The Effectiveness of E-Health Interventions Promoting Physical Activity and Reducing Sedentary Behavior in College Students: A Systematic Review and Meta-Analysis of Randomized Controlled Trials

Insufficient physical activity (PA) and excessive sedentary behavior (SB) are detrimental to physical and mental health. This systematic review and meta-analysis aimed to identify whether e-health interventions are effective for improving PA and SB in college students. Five electronic databases, including Medline, Web of Science, Embase, Cochrane Library, and ProQuest, were searched to collect relevant randomized controlled trials up to 22 June 2022. In total, 22 trials (including 31 effects) with 8333 samples were included in this meta-analysis. The results showed that e-health interventions significantly improved PA at post-intervention (SMD = 0.32, 95% CI: 0.19, 0.45, p < 0.001) compared with the control group, especially for total PA (SMD = 0.34, 95% CI: 0.10, 0.58, p = 0.005), moderate to vigorous PA (SMD = 0.17, 95% CI: 0.01, 0.32, p = 0.036), and steps (SMD = 0.75, 95% CI: 0.23, 1.28, p < 0.001. There were no significant effects for both PA at follow-up (SMD = 0.24, 95% CI: – 0.01, 0.49, p = 0.057) and SB (MD = −29.11, 95% CI: −70.55, 12.32, p = 0.17). The findings of subgroup analyses indicated that compared to the control group, interventions in the group of general participants (SMD = 0.45, 95% CI: 0.27, 0.63, p < 0.001), smartphone apps (SMD = 0.46, 95% CI: 0.19, 0.73, p = 0.001), and online (SMD = 0.23, 95% CI: 0.04, 0.43, p < 0.001) can significantly improve PA at post-intervention. Moreover, the intervention effects were significant across all groups of theory, region, instrument, duration, and female ratio. At follow-up, interventions in groups of developing region (SMD = 1.17, 95% CI: 0.73, 1.62, p < 0.001), objective instrument (SMD = 0.83, 95% CI: 0.23, 1.42, p = 0.007), duration ≤ 3-month (SMD = 1.06, 95% CI: 0.72, 1.39, p < 0.001), and all female (SMD = 0.79, 95% CI: 0.02, 1.56, p = 0.044) can significantly improve PA. The evidence of this meta-analysis shows that e-health interventions can be taken as promising strategies for promoting PA. The maintenance of PA improvement and the effect of interventions in reducing SB remain to be further studied. Educators and health practitioners should focus on creating multiple e-health interventions with individualized components.


Introduction
Inadequate physical activity (PA) and high levels of sedentary behavior (SB) are crucial risk factors for mortality and non-communicable diseases (NCD), such as cardiovascular disease, cancer, and diabetes around the world [1]. It is well known that regular PA can provide numerous metabolic, cardiovascular, and mental health benefits [2]. Previous identified that e-message intervention had no significant improvement in either PA or SB [37]. Based on the research status mentioned above, a quantitative meta-analysis of the existing trials is integral to validating the effects of e-health interventions.
As digital natives, college students have a high penetration of electronic devices and proficient internet skills [38], which can contribute to the widespread application of e-health interventions on campus. Therefore, validating the effectiveness of e-health interventions in increasing PA and reducing SB among college students will provide strong supporting evidence for developing corresponding interventions.
This review is the first meta-analysis to investigate the effects of e-health interventions on promoting PA and reducing SB in college students from a holistic perspective. The purpose of this study was twofold. First, to systematically summarize the effects of ehealth interventions for improving PA in terms of total PA (TPA), moderate to vigorous PA (MVPA), light PA (LPA), walking, steps, and SB among college students. Second, to investigate the potential moderators of e-health interventions' effects through exploratory subgroup analyses of participants' characteristics and intervention details.

Materials and Methods
The Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) statement [39] and Cochrane Collaboration Handbook recommendations [40] were employed as the rationales and methodological templates of this systematic review. This review has been registered on the PROSPERO platform (CRD42022352623).

