The review was limited to assessing the effectiveness of the interventions using a randomized controlled trial (RCT) design to gather only high-quality studies about health care interventions [25]. We included studies targeted at adolescents or young people in the age range of 10 and 24 years [26] who had been diagnosed with depression or had experienced symptoms of depression or anxiety [27]. We focused on interventions that aimed to support mental health among adolescents by preventing, identifying, or decreasing the symptoms of depression or anxiety, or through counseling. The interventions were delivered via ICT, including Web-based technology [28], which could be accessed by computers, tablets, or mobile phones. The review focused on published (or in-press) articles written in peer reviewed journals and published in English. The primary outcome used was depression, and the secondary outcomes used were anxiety, stress, moods and feelings, leaving the study early (attrition rate), and costs.

We excluded dissertations, letters, editorials, literature reviews, book reviews, and book chapters, in addition to studies with designs other than RCT. Study protocols of specific studies were searched for manually and used to verify possible risk of biases. Studies were excluded if the intervention was targeted at adults or persons under 10 or over 25 years old, parents, teachers, or health care staff. If the primary focus of the intervention was something other than depression, such as brain injury, eating disorder, or epilepsy, or if the intervention only included texting, the study was excluded.

We conducted a comprehensive literature search on September 1, 2015, and an updated search was done on February 10, 2017. Four electronic databases covering published research from the health and social field were investigated: MEDLINE, PsycINFO, CINAHL, and Cochrane. A combination of medical subject headings and text-based search terms [29] for the databases was used. The search was conducted with the help of an information specialist. Due to changes in user interface in databases between the original and updated search, a search for MEDLINE, PsycInfo, and Cochrane was combined via Ovid (in 2017). Electronic databases used, search terms, and number of hits are documented in Multimedia Appendix 1. For additional references, we consulted the references in the included studies. Relevant systematic reviews were identified through electronic searches to avoid overlap between previous studies.

When choosing the selection of studies [22], first, two authors (KA, MA) independently screened all titles of abstracts that were relevant to this systematic review. Second, the abstracts were screened for eligibility. Third, the full papers of the included abstracts were screened for inclusion and exclusion criteria. In cases of discrepancy, the papers were discussed with MV until a consensus was reached. On the basis of the assessment process, the abstracts were included first in the narrative synthesis and later in the meta-analysis (see Figure 1) based on specific criteria, which are recorded in Table 1.

We created the data extraction table matrix to collect and describe information, focusing our aims in the synthesis. Data extraction involved describing the included studies and interventions, as well as the excluded studies. For the descriptions of the included studies, the information was collated by authors, year of publication, country of origin, purpose of the study, setting, target group, age, total number of participants randomized, and the number allocated in each study group. Information was combined if the publication (a hit) was based on original data published in more than one paper and the identification was based on the study protocol number. These studies were described in the data extraction tables (matrixes) as one study. The data from 22 included studies (27 hits) for narrative analysis were entered into the specific data extraction grid, where each study was treated as a separate case, and descriptive characteristics of the studies were categorized manually (see Multimedia Appendix 2).

The interventions were extracted to a matrix table based on the TIDieR checklist and guide [24] (see Multimedia Appendix 3). The categorization was done under the following themes based on the thematic analysis of Braun and Clarke [41]: (1) materials and procedures; (2) provider and modes of delivery; (3) location, dose, and length of the program; and (4) tailoring of the intervention, modifications, and fidelity.

For the meta-analysis, a summary of outcome measures used in 15 studies is described in Table 2. The meta-analysis was undertaken using Review Manager RevMan version 5.3 (Nordic Cochrane Centre, Cochrane Collaboration, 2014) for preparing and maintaining Cochrane reviews [42]. For continuous outcomes, mean differences were compared between groups. When similar scales were used, presuming that there would be only small differences in measurement tools, measurements were combined. This decision was made to answer the overall question of whether there is evidence that Web-based interventions can be an effective intervention among young people with depression or depressive symptoms [22]. Standard deviations were used with the sample sizes to compute the weight given to each study. A random-effects analysis was used instead of a fixed-effect method, as the former allows the outcomes of studies to vary more than the latter does; a random-effects analysis can be seen as a more natural way of explaining outcomes [43].

In cases of missing or incomplete data, there was an attempt to contact the authors of the study in question. However, as no replies were received from these attempts, available data was used. Heterogeneity was assessed by calculating the I² index. If the estimated I² was greater than or equal to 50%, it was interpreted as indicating the presence of high levels of heterogeneity [22].

The quality of the 15 studies included in the meta-analysis was appraised by using a tool measuring risk of bias from Review Manager (RevMan) version 5.3 [60] with the following criteria [22]: random sequence generation, allocation concealment, blinding of participants and personnel, blinding of outcome assessment, incomplete outcome data, selective reporting, and other bias (Figure 2). For each study, KA and MV classified the domain as having a low, high, or unclear risk of bias. To minimize the risk of publication bias, bibliographic databases and trial registries were consulted to compare original review plans and outcomes reported. Any discrepancy between the two review authors was resolved through discussion (KA, MV, ML).

