Figure 1 shows the studies selection flow diagram. The initial search led to 7039 potentially relevant citations. After screening titles and abstracts, 60 publications were kept for further assessment, of which 10 articles were finally retained for the review (24, 42-50). An updated search ran in May 2018 resulted in the inclusion of three additional peer-review publications (51-53) and one thesis (54).
In total, 21 peer-review articles and one thesis —stemmed from 14 studies— were finally retained. The list of excluded publications and reasons for exclusion is provided in Appendix (Supplementary file 3).
The characteristics of the 14 included studies are presented in Table 1. Five studies were conducted in the USA (24, 43-45, 54), one in Japan (52) and the remaining eight in Europe (42, 46-51). Among European studies, three were conducted in the Netherlands (48, 49, 51), two in the UK (47, 53), and one in three countries (Greece, Spain and Sweden) (42). Furthermore, three studies presented different parts of their results in distinct publications: (i) Wijsman et al. (51, 55, 56); (ii) van het Reve et al. (50, 57, 58); and (iii) Peels et al. (48, 59-61).
When appraising the quality of the retained studies, we first noted that great variation existed with regard to sample size with a minimum of 14 and a maximum of 1729 participants. Seven studies had small samples (n<100) ― for a total number of 3645 participants aged between 50 and 88 years old. Also, in the majority of studies, the samples comprised more women than men. Second, as shown in Figure 2, the risk of bias was moderate to high across the studies, but the source of bias varied. The blinding (participants, personnel or outcomes assessor) bias was present at a high risk or unclear in most studies. Sequence generation and allocation concealment were variable among the studies, which means that the potential selection biases were foreseeable in half of the studies. Finally, potential risk of bias related to incomplete outcome data and selective outcome was low in a large majority of studies, meaning that there was a low risk of reporting bias. We also noted the use of a wide range of validated questionnaires (such as: Quality of life: RAND 36, Physical and Psychological well-being: SF-36; Wellbeing: SPF-IL scale) and non-validated rating scales (such as: behavioral change self-assessment, IT literacy, engagement in activity) to assess the impact of the interventions.
Focus area of the technology
All the included study interventions were primarily internet-based. These interventions were often compared with either paper-based interventions, interventions with a videophone component, mixed intervention; tailored or not. The technology devices that were part of the interventions consisted mainly of computers, tablets or mobile phones. In reference to the Center for Technology and Aging classification (39), three areas are represented in this systematic review: remote patient monitoring, remote training and supervision, and social networking. One study consisted of an educational program with telemonitoring of step count, blood pressure and body weight (52). Most studies aimed to detect, train and supervise patient remotely. One intervention was personalized with participants’ information provided during the use of the web-based intervention (53), other interventions included information provision to increase daily physical activity (50, 51), or through a Web site with a tailored advice to undertake strength and balance training (47). Finally, two studies evaluated social networking: one focused on Facebook and the use of an online diary (54), the second on an ICT-mediated social network (42).
As for the remaining studies, Cook et al. (24) focused more widely on health promotion goals (diet, physical activity, stress, tobacco use), whereas Slegers et al. (49) and van het Reve et al. (50) focused on computer training and internet usage. Lastly, Homma et al. (52) focused on information technology literacy.
With respect to the outcomes, the majority of included studies (11/14) focused on physical activity (PA) (24, 43-48, 50-53) with some focusing on the effect of physical activity on metabolic health and quality of life (51, 52) and another covering increasing healthy behavior (24). The three other e-Health interventions targeted multiple dimensions including cognitive function, wellbeing, social engagement or connections, quality of life or lifestyle modification (42, 49, 54).
Effects of e-Health on healthy behavior outcomes
The most often reported outcome in the included studies was physical activity (PA). Peels et al., comparing paper-based and web-based intervention on PA, concluded that the former was effective in increasing weekly days of sufficient PA (p=0.005) at baseline and 6 months later (p=0.042) (48). In similar vein, Irvine et al. showed that a web-based intervention to promote PA improved 13 of the 14 outcome measures and the intervention group maintained large gains on all 14 outcomes measured at 6 months (43). In the Mouton 2015 study, a mixed intervention (center- and web-based intervention) led to improvement in PA level (p=0.041), readiness for PA (p=0.001), and improved the awareness of PA (p=0.003) (46). In a trial using text messaging, Kim & Glanz contended that motivational text messaging (3 times/week) increases step count (679 vs. 398, p < 0.05) as well as perceived activity level (p < 0.05) (44). Using a tablet intervention, van het Reve et al. (50) showed improvement in physical performance for all groups (p: 0.02) compared to the brochure group in the single and dual task walking (p=0.03), as well as the falls efficacy (p=0.04) (50). Likewise, an internet-based moderate-to-vigorous PA intervention of Wijsman et al. (51) led to a significant improvement of weight and waist circumference (p=0.001). Finally, Homma et al. (52) reported an improvement in steps per day for both videophone intervention (interactive communication) and document groups (p < 0.01).