Search Strategy
A comprehensive and integrated literature search of randomized controlled trials involving the effects of e-health interventions in improving PA and SB without publication time and language restrictions was conducted for relevant literature published from the following databases: Medline, Web of Science, Embase, Cochrane library, and ProQuest. The search period is from the inception of the databases to 22 June 2022.
Boolean logical operators were used to perform an exhaustive search using the medical topic headings (MeSH) paired with free-text phrases. The leading search terms in the three topics domains are as follows: participants (e.g., college students, university students, tertiary school students); intervention (e.g., e-health, mobile health, smartphone apps, wearable activity trackers, Internet, text messages); outcomes (e.g., physical activity, exercise, sedentary behavior); and study design (e.g., randomized controlled trial, RCT). In addition, as a complementary search, we performed additional screening of top journals (e.g., JMIR Mhealth and Uhealth, Journal of Medical Internet Research, Health Psychology) in the domains of e-health, m-health, and health behaviors to avoid the omission of essential studies due to inclusion criteria. In the supplemental materials, specific search information for each database is provided.
All initial search results were imported into Endnote20 software (Thomson ISI Research Soft, Philadelphia, PA, USA). Duplicate studies were removed first. The titles and abstracts of all imported studies were then screened independently by two reviewers to identify the potentially relevant studies that met the inclusion criteria. After the first screening of abstracts and titles, an intuitive and backward snowball retrieval approach was performed to ensure the integrity of the included literature. Then, full-text reviewing was conducted by two reviewers independently to find studies that would be suitable for this review. Regarding disagreement and uncertainty regarding the inclusion of studies, an agreement was reached through consultation with the third reviewer.

Eligibility Criteria
Study eligibility was assessed based on PICOS criteria (participants, interventions, comparators, outcomes, and study design).

Participants
College students who lived alone away from their families were included in this review, which was not limited by gender, age, health status, region, or nationality. Participants were considered eligible if they could participate in the e-health intervention program set up by the researcher. Studies that included university employees among the participants were excluded.

Interventions
Interventions were conducted in college settings. E-health interventions refer to any interventions that include at least one of the following components: smartphone apps; wearable activity tracks; websites; e-messages (i.e., text messages, social media messages, email); telehealth (i.e., remote monitoring, real-time interactive, videotelephony, etc.); or videogame. Studies that comprised multiple group comparisons (i.e., e-health intervention versus multiple interventions) were enrolled, but only the comparisons between the ehealth group and the control group were included.

Comparators
Studies were included if neither e-health interventions nor other interventions were imposed in the control groups.

Outcomes
Studies with PA and SB measured with self-report questionaries or objective instruments (pedometers or accelerometers) were included. PA outcomes included TPA, MVPA, LPA, walking, and steps. SB outcomes were the duration of sitting time. PA outcome variables are defined by the individual studies that are included. All the included outcomes should be reported as minutes, hours, or steps per unit. This review also included studies that reported PA in other forms (e.g., energy expenditure, weekly counts, and times per week).

Study Design
Only published RCTs, including pilot RCTs, were considered, while quasi-experiments, cross-sectional surveys, and other qualitative studies were excluded.

Data Extraction
We employed Microsoft Excel to create the data sheets. Two authors (PSY and YF) conducted a separate double-blind investigation to check and extract the crucial information from the included studies. The critical information extracted is as follows: study characteristics (authors, publication year, region); participant characteristics (age, female ratio, health status); intervention details (intervention mode, theory, duration, instrument, outcomes); study design (RCT or not; per-protocol or intention-to-treat; sample size); outcomes of PA and SB. Disagreements in data extraction were resolved through discussion. Missing data were obtained by tracing the literature and emailing the corresponding author of relative studies.

