We performed a systematic review and meta-analysis summarizing the effects of digital online exercise (OE) versus no exercise (NEX) or conventional face-to-face exercise (FFE). Outcomes included markers of PA, physical function, and mental health. The review protocol was prospectively registered in the International Prospective Register of Systematic Reviews (PROSPERO) database (CRD42022338871). The PRISMA abstract checklist (Checklist 1) and PRISMA checklist (Checklist 2) were used to ensure that reporting was compliant.

Two independent investigators (AKB and JW) performed separate individual systematic literature searches using PubMed, Cochrane library, and Google Scholar. Searches were performed initially in August 2022 and updated in February 2024. A uniform search term was used to conduct the literature search across all 3 search engines, the term was: “(exercise OR “PA”) AND (internet* OR online* OR web* OR e-health OR digital OR tele* OR virtual) AND home*.” Searches were limited to articles published from January 1, 2000 (Multimedia Appendix 1). The decision to start the range at the year 2000 was made to accommodate for the lack of the necessary online technologies. Such technologies include but are not limited to high-speed internet, streaming platforms, and advanced audio-visual hardware required for effective and interactive online exercise programs. In addition to database searches, the reference lists of all included studies were checked to identify further potentially eligible articles [31]. Articles that met the initial screening criteria were entered into a Microsoft Excel spreadsheet for title and author screening. Duplicate and suspected duplicate articles were individually downloaded and compared. At this stage, if an article was confirmed as a duplicate, it was removed from the spreadsheet.

Studies were considered eligible for inclusion based on the following criteria (Textbox 1).

Eligibility checks were performed by two independent investigators (AKB and JW), and disagreements were resolved by discussion.

Study quality was evaluated by means of the PEDro scale [32,33]. Two examiners (AKB and JDP) independently rated each study, and in case of disagreement, meetings were held to achieve consensus [34]. For the identification of a potential reporting bias, we used visual inspection of funnel plots. To classify the certainty about the evidence as very low, low, moderate, or high, the criteria of the GRADE (Grading of Recommendations, Assessment, Development, and Evaluations) working group were applied (Multimedia Appendix 2 [35]). Briefly, the evaluation starts with the assumption of a high certainty about the evidence due to the randomized controlled trial design. The GRADE framework then suggests adjusting the certainty as follows: in case of risk of bias, imprecision, inconsistency of results, indirectness of the evidence, or publication bias, one point is subtracted for each weakness. In contrast, a large magnitude of the effect or a dose-response gradient leads to an upgrade of the certainty rating [35].

We extracted the following data: sample size, participant characteristics, interventions, measurements, and results (pre-post changes plus SDs of each intervention arm). The primary outcomes of the meta-analysis were PA, motor function (strength, endurance, balance, flexibility, gait, and body composition), and mental health (depression, anxiety, mood or emotion, psychological well-being, sleep, and self-efficacy). If a study performed more than one test assessing the same outcome, all effect sizes (ES) were obtained.

For each trial arm, we extracted the mean baseline and postintervention values, pre-post changes, and the related SD. In case of incomplete reporting (ie, missing SDs of the changes from baseline), the corresponding authors of the trials were contacted. If no values could be obtained, missing data were determined from figures or imputed according to Cochrane recommendations:

where Corr=0.7. The value chosen for Corr represents a conservative estimate of the correlation between the baseline and posttreatment SDs [36].

Robust variance estimation random [37] was used to pool the standardized mean differences (SMD) and 95% CI for OE versus NEX as well as OE versus FFE. Dependency of ES was considered by nesting the term “study” as a random factor in the model. The between-study variance component was determined by means of Tau2, using the method-of-moments estimate; for within-study variance (more than one dependent effect size), ω2 was calculated [38]. We interpreted the resulting ES as follows: small (SMD=0.2), medium (SMD=0.5), or large (SMD=0.8). The used software was R (R Foundation for Statistical Computing, Vienna, Austria), packages meta (G Schwarzer) and robumeta (version 2.0) [39].

Ethical approval for this study was applied for and obtained before the commencement of data collection. Approval was issued by the Institutional Research Ethics Committee (IREC) at the Durban University of Technology (DUT; IREC 090/20).

The initial database searches yielded a total of 6335 articles (Figure 1). After removal of duplicates and application of inclusion criteria, 18 studies were included.

