Study selection
A total of 14081 and an additional 13 articles were retrieved from databases. After removing duplicates, 4021 records were screened for the titles and abstracts. A total of 86 full-text articles were assessed for eligibility. Of these, 23 articles24-48 were included in the systematic review (see Fig. 1). List of studies excluded at the full-text screening stage with reasons is in Supplementary Table 1. Three articles35,47,48 were excluded for the meta-analysis due to nonstandard data reporting, resulting in 20 articles being included in the meta-analysis.
Characteristics of included studies
The characteristics of all included studies are described in Table 1. A total of 23 studies involved 3824 patients, of which 60% of the participants were female, were included in the systematic review. One study27 observed the effect in the female population only, and others recruited participants of both sexes. Two (9%) studies were conducted in Europe, five (22%) in North America, five (22%) in Asia, nine (39%) in Australia, and two (9%) in Africa. The age of included participants was 61 (SD 3.9) years. The study period varied between 4 and 96 weeks, with seven trials (30%) performing the telehealth-based intervention for less than 3 months, six (26%) between three to six months, and 9 (39%) more than 6 months. 13 (57%) studies received telehealth-supported exercise programs, four (17%) received a physical activity program, and six (26%) received both treatments in combination. These studies tested different digital technologies, including mobile applications (k = 4), telephone (k = 4), Internet (k = 4), SMS (k = 2), and combinations (k = 9). Moreover, studies tested different additional components, such as reminder alone (k = 4), remote coaching alone (k = 3), remote monitoring alone (k = 3), combined remote reminder and monitoring (k = 2), combined remote coaching and monitoring (k = 7), with fewer studies focused on combined remote reminder and coaching (k = 1) and combined all (k = 1). For models of telehealth delivery, 8 (35%) studies used virtual contact, 5 (22%) studies used no interacting contact, 5 (22%) studies used mixed models, and one study (4%) used in-person delivery.
Risk of bias
Risk of bias appraised with the Cochrane Collaboration’s risk of bias tool and additional PEDro scale are presented in Supplementary Fig. 1 and Supplementary Table 2. Risk of bias assessment revealed the following sources of bias: the absence of adequate random sequence generation in five (22%) studies31,35,45,46,48; no details of allocation concealment in five (22%) studies29,32,35,46,47; participants and personnel were unblinded in all studies (100%)24-48 limited by the intervention; the absence of the blinding of outcome assessment in nine studies (39%)25,27,31,35,38,41,46-48; insufficient strategies for dealing with incomplete outcome data in six (26%) studies24,27,28,32,46,47; no reporting bias in all studies and the presence of other bias in four studies (17%) (including insufficient power27 and without clinical registration46-48).
Main analyses
Of these studies, 20 (87%) studies contributed data that were considered appropriate for meta-analysis. The GRADE summary of findings can be found in Supplementary Table 3. Meta-analysis results of the effects of telehealth-based exercise/physical activity programs on primary outcomes are presented in Table 2.
Effectiveness on pain
For pain relief, there was a significant difference and a small effect size across 19 studies (n = 2512; g = −0.39; 95% CI −0.67 to −0.11; p = 0.0004; forest plot Fig. 2A) with high heterogeneity (I2=83%) for favouring the telehealth-based intervention. The calculated MID of pain was 1.3, which was lower than the reported MID (2.0 units for the Western Ontario and McMaster Universities Osteoarthritis Index (WOMAC) pain subscale).49 Overall, there was low-certainty evidence that the telehealth-based exercise/physical activity programs led to a small, significant, but not clinically meaningful reduction in pain.
Effectiveness on physical activity
The meta-analysis result across nine studies for physical activity favoured the telehealth-based intervention (n = 1570; g = 0.13; 95% CI 0.03 to 0.23; p = 0.01; forest plot Fig. 2B) with low heterogeneity (I2 = 0%). The calculated MID of physical activity was 9.0, which was lower than the reported MID (46.0 units for Physical Activity Scale for the Elderly).50 Overall, there was low-certainty evidence that the telehealth-based exercise/physical activity programs for the treatment of KOA might provide a significant, but insufficient, and not clinically meaningful improvement in physical activity.
Effectiveness on physical function
For the improvement of physical function, the meta-analysis of 18 RCTs favoured the telehealth-based intervention with g = −0.51 (n = 2373; 95% CI −0.98 to −0.05; p = 0.0004; I2 = 87%; forest plot Fig. 2C). The calculated MID of physical function was 5.3, which was lower than the reported MID (10.1 units for the WOMAC physical function subscale).51 Overall, there was low-certainty evidence that telehealth-based exercise/physical activity programs might lead to a moderate, significant, but not clinically meaningful improvement in physical function.