Risk of Bias (ROB) and Quality Assessment
The Cochrane Collaboration risk of bias tool [41] was employed to assess the risk of bias for the included studies using seven domains: (1) random sequence generation, (2) allocation sequence concealment, (3) blinding of participants, (4) blinding of outcome assessment, (5) incomplete outcome data, (6) selective outcome reporting, (7) other undefined biases (such as small sample size and conflict of interest). Each domain was graded as low, unclear, or high risk of bias, respectively, in each study. Each study was classified as low, unclear, or high risk of bias based on a combination of the seven domains. The study was assessed as high risk if more than one item was high risk. If most of the study (over three items) was unclear and there were no high-risk items, the study was assessed as unclear. When there was no high risk or less than three items for unclear, the study was assessed as low risk. Review Manager software (Revman 5.4; The Cochrane Collaboration, The Nordic Cochrane Centre, Copenhagen, Denmark) was used to create the figures of ROB. Two reviewers assessed the ROB of the included studies, and disagreements were resolved by negotiation or consulting the third author.

Statistical Analyses
This review took various outcomes of PA and SB (such as minutes per day or week of TPA, MVPA, and LPA; minutes or times of walking; steps per day; minute per day of sitting time or sedentary time) as the data sources for statistical analyses. The mean (M) and standard deviation (SD) of each outcome at baseline, post-intervention, and followup in endpoint were drawn for different calculations of numerical variables to calculate effect sizes based on Cochrane Collaboration Handbook recommendations [40]. First, the effect sizes were pooled using the inverse variance statistical method and random effect models to assess the principal impact of e-health. Standardized mean difference (SMD) representing the pooled effect sizes were supplied, along with 95% confidence intervals (CIs). When pooling the effect sizes of different measurements of PA, each comparison group's mean and standard deviation were used to determine SMD. If the M and SD were not provided, we performed statistical transformation or requested data from the original authors. Considering the consistency of SB measurements, mean difference (MD) was employed to pool the effect sizes. Second, subgroup analyses of eight moderators conducted in this review are presented as follows: (1) outcome (TPA, MVPA, LPA, walking, steps), (2) participant (inactive vs. general) (inactive refers to participants self-reporting less than 150 min of moderate PA or 75 min of vigorous PA per week; generally refers to participants without any PA level limitation), (3) theory (yes vs. no) (yes: explicitly mentions the adopting of health behavior theories as guidance for the interventions; no: no mention of theories as guidance for interventions), (4) intervention mode (smartphone app, social media, accelerometer or pedometer, online), (5) region (developing vs. developed), (6) instrument (objective vs. subjective), (7) intervention duration or follow-up (>8 weeks vs. ≤8 weeks or >3 months vs. ≤3 months), (8) female ratio (all vs. partial) (all: all participants are females; partial: participants include males). Moreover, I 2 statistics and Cochran Q-test were used to determine the statistical heterogeneity. When I 2 was below 25%, between 25% and 50%, between 50% and 75%, and above 75%, it was classified as very low, moderate, medium, and high heterogeneity, respectively, and p < 0.1 for Q test was assessed as statistically significant [42]. To identify publication bias, funnel plots and Egger's test were adopted [43]. Additionally, a sensitivity analysis was performed to ensure the robustness of the pooled effect size.
All data calculations (such as effect size syntheses, publication bias evaluation, subgroup analysis, heterogeneity tests, and sensitivity analysis) were performed using the statistical software STATA 16.0 (Stata Corp, College Station, TX, USA).
Regarding the intervention modalities of e-health, we grouped all studies into four categories based on the most dominant interventions in the study. The first is smartphone apps [36,46,48,52,53,58,60,61], which are interventions that mainly set up intervention programs through specific software or deliver intervention content through messages. The second is social media [47,49,50,59], including Facebook, WeChat, etc., where the intervention content was imposed through social media interactions. The third is wearable devices [45,52,54,56,63], such as accelerometers and pedometers, which impose interventions through monitoring and feedback functions on movement. The fourth is online interventions [37,44,55,57,62], which impose interventions through information interactions
Regarding the intervention modalities of e-health, we grouped all studies into four categories based on the most dominant interventions in the study. The first is smartphone apps [36,46,48,52,53,58,60,61], which are interventions that mainly set up intervention programs through specific software or deliver intervention content through messages. The second is social media [47,49,50,59], including Facebook, WeChat, etc., where the intervention content was imposed through social media interactions. The third is wearable devices [45,52,54,56,63], such as accelerometers and pedometers, which impose interventions through monitoring and feedback functions on movement. The fourth is online interventions [37,44,55,57,62], which impose interventions through information interactions on specific websites. The duration of the intervention ranged from 1 week to 3 months, and the follow-up period from post-intervention to endpoint ranged from 8 weeks to 15 months. A little over half of the included interventions (12/22, 55%) [37,44,45,[48][49][50][51][54][55][56]58,60] were designed based on at least one behavioral theory.
Most of the studies employed a non-intervention control group, and five studies provided their control group with general health information and instructions through sessions [36,48], mental counseling [55], or physical education course [46,56]. The control group of one study [50] needed to report their daily PA duration.
Most of the studies employed a non-intervention control group, and five studies provided their control group with general health information and instructions through sessions [36,48], mental counseling [55], or physical education course [46,56]. The control group of one study [50] needed to report their daily PA duration.

