Depression is a major affective disorder with significant socioeconomic cost [1], affecting over 300 million people worldwide [2] across the life span [3]. However, access to treatment is problematic [4] given the acknowledged barriers such as high treatment cost, time constraints [4], geographical location [5], and stigma [4-7]. With over 90% worldwide penetration [8], mobile phones have significant potential to scale up the provision of interventions targeting depression [9]. They are especially useful to reach users who do not normally seek professional support, such as adolescents [10]. Prior work has already indicated a high user acceptance and effectiveness of mobile-delivered interventions for depression [11,12]. The number of mobile apps available on marketplaces offering treatment for depression has also been growing rapidly [9,13].

The apps available on mobile phone marketplaces provide access to a range of interventions targeting depression [14-16], which people can select and download to fit their needs [17]. Yet, users acting independently can only select apps based on information that is available at the point of download, ie, popularity, user ratings, or app descriptions provided on the marketplaces. Evidence for supporting assessment of the quality of an app, ie, structured description of its main features, evidence-based functionality, and potential risks, is not reflected in user ratings of apps [18,19]. Additionally, marketplaces do not require app developers to provide such information [20,21]. As a result, concerns have been raised regarding the lack of an evidence base for mental health apps [15,19,22] and poor regulation of the major mobile marketplaces [23-25] hosting them. Prior work [26] has also suggested the importance of having controlled clinical trials to determine the efficacy of new therapeutic treatments. In this newly established field of mobile health (mHealth) apps, most apps claim to be informed by evidence-based treatments rather than presenting rigorous evaluations of the app itself.

Besides efficacy, understanding patients (eg, their characteristics, needs, and behaviors) is also key for improving the uptake of apps [26,27]. Most human-computer interaction (HCI) studies on understanding [28-30] or supporting depression have focused on designing and evaluating mobile technologies in research contexts rather than marketplaces [31-33]. Scholarly work has also called for the evaluation of commercial apps for depression to support the effective development of the rapidly growing market of commercial apps [10,13,15]. However, such evaluations tend to focus in isolation on specific aspects such as ethics [34] and safety [35] or on specific interventions such as cognitive behavior therapy (CBT) or acceptance and commitment therapy (ACT) [10,19]. Moreover, previous evaluations tend to analyze app information from marketplaces without the actual experience of using the apps [15].

This paper addresses these limitations by focusing on a broader range of interventions and functionality of the top-rated apps for depression. Thus, we focused on the following research questions:

This paper focuses on apps selected in spring 2019 from 2 major marketplaces, iOS and Android, whose analysis triangulates (1) reviewing app ratings on marketplaces to identify the top-rated apps for depression, (2) reviewing app descriptions on marketplaces, and (3) experimental evaluation through author interaction with the apps as expert HCI researchers [36,37].

We now describe the selection process (Figure 1). The apps were initially identified through the 2 keywords “depression” and “depressed” entered into App Crawler and Google Play search engines. A script was used [38] to extract all the apps shown in the search results. The script automatically downloaded information for each app from its marketplace, including name, category, marketplace description, price, review score, and number of reviewers. This resulted in 482 apps, and after removing duplicates, 444 apps were included in the later selection.

The strategy for app selection outlined in Figure 1 aimed to include top-rated publicly available apps targeting primarily depression. From the initially identified 444 apps, we excluded those that (1) had less than 100 reviews; (2) were inaccessible at the time of selection; (3) belonged to irrelevant marketplace categories such as social, casual, business, news, or book; and (4) had average user review scores lower than 4.0 (out of 5.0). The application of these criteria on the initial set of 444 apps resulted in 94 apps for consideration.

From these apps, we further excluded those that did not focus primarily on depression by employing the following criteria: (1) the words “depression” or “depressed” do not appear in the app’s title or marketplace description of the app, (2) the primary target is not depression (eg, yoga tracker), and (3) their marketplace description mentions that people with depression should not use the app. These criteria led to 31 apps, from which we further excluded 2 more apps as their functionality was limited to the provision of therapy sessions to be purchased in-app. The remaining 29 apps were analyzed in this review (see Multimedia Appendix 1).

