System design of a text messaging program to support the mental health needs of non-treatment seeking young adults

Abstract

Young adults (ages 18–25) experience the highest levels of mental health problems of any adult age group, but have the lowest mental health treatment rates. Text messages are the most used feature on the mobile phone and provide an opportunity to reach non-treatment engaged users throughout the day in a conversational manner. We present the design of an automated text message-based intervention for symptom self-management. The intervention comprises: (1) psychological strategies (i.e., types of evidence-based techniques leveraged to achieve symptom reduction) and (2) interaction types or the form that intervention content takes as it is delivered to and elicited from users.

1. Introduction

Young adults — those between 18 and 25 years old — are the adult cohort in the United States with the highest prevalence of mental health conditions [1]. Added stressors such as the COVID-19 pandemic have further exacerbated mental health conditions in this group [2]. Data from 2020 reveal that 17% of young adults experienced a major depressive episode in the past 12 months, an 11.8% increase from 2019 and a 98.8% increase from 2010 [1]; and while prevalence of generalized anxiety disorder (GAD) is harder to estimate due to different surveillance methodologies, point prevalence estimates for adults ages 18–39 were 16.6% [3], contrasting with 7.4% of adults aged 18–29 in 2019 [4]. In addition to the burden of symptoms these conditions place on young people, these affective conditions can have profound effects on individuals’ functioning across social, emotional, economic, and health domains [5–7]. This accelerating mental health crisis among young people in the United States recently garnered a national advisory from the office of the U.S., Surgeon General [8].

Despite the high prevalence of mental health conditions among young adults, most do not want or cannot access formal treatment [9–11]. Over half of young adults with a mental health condition in the last year did not receive any mental health services, the greatest proportion of any adult age group [1]. When asked, young adults cite both structural and attitudinal barriers for not seeking treatment [9–11]. Structural access barriers can include low treatment availability due to mental health provider shortages [12], financial or insurance restrictions [1], and transportation limitations. Attitudinal barriers — which tend to be more prevalent than structural barriers [10] — can include beliefs that treatment is not needed, preferences for self-management, stigma, or concerns that treatment may be ineffective. Unfortunately, without intervention, many mental health conditions worsen and can follow a chronic course [13–15].

By leveraging young adults’ routinely used devices (e.g., smartphones) [16] and services (e.g., apps, social media) [17], digital mental health (DMH) interventions have potential to address the significant treatment gap in young adults, circumventing key structural and attitudinal barriers that individuals often encounter when seeking in-person care. DMH interventions can be private, including preventing the need for individuals to disclose their concerns to others who might moderate their access to care (e.g., parents/guardians). Moreover, resources are accessible on-demand in an individual’s daily life, supporting building and applying skills and strategies when and where they are needed, and eliminating the need to travel for care. Young adults generally find digital interventions acceptable, with many preferring self-guided digital tools to in-person treatment, reflecting considerations like cost, convenience, self-pacing, and potential for anonymity [18].

Despite the benefits of many DMH interventions in reducing symptoms in clinical trials [19–21], these interventions have low rates of sustained engagement in real-world contexts [22]. Some estimates suggest that over 96% of users stop opening downloaded DMH apps within the first 15 days of installing them [23]. When sustained engagement is this low, real-world users may not be able to realize symptom improvement [24]. Thus, enhancing engagement has been identified as a high priority focus for DMH [24].

Several contributors to low engagement suggest a need to reconceptualize how DMH tools are designed and delivered. First, most DMH interventions on the consumer market are mobile apps or browser-based interventions; these require a user to open an app or tool that is for the single purpose of mental health management, presenting a barrier for users whose symptoms include decreased motivation and increased avoidance. Second, these DMH tools are often one-size-fits-all and inflexible, offering users limited opportunities for customizing or shaping their own experience. Interventions often “tunnel” users through the same content, meaning that the user must progress in a sequence that is set by the system, completing a specified module before moving on to the next. While this structure can progressively build knowledge without overwhelming users [25–27], strict tunnelling may not be conducive to users’ agency, may not match preferences for variety and self-experimentation [28], and may not accommodate the diversity of users, including their differing symptoms and self-management styles [26,29,30]. Moreover, a strict tunnelling approach may work against efforts to provide person-centered care to users by catering to what works best for individuals on average, potentially exacerbating existing mental health disparities among individuals who are younger, lower income, or minoritized [31,32]. Tailoring interventions to an individual and maximizing user agency and choice in addition may help increase access to individualized care.

