Out of the 3895 reviews initially located and downloaded, 587 doublets were removed. During title and abstract screening, additional 3233 studies were excluded, with 75 studies remaining for full text screening. Sixty-four of these articles were excluded due to above mentioned exclusion criteria. This resulted in a total of 11 systematic reviews and meta-analyses which were included in this umbrella review [46–56] (for more details see Flow-Chart in Figure 1). The updated search located 472 additional articles which were all excluded after title and abstract screening.
*** Insert Figure 1 around here ***
Description of the included studies
The 11 reviews included a total of 195 publications (182 studies) published between 1998 and 2018, with 167 of these publications being included once throughout the reviews and 13 publications being included in two or three reviews, accounting for 28 publications.
The included original research studies were mainly conducted in USA and Canada and Europe, and the most common study designs were randomized control trials (RCTs). The duration of interventions ranged from one session to 24 months, with the majority (92%) of interventions lasting at least four weeks. Sample sizes ranged from 458 [50] to 73,417 participants [52] for the reviews and added up to 114,430 participants throughout all studies. The full details of study characteristics of articles included in the umbrella review are displayed in Table 1.
*** please insert Table 1 about here ***
Two reviews focused on children and adolescents [46, 55], four focused adults [47, 48, 52, 53] and five included participants of all ages [49–51, 54, 56]. Five systematic reviews focused on PA outcomes [46, 48–50, 55], one meta-analysis focused on SB outcomes [47], and one meta-analysis and one systematic review included both PA and SB outcomes [51, 52]. HE was the main outcome in two meta-analyses [53, 54], and one systematic review included PA, SB and HE as main outcomes [56].
Eight reviews reported the use of theoretical frameworks [46, 48–50, 52, 53, 55, 56], and 78/125 (62%) publications in these reviews reported the use of a theoretical foundation. The most common reported theories were social cognitive theory (n = 29), transtheoretical model (n = 16), theory of planned behavior (n = 10), self-determination theory (n = 10) and I-change model (n = 7). Four reviews [47, 51, 52, 54] coded the use of BCTs using a taxonomy of behavior change [19, 20] and two reviews [46, 50] reported BCTs without coding them. The BCTs, which were most frequently reported by the reviews, were goal setting (n = 5), self-monitoring (n = 4), social support (n = 4), prompts/cues (n = 4), feedback on the behavior (n = 3) and instruction on how to perform the behavior (n = 2). Since the BCTs were neither coded nor reported in a comparable way by the reviews, a more detailed summary was not feasible.
The majority of intervention studies were socially embedded (111/182, 62%). School, university or college settings were mentioned in 45 studies, workplace in 37 studies, home and/or community-based study populations were reported in 17 studies, while two studies reported a combination of workplace and home setting. A social media setting was mentioned in eight studies, and supermarket and online setting in one study each. Two reviews [55, 56] examined whether the interventions involved social support from the setting or solely took place in this context. Social support through peers and/or friendly challenges was described in six studies [56] and parental involvement in three studies [55]. None of the reviews reported about the use of JITAI or EMI.
Overall effectiveness
The heterogeneity of the included studies concerning study type, outcome parameter, and assessment method was high. Thus, the overall effectiveness reported in the reviews is displayed in the following paragraph for any significant differences, which were found for the e/mHealth interventions over time or vs. a control group. Of all included studies, 10/182 did not report intervention effectiveness. The remaining 172 studies found a significant benefit for the intervention group over time and/or vs. a control group in 101/172 (59%) cases. No significant differences were found in 68/172 (40%) studies, and 3/172 (2%) resulted in a significant deterioration of the parameter over time and/or vs. control (see Table 1).
Effectiveness vs. Control
The between group differences for the included systematic reviews are displayed in the following chapters and the results of the included meta-analyses are reported in further detail.
PA
PA (i.e. time spent in different PA intensities, step count, PA frequency, PA goal achievement, school related PA, and leisure time PA) was assessed by seven systematic reviews (PA outcome in 106 studies) [46, 48–50, 52, 55, 56] and one meta-analysis (PA outcome in 20 studies) [51]. Of the 126 studies included in these reviews, 58 studies used device-measured outcomes, 52 used self-report (1 not validated), and 16 used a combination of both measures.