Descriptive characteristics of the apps were extracted from the information provided on the marketplace. These included category, costs, target audience, whether they claimed to be evidence-based (including explicit scientific underpinning and clinical input), and data supporting analysis of ethical aspects such as the privacy policy.

To extract data on app functionality, between June and October 2019, 2 rounds of experimental evaluation [36,37] were used in which the authors as HCI experts interacted with the apps using both Android and iPhone mobile devices (ie, Samsung tablet and Xiaomi phone for Android apps and iPhone for iOS apps). The entire set of apps was evaluated by 2 authors (CQ and CD), and 21% (6/29) of the apps were evaluated by all authors. The coding scheme was iteratively revised until agreement was reached among all the coders. The coding process was hybrid, integrating both deductive and inductive coding. Informed by prior work on the classification of mHealth apps [14], the deductive codes consisted of 3 main types of functionality of depression apps: screening, tracking, and provision of interventions (Table 1). The inductive coding [39] allowed the identification of specific subcodes under each of the main functionality described above. For instance, the screening function was broken down into subcodes such as symptom monitoring, self-diagnosis, and basis for personalization.

The description of findings is organized into 3 parts. The first outlines a broader picture focusing on descriptive app characteristics (eg, categorization). The second part covers ethical considerations. The third part looks in more depth into specific functionality such as screening, tracking, and provision of interventions.

This section describes the characteristics of the selected apps, for example, the main categories under which depression apps are classified on marketplaces, their target audience, costs, evidence base, medical disclaimer, and whether involving of clinicians’ guidance while using the apps.

This section describes the ethical considerations raised while reviewing selected apps.

We now discuss the functionality of reviewed apps such as screening, tracking, and providing interventions.

This paper indicates that the current top-ranked apps for depression provide various features to benefit users across different age groups. The potential of this newly established marketplace is promising, especially for reaching subgroups of users such as adolescents, who are less likely to seek professional support offline and thus could benefit from appropriately designed mHealth apps. For this purpose, we discuss the need and opportunity for regulating the marketplace to safeguard users and to ensure a positive impact from the use of apps.

We begin by considering the ethical principle of nonmaleficence [52] within the top-rated apps for depression. First, a clearer definition of age restrictions on the marketplace could better support users in general and younger users in particular to select age-appropriate apps. We found age to be handled insufficiently and inconsistently in current commercial apps, given that the age ratings on the marketplace generally indicate the maturity of app content rather than the targeted users for the app, and we also found that these ratings were generally inconsistent with information regarding the targeted age group. This risk is further heightened by the conditions within the reviewed apps’ privacy policies including the sharing of users’ data with third parties for commercial purposes.

A recent systematic review of HCI work on affective health technologies also identified potentially harmful aspects of tracking apps such as the provision of negative mood or thinking patterns with insufficient professional support, inadequate screening, and insufficiently founded diagnosis claims based on tracked data [30]. With respect to communicating negative content, we see apps supporting the consumption of publicly shared emotional expressions of depression generated by others (A6 and A12). We further advocate that developers should consider the presence of negative content when selecting an age rating on the marketplace, as consumption of such content may lead to harmful behavior among adolescent users.

In addition, this paper systematically reviewed and analyzed the apps’ functionality. The result inspires recommendations to guide developers to further leverage digital affordances to mitigate harm, to deliver personalized depression treatments, and to track multimodal content. For instance, for apps that provide screening functionality, there may be a tendency to overclaim symptom screening informed by nonvalidated screening tools rather than using validated ones, eg, developers of A24 and A29 prominently state their apps’ effectiveness in clinical practice on the marketplace but do not provide scientific validation for the screening tools employed. In addition, with regard to the increased vulnerability of depressed individuals, we find limited direct access to professional help when screening results are communicated to users. For instance, in general, 76% (22/29) apps do not provide immediate access to suicide prevention or online counseling helplines (Multimedia Appendix 3).