Primary Outcomes
A meta-analysis of the random effect model including 22 studies (31 effects) yielded a significant improvement in PA in the e-health intervention group at post-intervention compared to the control group (SMD = 0.32, 95% CI: 0.19, 0.45, p < 0.001) (see Figure 4), but not a significant improvement at follow-up (SMD = 0.24, 95% CI: −0.01, 0.49, p = 0.057) (see Figure 5). At post-intervention, the effect sizes ranged from −0.87 to 1.53; at follow-up, the effect sizes varied between -0.36 and 1.36. A funnel plot paired with the Egger test (at post-intervention p < 0.001; at follow-up p = 0.035) indicated that publication bias might be present (see Figures S1 and S2, available in the Supplementary Materials).

Primary Outcomes
A meta-analysis of the random effect model including 22 studies (31 effects) yielded a significant improvement in PA in the e-health intervention group at post-intervention compared to the control group (SMD = 0.32, 95% CI: 0.19, 0.45, p < 0.001) (see Figure 4), but not a significant improvement at follow-up (SMD = 0.24, 95% CI: −0.01, 0.49, p = 0.057) (see Figure 5). At post-intervention, the effect sizes ranged from −0.87 to 1.53; at followup, the effect sizes varied between -0.36 and 1.36. A funnel plot paired with the Egger test (at post-intervention p < 0.001; at follow-up p = 0.035) indicated that publication bias might be present (see Figures S1 and S2, available in the Supplementary Materials). A meta-analysis of the random effect model for five effect sizes reporting SB-related outcomes found that the e-health intervention did not reduce SB significantly at post-intervention compared to the control group (MD = −29.11, 95% CI: −70.55, 12.32, p = 0.17) (see Figure 6). Considering the small number of studies reporting SB, no test for publication bias was performed.

Subgroup Analysis of PA
The subgroup analyses of eight moderator variables at post-intervention and followup are shown in Table 1   A meta-analysis of the random effect model for five effect sizes reporting SB-related outcomes found that the e-health intervention did not reduce SB significantly at postintervention compared to the control group (MD = −29.11, 95% CI: −70.55, 12.32, p = 0.17) (see Figure 6). Considering the small number of studies reporting SB, no test for publication bias was performed.
There was no subgroup exploration for the effects of SB due to only five included studies.

Robustness of the Results
Sensitivity analyses were performed to assess the reliability of the results, which were conducted using Stata 16.0. The specific method is to eliminate the literature one by one and then combine the effect sizes to observe whether the results have changed significantly. The sensitivity analysis results showed that the effect sizes did not alter much for PA both at post-intervention and at follow-up, as well as for SB, indicating that the findings of the meta-analysis were robust (see Figures S3-S5, available in the Supplementary Materials).

