All analyses were conducted using data from previously published studies. Therefore, no ethical approval or patient consent was required.

This review was conducted in accordance with the PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analyses) statement [20] and its associated checklist (Multimedia Appendix 1). The study was registered with PROSPERO (CRD42023484819). This systematic review was executed in strict alignment with the PROSPERO protocol, maintaining complete consistency in search strategies, inclusion and exclusion criteria, outcome definitions, and statistical approaches as originally registered.

A literature search was performed until October 25, 2023, using the following databases: PubMed, Embase, Cochrane Library, Web of Science, EBSCO, and SPORTDiscus. The keywords used in the search were “osteoarthritis,” “exercise therapy,” “Web-based Intervention,” and “Self-Management.” Boolean operators “AND” and “OR” were used to combine keywords for a broader systematic search. Details of the search are presented in Multimedia Appendix 2.

Two independent reviewers (JL and JY) retrieved records from 6 databases and imported them into EndNote X9 (Clarivate Analytics). Duplicate records were systematically identified and removed using EndNote X9’s deduplication function, followed by manual verification to ensure accuracy. Following duplicate removal, a manual field-based screening was conducted in EndNote X9 to exclude records violating predefined inclusion criteria. This involved filtering the “Reference Type” field to exclude conference abstracts (eg, using “Conference” as a keyword) and searching titles for keywords to exclude systematic reviews, meta-analyses, trial protocols, questionnaire studies, and pilot studies (eg, “review,” “meta-analysis,” “protocol,” “questionnaire,” and “pilot”).

The study selection process was conducted in two stages. First, titles and abstracts were screened independently by 2 reviewers (JL and JY) to exclude irrelevant studies. Second, the full texts of the remaining potentially relevant studies were reviewed independently by the same reviewers (JL and JY) to determine eligibility. Any disagreements during both stages were resolved by a third reviewer (YY). Finally, backward searches of reference lists from included studies and relevant systematic reviews were performed by the initial reviewers (JL and JY) to ensure comprehensiveness. During the initial screening, no restrictions were placed on the specific pain measurement tools used. The primary criterion for inclusion was the reporting of pain measures, physical function measures, or both, regardless of the specific scale used Textbox 1.

The primary outcomes evaluated were pain intensity and physical function. Pain intensity was measured using the numeric rating scale (NRS), a widely used tool for pain assessment. The NRS is an 11-point, unidimensional, horizontal scale where “0” indicates “no pain” and “10” indicates “worst possible pain.” Patients rate their pain by choosing a number that best reflects their current pain level. The NRS is easy to administer and score and is sensitive to changes in pain levels, making it popular in both clinical and research settings [21,22]. Physical function was measured using the Western Ontario and McMaster Universities Osteoarthritis Index (WOMAC), a standardized, self-administered questionnaire widely used to assess the impact of OA on patients’ lives. The WOMAC consists of 24 items divided into three subscales: pain (5 items), stiffness (2 items), and physical function (17 items). Each item is scored on a 5-point Likert scale, with higher scores indicating greater severity of symptoms and functional limitations [23,24].

Two reviewers (JL and JY) independently extracted data using a standardized form, including study characteristics (design and sample size), participant demographics, intervention and control details, outcome measures (mean and SD for the NRS and WOMAC immediately after the intervention and at follow-up), and discrepancies were resolved through discussion. Corresponding authors were contacted for missing data. Upon reviewing the data, we found that most studies included in our analysis used the NRS to measure pain. However, one study [25] reported pain using the WOMAC pain score. To maintain consistency in our meta-analysis, we decided to include only the NRS data. This decision was made to ensure uniformity in the quantitative synthesis of pain outcomes across studies. The study using the WOMAC pain score was still included in the overall review but was excluded from the meta-analysis of pain outcomes.

Quality assessments were performed with the Physiotherapy Evidence Database (PEDro) scale [26], which is based on the Delphi List criteria and is considered valid and reliable to evaluate the methodological quality of included studies. All included trial reports were checked in the PEDro database to confirm their PEDro scale score. Considering that criterion 1 was not used to calculate the score, the sum of the other criteria could have a maximum of 10 points. Trials with a score ≥6 points were classified as “good,” whereas those with a score ≤5 points were classified as “poor” and excluded from the analysis. If a PEDro score was not available in the database, 2 authors independently assessed study quality using the PEDro scale and its associated notes. Any disagreements were resolved through discussion until a consensus was reached [27].

Based on the availability of the reported data, the primary outcomes selected for meta-analysis were the NRS score and the WOMAC score. These outcomes were analyzed immediately after the intervention and at follow-up). As the data were continuous, the mean and SD were used to calculate the mean difference (MD) with a 95% CI. The primary meta-analysis was performed using Review Manager 5.4 software (The Nordic Cochrane Centre). Fixed-effects statistical models were applied as the default in the meta-analysis. If heterogeneity was found, random-effects models were applied. The significance level was set at P<.05, and results were presented using forest plots [28-30].