A sensitivity analysis was conducted by excluding studies (1) with a sample size that vastly differed from other studies or (2) if one or more of the studies had low-quality issues affecting the study results.

We screened 2087 hits of abstracts, which were identified through electronic databases. On the basis of the screening, we removed 1458 duplicates. After assessing their eligibility, 1398 abstracts were excluded, which left us with 60 abstracts. Sixty full papers were then retrieved for potential inclusion in the study, and their eligibility was assessed based on our inclusion and exclusion criteria. We excluded 33 abstracts, which left us with 27 paper hits (22 studies) included in the narrative synthesis. Overall, 16 paper hits (15 studies) were included in the meta-analysis (see Figure 1).

In 22 eligible studies (27 articles), a variety of Web-based interventions were used (see Multimedia Appendix 3). The most common background approach used was CBT. Related to materials and procedures, the interventions were composed of modules, sessions, or lessons, with a variety of themes of background theories. ICT-based interventions used interactive games, online chats, mobile phone apps, and emails. The participants were offered activities to support their progress, such as homework assignments or exercises, skill training, workbooks or guided work, quizzes, and questionnaires. Interventions were provided by various professionals such as teachers, school counselors, research team members, project coordinators, or health care personnel. The programs were delivered on websites through computer software via compact disc read-only memory, mobile phone apps, or emails. They were offered at schools, in health care services, or in community settings. The interventions could include up to 14 modules that lasted from 3 to 10 weeks, typically done once per week. The time spent on the programs ranged from 20 min to 3 hours per week. Regarding the fidelity of the intervention, the participants’ completion of the intervention varied between 10% and 94%.

Most studies (12/15) included in the meta-analysis had a low risk of selection bias in random sequence generation. Half of the studies (7/15) had a low risk in allocation concealment, whereas in 4 studies (4/15), the risk was high. The risk involved with blinding participants and personnel (13/15) and outcome assessment (11/15) was low in most studies. More concern was raised regarding attrition bias (6/15 had a high risk) and especially, selective reporting (8/15 had a high risk). Out of 15 included studies, a published trial registration or a protocol was not found for 3 studies (Figure 2).

For the primary outcome, a meta-analysis was performed involving 10 studies [44,47-51,53-55,57]. We compared Web-based interventions with the control groups of the studies by investigating the short-term effects of the interventions on depressive symptoms. This analysis (postintervention measurement) showed statistically significant improvements in the Web-based intervention groups (P=.02, median 1.68, 95% CI 3.11-0.25). However, only 4 of the studies [36,38,40,54] compared the effects on depressive symptoms regarding mid-term effects (follow-up measurements after 3-5 months). No statistically significant improvements in the Web-based intervention group were found in these comparisons (P=.08, median 2.91, 95% CI 6.19-0.36).

We further assessed the long-term effects of the Web-based interventions. Out of 10 studies, we found 3 studies [44,49,54] that assessed the long-term effects (6 months or longer). As for short-term effects after intervention, statistically significant improvements were found in adolescents’ depression scores in the Web-based intervention group (P=.01, median 1.78, 95% CI 3.20-0.37).

Substantial heterogeneity was found in the short-term and mid-term effects, but regarding the long-term effects, heterogeneity was only at a moderate level (see Figure 3).

Anxiety symptoms were assessed in 8 studies comparing short-term effects of Web-based interventions to control groups [44,48,49,51,53,55,57,58]. Statistically, significant improvements were found in the Web-based intervention group in the short term (P=.001, median 1.47, 95% CI 2.36-0.59). However, for the mid-term assessment (follow-up measurements after 3-5 months), only 2 studies evaluated the effectiveness of a Web-based intervention for anxiety symptoms [48,53], and no statistically significant improvements in the symptoms were found (P=.36, median 1.42, 95% CI 4.45-1.62; see Figure 4). None of the studies measured the effectiveness at the 6-month mark.

Tests evaluating heterogeneity showed that for short-term and mid-term effects, heterogeneity was on a moderate level (see Figure 4).

An analysis of stress outcomes was performed (3 studies) [49,51,55] to compare the effectiveness of a Web-based intervention on stress symptoms with that of a control group. The postintervention comparison showed no statistically significant short-term improvements in the intervention group (P=.14, median 1.06, 95% CI 2.44-0.33; see Figure 5).

Heterogeneity tests showed that heterogeneity was on a moderate level in short-term effects (see Figure 5).

A meta-analysis was performed on 2 studies [53,58] to compare Web-based interventions with control groups with regard to short-term effects on moods and feelings. These comparisons (postintervention measurement) showed some statistically significant improvements in the Web-based intervention groups (2 studies; P=.04, median 5.55, 95% CI 10.88-0.22). Heterogeneity tests showed that heterogeneity was at a considerable level in the short-term effects (see Figure 6).