Systematic reviews concerning PA did not report group differences or did not use a control group in 14/106 studies. The remaining 92 studies found significant group differences in favor of the intervention group in 19/92 (21%) studies, temporary significant group differences in favor of the intervention group in 25/92 (27%) studies and 49/92 (53%) showed no significant differences between the groups. One meta-analysis [51] included participants aged from 8.4 to 71.7 years and found no significant pooled effects using a random effect model between the eHealth and a usual/minimal care group for total PA (seven studies, SMD = 0.14, 95 % CI [−0.12, 0.41]; Ι2 = 60 %), MVPA (nine studies, SMD = 0.37, 95 % CI [−0.03, 0.77]; Ι2 = 78%) and measures of walking (eight studies reporting steps/day and walking duration/day, SMD = 0.14, 95%CI [−0.01, 0.29]; I2 =0 %). Subgroup analysis between device-measured and self-reported results showed no significant differences in the eHealth group for total PA, MVPA and walking.
SB
SB (i.e. sitting time (overall and occupational), sedentary time (overall and occupational), screen time, and computer activity) was assessed by two systematic reviews (SB outcome in 13 studies) [47, 56] and two meta-analyses (SB outcome in 20 studies) [51, 52]. Of the 33 studies included in these reviews, 15 studies used device-measured outcomes, 16 used self-report (one not validated), and two used a combination of both measures.
The systematic reviews concerning SB included 4/13 studies which did not report group differences or did not involve a control group. The remaining nine studies showed a significant group difference in favor of the intervention group in 2/9 (22%) studies, 6/9 (67%) studies with no significant differences between the groups, and 1/9 (11%) reported a significant group difference in favor of the control group. The first meta-analysis (five studies) [51] which included participants aged from 8.4 to 71.7 years found a significant reduction of SB in favor of the intervention group using a random effect model. This pooled effect was negative and small (SMD = −0.26, 95 % CI [−0.53, −0.00]; I2 =0 %) with no evidence of heterogeneity. Subgroup analysis between device-measured and self-reported results showed no significant differences for the intervention group in SB. The second meta-analysis on SB (15 studies) [47] included only adults (20.4 to 64.1 years) and showed a significant pooled reduction of SB with a substantial heterogeneity (-41.28 min/day, 95% CI [-0.99, −21.58], I2 = 77%; n = 1402) in favor of the intervention group at the end point follow-up measurement using a random effect model. Analysis for device-measured (eight studies) results showed a significant pooled reduction of −35.07 min/day with a low heterogeneity (95% CI [-46.57, −23.57], I2 =21%; n = 595), while self-reported measures (seven studies) led to a significant reduction of −52.66 min/day with a considerable heterogeneity (95% CI, [−93.63, −11.69], I2 =88%; n = 807) at end point. The comparison between device-measured and self-reported results has not been conducted by this meta-analysis. The additional analysis of short-term measures for overall SB (less than 3 months, 10 studies) showed a significant mean reduction of −42.42 min/day with a substantial heterogeneity (95% CI [-63.21, −21.63], I2 =61%; n = 760), the medium-term measures (three to six months, five studies) showed a significant mean reduction of −37.23 min/day with a considerable heterogeneity (95% CI [-73.70, −0.75], I2 =85%; n =691) and the long-term measures (over six months, three studies) showed no significant mean reduction with a low heterogeneity (−1.65 min/day, 95% CI [-14.77, 11.47], I2 =23%; n =670).
HE
HE (i.e. FVI, vegetable intake, and healthy dietary choices) was assessed by one systematic review (HE outcome in 13 studies) [56] and two meta-analyses (HE outcome in 33 studies, focus on FVI) [53, 54]. All of the 46 studies included in these reviews used self-reported results, 10 of which were not validated.