The characteristics of the 18 included studies, which included a total of 3531 cumulative participants, are displayed in Table 1. Half (n=9) of the studies had participants that were older than 65 years, the remaining half (n=9) examined young or middle-aged adults. Five studies compared OE to FFE, while 12 investigated OE versus NEX, and one compared OE to both FFE and NEX. Motor performance outcomes were assessed in 11 studies, PA was measured in 5 studies, and mental health was examined in 8 studies. The mean (SD) duration of the OE interventions was 12.5 (10.34) weeks and the mean (SD) training frequency was 2.7 (1.94) sessions per week.

Ratings on the PEDro scale ranged from 3 to 10 with a mean of 7.2 (SD 2.3) out of 10 points (Table 2).

All studies reported between-group statistical analyses for at least one key outcome measure. Almost all (94%, n=17) articles provided clear eligibility criteria; had intervention and control groups that were similar at baseline with regards to key prognostic indicators; and reported point and variability measures for at least one key outcome measure. A clear majority (83%, n=15) used randomized groups allocation, while 78% (n=14) ensured concealed group allocations. A slightly smaller share (72%, n=13) reported all participants receiving the intended intervention and the use of intention to treat in cases where participants did not receive the intended interventions; and reported outcome measures for at least 85% of the initial group allocations. In terms of blinding, 61% (n=11) of the studies reported participant blinding, 44% (n=8) indicated assessor blinding, and only 17% (n=3) stipulated therapist blinding.

Visual inspection of funnel plots (Multimedia Appendix 3) yielded indications for a reporting bias regarding strength measures in OE versus FFE and for mood or emotion in OE versus NEX.

Compared to NEX (Table 3), OE had a moderate to large beneficial effect on measures of strength (SMD=0.61, 95% CI 0.06-1.15, P=.04; GRADE: moderate certainty about the evidence), balance (SMD=0.52, 95% CI 0.06-0.99, P=.04; moderate certainty), endurance (SMD=0.85, 95% CI −0.01 to 1.70, P=.05; low certainty), and physical activity (SMD=0.46, 95% CI 0.05-0.87, P=.04; moderate certainty). No difference was found for gait speed (moderate certainty) and flexibility (low certainty).

With regard to mental health (Table 4), OE was moderately to strongly superior to NEX for depression (SMD=1.08, 95% CI −0.01 to 2.16, P=.05; moderate certainty), mood/emotion (SMD=0.47, 95% CI 0.05-0.90, P=.04; low certainty), mental well-being (SMD=0.79, 95% CI 0.06-1.52, P=.05; moderate certainty), and self-efficacy (SMD=1.1, 95% CI 1.03-1.17, P=.06; moderate certainty). No effect was found for anxiety and sleep (P>.05, moderate certainty).

Regarding OE versus FFE (Table 5), no difference was found for most measures (strength: very low certainty, endurance: moderate certainty, body fat: low certainty). However, gait speed improved slightly more in OE (SMD=0.25, 95% CI 0.24-0.26, P=.002; moderate certainty).

The manifold benefits of exercise have been repeatedly presented in the literature [1,2]. Recently, there has been a shift toward the use of online platforms aiming to provide easily accessible exercise opportunities to the general population [23,26]. To the best of our knowledge, this article is the most comprehensive and up-to-date quantitative summary of the available evidence. We show that OE is non-inferior to FFE and superior to NEX in a variety of measures of physical function, mental health, and PA. Regarding motor performance, OE induced moderate-to-large improvements in strength, endurance, and balance, which accords with data from older reviews [23,26]. In contrast, gait speed and flexibility did not change. This may be related to a lack of power as both outcomes were investigated in only two studies. For mental health outcomes, the very large positive effects on depression (SMD=1.08) and self-efficacy (SMD=1.1) were striking. A beneficial impact, although of smaller magnitude, was also observed for mood or emotion and mental well-being. While the sizable effects of OE on most psychological outcomes are in line with studies examining face-to-face exercise [58,59], the lack of a change in anxiety is a bit surprising because anxiety had been reported to improve in response to conventional exercise [60]. Again, the nonsignificant effect size may be due to the small number of studies (n=2).

Our results have a variety of practical implications. It would be reasonable to assume that exercise performed face-to-face would be the most effective way to improve health and performance. Yet, as we demonstrated noninferiority of OE regarding most outcomes, the selection of the exercise mode may be left to the individual preference. This is particularly relevant because under some circumstances, OE may be the more feasible option. For example, there is a distinct need for the development and implementation of innovative and convenient exercise strategies to combat physical inactivity, and some individuals living in scarcely populated areas may have difficulties finding FFE. Also, engagement in OE may save travel time and costs. The use of OE can hence deliver an opportunity within certain populations where inactivity is rife [61]. A study showed that digital PA programs can be integrated into workplace interfaces to address inactivity that occurs during work hours [62].