Regarding the secondary outcome, leaving the study early, a meta-analysis was performed on 11 studies [44,47-50,52-55,58,59]. Postintervention measurement comparisons of short-term effects on leaving the study early regarding showed statistically significant results favoring the control group (P=.007, median 1.31, 95% CI 1.08-1.58). In an assessment of the mid-term effects (follow-up measurements after 3-5 months), 3 studies were compared [48,50,53]. Again, a statistically significant result favored the control group, showing that young people left the study earlier in the intervention group (P=.02, median 1.65, 95% CI 1.09-2.49). Heterogeneity tests showed substantial heterogeneity both in short and mid-term effects (Figure 7). An analysis of long-term effects was not possible because of missing data.

Our review showed fluctuations in the results of the effectiveness of the Web-based interventions; there was statistically significant improvement in the short-term and long-term (over 6 months with 2 studies) effects regarding adolescent depression but not in mid-term effects. Furthermore, in scores regarding anxiety and moods and feelings, a statistically significant improvement was found after the intervention but not in follow-up measurements. The fluctuation in the depression scores and a lack of significant findings may be a result of the small number of studies that included follow-up measurements. More studies with longer follow-up periods should therefore be conducted to produce clinically significant evidence on the long-term effectiveness of Web-based interventions.

This review is pertinent because the effective interventions for supporting adolescent health are an investment in public health and the future (63). As 92% of adolescents use the Internet daily (5), Web-based interventions could offer solutions to problems in seeking help with depression (1). In line with previous studies related to psychological therapies [61], we assumed that Web-based interventions might have positive effects on stress levels among young people. However, we were unable to fully confirm our hypothesis. Previous studies have found relief in adolescents’ stress, if the intervention included psychological therapies with face-to-face contact [62]. Contrary to personal contact, Web-based interventions are often self-directed [9,44,52] or self-guided [33]. We can, therefore, ask whether the lack of regular human contact produced less-effective results. Indeed, interventions in our study that favored the intervention group included face-to-face guidance, monitoring of engagement, or follow-up telephone calls by teachers and health professionals (eg, [44,46,49]). We also found that participants in the intervention groups left the study early more often, indicating that they may not have been fully engaged in these Web-based interventions [9].

A comparison of costs of stand-alone online interventions and that of personal communication has not been analyzed thoroughly enough. In general, the use of effective interventions for supporting adolescent health has been seen as an investment in the future of public health [63]. The promises of cost savings with the use of information technology could rely on the fact that most adolescents are already frequent Internet users [5], which could save high investment costs. Its features could be translated into health services and offer easy access [1,3], safeguarded anonymity [4], and opportunities to receive help without the fear of stigmatization [6]. However, the Web-based interventions found in our narrative analysis varied greatly, with diverse background approaches, materials and procedures, providers, delivery types, dosages, and lengths of the intervention. Descriptions of interventions have been of poor quality, which has limited the possibility of comparisons, intervention replication, and the usability of study results in practice [24].

Many studies rely on the opportunities of health technologies for better health outcomes and decreased health costs [7,8]. Our review, however, did not reveal any studies assessing the costs of Web-based interventions. This result is noteworthy because huge investments are currently being made in the development of technological solutions for health services. The World Bank [64] has already screened more than 500 mobile health studies and concluded that evidence regarding the best strategies for effectiveness of the interventions and engaging the users in these interventions is still missing. After our review, the knowledge about the impact of the Web-based interventions remains controversial. Therefore, there is a need for a comprehensive impact evaluation that would show the costs and benefits of Web-based technology in the health sector.

In addition to follow-up periods, larger sample sizes and more rigorous study designs could increase the quality of the research. Specific Web-based interventions instead of packages of intervention could also provide a more feasible and accurate conception of the factors impacting the outcomes. Furthermore, more studies are needed in the future to gain a deeper understanding of why adolescents are eager to leave the study and why their engagement in information technology interventions is low.

The results of this study should be considered in the light of its limitations. We only included papers from scientific journals that had been written in English, which may have caused relevant studies written in other languages or existing in gray literature to have been left out [65]. Our review is, therefore, biased toward positive results and western countries. Publication bias may have also affected our results. This review may potentially favor results that have been deemed statistically significant. In addition, a number of studies included in the meta-analysis focused on specific outcomes, which may affect the reliability of some results. Moreover, the heterogeneity of our meta-analysis was high (I² ranged between 0% and 89%). Although I² is not a measure for absolute heterogeneity, it may refer to the high variation in some outcomes between studies [66]. Another point to consider is that the interventions used were more like “packages of interventions,” which included many different elements. As an outcome, it may be difficult to identify which specific elements influenced the effectiveness of the interventions. Furthermore, based on the sensitivity analysis, problems pertaining to the heterogeneity of interventions were identified. We must consider that participants or severity of depressive symptoms could have varied greatly among the studies included in the meta-analysis, which jeopardizes the results of the review [67]. All these issues should be taken into consideration when interpreting the results.

In conclusion, the principal finding of this review supports the evidence that Web-based interventions are effective in the short term in decreasing depressive and anxiety symptoms and improving moods and feelings among adolescents and young people. The review also indicates that adolescents are not fully engaged in using Web-based interventions. Instead of simply stating that “more studies should be done in this area,” we assert that more critical thinking is needed to understand to whom information technology might be useful, which components or characteristics of interventions make it more effective, and what role human contact in conjunction with information technology may play in engaging and supporting young people with mental health concerns.