Risk of harm can be identified with respect to the viewing of strongly negative content from others within the emotional expression apps for depression. Our findings highlight strong ethical concerns around these apps. Although arguably beneficial for people creating it [53], such content might have a negative effect on those viewing it, especially given that depressed individuals have a tendency toward rumination [54]. We suggest that such apps should include safeguards for users viewing highly negative content. Moreover, developers of such apps could limit views of negative content, especially given that these 2 apps (A6 and A12) are also accessible to adolescent users, who are susceptible to engage in problem or at-risk behaviors [40]. One deployed strategy was to automatically cover negative keywords within app-provided content and to offer a pop-up window with free psychological counseling helpline every 3 times when users choose to reveal the hidden negative words (A23).

In addition, apps not specifically designed for children and adolescents, but with a child-friendly age rating on the marketplace, should consider introducing customizable designs for nonadult users. It has previously been suggested that providing support and treatment sessions with parents, teachers, and siblings should be seriously considered when administering treatment to children with depression [40]. Therefore, we suggest that designers of such apps should consider mechanisms to engage parental support or supervision while children or adolescents are using these apps.

An interesting issue with respect to apps supporting the tracking of mood and thought patterns is the unfiltered presentation of these data when predominantly negative content is being tracked. Apps tracking thoughts only provide access to tracking logs in chronological order, and this presents a 2-fold limitation. First, such visualizations can be browsed but not queried to retrieve a specific entry. Second, browsing such logs may trigger vivid recall when they capture negative content and may increase the risk of rumination [29].

The suggestions discussed in this section particularly target the developers of marketplaces hosting apps for depression. Previous findings suggested that the regulation of such apps regarding data privacy remains inadequate [25,32,35] and reported the prevalence of health-related apps selling users’ data to third parties. Survey studies have also indicated that the general public is less inclined to share their health care data with technology companies [32]. The identified limitations of the privacy policies for the reviewed apps illustrate that these concerns can be better addressed; 24% (7/29) of the apps failed to provide any privacy policy in English or in a reliable source (Multimedia Appendix 2). In addition, aligned with prior studies [26,55], the current privacy policies may be difficult to comprehend by typical users. We, thus, call for developers to improve the readability of privacy policies and support the suggestion of making them easy to read at a sixth-grade reading level [26].

Another concern is protecting the privacy of users’ health data and, in particular, the data of young people while using depression apps. First, more than half of these apps (24/29, 83%) fail to provide privacy policies that specify strategies to protect children’s data (16/29, 55%). Second, our findings also show that although most of the apps are free to download, they normally come with in-app purchases for additional features or advertisements. Regarding advertisement, we found that 80% (8/10) of apps that use advertisements declare that they share users’ data for commercial purposes.

All of the reviewed apps are rated as suitable for children and adolescents on the marketplace, whereas one-fifth (7/29, 24%) of the apps specifically claim to restrict access from young users. This finding demonstrates the need for developers of marketplaces that host depression apps to increase the transparency of their standards. For instance, Google specifies that [56] their age rating is not for describing the apps’ target user group but rather for describing the minimum maturity level of content in apps such as violence, drugs, and profane language.

Surprisingly, however, no statement regarding data sharing or targeted users’ age range could be found on the app descriptions in the marketplace to support users making an informed decision at the point of downloading the app. The age rating may be specifically misleading to parents when they are selecting age-appropriate apps for their children as developers only claim age restrictions in the privacy policy. We advocate a clearer definition and regulations for age rating of depression apps on marketplaces.

In addition, we argue that users should be informed upfront of the risk of having their sensitive data shared with third parties for commercial purposes. The prevalence of health-related apps selling users’ data to third parties has been previously reported [25,35,57]. Thus, we argue for the responsibility on the marketplaces’ developers to ensure consistency of privacy-related information in the app description on the marketplaces when compared with its privacy policy or to ensure that the privacy policy is included directly within the app.