Effectiveness on secondary outcomes
Better secondary outcomes, such as quality of life (n = 1301; g = 0.25; 95% CI 0.14 to 0.36; p <0.00001; I2 = 5%; Supplementary Fig. 2), self-efficacy for pain (n = 1337; g = 0.72; 95% CI 0.53 to 0.91; p < 0.00001; I 2= 4%; Supplementary Fig. 2), and global improvement (n=1042; odds ratios [OR] = 2.69; 95% CI 1.41 to 5.15; p=0.0005; I2 = 79%; Supplementary Fig. 2) were observed with the intervention groups compared with control groups, but with a non-significant trend and moderate heterogeneity between studies for self-efficacy for physical function (n = 578; g = 0.14; 95% CI −0.26 to 0.53; p = 0.50; I2 = 52%; Supplementary Fig. 2). Overall, there was moderate-certainty evidence that telehealth-based exercise/physical activity programs improved the quality of life and self-efficacy for pain, and low-certainty evidence that telehealth-based intervention might provide benefit in global improvement, but not significant in self-efficacy for physical function.
Sensitivity analyses
After removing one study29, the overall significant effect size for pain relief was still small with g = −0.28 (n = 2448; 95% CI −0.44 to −0.11; p=0.0003; Supplementary Fig. 3) and a lower heterogeneity (I2 = 69%), which was a robust result. Moreover, after omitting the study29, the overall significant effect size for physical function improvement was small with g = −0.30 (n = 2309; 95% CI −0.47 to −0.13; p = 0.0002; Supplementary Fig. 3) with a lower heterogeneity (I2 = 69%).
Subgroup analysis
The result of the meta-regression test is listed in Supplementary Table 4. According to the World Health Organization (WHO) classification, studies were divided into (1) interventions for clients subgroup, (2) interventions for healthcare providers subgroup, and (3) interventions for clients and healthcare providers subgroup. In subgroup that included interventions for clients and healthcare providers, the effects of the telehealth-based intervention on pain (g = −0.29; 95% CI −0.49 to −0.09) and physical function (g = −0.36; 95% CI −0.63 to −0.08) were significant. However, no significant differences were found in both pain (g = −0.73; 95% CI −1.47 to 0.01) and physical function (g = −0.98; 95% CI −2.21 to 0.26) in interventions for clients subgroup. The interventions for healthcare providers subgroup showed that participants in the intervention group had no improvement in pain relief (g = 0.21; 95% CI −0.20 to 0.62) and physical function (g = 0.20; 95% CI −0.21 to 0.61) compared with the control group, either. The subgroup differences were significant in pain (p = 0.039) and physical function (p = 0.04), although heterogeneity was still significant in some of these sub-groups (I2>50%, Fig. 3).
As for the type of digital technologies in intervention group, the subgroup differences were significant in pain (p = 0.003) and physical function (p = 0.012) with high heterogeneity in mixed and SMS subgroup (I2 > 50%, Fig. 4). The subgroup analysis showed significant relief in pain (mobile application subgroup: g = −0.59 (95% CI −1.01 to −0.16); Internet subgroup: g = −0.25 (95% CI −0.43 to −0.07); and mixed type: g = −0.29 (95% CI −0.57 to −0.01)), and significant improvement in physical function (mobile application subgroup: g = −0.73 (95% CI −1.10 to −0.36); Internet subgroup: g = −0.42 (95% CI −0.80 to −0.04); and mixed type: g = −0.28 (95% CI −0.54 to −0.02)). Nonetheless, interventions with telephone and SMS subgroup showed both non-significant results in pain (g = −0.08; 95% CI −0.34 to 0.18) and g = −1.41; 95% CI −4.54 to 1.73), respectively) and physical function (g = 0.01; 95% CI −0.22 to 0.23) and g = −2.45; 95% CI −7.39 to 2.48, respectively).
The control treatments in included studies were divided into active controls (ie, exercise, physical therapy, and self-management) or inactive controls (ie, education, usual care, and waitlist) with subgroup differences (pain p = 0.021; physical function p = 0.065, Fig. 5). Compared with the active control groups, no statistically significant difference in pain (g = −0.08; 95% CI −0.21 to 0.05) and physical function (g= −0.07; 95% CI −0.21 to 0.07) was determined. Compared with inactive control groups, statistically significant pain reduction (g = −0.63; 95% CI −1.08 to −0.18) and function improvement (g = −0.79; 95% CI −1.54 to −0.03) were found in the intervention group.
Publication bias
Visual assessment of funnel plots for studies reporting pain relief, and improvement in physical activity and function did not suggest publication bias (see Supplementary Fig. 4) and these were supported by Egger’s regression test for plot (see Table 2).