Discussion
This systematic review and meta-analysis aim to identify and quantify the valid evidence of the e-health interventions for improving PA and SB among college students. The results indicated that e-health interventions have a significant small-to-moderate effect on PA at post-intervention (SMD = 0.32, 95% CI: 0.19, 0.45, p < 0.001) according to Cohen's criteria [64], whereas the maintenance of PA improvement was not observed because there was no significant effect at follow-up (SMD = 0.24, 95% CI: −0.01, 0.49, p = 0.057). Regarding reducing SB, the e-health intervention group contributed to a mean reduction of 29.11 min per day in SB time compared to the control, but the effect was not statistically significant.
The finding of e-health interventions positively affecting increasing PA at post-intervention is powerful support of recently published reviews [28,[30][31][32][33][34]. Previous relevant studies mainly focused on populations such as adolescents [30,31], patients [32], women [28], and older people [33], while only this review targeted college students. A meta-analysis by Champion et al. [31] showed that school-based e-health interventions could improve PA in adolescents, and Kwan et al. [33] also found the same effect in older people. Cotie et al. [28] observed large effect sizes in working-age women, and the recent review conducted by Duan et al. [32] has the same findings in NCD patients. Although this review found only small-to-moderate effect sizes of all PA outcomes in college students, the pooled effect size for steps was also close to the large effect size by subgroup analysis of PA outcomes. Based on the extensive validation of e-health interventions' effects on different populations, such interventions will have a promising prospect of improving PA and SB. Given that different PA outcomes and measures may lead to high heterogeneity, which would impede accurate comparison and interpretation of results [65], the experimental design of future studies should take this into full consideration.
College students have more freedom and independence, often leading to a high risk of developing poor health behaviors due to their lack of self-control and self-efficacy [65,66]. Considering this point, many e-health intervention trials have used self-efficacy as the core theory component [44,48,49,[54][55][56]58,60]. Several meta-analyses [32,67,68] also found that theory-based interventions were more effective, which, unfortunately, was not found in this study. Although the college settings and the literacy that college students possess are well-suitable for e-health interventions in improving PA and SB, confusing goal setting and non-targeted intervention content may discourage college students from engaging in some complex PA (e.g., MVPA). This may be one of the reasons for the small effect sizes of MVPA in this review. The design of future e-health interventions should focus on adding specific goals and plans for PA and precisely matching BCTs to goals rather than just general health behavior education or general counseling advice.
High heterogeneities were observed in all three effect estimates. Still, the robustness of the pooled effect size for this review was determined by sensitivity analysis, which also indicates that the results are reliable. Subgroup analysis only found that differences in PA outcomes at follow-up may be a potential source of heterogeneity [69]. Furthermore, in addition to identifying significant intervention effects in many subgroups, we also observed some within-group comparisons of intervention effects where one group had a significantly larger effect size than the other.
From the subgroup analysis of intervention modes, smartphone apps and online interventions had significant intervention effects at post-intervention, with the highest amount of smartphone app intervention effects. Based on the rapid development of technology, smartphones are becoming more and more functional, and the open mobile app development platforms offer many convenient conditions for PA and SB interventions [68,70]. There are more comprehensive information and entire interaction in the interventions through websites, which is perhaps the main reason for their effectiveness. Several studies [34,71,72] confirmed that hybrid intervention modes were better than single intervention modes. However, integrating different intervention modes must be matched with corresponding BCTs so that the key roles' variables can be easily identified [73,74].
Interestingly, this review found visible differences in the effects of the e-health intervention across different participants. The effect size was larger for the general college students than the inactive college students. The lower level of PA motivation of inactive college students may have impeded the intervention effect [75]. In addition, the intervention effect was better in the group of all female college students, especially at follow-up, where the effect size almost reached the level of a large effect size, which indicated that female college students have better adherence to the e-health interventions. Therefore, the behavioral-psychological characteristics should be fully considered when applying e-health interventions to different participants.