Prior studies [57] have reported the tendency of commercial depression apps to blur the line between depression as a lack of wellness or as a mental disorder, which aligns with our findings. In addition, none of the apps examined claim to target a specific level of depression severity. Although apps may potentially reach a wider range of users by following such a strategy, it may be more difficult to formulate appropriate safeguards for users whose depression leaves them with higher levels of vulnerability [57]. In addition, we found that most depression apps tend not to undergo a rigorous evaluation of their intervention components but instead rely on designing the app based on evidence-based theory [26]. Apps with insufficient evidence of efficacy present challenges as they may risk misinforming patients [57]. We advocate clear communication of the targeted user groups for mHealth apps and marketplace guidelines to match the required level of evidence for each app as well as the condition and risks of their specifically targeted user group.

App-based depression assessment is potentially valuable in supporting individuals with depression concerns to seek help and share their electronic health information with health professionals [15,26]. In addition, health data collected by users could support professionals’ understanding of users’ symptoms, which could support diagnosis and the delivery of clinical treatment. Despite these potential benefits, the top-rated depression apps reviewed seldom support this usage. Only 1 of 8 apps offered the option of generating reports of screening outcomes for sharing with mental health professionals.

Although PHQ-9 is the most used tool for depression screening, 3 out of 8 apps use nonvalidated screening tools, and information about screening tools and their scientific underpinning is seldom provided within app descriptions. We recommend that app developers use validated screening tools and provide basic information about the tools and their validity.

In addition, findings indicate that screening tools employing periodic repeated measures such as PHQ-9 [41] also tend to be used within apps during daily tracking. However, the latter may be better suited to more lightweight ecological momentary assessment measures [58] rather than depression diagnosis measures. We also found a few emerging practices addressing this concern by suggesting an appropriate frequency for screening or even limiting the frequency of access to screening tools (A16 and A28). Thus, we suggest that app developers decouple the use of periodically repeated measures such as PHQ-9 for the purpose of depression screening and the use of ecological momentary assessment for more frequent daily tracking of mood, thoughts, behavior patterns, and symptoms of depression [59].

An important challenge of mobile apps for depression is attrition [29,60]. Previous work suggested the value of personalization for improving users’ engagement with apps [19,29,61] and the value of accessing social support [19] and involving concepts from gamification [51]. In the future, this may involve the provision of real-time adaptive personalization of intervention content to the tracked thoughts or emotions [59]. However, despite the potential of mobile technology to deliver personalization, apps supporting it are limited. Exceptions here include the use of AI chatbot conversational agents (A2 and A28) to respond in real time to users’ currently recorded thoughts, instead of generic (not personalized) psychoeducational content. Personalization can also be extended to the schedule of activities within an app-delivered intervention. However, only 1 of the reviewed apps (A11) offered a personalized intervention plan based on users’ screening results. There is an opportunity to better leverage digital affordances for personalization when designing apps for depression.

Findings also indicate that tracking within depression apps is focused on capturing users’ mood patterns or thought patterns and their engagement with app-delivered interventions. However, these distinct types of tracked content are seldom available together in a single app. We argue for the value of simultaneously capturing both thinking and emotional content as these can support better encoding at the moment when an event occurs and better retrieval later [62,63]. We also suggest that integrating such tracked content with a record of progress through the intervention and completion of intervention activities could better allow users to understand the value of the app for their well-being. Such combined visualization could further support users’ engagement and motivation to continue to use the app-delivered intervention.

The rapid increase of mobile apps for reducing depression can benefit from a closer look and evaluation of the functionality such apps actually deliver and the potential ethical issues that they raise. From a systematic analysis of 29 top-rated depression apps on the major marketplaces, we suggest that developers of marketplaces should regulate depression apps to mitigate ethical risks, including missing, inadequate, or inconsistent privacy policies, ie, sharing data with third parties, child data protection, and safeguarding of vulnerable user groups. In addition, the analysis of app functionality provided new insights into opportunities for mitigating harm regarding the consumption of the negative content, unrestricted access by children (with related privacy concerns), and the provision of screening tools with less scientific validation.