To address these barriers to sustained engagement, some recent DMH interventions emphasize highly accessible modes of interaction and allow users to experiment with eclectic content offerings [e.g., 30,33]. As far as accessibility, text messaging is the dominant communication medium among young adults [17]. As such, messaging notifications are frequently attended to in a short amount of time [17,34]. While a relatively simple technology compared to apps, messaging allows for reliably reaching users in their day-to-day life and can be delivered inexpensively at a very large scale [35]. Additionally, the interactive nature of text messaging is well suited for supporting user choice and personalization. Whether via branching logic or natural language processing techniques, interactive dialogues map each user message onto tailored responses from the system, supporting a conversational interaction. Dialogues may also be designed to allow users to exercise agency by allowing them to make explicit choices about the content they see (e.g., selecting between possible offerings, or rating and ranking content to inform future selections). Such approaches fit within a “self-experimentation” framework, scaffolding the process of selecting, trialling, and evaluating content to find a data-driven match with one’s needs and preferences [30,36]. Tools that are sufficiently interactive, and support user agency, can encourage deep cognitive, behavioral, and affective engagement with content [37]. In turn, this engagement can activate the changes needed to produce a reduction in symptoms [24].

In this paper, we describe the overall design and structure of the Small Steps SMS intervention. We sought to meet the needs of young adults through an accessible automated text messaging intervention with diverse content. By reaching individuals through text messaging, a communication channel in which young adults are already immersed, our intervention aims to overcome issues of low engagement and high attrition while supporting low-stakes experimentation with a range of content types. The content is diverse in terms of its clinical approach, with strategies for depression and anxiety symptom self-management drawn from evidence-supported therapies such as cognitive behavioral therapy, acceptance and commitment therapy, positive psychology, and social rhythms therapy. It is also diverse in terms of the types of interactions it offers (e.g., prompts, stories, exchanging messages with others), so that users can find what works for them.

2. Intervention Design

Drawing on a series of design activities conducted with non-treatment seeking young adults [29,38–40], we designed a text messaging tool that would offer young adults a convenient, engaging way to learn and apply psychological strategies through eclectic interactions. This approach is well-suited to meet the needs of heterogeneous users, accommodating variability in users’ symptom presentations, preferences, backgrounds, and contexts. Design activities are detailed elsewhere, but included: (a) two asynchronous remote community (ARC) elicitation workshops (n=28, n=22) in which young adults with symptom severity scores of 10 or higher on the PHQ-9 or GAD-7 responded, on a private online forum, to a series of researcher-posted prompts about their needs and preferences for a text-messaging system to support symptom self-management [39,40], and (b) a series of synchronous co-design workshops with young adult users who had participated in one of the two ARC workshops (n=9) to refine designs and features of the text messaging system [29]. Designed components included: (a) a set of evidence-based psychological strategies or techniques to self-manage symptoms, and (b) the format and content of text messaging dialogues supporting these strategies.

2.1. Psychological strategies

Based on users emphasizing the appeal of experimenting with diverse strategies that may help them manage depression and anxiety [29], we composed an eclectic set of 11 psychological strategies, drawing from several theoretical frameworks (see Table 1).

Table 1.

Included psychological strategies.

Psychological strategy	Brief description and application
Behavioral Activation [41]	Increases contact with rewarding activities. Messages may prompt the user to: 1) identify an activity that has potential to bring pleasure or a sense of accomplishment, 2) make a plan to do that activity, and 3) notice changes in mood that accompany doing the activity.
Cognitive Restructuring [42]	Centers on noticing and changing negative thought patterns. Messages may prompt the user to: 1) notice which automatic negative thoughts recur, 2) create a record of those thoughts, and 3) come up with re-framings of negative thoughts.
Social Connectedness [43]	Increases outreach and bids for connection with others to elicit social support and strengthen relationships. Messages may prompt the user to: 1) identify people who bring positive value into their life, 2) show appreciation to those people, and 3) schedule activities that involve positive social contact.
Problem Solving [44]	Focuses on identifying problems and systematically choosing optimal solutions. Messages may prompt the user to: 1) identify and articulate a specific problem, 2) generate possible solutions, 3) evaluate solutions relative to potential gains and drawbacks.
Willingness [45]	Aimed at increasing a user’s ability to experience a full range of human experiences without selectively avoiding unpleasant experiences. Messages may prompt the user to: 1) identify experiences that have been actively avoided due to fear of discomfort, 2) generate ways to approach this experience with openness and non-judgmental observation, 3) evaluate efforts to practice a willing stance.
Valued Living [45]	Assists uses in identifying and moving toward overarching forces or themes users find meaningful. For example, messages may focus on: 1) psychoeducation on the distinction of values from goals, 2) ways to articulate and identify meaningful themes or forces, 3) practice constructing concrete objectives and actions consistent with identified values.
Social Rhythms [46]	Helps users establish and maintain daily routines. Messages may prompt the user to: 1) identify any daily routines that work well for them, 2) try several types of new routines centered around sleep, meals, or other daily occurrences, 3) evaluate the effect of practicing routines on their affective states.
Gratitude [47]	Involves practicing an appreciative stance. Messages are designed to help a user 1) notice things, people, and experiences that bring a user a sense of gratitude, 2) notice how gratitude affects mood and anxiety, 3) document and share aspects of their lived experience that they appreciate.
Relaxation [48]	This strategy is designed to help users reduce physical and emotional tension. Messages are designed to 1) familiarize users with basic brief relaxation techniques, 2) notice how relaxation impacts mood and anxiety symptoms, 3) incorporate targeted relaxation practices to manage stress and tension.
Self-Compassion [49]	Centers on building awareness of self-criticism and increasing users’ psychological flexibility with regard to processes of change and growth. Messages prompt a user to 1) notice harsh self-critical thoughts, 2) practice a gentler self-narrative, 3) track progress over time.
Help Seeking [50]	Aims to decrease some barriers to seeking formal and informal mental health support. Messages may 1) provide information on types of treatment or available support, 2) link to and show ways of using informational resource hubs that connect would-be-patients to providers, 3) provide psychoeducation and practice prompts for how to identify trusted others and confide mental health concerns for the purpose of eliciting support and connection to care.