Figure 1 depicts the study selection process. Initially, 357 publications were retrieved through electronic database searches. After the removal of 106 duplicates, 251 articles were assessed using manual field-based screening, leading to the exclusion of 122 articles for various reasons, including conference papers, systematic reviews or meta-analyses, trial protocols or designs, pilot studies, and questionnaire-based studies. Based on predefined inclusion and exclusion criteria, the titles and abstracts of 129 articles were further evaluated. Among these, 116 articles were excluded as they did not investigate or compare the combination of digital exercise therapy with conventional treatment in patients with OA. After a detailed review of the remaining 13 full-text articles, 4 additional articles were excluded for the following reasons: ineligible outcome measures, irrelevant outcomes, or an ineligible study design. Consequently, only 9 articles were deemed suitable for further review and analysis. This meticulous selection process ensured that only relevant and high-quality studies were included in the subsequent analysis, thereby enhancing the validity of the findings [18,25,31-37].

All 9 randomized controlled trials (RCTs) included in the meta-analysis were written in English between 2016 and 2023 [18,25,31-37]. A total of 1604 participants were enrolled, with 798 (49.7%) in the digital exercise therapy group and 806 (50.2%) in the conventional treatment group. Detailed characteristics of the eligible studies are presented in Table 1.

All 9 studies involved adult participants, with 4 studies including participants aged ≥45 years [18,31,33,36], 3 studies including participants aged ≥50 years [25,32,35], 1 study including participants aged >18 years [34], and 1 study including participants aged 40-80 years [37]. Overall, 7 studies focused specifically on knee OA [18,25,31-33,35,36], 1 study focused on knee OA or hip OA [37], and 1 study focused on hand OA [34]. The inclusion of participants from different age groups and the focus on various types of OA contributed to the diversity and representativeness of the included studies.

Table 2 summarizes the 9 included trials, each featuring unique digital exercise therapy protocols. These studies varied in intervention methods, evaluation indicators, and effects. The digital exercise therapy groups used online exercises, while the conventional groups in 6 studies lacked online support [25,32,34-37], and 3 did not include an exercise intervention [18,31,33]. Diverse therapies were used: Nelligan et al [33] offered a 24-week program via a website and SMS text messages; Rodríguez Sánchez-Laulhé et al [34] introduced a 12-week app-based program for hand OA; Hinman et al [31] added physiotherapist support into a telephone service; Bennell et al [18,32] studied internet-based yoga and telephone coaching for knee OA; Allen et al [25] developed an internet-based exercise training program with tailored exercises and progress tracking; Baker et al [35] created the BOOST telephone-linked communication program focusing on exercise behavior; Gohir et al [36] developed a 6-week app program for leg strength and balance; and Kloek et al [37] combined online and face-to-face sessions in the e-Exercise program (refer to Table 1 and Table 2 for further details).

Table 3 presents the effect sizes of digital exercise therapy interventions for both immediate and follow-up findings, categorized by digital resource type (phone interventions, app-based interventions, and website-based interventions). A total of 4 studies [31-33,35] used phone interventions (SMS text messages), showing mean effect sizes of –0.44 (95% CI –0.75 to –0.14) for NRS pain and –0.24 (95% CI –0.39 to –0.09) for WOMAC physical function immediately after the intervention. During follow-up, the mean effect sizes were –0.11 (95% CI –0.43 to 0.22) for NRS pain and –0.29 (95% CI –0.62 to 0.03) for WOMAC physical function. Two studies [18,25] used websites, with mean effect sizes of –0.29 (95% CI –0.57 to –0.01) for NRS pain and –0.13 (95% CI –0.25 to –0.01) for WOMAC physical function immediately after the intervention. During follow-up, the mean effect sizes were –0.01 (95% CI –0.09 to 0.08) for NRS pain and –0.11 (95% CI –0.37 to 0.15) for WOMAC physical function. Three studies [34,36,37] used apps, revealing mean effect sizes of –0.43 (95% CI –0.63 to –0.23) for NRS pain and –0.47 (95% CI –0.87 to –0.07) for WOMAC physical function immediately after the intervention. During follow-up, the mean effect size was 0.04 (95% CI –0.30 to 0.38) for NRS pain.

The PEDro scale assesses the methodological quality of RCTs, with higher scores indicating better quality. All 9 studies included in the review scored high on the PEDro scales, with scores ranging from 7 to 9, indicating sufficient quality among the included studies (Table 4). An identified risk of bias across all studies was the lack of blinding for participants or therapists during the intervention, along with inadequate blinding of outcome assessors in 4 studies. Table 4 provides detailed information on PEDro scores and specific aspects of blinding for each study.

