Mental disorders are a challenging public health issue. This is due to the high impact of these disorders on people, which limit their participation in all aspects of personal life, family life, and society. Mental disorders affect approximately one in four adults across the world at some point during their lifetime [1]. The high prevalence of mental disorders reveals a high demand for timely mental health care and intervention for people with mental disorders. However, the enormous cost of mental health care, the shortage of mental health professionals, and the barriers to accessibility make both diagnosis and treatment delayed or unavailable [1-3]. Therefore, how to provide affordable, in time, and accessible mental health care for those in need has become a critical and urgent issue.

With the rapid development of technology, the computing capacity of mobile technologies has advanced to the point that today’s mobile devices function like handheld computers and are highly integrated into our daily lives. According to surveys [4,5], over two-thirds (64%) of US adults own a mobile phone, and users, on average, check their phones 46 times a day. The prevalent ownership and use of mobile devices make mobile apps a promising venue with which to engage users into beneficial activities or therapeutic sessions in the context of mental health [6,7]. For instance, many mental health apps with self-monitoring mechanisms enable users to track their moods and emotions over time [8]. The personal use of mental health apps also provides confidentiality for users’ engagement, which may further encourage their adoption by young people and users who have a high sense of autonomy for seeking self-help [9]. With the advantage of continuous and ubiquitous access, mobile apps have the potential to decrease barriers for help-seeking and make therapeutic activities more accessible and less stigmatic [6,10-12].

Although mobile apps enhance the deployment of mental health interventions, there are still significant challenges in increasing their adoption and incorporating them into users’ daily lives in the real world. Thus, understanding users’ adoption of mental health apps becomes an important step toward designing and utilizing them as effective intervention approaches.

Mental health apps can encompass various functions ranging from guiding recovery for mental disorders to encouraging beneficial behaviors for improving emotional health [13,14]. For example, many mental health apps can assist clinical practice, engage real-time communication, or provide psychoeducation [15]. However, the adoption of mental health apps is rather distinctive from other types of apps because of its sensitive nature. The sensitivity of mental health issues can be attributed to the long-existing social stigma, which is the most common reason given for hindering people seeking mental health care or support [16,17]. Previous research has found that avoiding a social stigma becomes a significant reason for young people to use mental health apps [18]. Nevertheless, many available mental health apps target specific users and label them by diagnosis [14] that not only may exacerbate the stigma [6,8,14] but also affect users’ adoption of mental health apps.

As there is no clear guideline, regulations (eg, Health Insurance Portability and Accountability Act [HIPPA] or Food and Drug Administration [FDA]), or recommendation for users to select mental health apps, another challenge in mental health app adoption is that users may find it difficult and overwhelming to select the appropriate app from hundreds of choices available on the app market [19]. For instance, while browsing apps on the Android app store, users can only filter apps by rating or price. The filtering mechanism on the Android app store is limited. If a user wants to find an app to help alleviate her or his anxiety, she or he may need to go through all of the apps on a search results page. Another option for users is to randomly select an app, which may not correspond to the user’s needs. With more and more mental health apps available on the market, a critical and pressing issue is how to help users select and identify the “right” app for their mental health wellness. However, it is critical to understand the adoption of mental health app from the users’ perspective before designing reliable mechanisms to assist users’ app adoption.

App adoption can be regarded as a selection process in a computer-mediated context where users make their decisions by relying on a variety of information cues [20,21]. While making decisions, individuals often use a heuristic approach to process information instead of a systematic approach due to the human’s bounded rationality [22-24]. Heuristics are “process-oriented strategies” that allow individuals to make decisions in a faster and more frugal way by reducing cognitive efforts [24,25]. That is, when utilizing a heuristic approach, individuals are inclined to examine and integrate fewer information cues and simplify their weighing principles for cues [26].

In the selection process, heuristics comprise three stages: (1) searching, (2) stopping, and (3) deciding [23,25]. For example, when selecting an app, users may search for app titles that they recognize and stop the search once they categorize the available apps into either recognized or unrecognized titles. However, individuals may rely on multiple information cues rather than just a single cue during a heuristic process [27]. Thus, the consideration of multiple effects of information cues on users’ app adoption becomes important.

Previous studies have identified a variety of information cues that can affect users’ selection in app adoption, including prices, ratings, reviews, rankings, installs, titles, descriptions, functions, and privacy issues of apps [15,20,28-36]. A study by Dogruel et al [20], which is most relevant to ours, further points out that when users know what type of apps they need, they often employ the simple “take the first” heuristic approach, which is predominantly influenced by apps’ titles and crowdsourcing-based cues such as ratings and rankings of apps. Users only seek additional information (eg, reviews and functions) when they are uncertain about rating and ranking cues [20].

