Electronic searches were conducted in MEDLINE, PsycINFO, and the Cochrane Controlled Trials Register. Medical Subject Headings (MeSH) and relevant text word terms were used to identify relevant studies (see Search Strategy). The last search was run on September 1, 2016. Reference lists of systematic reviews and articles identified were manually checked for relevant entries.

Search terms for depression were used to scan all trials registers and databases outlined previously. Additional terms for a range of delivery methods (eg, online, Internet, Web, computer, phone) and terms that specify the type of intervention (eg, cognitive behavioral, psychodynamic, interpersonal, psychoeducation) were applied. Further search terms were utilized to limit the search to studies of therapeutic interventions (eg, therapy, psychotherapy, intervention, treatment) and to RCTs. Figure 1 gives an overview of the terms used in this literature search.

As a consequence of the protocol also including the evaluation of other mental disorders, the search strategy was refined during the course of the review to limit our study to depression. A compilation of the preliminarily defined search terms is given in Figure 1.

The process of study selection required an eligibility check for each article identified. Eligibility of studies was assessed by reviewing the abstracts of the references identified by the search strategy. Full texts were additionally screened when necessary. In case of doubt, any disagreements and ambiguous articles were discussed among the authors, and eligibility of studies was decided by consensus.

Data collection tables were predeveloped and subsequently refined during the process of data extraction. The following information were collected from every article.

Each study was summarized in detail in the predeveloped data extraction table. The primary outcome of the BDI and PHQ was assessed as a continuous measure of effect in an additional table. Because moderate to substantial heterogeneity among the interventions was expected, mean effect sizes were calculated using a random-effects meta-analysis according to DerSimonian and Laird [69]. Review Manager 5 (RevMan 5) was used to conduct this systematic review [77]. In general, a P value of <.05 was considered statistically significant.

The searches in MEDLINE, PsycInfo, and the Cochrane Controlled Trials Register identified a total of 6387 citations (articles and abstracts) published after 2000. After the adjustment for duplicates and the exclusion of noneligible trials based on titles and abstracts, 491 studies remained. Forty-two additional possibly eligible trials were identified by checking the reference lists of relevant articles already identified. A more detailed review of the full text of the remaining citations led to the detection and exclusion of 130 publications. Thirty-four trials were excluded because of the lack of appropriate reporting of outcomes. It was decided by consensus to exclude two additional studies that included an active control group that only differed from the study group in the use of a program component that was not relevant to this review. In total, 45 RCTs were included. Of these, seven trials analyzed patients with depressive symptoms comorbid with anxiety. Figure 2 summarizes the study selection process.

Tables 3 and 4 provide information on the setting and outcomes of the 45 trials finally included in the analysis. These RCTs contributed a total of 7326 randomized and 4519 analyzable patients. The majority of trials studied CBT in adult patients with mild-to-moderate depressive symptoms.

The risk of bias at the level of the individual trials is addressed in Table 2 by reporting the modified Jadad score [19] and whether analysis was performed according to the intention-to-treat principle (see Assessment of Methodological Quality). None of the trials were described as “double-blind RCT” and we also noted a relatively high risk of bias due to the insufficient blinding of participants. Typically, the method of blinding was described insufficiently. This, however, is not uncommon for psychotherapy trials [67]. In general, achieving a successful blinding in psychotherapy trials is regarded to be more demanding than in pharmacological trials [89]. Study participants can easily identify the discrepancies in the contents of the treatment and control arms, and it is probably not very likely to successfully blind the participants or the therapists. Thus, the methods used to achieve proper blinding are rarely reported in psychotherapy trials [67]. The risk of bias introduced by selective reporting was small because all outcomes of interest were adequately described in the vast majority of the included RCTs.

In total, a set of 15 system components was identified based on occurrence in reviewed literature. These were either defined, hypotheses-driven, or derived from Oinas-Kukkonnen’s model of persuasive systems design [18]. For each study included in the systematic review, we determined the presence of defined system components for later analysis. In the following, we present an overview of identified components, together with the underlying inclusion rationale.

Table 5 contains detailed information on the settings and system component configuration of the interventions.