The systematic review concerning HE did not report group differences or did not involve a control group for 1/13 studies. The remaining 12 studies found a significant group difference in favor of the intervention group in 2/12 (17%) studies, a temporary significant group difference in favor of the intervention group in 3/12 (25%) studies and 7/12 (58%) showed no significant differences between the groups. One meta-analysis [53] included young adults (M =20.8 years) and showed a significant increase in FVI (eight studies) calculated by a random effect model with a small pooled Cohen’s d of 0.22 (95% CI [0.11, 0.33]) and a substantial heterogeneity (I2 = 68.5%). Effects for vegetable intake alone were also assessed (five studies) and the pooled effect showed a negligible effect with low heterogeneity (Cohen’s d = 0.15, 95% CI [0.04, 0.28], I2 = 31.4%). The second meta-analysis [54] included participants of all ages (4.5 to 57.75 years) and found a significant increase of FVI in favor of the intervention group using a random effect model with a small Hedge’s g and substantial between study heterogeneity (g = 0.26, SE = 0.05, 95% CI [0.17, 0.35], I2 = 62.77). Subgroup analyses revealed that computer-based eHealth interventions (three studies) showed the largest effect (g = 0.44), followed by SMS interventions (three studies) with a Hedge’s g of 0.41, while internet-based interventions (nine studies) showed a Hedge’s g of 0.19 and CD-ROM, mobile apps and video game interventions (four studies) showed no significant improvements. The subgroup analysis relating to age groups yielded no significant differences between adults (11 studies), adolescents (four studies) and children (four studies). Interventions including adults and adolescents showed significant improvements in favor of the intervention group with Hedge’s g of 0.26 and 0.35 respectively, while interventions conducted with children showed no significant effects.
Determinants of effective Interventions
The extraction of effect sizes regarding the influence of theoretical foundation/BCTs, social influences and EMI/JITAs on the efficiency of e/mHealth interventions was not feasible so that only descriptive results were reported in this umbrella review (see Table 2).
*** please insert Table 2 about here ***
Theoretical foundation and BCTs were mentioned in all the included reviews [46–56]. One review [49] related to PA noted that 5/5 (100%) studies based on social cognitive theory led to significant differences over time or vs. control compared to 1/2 (50%) for theory of planned behavior and 1/1 (100%) showing a temporary significant difference directly after the intervention for transtheoretical model. Another review concerning PA [48] also found theory-based interventions more effective than those without a theoretical foundation. A third review concerning PA [55], which found that 6/9 (67%) theory-based interventions showed significant improvements of the intervention group over time or vs. control, while only 2/5 (40%) without a theoretical foundation led to such improvements, is in line with these findings. The inclusion of BCTs was associated with higher effectiveness of PA, SB and HE interventions in one review [56]. However, the question which BCTs are linked to effectiveness has not been answered by this review. Two meta-analyses [47, 51] reported the usage of BCTs for PA and SB interventions, but did not link the use of BCTs to effectiveness due to the small number of studies included. For healthy eating behavior, the use of BCTs was one key component of successful interventions, while the impact of using multiple BCTs remained unclear [53]. Further, a more recent meta-analysis [54] revealed that the inclusion of seven to eight BCTs (four studies) resulted in a statistically significant larger effect size (SMD = 0.42, SE = 0.10, 95% CI [0.21, 0.62], p < .001) than those involving four to six BCTs (seven studies) and one to three BCTs (seven studies). In a next step, the meta-analysis found no statistically evidence for specific BCTs yielding larger effect sizes.
The influence of a social settings concerning effectiveness has not been reported in detail by the included reviews and two reviews [55, 56] reported on the matter at all. The integration of eHealth interventions in school settings was reported to lead more often (6/9, 67%) to positive effects on PA or weight reduction in comparison to home-based interventions (2/5, 40%) [55]. Another possible influence on effectiveness mentioned in this review was parental influence [55]. The second review about mHealth interventions points out that efficient interventions often include social support related to peers and friendly team challenges among many other facets [56]. However, since both reviews did not report effect sizes, and there were a variety of other possible facets contributing to effectiveness, the magnitude of the potential influence for social settings remains unclear.
Since none of the reviews reported the use of EMI/JITAIs, the question concerning their effectiveness has to be left unanswered by this umbrella review.
Study Quality
Mean study quality of the included reviews as assessed by the AMSTAR tool was medium (M = 5,9/11) while one review scored high (9/11) [53]. The most common weakness was the reporting of the conflict of interest of the included studies (0/11) and providing a list of all included and excluded studies (1/11). For the score of every criterion see additional file 2. Risk of bias ratings conducted by the authors of the included reviews was mainly medium to high with some studies of low risk (see Table 3).
*** please insert Table 3 about here ***