Existing research has indicated several influential cues involved in the process of app selection and adoption. However, the literature about how users select and adopt mental health apps is still rather scarce. As users’ app adoption varies by the kinds of apps, we still have no knowledge about the type of information cues that have relational impact on users’ adoption of mental health apps. Considering the sensitivity of the mental health context, we are curious whether predominant cues (eg, apps titles, ratings, and rankings) indicated by previous studies remain influential in app adoption. To the best of our knowledge, this study is the first work focusing on examining the relationship between information cues and mental health app adoptions.

The aim of this study was to investigate how exhibited information cues in an app store are related to mental health app adoptions. As mental health is a realm too broad to investigate, we focus in particular on one mental health condition: anxiety, which is also one of the most common mental health issues among US adults [37]. In addition, the delivery of psychological intervention via mobile devices is well suited for anxiety disorders because it allows users to receive in-time treatment during their daily routine, which makes anxiety disorders an ideal research topic.

We selected Android app store, Google Play, as our research site because it is currently one of the leading app marketplaces [38]. As exhibited in Figure 1, the anxiety app adoption flow starts from the keyword search. Users input the keyword and get the first search result page that mainly displays apps’ titles, ratings, and prices. If users are interested in one of the anxiety apps, they can click the app for more detail, such as the number of installs, reviews, and descriptions of the apps. After browsing the information, users can decide either to install the app or go back to their search results page for more options. If users decide to install the anxiety app, the permission consent dialogue will pop up to notify users what permissions are requested by the app. Apps will be installed if the users accept the app permission request.

As pointed out by previous studies [20,27], users’ app adoption is affected by multiple information cues. However, this type of multiple effect is difficult to measure or observe directly from the users’ self-report because of bounded rationality and self-bias. Therefore, we propose to use the observational data of apps for examining the multiple effects of information cues on app adoption instead of using users’ self-reported data. Based on prior research, we examined 8 types of information cues exhibited on Google Play, including titles, prices, ratings, reviews, rankings, installs, category, and app permission. We conducted several statistical analyses to explore the connections between these information cues and anxiety app adoptions, which are described in the Methods section.

To simulate the users’ app search process, we used keyword search strategies to identify apps that most likely would be adopted by users seeking anxiety-related apps, which is similar to the approach employed by Ramo et al [39]. According to the Diagnostic and Statistical Manual of Mental Disorders, fifth edition (DSM-5) [40], we first identified 3 main keywords related to anxiety disorders, including anxiety, fear, and avoidance. Each term reported 250 results on Google Play. We decided to drop the term “avoidance” because its search results did not report anxiety-related apps. In order to identify other potential keywords, a search for the word “anxiety” was performed at the website UrbanDictionary.com. A total of 27 commonly used terms were listed, and two of the words most compatible with anxiety and fear, “anxious” and “worry,” were selected. Four keywords were used in our final search terms on Google Play, including anxiety, anxious, fear, and worry. We used “anxiety” as our primary search term and the other three keywords for supplementary searches.

We conducted a 2-phase app search. The first app search using these keywords was conducted on Google Play from July to September 2016. Researchers collected metadata information for all of the apps and selected the anxiety-related apps based on the apps’ descriptions. After selection, a second round of app searches by keywords was conducted on October 7, 2016, by 3 researchers. Twenty-four new apps were found and 14 apps no longer existed. A total of 274 apps were selected for analysis. Figure 2 displays the search process for anxiety apps on Google Play.

The title of an app gives users their first piece of information on what the app is about, which can further influence the users’ apps selection [29]. However, the functionality of apps, such as treatment, psychoeducation, and diagnosis, may not always appear in the title. Instead, they may emphasize specific disorders, symptoms, or activities. We are interested in whether anxiety apps tend to use certain terms in their titles. To classify the titles of anxiety apps, we adopted the recommendation proposed by [14] and generated six criteria from both clinical and nonclinical perspectives. The clinical criteria include anxiety disorders, symptoms, and treatments, and the nonclinical criteria are self-help approaches, mindfulness activities, and the self-tracking or management tool. We reviewed the title for each anxiety app to see if they are related to these six aspects. The details of the six criteria are exhibited in Textbox 1.

To capture the connections and multiple effects of information cues on anxiety app adoptions, we employed different statistical techniques. We first examined the relationship among metadata information cues and app adoption by correlational analysis. We then tested the normality of each information cue using the Kolmogorov-Smirnov test and found that these variables were not normal distribution. Therefore, nonparametric approaches including the Kruskal-Wallis test and the Mann-Whitney test were adopted to examine the differences of app adoption by app categories and titles.

In order to test predictable effects of information cues on anxiety app adoptions, we applied a nonparametric regression technique known as generalized additive model (GAM) proposed by Hasite and Tibshirani [42]. The GAM framework assumes that the contribution of each predictor is additive, which is similar to the concept of linear regression that each variable is estimated independently. The dependent variable is an additive combination of arbitrary functions of predictors. This additive modeling technique captures the underlying predictive patterns of data by smooth functions, especially when the model has nonlinear effects [43]. In this study, the flexibility of GAM allows us to predict the nonlinear impact of each information cue on app adoption. We implemented GAM by the function gam() with regression splines in the R package “mgcv.”