Approximately 45% (20/45) of all included studies reported using email reminders and 10 of 45 studies (22%) reported providing SMS text message reminders, mostly as an alternative to reminders sent by email. In the included trials, reminders were typically intended to increase motivation and adherence to therapeutic interventions. As explained earlier, reminders play a decisive role in the engagement of users in mental health interventions. Most of the RCTs made use of the Internet for delivering mental health interventions for depression or depression comorbid with anxiety. The majority of RCTs included in this systematic review (91%, 41/45) did not make use of mobile phones or tablets. Typically, the interventions required interaction with the system, and many also included interaction with a therapist (face-to-face or online) and/or peers on the Web. In all, 80% (36/45) of included programs were based on CBT or used elements of CBT. Tunneling, which refers to the stepwise delivery of content, is typically found in technology-based interventions for depression [18], and was also used in the majority of studies (87%, 39/45) included in this work. Twenty-seven of 45 included studies used tailored content, tailored feedback, and/or tailored reminders.

Only a small number of included studies made use of self-monitoring components, such as symptom tracking and tracking reminders, yet they are seen as key features of psychotherapy in particular [40,96], and in behavior-change interventions in general [12,115-118].

Although social support is widely recognized as an important feature in behavior change [119,120], in this analysis, only seven studies (16%) used peer support. In the included studies, social support consisted of the use of Web-based discussion boards (asynchronous social support), which was intended to provide the ability to connect with other patients of the same intervention.

Regarding visual attractiveness, 44% of all included studies analyzed an intervention that included visually appealing content, such as graphs, illustrations, comics, or photos, which often serve a motivational purpose. Only 11 of 45 studies (24%) described the utilization of audio and/or voiceovers. Video footage, often containing case-enhanced learning, was also found in 11 RCTs (24%).

In addition, this analysis showed that game elements were only used in three RCTs and, if gaming was included, it was in the form of knowledge quizzes. See Figure 3 for a quantitative overview of components among included studies.

Data from 45 trials (4519 patients) that reported on BDI or PHQ scores before and after the treatment were combined to estimate the overall effect of technology-based interventions on depressive symptomatology. Technology-supported treatments for depression showed a trend toward reduced depressive symptoms (SMD=–0.58, 95% CI –0.71 to –0.45; P<.001) (Figure 4). The chi-square test for heterogeneity (χ²₄₄=183.3), across all patient types, was significant (P<.001) and total variation across trials due to heterogeneity (I²) was 76%, a strong indication for inconsistency across included RCTs.

The funnel plot drawn for the main analysis on the effectiveness of technology-delivered interventions for depression showed an asymmetry, which is evidence of missing studies suggesting publication bias (Figure 5). Additional testing utilizing the Begg and Mazumdar rank correlation method [88] confirmed that there is, in fact, evidence of publication bias (P=.03).

As a consequence of this finding, we also explored publication bias in the 22 RCTs that used a WLC group. The funnel plot of this subgroup of included trials showed no evidence of bias, which was confirmed by the Begg test (P=.08).

Subgroup analyses by study quality, treatment duration, provision of face-to-face contact, utilization of CBT techniques, severity of baseline depression, age of participants, and type of control (eg, TAU or WLC) were prespecified and served to test our hypotheses (see Subgroup and Sensitivity Analyses ). The individual forest plots per subgroup analysis can be found in Multimedia Appendixes 1-7.

Differences in study findings could also be explained by biased results due to differences in quality of individual studies. Thus, a subgroup analysis based on the methodological quality of included trials was performed. It could be shown that the effect of technology-based interventions on symptoms of depression is consistent in trials of higher (quality score >3.5) and lower quality (quality score ≤3.5). In high-quality trials, technology-based interventions were associated with an SMD of –0.60 (95% CI –0.76 to –0.45, P<.001; I²=77%) and a SMD of –0.53 (95% CI –0.77 to –0.29, P<.001; I²=76%) in low-quality studies. It seems that methodological quality of trials only showed a small impact on effect sizes, but overall had a moderate effect (χ²₁=0.3, P=.60; I²=0%).