Through subgroup analyses of the two instruments for PA measurement, an important finding of this review was that the objective instrument group had significantly larger intervention effects at both post-intervention and follow-up than the self-report questionnaire group. Self-report is a low-cost, feasible, and convenient method for data collection [76]. Previous studies also verified that the results of self-report questionnaires have high correlations with those of objective instruments [77,78]. However, objective instruments should be promoted to ensure the accuracy and precision of the measurements. Combining accurate algorithms and the portability of measurement tools will facilitate PA-related health behavior studies.
In addition, the effects of two follow-up durations on PA were significantly different, while the difference between the two intervention durations' effects was marginally significant. The effect size of short-term intervention was larger than that of long-term intervention, which was most pronounced at follow-up. This finding is consistent with Moenninghoff and colleagues' findings [79]. A possible explanation is that prolonged intervention can lead to losing personal interest and increased objective barriers. Based on this, we suggest that reinforcements should be added to e-health intervention at regular intervals, especially during the follow-up period, which has been verified to be an efficient approach to avoid attenuation [80][81][82].
Regarding SB, this review did not find a significant effect of e-health interventions on SB, which is not sufficient to deny the impact of e-health due to the limited quantity and quality of included studies. This meta-analysis still found a mean reduction of 29.11 min per day in SB after the intervention. A recent meta-analysis by Castro et al. [83] found an increasing trend in sedentary time among college students over the last decade, with an average of 9.82 h per day measured by accelerometers. Therefore, Reducing SB in college students through effective interventions is urgent. To improve the effectiveness of e-health interventions for less SB, providing good monitoring and feedback measures (e.g., setting regular reminders) may be a practical approach [84,85]. Furthermore, future studies should implement trials targeting SB reduction to find key intervention factors that influence SB.
This review is the first meta-analysis to examine the effectiveness of e-health interventions for PA and SB among college students. The included studies are all RCTs that were conducted in over ten countries. The findings of this study can be used as an essential theoretical basis and practical guidance to improve PA and SB among college students. As interventions are highly in tune with intelligent technology, e-health interventions are convenient, efficient, and inexpensive, making them suitable for dissemination and implementation in college settings. Future health promotion projects, especially campus health projects, should employ e-health in their priority list of interventions, which will contribute to the prevention of NCDs and improve the health and well-being of college students.
Despite the innovation and strength of evidence in this study, there are still the following limitations. First, only RCTs were included in this study. Thus, many other relevant trials and investigations have been omitted. Future research should enlarge the search scope to include exhaustive studies for more comprehensive explorations. Second, pooling the effect sizes of different PA outcome variables is a challenging attempt. Although this review has been registered in PROSPERO, the high heterogeneity from numerous potential moderators (i.e., PA outcomes, participant characteristics, intervention modes, durations for intervention implementation and follow-up, and outcome measurements) and not enough included studies contribute to the cautious interpretation of the synthesized results. Third, based on the characteristics of e-health interventions, participant blinding is not possible to perform, which should be the potential reason for downgrading the quality of ROB assessment. Finally, this review has not provided insight into the correlations and mechanisms of action between factors associated with BCTs and the effects of e-health interventions. Further research should focus on these crucial issues.

Conclusions
The findings of this systematic review and meta-analysis identified that e-health interventions have a significant impact on increasing PA, especially TPA, MVPA, and steps at post-intervention. However, the maintenance of PA improvement at follow-up and the effect of interventions on improving SB remain to be further studied. In addition, the current review provided valuable evidence that the effects of e-health interventions vary in the light of different outcomes and moderators. As promising strategies, e-health interventions have become a new trend in college settings in recent years. Educators and health practitioners should follow this trend and delve into the vital psychological variables of college students' health behavioral change, integrating smartphone apps, the Internet, monitoring tools, and social media to create multiple e-health interventions with individualized components.
Supplementary Materials: The following supporting information can be downloaded at: https:// www.mdpi.com/article/10.3390/ijerph20010318/s1, Figure S1: funnel plot of PA at post-intervention; Figure S2: funnel plot of PA at follow-up; Figure S3: sensitivity analysis of PA at post-intervention; Figure S4: sensitivity analysis of PA at follow-up; Figure S5: sensitivity analysis of SB.