Half of the studies reviewed (n=9) used older participants (>65 years). Of note, the projected change in demographics leaves older individuals at a higher risk for decreased PA [13]. Research suggests that the age and PA are inversely proportionate in that an increase in age tends to result in decreased levels of PA. This has been attributed to several factors, decreased physical capability, socioeconomic barriers, and a lack of social support [63]. For this particular population, the use of online PA can potentially increase engagement as individuals could access programs remotely [64]. Thus, OE increases the potential to provide these older adult persons with a safe, effective, and accessible PA opportunity. As much as there is the limitation of digital literacy [65], research suggests that this limitation is improving [65]; hence, as the years unfold, the use of online PA interventions may become more beneficial and used in the older population [66].

As indicated, OE may be of particular interest in rural areas where public PA infrastructure (eg, availability of sports clubs, or gyms) is limited [67]. In these instances, the distance between residential dwellings and traditional PA establishments is too large for individuals to travel [68]. Hence, online PA programs, especially in light of growing internet accessibility throughout the world, could offer the benefits of PA without the need for increased traveling [69]. As the share of users with internet access is increasing steadily around the globe [70], OE may be of help in the fight against the pandemic of inactivity.

Our results on mental health outcomes showed that OE was superior to NEX for depression, mood/emotion, mental well-being, and self-efficacy. This implies that OE could be of value in cases of mental illness where PA facilities are either unavailable or undesired by patients. Further to this, in a clinical setting, OE incorporating group interactions into the structure of the platform can present the opportunity for social interaction [71]. This adds a social aspect to the benefits of online PA [72]. The use of social integrations in online PA programs allows for individuals with similar backgrounds, such as those with mental illness, to interact [73]. Thus, increasing motivation and desire to remain consistent with OE, which enhances long-term program engagement [74].

Finally, OE can also help to reduce the direct costs of exercise as streamed contents would theoretically be available to an unlimited number of individuals rather than being limited to those who can afford memberships in face-to-face PA facilities [75]. However, some caveats also require consideration. A large benefit of FFE is that coaches have a 3D view on exercising individuals, which allows for an easier correction and monitoring of movement execution. While this is less a concern for healthy individuals, it may become relevant for older adults or people with disease [76,77]. Another issue relates to interpersonal factors, as sports and exercise cannot be reduced to their direct physical effects. When exercising in a group, social interaction can increase motivation or build friendships that could be of value in other aspects of life [78,79].

A particular strength of our review is that we only included randomized controlled trials, as this study type is viewed to be the most valid in the assessment of intervention effectiveness [80]. In addition, the mean PEDro score of 7.2 indicates a good overall quality, which adds to the validity of the findings [50]. A further strength was the use of GRADE ratings to assess the quality of evidence analyzed in this manuscript. This adds to the overall quality of the review by creating a more accurate representation of data. Finally, to reduce heterogeneity, we limited inclusions to studies with healthy participants. There are certain limitations identified within our study that should be taken into consideration in the planning and execution of any future studies. Firstly, our study used an identical search across three different search engines. Although this is in keeping with regular review protocols, future studies may benefit from the use of additional search engines to ensure that results are extensive and exhaustive. Secondly, the filter applied to exclude studies published before January 2000 was included to streamline the search results; however, it may have resulted in exclusion of some relevant data. Hence, we recommend that future reviews use wider timeframes. Thirdly, in about one third of the outcomes, the certainty about the evidence was low, which was probably due to the small number of studies resulting in large confidence intervals, reporting bias, and heterogeneity. We therefore reinforce the need for additional studies to fully gauge the long-term effects of OE [41,53,54]. This particularly applies because some of the studies included in our review were conducted during the COVID-19 pandemic, and it is unknown how the specific conditions affected the overall result of the meta-analysis. It is imperative that more studies are conducted in a less volatile period to observe unbiased effects in relation to the everyday lives of individuals. Fourth, the search criteria did not allow for online PA interventions that assessed mental outcome measures only and instead focused on a coupling of physical and mental outcome measures. Further investigations are required to report specifically on the effectiveness of online PA intervention on mental health, against active controls and inactive controls. Finally, our study included only healthy participants, although this was specifically conducted to encourage heterogeneity. Further investigations are required to report specifically on the effectiveness and safety of OE in unhealthy participants.

There is mostly moderate certainty evidence that OE improves PA, motor performance, and mental health when compared to NEX. In addition, OE seems non-inferior to FFE. However, additional studies including larger sample sizes, longer study durations, and long-term follow-up measurements are warranted in order to better delineate the benefits of OE interventions.