As statistical heterogeneity was found and to exclude the possibility of heterogeneity due to differences in the duration of the interventions, trials of different durations were compared to one another. Regarding depressive symptomatology, technology-based interventions showed to be effective, irrespective of treatment duration. In treatments with a duration of 10 weeks or less, the intervention was associated with an SMD of –0.60 (95% CI –0.76 to –0.44, P<.001; I²=79%). Similarly, treatments with a duration of longer than 10 weeks resulted in an SMD of –0.52 (95% CI –0.70 to –0.33, P<.001; I²=48%). There was no evidence for an association between duration of treatment and the effect of treatments on depressive symptomatology (χ²₁=0.4, P=.50; I²=0%).

As statistical heterogeneity was found in our analysis of treatment effectiveness and in order to exclude the possibility of heterogeneity due to the provision of face-to-face contact, trials that incorporated face-to-face support were compared to interventions that did not use live contact with a therapist. We hypothesized interventions that included face-to-face sessions would show greater reductions in depressive symptoms than treatments that did not. Contrary to what was expected, effect size in treatments that did not use face-to-face support was larger (SMD –0.65, 95% CI –0.79 to –0.50, P<.001; I²=78%) than in interventions that offered live support (SMD –0.28, 95% CI –0.52 to –0.03, P=.03; I²=50%). An association between the effectiveness of treatments and the provision of face-to-face contact was found (χ²₁=6.34, P=.01; I²=84%).

We made the hypothesis that technology-based interventions using CBT techniques are more effective than treatments that do not use components of CBT, reflecting on the predominance of CBT in the literature. Effect size varied only slightly between trials that were based on CBT (SMD –0.58, 95%CI –0.72 to –0.45, P<.001; I²=71%) and interventions that did not use a CBT protocol (SMD –0.56, 95% CI –0.96 to –0.16, P=.006; I²=85%). Also, according to the test for subgroup differences, there was no association between the effect of technology-supported treatments on depressive symptoms and utilization of CBT methods (χ²₁=0, P=.90; I²=0%).

Comparing trials in which patients showed higher depression scores (BDI ≥25 or PHQ ≥15) with trials in which patients had lower scores (BDI <25 or PHQ <15), technology-based treatment showed to be effective in both groups (higher severity: SMD –0.61, 95% CI –0.79 to –0.44, P<.001; I²=77%; lower severity: SMD –0.54, 95% CI 0.75 to –0.33, P<.001; I²=75%). The test for subgroup differences resulted in a chi-square value of 0.3 (df=1, P=.59; I²=0%).

Age did not show a significant impact on effectiveness of technology-based interventions for the treatment of depression. Although effect sizes varied between adults (SMD –0.63, 95% CI –0.79 to –0.46, P<.001; I²=79%), adults excluding older adults (SMD–0.48, 95% CI –0.69 to –0.27, P<.001; I²=55%), older adults (SMD –0.53, 95% CI –1.52 to 0.46, P=.29; I²=84%), and adolescents (SMD –0.28, 95% CI –1.45 to 0.88, P=.63; I²=91%), subgroup analyses pointed in the same direction (see Table 1 for a description of the different age categories). The effect of technology-based interventions on symptoms of depression was not associated with patient age (χ²₃=1.4, P=.71; I²=0%).

Comparing technology-based treatments to TAU resulted in a moderate effect (SMD –0.48, 95% CI –0.78 to –0.18, P=.002; I²=62%) and the comparison of technology-supported interventions for depression to WLC showed a large effect (SMD –0.79, 95% CI –0.93 to –0.64, P<.001; I²=51%). The test for subgroup differences resulted in a chi-square value of 3.3 (df=1, P=.07; I²=69.3%).

Forty-five publications with a total number of 7326 randomized and 4519 analyzable participants were included in this systematic review, and most of the interventions analyzed were able to reduce symptoms of depression. The majority of included studies were of fair (60%) to good (33%) quality, and almost every study included analyzed an intervention that was modular in setup and typically lasted for approximately 10 weeks. Thirty-six studies (80%) deployed a CBT approach. This extends the systematic review by Saddichha et al [8], which consisted of 29 RCT studies using CBT. Usually, the programs were aimed to be used about once a week. This is in line with traditional CBT, which is seen as a step-by-step, short-term treatment with weekly or biweekly therapist sessions for 10 to 20 weeks [17].