We examined the relationship between app adoptions and exhibited metadata information by correlational analysis. As exhibited in Figure 4, the ranking of apps have a significant positive correlation with app price, indicating that the higher price apps have, the lower ranking they have in the search results. The results also showed that the app ranking has a negative correlation with app review, rating, and installation. These results suggest that apps with a higher ranking also have more reviews, installs, and higher ratings.

Another interesting result is that the total number of app permission requests positively correlates with installs. This suggests that apps requesting more app permissions from mobile devices have more installs. The results also show that apps with lower prices request more app permissions from the devices. Furthermore, the ratings and reviews of apps have significant positive correlations with the number of installs, and the price of apps has significant negative correlations with installs (Figure 5). This means that apps with higher ratings and reviews at lower prices have a higher number of installs. We want to note in particular that although these results have significant correlations, the coefficients are only at moderate or low level. It is also important to note that significant correlations only show the relationship between these observational cues that do not represent the cause and effect relations.

We applied the Kruskal-Wallis test to examine whether different categories of anxiety apps showed significant differences in ratings, reviews, and installs. As shown in Table 3, the results show the significant differences in price, rating, review, and install. According to the post-hoc test, apps in books and reference have significantly lower ratings (P=.05) and installs (P=.01) than apps in other categories. Also, apps in books and reference have a significantly lower amount of reviews than apps in health and fitness (P=.04) and others (P=.002). There is no significant difference among categories in app rankings.

We adopted GAM to predict the impact of 5 metadata information cues (price, rating, review, ranking, and permission) on adoptions of anxious apps. To select the smooth function of parameters for minimizing the prediction error, we used deviance and generalized cross validation score [44] as indicators. Deviance (D) represents the discrepancy between observations and fitted values, and generalized cross validation (GCV) estimates the expected fit of a model to a dataset. The smaller these scores are, the better fit the model has. Thus, we decided to smooth all variables because the model produces the lowest deviance (D=316.24) and GCV score (GCV=1.51). As shown in Table 4, the price, rating, and review have significant predictable effects on app adoption. Note that although ranking cue does not reach a significant level, it is still very close.

Figure 6 displays the direction of predictive impact on app adoption. We found that the expected number of app installs will decrease drastically at first when the price of the app increases. However, the decrease of expected installs becomes gradual while the price increases. This may suggest that when prices of apps are over a certain amount, the number of app installs will remain similar. For rating, the app installs will increase when the rating also increases until the rating reaches nearly 4.4. After this point, the number of installs will start decreasing as the rating increases. This also implies that the high rating apps may not always get high installs. Furthermore, the result shows that the app installs will increase when the number of reviews also increases. Yet, when the app review reaches approximately 50,000, the number of installs will slightly decrease until the number of reviews is over 100,000. For ranking, the higher ranking the app has, the more expected installs the app will have.

An important function of the app title as an information cue is to inform users about the apps. We examined whether the linguistic cues in an app title are related to the users’ adoptions of anxiety apps. Considering that app installs, ratings, and reviews are not normally distributed, we applied the nonparametric Mann-Whitney test to examine the difference. Our results show that app titles using the terms relating to anxiety disorders have significantly fewer installs and reviews (Table 5). We also found that apps with titles relating to symptoms have significantly lower installs, ratings, and reviews than those without symptom-related titles. On the other hand, apps with titles relating to mindfulness activities have higher installs, ratings, and reviews than those without mindfulness-related titles. Overall, these results suggest that apps with titles not directly related to anxiety disorders or symptoms, but to mindfulness activities, are more adopted by users.

In this study, we only examined anxiety apps on Google Play, which may limit our findings to a specific mental health context and app market. We recommend future studies apply the same approach to other mental health conditions and app markets and compare their results with this study. Although we found the correlational impact of information cues on app adoptions by using observational data of apps, we were unable to identify their cause and effect on app adoption. Therefore, we suggest future studies conduct empirical work based on the assumptive inferences we proposed in this study and investigate the cause and effects of information cues on users’ mental health app adoption.

Mental health apps can be a powerful instrument for mental health care and intervention. How to design an app-searching system that can lead users to select the right apps for their mental well-being becomes a challenging issue. Since app adoption is a heuristic process, information cues play important roles. To clarify the impact of information cues on users’ app adoption, we examined the relationship between information cues and users’ app adoptions by using observational data in an app market. We discovered that app price, rating, and review are important indicators for anxiety app adoptions. On the other hand, the impact of app permission and category cues remain unclear. Most importantly, our findings revealed the importance of app titles on users’ selection and adoption of anxiety apps. Although there is still a long way to go for designing an effective search mechanism for mental health apps, our work demonstrates interesting phenomena and provides insight into the information cues and anxiety app adoptions, which will help provide better solutions for the future design of search mechanisms for mental health apps.