This systematic review and meta-analysis aimed to evaluate the effectiveness of digital exercise therapy in improving pain and physical function among patients with OA. The meta-analysis included 9 studies, involving a total of 1604 patients with OA. These studies exhibited high methodological quality, as evidenced by PEDro scores exceeding 7. The results indicate that digital exercise therapy can significantly alleviate pain compared with conventional treatment immediately after the intervention, with phone-, app-, and website-based interventions demonstrating consistent pain reduction. Similarly, improvements in WOMAC physical function were observed across all modalities. Subgroup analysis revealed that the duration of the intervention did not influence the effects on pain outcomes. However, follow-up results showed no statistically significant differences in pain or physical function scores between groups, suggesting limited sustainability of benefits. Our findings align with a previous review of digital therapeutic tools for exercise programs in knee OA [38], which reported significant pain improvements in 10 of 11 studies but did not address long-term outcomes. This underscores the potential of digital modalities for short-term symptom management while highlighting the need for strategies to prolong therapeutic effects.

The loss of effect at follow-up is an important finding that warrants further exploration. Based on our analysis of the included studies, we found that changes in pain and physical function scores during follow-up were not uniform across studies. Specifically, some studies reported improvements in pain and physical function in the digital exercise therapy group compared with the control group, while others showed no significant changes or even slight worsening in some cases. Our analysis suggests that this phenomenon is multifactorial and cannot be solely attributed to increased pain in the digital exercise therapy group or decreased pain in the control group. Instead, it appears to be influenced by a combination of factors, including patient adherence to the digital exercise program [31], the duration and intensity of the intervention [25], and the natural course of OA [32,37]. Future studies should aim to investigate these factors more thoroughly to better understand the sustainability of digital exercise therapy benefits over time.

This variability suggests that the loss of effect at follow-up may be influenced by multiple factors. One possible explanation is that digital interventions primarily influence cognitive and behavioral aspects of pain management without directly addressing underlying pathological changes in OA. This hypothesis is supported by studies demonstrating that digital interventions can improve pain perception through cognitive manipulation but may not directly impact tissue damage [39,40]. Future research should focus on integrating digital exercise therapy with other therapeutic approaches, such as pharmacological treatments or lifestyle modifications, to enhance long-term efficacy. Another potential explanation for the loss of effect at follow-up is the lack of sustained engagement with digital exercise programs. Adherence to home-based exercises is a critical factor influencing long-term outcomes, and digital interventions may face challenges in maintaining participant motivation and compliance over extended periods [41-43]. Future studies should incorporate assessments of participant adherence and explore strategies to enhance long-term engagement with digital exercise therapy. Additionally, standardizing intervention protocols and incorporating personalized elements may improve the effectiveness and sustainability of digital exercise therapy.

Despite the limitations in long-term efficacy observed in this study, digital exercise therapy remains a promising approach for managing OA, particularly in settings where access to conventional rehabilitation services is limited. The COVID-19 pandemic has highlighted the potential of telehealth interventions to provide accessible and convenient care for patients with chronic conditions [13-15]. Digital exercise therapy offers flexibility and convenience, allowing patients to engage in rehabilitation from home, thereby reducing barriers related to time, transportation, and geographical constraints [44,45]. Integrating digital exercise therapy into clinical practice may enhance treatment accessibility and efficiency, particularly for patients in rural or remote areas or those with limited mobility.

However, several practical considerations must be addressed to optimize the implementation of digital exercise therapy. First, the safety and cost-effectiveness of digital interventions need to be better understood. While no increase in harm was reported in the included studies, comprehensive safety assessments and cost-effectiveness analyses are necessary to support widespread adoption [46-49]. Additionally, future research should focus on developing standardized outcome measures and incorporating multidimensional factors, such as mental health status, social support, and personal preferences, to provide a comprehensive evaluation of digital exercise therapy’s potential benefits and limitations.

This review has several limitations. The considerable heterogeneity in digital resource types and research designs limits the generalizability of the findings. Future studies should aim to standardize intervention protocols and outcome measures to facilitate more consistent comparisons. Additionally, the lack of long-term follow-up data and the absence of blinding methods in the included studies pose challenges in interpreting the sustained effects and safety of digital exercise therapy. Future research should prioritize long-term follow-up assessments and methodological improvements to reduce bias.

Another limitation is the relatively homogeneous sample of participants, primarily comprising middle-aged individuals from higher-income countries. Future studies should broaden the geographical and demographic scope to enhance the external validity of the findings. Furthermore, the potential impact of socioeconomic status, educational background, and digital literacy on the effectiveness of digital exercise therapy should be explored in diverse populations.

Digital exercise therapy has demonstrated certain therapeutic effects in alleviating pain and improving physical function in patients with OA, and it can serve as an auxiliary means in the rehabilitation treatment of arthritis. Due to the limited sample size in this study, more high-quality, large-sample RCTs are needed in future research to further explore the efficacy of digital exercise therapy for OA.