Subgroup analysis showed that study quality, treatment duration, provision of face-to-face contact, utilization of CBT, and age of participants had relatively small impact on the outcome of the interventions. For study quality, it is plausible because the intervention quality is not inevitably reflected by the study quality. Likewise, there are reasonable explanations that the age of the participants and treatment durations did not have a decisive impact on treatment outcome. We probably overestimated the influence of technology literacy in older participants because our results confirm findings of literature showing comparable treatment results for all age groups. The small difference in effect for treatment durations can also be explained taking into account that traditional therapy has a similar range of time spans to deliver the same amount of structured information and is chosen depending on, for example, severity of illness and level of support [122]. We were surprised that provision of face-to-face therapy did not show significant effects on the treatment outcome. In fact, this confirmed a recent study comparing traditional therapy with a computerized intervention [65].

Interestingly, the use of CBT components also did not show significant effects on treatment outcome compared to interventions that were not based on CBT. Although there exists evidence that other types of intervention can be equally effective, predominance of CBT in traditional therapy as well as in technology-mediated MHIS urged us to test its superiority in our analysis.

As explained earlier, reminders play a decisive role in the engagement of users in mental health interventions. In this systematic review, approximately 45% of all included studies reported to use email reminders and 10 studies (22%) reported providing phone reminders, mostly as an alternative to reminders sent by email. In the included trials, reminders were typically intended to increase motivation and adherence to therapeutic interventions.

With respect to the design and effective use of reminders, research indicates that a variety of factors show an impact on the efficiency of these alerts. Firstly, it was found that there is a high risk that motivational emails provided within a workplace setting are easily ignored as a consequence of a full email inbox [123,124]. Thus, it might be of importance to offer the possibility to choose between mobile phone or email reminders and to customize time points at which users will be reminded to complete program modules. Secondly, regarding the content of expert-initiated contact, it is postulated that contacts delivering behavior-change techniques might be of greater effectiveness than simple messages that prompt users to access the intervention [10]. Thirdly, it is suggested that reminders containing short motivational messages, quotes, or facts might counteract negative feedback cycles that maintain perceptions of low self-worth and associated depressive symptomatology [11,125]. Lastly, it could also be shown that the personalization of reminders and sending them out frequently enhances the effectiveness of treatment [98]. Nevertheless, nonspecific factors, such as encouragement, empathy, and hopefulness of improvement, may also independently enhance therapeutic gains. To conclude, it seems evident that creating a sense of being continuously supported and encouraged is crucial for user engagement, treatment adherence, and buffering against the development of negative feedback cycles. The regular and consistent receipt of well-designed reminders, motivational messages, and tips may thus be a very powerful means of reminding patients that they are actively working on gaining control over their symptoms [11].

Most of the RCTs included made use of the Internet for delivering mental health interventions for depression or depression comorbid with anxiety. The majority of RCTs included in this systematic review (93%, 42/45) did not make use of mobile phones or tablets, which is surprising because the benefits with respect to user engagement and adherence seem apparent. Typically, the interventions required the interaction with the system and many also included the interaction with a therapist (face-to-face or online) and/or peers on the Web. In all, 80% (36/45) of included programs were based on CBT or used elements of CBT. Given the fact that therapeutic interventions for depression are commonly based on CBT techniques and psychoeducation, which follow a stepwise approach and are usually delivered in person by a therapist, these findings support the authors’ premise. Twenty-seven of 45 included studies used tailored content, tailored feedback, and/or tailored reminders. In the opinion of researchers, the adaptation of information to factors that are relevant to one individual or a group of individuals is an important feature in effective health communication [93,126,127]. In fact, van Genugten et al [128] found that interventions that are more flexible in use (easy to handle for both advanced as well as novice users), provide structure (which is comprehensible to the user and that the user knows at what point he or she is in the process), and use default settings are more likely to be effective. With respect to treatment exposure, it is suggested that personally tailored feedback and goal setting are among the important factors related to the use and exposure to Web-based interventions. In the study by Brouwer et al [129], exposure was regarded as the time spent on the website, page views, and the number of times the user logged on. Although there is a relationship between exposure, adherence, and treatment outcomes, focusing on exposure only gives a limited insight into the pattern of usage and adherence [97].