2.2. Interaction types

The strategies above are each supported by four broad types of interactions (i.e., “dialogues”) that may be launched on a given day during the intervention.

First, prompt dialogues present a psychological strategy and guide the user in applying it. Unfolding over the course of an entire day through several points of contact, prompt dialogues first introduce the psychological strategy, offer foundational psychoeducation, and provide a link to complementary web content so the user can read more at their discretion and convenience. Self-reflection is encouraged through open-ended questions asking the user to consider how the strategy applies within their life. Prompt dialogues also support a user applying the strategy by giving a suggestion of a small action to take in line with the strategy, followed in the afternoon by a brief reminder and statement of encouragement. At the end of the day, the user is asked if they used the strategy that day, followed by a tailored message of either positive reinforcement (for those who took the step), or encouragement to try the strategy later (for those who did not). To build choice into prompt dialogues, participants are asked if they wish to continue receiving content on the same strategy or switch to something new. Figure 1 shows an excerpt from a prompt dialogue (a), a brief summary of the key objectives of prompts (b), and an exemplar reaction to prompt dialogues (c).

Fig. 1.

(a) Prompt-style interactions; (b) goals of prompts; (c) exemplar feedback on prompts.

Second, story dialogues include brief first-person narratives illustrating how a peer addressed a challenge by applying a psychological strategy. Stories were composed by lab members based on their real experiences and crowdsourced from individuals with mental health concerns. Story writers were advised to address a common issue faced by young adults (with the ability to pick from a list of issues or to specify their own), and to select a psychological strategy that helped them begin to address the issue (with the ability to pick from a list of strategies or specify their own). Based on guidance from end-users, we composed, selected, and edited stories to ensure that they provide details that establish authenticity and that they demonstrate that change is possible without trivializing the effort and time needed to change behaviors [51]. Stories were edited for clarity, flow, brevity, and to connect the story clearly to the strategy being illustrated. To build choice into story dialogues, variations 1) allow users to select the topic upfront, or 2) launch a story and then asked the user to choose either to continue or switch to a different story. See Figure 2 for an example of stories (a), objectives of stories (b), and exemplar feedback to stories (c).

Fig. 2.

(a) stories-style interaction; (b) goals of story interactions; (c) exemplar feedback on story interactions.

Third, writing dialogues ask users to write messages for the purpose of supporting individuals when they feel down, low, or depressed, thereby activating “helper therapy” [52]. After responding to a variation of a writing prompt (e.g., “Please help us write a short message that might motivate someone who feels depressed. What would you say to help them get through a bad day?”), follow-up questions give users the option to share their message 1) with peers and 2) with themselves later on if and when they are struggling. Responses designated for sharing with peers can be banked, vetted by the research team to screen for biased or stigmatizing language, identifiers, or any content that might be harmful for others (e.g., explicit details around self-harm or suicide). Figure 3 displays an example of a writing dialogue (a), the goals of writing dialogues (b), and exemplar feedback on writing dialogues (c).

Fig. 3.

(a) writing-style interaction; (b) goals of writing interactions; (c) exemplar feedback on writing interactions.

Based on our prior research and end-user feedback suggesting that writing is highly effortful and potentially overwhelming during moments of low mood [53], most writing dialogues were sequenced such that they directly followed a brief additional dialogue gathering self-reported mood and energy ratings, with their responses leading users down two possible branches, either 1) prompting the user to compose a message in cases where their mood is high, or 2) delivering a user-generated message when their mood is low or not reported, with users randomized to receive either a peer-composed message (that had been vetted by the research team) or a self-authored message from a prior Writing dialogue, if available.