In a trial testing the addition of a monetary incentive to an Internet intervention, Kurti & Dallery concluded to a higher percentage of goals achieved (87%) in the group that received the monetary motivation (45). Nevertheless, some studies were unable to find any significant difference in the PA outcomes targeted. For instance, Lara et al.’s pilot study showed weak and non-significant differences between both groups for PA (53). However, we should not conclude in the absence of effect for this intervention, as the study was not sufficiently powered.
Effects of e-Health on clinical parameters outcomes
The study by Wijsman et al. comparing Internet-based PA intervention versus no intervention, concluded to a significant improvement in clinical parameters, including insulin and HbA1c (p < 0.001), this for moderate-to-vigorous PA (p = 0.001) (51). Likewise, Homma et al. found significant improvements for blood pressure, HbA1c, and albumin when comparing videophone intervention group to document group (52).
Effects of e-Health on psychological outcomes
Regarding the psychological outcomes, in the Nyman et al. study (47), receiving a web-based tailored advice led to higher ratings of the advice relevance (p = 0.017) and goodness of fit of activities (p = 0.047). Besides, Wijsman et al. (51) demonstrated that the Internet-based PA intervention improved the emotional and mental health (p: 0.03) and health change (p < 0.01) in their measure of quality of life. In the Slegers et al. study, however, using computers and the internet did not influence quality of life, well-being and mood, nor the social network of healthy older individuals (49).
For their part, Ballesteros et al. found that an ICT-mediated social network improved the affective dimension of wellbeing in their quality of life scale at post-test (p < 0.05) (42). Similarly, Myhre et al.’s Facebook intervention improved Knowledge (p < 0.01), as well as Letter Memory task (p < 0.01) (54).
Effects of e-Health on other outcomes
Cook et al. (24) showed that their web-based multimedia program (information and guidance) had a significant effect on diet behavioral change self-efficacy (p = 0.05), planning healthy eating (p = 0.03), eating practices (p = 0.03), exercise self-efficacy (p = 0.03), exercise planning (p = 0.03), and aging beliefs (p = 0.01). In the Peels et al. study (48), the process outcomes showed that the printed group significantly performed better in reading (92.7–98.2%), keeping (70.1–76.5%), and discussing (39.9–56.8%) the advices received. Furthermore, the printed intervention was better appreciated than the web-based intervention (scores 6.06–6.91 versus 5.05–6.11, respectively on a scale of 1–10) (48). Moreover, Homma et al. (52) showed a significant positive change in self-assessment of PA (p = 0.004), diet (p = 0.002), and lifestyles (p = 0.005). Participant satisfaction using IT-related devices was significantly higher in the intervention (videophone) group than in the control group (printed documents) (40% vs 15%).
Outcome synthesis and assessment of the certainty of the evidence
Due to the important heterogeneity in the studies, it was not possible to conduct a meta-analysis for the outcomes of interest. However, following the SWiM guidance (37), we computed the effect estimates for PA as it was the most frequent outcome reported in the studies. Figure 3 shows the effect size and corresponding 95% confidence interval (CI) for the studies that documented the effectiveness of eHealth on PA. However, some of these studies did not provide sufficient information to calculate the effect size (24, 53) or the CI (46, 50).
We assessed the certainty of the evidence based on the GRADE approach (64) considering the within‐trial risk of bias, indirectness, heterogeneity, imprecision and other considerations (Table 2). As it was not possible to pool the data for most outcomes, we considered the evidence provided by the individual trials as a whole to illustrate the level of evidence for each main category of outcomes. For all outcomes, the certainty of evidence is considered to be very low, mostly due to the risk of bias in individual trials and the imprecision of the estimates given the small sample sizes.