Doherty et al [130] showed that user engagement also depends on depression severity and that users with minimal symptoms engage much less than other groups. To serve the needs of this patient group, users might benefit from more flexibility (eg, in the elements of the intervention they wish to focus on). Therefore, new approaches such as tailoring of interventions that are more lightweight are needed. However, due to the severity of the disorder, patients with more pronounced symptoms face specific difficulties in engaging with the program. Because these patients require more intensive support than patients with less severe symptoms, they should also be given the choice to change the means of support throughout the treatment. Thus, mechanisms are needed that allow for requesting or being offered face-to-face contact even if the user has initially commenced online treatment. We assume that customization of programs is necessary to enhance long-term adherence and outcomes of interventions.

Only a small number of included studies made use of self-monitoring components, such as symptom tracking and tracking reminders, yet they are seen as key features of psychotherapy [11,96] and in behavior-change interventions in general [99,115-118].

Although social support is widely recognized as an important feature in behavior change [119,120], in this analysis only seven studies (16%) used peer support. In the included studies, social support consisted of the use of Web-based discussion boards (asynchronous social support), which intended to provide the ability to connect with other patients using the same intervention. However, the literature shows that there is disagreement with regard to use and benefit of discussion forums and chat rooms. Although many studies support the use of these components [123,131,132], others do not [133-135]. The effectiveness of peer support depends on individual factors, such as perception of the credibility of Internet-based peer advice and perceived quality of interaction [10]. Furthermore, effectiveness might also rely on user involvement. It could be shown that users that actively post and respond to messages are more likely to benefit than users that participate only passively [132].

Regarding visual attractiveness, 44% of all included studies analyzed an intervention that included visually appealing content, such as graphs, illustrations, comics, or photos, which often serve a motivational purpose. Only 11 studies (24%) described the utilization of audio and/or voiceovers. Video footage, often containing case-enhanced learning, was also found in 11 RCTs (24%).

In addition, this analysis showed that game elements were only used in three RCTs and if gaming was included, it was in the form of knowledge quizzes. Relatively few studies have incorporated games as part of their persuasive design.

Although virtual reality has shown to be effective in the treatment of anxiety and pediatric disorders [136], so far there is no study utilizing this technology for the treatment of depression and this might open a promising direction [137]. With respect to virtual agents and synthesized speech, Morrison et al [10] were also not able to show an association between digitized speech and improved outcomes in depression. To date, technological advances and improved design of animations and avatar-based systems are likely to permit the development of sufficiently sophisticated services to simulate real interaction [10] and, therefore, might show more impact on treatment outcomes. Future research is needed that concerns the effectiveness of serious gaming and virtual reality in technology-supported mental health interventions currently underrepresented in literature.

The list of factors that influence user friendliness as well as the different platforms for delivery included in this analysis is not exhaustive. In fact, the majority of RCTs included in this systematic review did not make use of mobile phones or tablets. It is expected that especially newer studies could use different channels of service delivery (eg, mobile phone or tablet delivery). Consequently, studying future interventions that make use of these delivery channels would give an interesting insight into the influence of different modes of delivery on treatment effectiveness.

Further limitations are related to the strict process of study selection applied in this systematic review. Many trials were excluded because (1) they were not described as being randomized, (2) participants showed no symptoms of depression at baseline, (3) they included other mental health disorders, and (4) they did not assess one of the outcomes of interest. In fact, the decision to only include RCTs might lead to potential limitations of this systematic review. Even though RCTs are regarded as the “gold standard” of reliable evidence, the criteria to only include RCTs might lead to the exclusion of relevant articles that examined the effectiveness of MHIS, but used a different study design. Primarily, the exclusion of non-RCTs in this review lead to a facilitated analysis of studies because differences in methodological quality are, although not completely removed, limited. A possible consequence of this decision might be that we missed studies of newer interventions, which might not yet be evaluated in an RCT study format because they are undergoing their piloting phase at the current time [15,104].

