Statistical analysis: Kasza, Crofts.

Obtained funding: Hinman, Bennell.

Administrative, technical, or material support: Nelligan.

Supervision: Hinman, Kasza, Bennell.

Conflict of Interest Disclosures: Ms Nelligan reported a PhD scholarship from the Australian Government Research Training Program during the conduct of the study and personal fees from The University of Melbourne outside the submitted work. Dr Bennell reported grants from the National Health and Medical Research Council (NHMRC) during the conduct of this study and consulting fees from Wolters Kluwer for production of UpToDate Knee OA Clinical Guidelines and grants from Medibank Private for osteoarthritis research outside the submitted work. Dr Hinman reported grants from the NHMRC during the conduct of the study and grants from Medibank Private and the NHMRC outside the submitted work. No other disclosures were reported.

Funding/Support: Funding was provided by the NHMRC (Grant No. 1091302). Drs Bennell and Hinman are supported by NHMRC Fellowships (No. 1058440 and No. 1154217, respectively). Ms Nelligan is supported by an Australian Government Research Training Program Scholarship.

Role of the Funder/Sponsor: The funder had no role in the design and conduct of the study; collection, management, analysis, and interpretation of the data; preparation, review, or approval of the manuscript; and decision to submit the manuscript for publication.

Data Sharing Statement: See Supplement 3.

Additional Contributions: We would like to acknowledge and thank Rob Spencer, a study participant enrolled in the intervention, for providing feedback on drafts of the manuscript; Sarah Schwartz, MPT, and Penny Campbell, BAppSci, research assistants at The University of Melbourne, for randomizing participants into this trial; Lou Atkins, PhD, senior teaching fellow and consultant at University College London, Centre for Behaviour Change, for her involvement in the development of the text messaging system; and Wen Wu, PhD, research fellow at The University of Melbourne, for assisting with organizing website analytics data. Written permission has been obtained from all contributors and no compensation was provided.