In addition, as a consequence of considering a wide range of MHIS, included trials differed considerably in the type of therapeutic programs they used, baseline depression severity, age of participants, duration of treatment, type of control condition, methodological quality, and the various system components they utilized to enhance user engagement, motivation, and effectiveness of the intervention. As a consequence of this moderate-to-high level of heterogeneity across included trials, comparability is restricted and results should be handled with care. Subgroup analyses demonstrated that there is, in fact, a significant association between the effectiveness of interventions and the provision of face-to-face contact as well as the type of control they used. As previously noted, the inclusion criteria stated in the protocol also included other mental disorders such as anxiety disorders. However, the vast number of articles identified in the electronic search posed an additional challenge. Consequently, it was decided to focus on depression and depression comorbid with anxiety only. Considering that many mental health problems often co-occur [138], study findings might be constrained as a result of this relatively strict inclusion/exclusion of certain mental disorders.

Because technology-based psychological interventions adapt established methods of treatment and only the means of delivery are altered, the issue of noninferiority plays a major role in this review. With respect to the overall effectiveness of technology-based interventions, it is of utmost importance to review the literature from the perspective of noninferiority trials that compare an established evidence-based treatment (eg, CBT) with a new one (eg, technology-based CBT). It also needs to be clarified that the absence of a significant difference between two interventions in an RCT cannot be equated with noninferiority and that the comparison of treatment effects between studies are only appropriate if the new and existing treatments are compared against a reference that does not substantially differ in methods and population [2].

In addition, identifying the points of disengagement and gaining deeper insight into the patterns of program usage is crucial for the refinement of system components that are most strongly associated with user engagement and symptom improvement. Data on patterns of use further offer an opportunity to refine content, means of delivery and to adapt both to the needs and preferences of specific groups of users [130]. Additional research is needed to overcome these shortcomings by assessing the association of patterns of component use and the improvement of symptoms by means of advanced statistical analyses. Furthermore, it is suggested that RCTs assess symptom reduction more frequently to obtain improved information about nonlinear relationships between patterns of usage and therapeutic gains. To enhance the understanding of when and how to choose and use different system components, it also is important to comprehend the dose-effect relationship of different components [128]. Future studies should be aimed to clarify the causal relationship between patterns of program usage and symptom improvement by assigning participants to different system components. In addition, it is likely that other predictors such as age, gender, and education affect the relationship between usage of system components, engagement, and outcomes. Therefore, additional studies are needed to assess whether differences exist between those populations.

In regard to designing mental health interventions, it is of utmost importance to understand what users need and expect from computerized or Web-based mental health interventions and how individuals rate different system components with respect to usefulness, practicality, connectivity, time demands, professional support, social interaction, convenience, novelty, reliability, confidentiality, trustworthiness, motivation, and engagement. Qualitative feedback offers a solution to find answers to the proposed questions. Likewise, user feedback might disclose disparities between user expectations and actual results. It can be assumed that user preferences vary greatly from individual to individual. This, in turn, supports the rationale of customization and tailoring of programs to create unique user experiences based on client’s preferences without losing the effectiveness of interventions.

Moreover, to clarify the clinical feasibility of computer- and Web-supported mental health interventions, it also is important and worthwhile to repeatedly listen to the opinion of therapists. Apart from the time and cost savings, there is a need for a thorough understanding of the program, which could be achieved by the provision of a protocol, printed manual, and an overview of the program to the therapists involved. In addition, more detailed information on their practices and how to deal with clients who are not engaging with the program should be provided [130].

Although clinicians tend to be very self-protective about their time commitments and skeptical about technology [139], value in health care should always be defined around the customer. Nonetheless, because value in health care is measured by the outcomes achieved, value depends on results not input [140]. First, computer-supported interventions should be designed and modified to optimize clients’ benefits. Although value in health care is measured by the outcomes achieved, only assessing the effect on symptom reduction fails to include the value of mediators that might explain behavior change (eg, factors such as skills, attitudes, and self-efficacy) [128]. Thus, a further shortcoming of the current meta-analysis is the lack of analyses of potential mediators of the effect.

From a system component perspective, there is a strong need to counteract the decrease of program usage over the course of the intervention that is typically found in unguided technology-mediated interventions for mental health [118,141,142]. Thus, it is of great importance to implement the program components that are associated with improved and regular program engagement. One of the factors that have been linked to a decrease in module completion rate is obliging users to complete module sessions in a predetermined sequence [143]. On the one hand, the delivery of module sessions in a tunneled format bears many advantages, such as a greater number of website pages accessed, greater time spent on the website, and greater knowledge gained from the website [143]. On the other hand, tunneling might lead to a decrease in module completion because users are required to complete the between-session homework assignments before being able to start the next module session [11]. Therefore, developers have to enable a high level of flexibility in the choice of relevant modules as well as the speed with which users proceed through the modules to keep users engaged. One approach to allow for a greater flexibility, if not violating the underlying psychological theory, is to make homework assignments optional [11] or to leave it to the user at what stage they prefer to complete the homework tasks. Future studies are needed to confirm whether the provision of more user-driven programs might offer an advantage over the frequently used linear delivery of interventions in mental health interventions [139].

In addition, little is known about the synergistic effects of behavior-change components, modes of delivery, and user friendliness. Van Genugten et al [128] found synergistic effects in interventions that made use of a specific CBT components in combination with the provision of rewards. In general, little synergistic effects were found. Thus, there is a need to further assess the cumulative effect of different system components on treatment effectiveness and to analyze how specific combinations affect behavior change. To fully determine the most optimal delivery mode, further studies are needed that randomly assign participants to different platforms for delivery [128].

Although no consensus with respect to effectiveness of online peer support was reached yet [94,95], anonymous online support groups and discussion forums might help users to overcome the feeling of being stigmatized by connecting patients with others. A further advantage of these social support components is that time and location are no longer obstacles for active participation [96]. Nevertheless, further research is needed concerning the type of online support, such as expert-led or user-driven, moderated or nonmoderated, and synchronous (eg, chat rooms) or asynchronous (eg, discussion forums) [96]. From a design and engagement perspective, a nuanced view on target groups suggests that complementarity between content of interventions that target different mental disorders is crucial when designing computerized and Web-based mental health interventions. Patients with multiple disorders present a considerable challenge in the design of technology-supported interventions and, as a consequence, are often excluded from studies even though they might profit from certain components. One reason for the exclusion of comorbid and multimorbid patients is the pressure toward relatively stringent and precisely defined interventions, which are amenable to RCTs [130]. Regarding the high comorbidity of mental disorders, it is of greatest importance that this topic is given more attention in the design of technology-amenable interventions.

The development of MHIS targeting the change in health behavior requires great expertise and a thorough understanding of the problem area, underlying therapeutic strategies, and the design of persuasive systems. The findings of this systematic review contribute to the body of knowledge on the effectiveness of technology-supported therapeutic interventions for the treatment of depressive symptoms. This work is intended to provide a basis for the assessment of the impact of specific system components on treatment effects in RCTs of technology- and Web-based interventions for depression. Thus, the overall goal of this review was to identify such components and to enhance the understanding of the mechanisms through which technology-enabled interventions exert their therapeutic benefits by means of such.

Further quantitative studies are needed to assess the impact of identified components and to identify other system components that are relevant for the design of future technology-mediated MHIS for the treatment of depression and other mental disorders.

Because of the high relevance of the anatomy of MHIS, attention should be paid to design issues when developing new eHealth services in the future. To enhance dissemination and utilization of technology-based MHIS, the focus needs to be not only on how the interventions affect users, but also on how patients use and interact with technology and one another through them. Therefore, future studies are needed that add to the body of knowledge of technology-supported interventions for the treatment of depression by assessing patterns of program usage and user engagement.

To conclude, health information technology is a fast-growing field of research, which has the potential to effectively treat people suffering from mental disorders. Despite that, there is still room for improvement in the design of technology-based interventions for the treatment of depression. The delivery of interventions via technology is a promising and cost-effective approach to diminish the significant treatment gap and the various barriers associated with the disorder.