Finally, modular dialogues involve brief, self-contained interactions that may take the form of reflection questions, supportive messages, peer stories, or actionable prompts. Each interaction supports a psychological strategy. Whereas the above dialogues unfold across multiple contact points and conversation turns, modular dialogues can be delivered in a matter of minutes at one contact point, and do not reference messages earlier or later in the day. Variations of modular dialogues allow users to cycle through many brief interactions in a short period of time, such that new content is delivered as long as the user keeps responding. For example, users may be asked if they would like a suggestion of something they can do for their mood or mental health. If they indicate yes, then a short message with an actionable prompt (e.g., a 1-minute breathing exercise) is delivered. If the user says no, or completes the prompt, the system offers up a new interaction, such as a peer message, story, or reflection question. Additional variations allow users to launch on-demand modular dialogues by sending “text me.” These dialogues can therefore leverage moments when users have the greatest motivation and availability to engage, and can allow for rapidly trialing many types of interactions. Figure 4 displays an example of a modular dialogue (a), the goals of modular dialogues (b), and a representative user reaction to modular dialogues (c).

Fig. 4.

(a) modular-style interaction; (b) goals of modular interactions; (c) exemplar feedback on modular interactions.

2.3. Overall system composition

Our overarching scheduling approach sought to accommodate non-treatment seeking young adults’ desire to experiment with diverse content, but without overwhelming them with choices. Thus, we sought to find a balance between system-selected content, and user choice. For example, a user may be scheduled by the system to receive up to three days of messages on a particular psychological strategy (e.g., valued living), but at the end of each day, the user is asked if they want to continue receiving content on the same strategy or switch to something new. We also further emphasized self-experimentation by describing the temporary basis of each new strategy, normalizing that users might like or not like it, and noting that their preferences would be gathered to shape how the system worked.

We sought to further increase variety by supporting each psychological strategy with diverse interaction types (see Figure 5 for diagram of the overall system design). On each day of the program, users receive a dialogue representing one of the four types of interactions described in the section above (prompts, stories, writing, or modular). To build foundational understanding of psychological strategies, prompts are emphasized in earlier days of the program. We created multiple day-long prompt dialogues for each psychological strategy, such that the user can practice and build on their knowledge of each strategy over several days. While these prompt dialogues have a defined order, they are not always shown on sequential days, with some intervening days being dedicated instead to stories, writing, or modular dialogues (all related to the same strategy), weaving variety through the entire program. After foundational understanding is built, later program days emphasize modular content, bite-sized pieces of eclectic content that they can engage with at their discretion. Based on patterns of engagement and explicit feedback (e.g., ratings), algorithms can be integrated to deliver more of what users find engaging or helpful [e.g., 54]. For example, reinforcement learning algorithms can collect users’ responses to various content types and psychological strategies, and then adapt the rates of content delivery in real time. These algorithms can also consider how information about an individual user (e.g., baseline symptoms) and their contexts (e.g., self-reported mood, time of day) may shape the performance of various content types.

Fig. 5.

Anticipated overall system design.

Algorithms can also be incorporated to personalize multiple additional dimensions of messages within a dialogue, including their “framing,” or how they present equivalent information. As an example, in some dialogues, we present users with either a directive or non-directive framing of an action prompt. A directive framing may be, “Today, try noticing your negative thoughts. See if there are any thoughts that automatically pop into your head and bring you down.” While the non-directive version may be, “Today, you could try noticing your negative automatic thoughts. You might see if there are any thoughts that repeatedly pop into your head and bring you down.” In addition to directiveness, users may be assigned to gain or loss framing, and varying motivational styles (e.g., inspirational messages, affirming messages, and validation). The system can, over time, learn users’ preferences to deliver more of the framings that work well overall, for particular users, and in particular contexts.

Across all dialogues, our approach sought to strike a balance between non-interactive and interactive messaging. non-interactive messaging – or messages that are not aimed to elicit a response from a user – can be helpful as a way to sustain contact in a low-demand way, potentially reaching individuals and recapturing attention even when they may not remember or have high motivation to engage. This type of messaging does not call for a text-based user response, but may involve glancing at the phone to read the new message. Other messages are interactive, allowing deeper engagement at the user’s discretion. While conversational dialogues are arguably a higher burden on users, they fit within a broader emphasis on user choice and agency [29]. User responses are not required to continue to receive messages later in the day. For example, within a “prompt” dialogue supporting behavioral activation, users might be asked in the morning to reply with a plan of a positive activity. Those who reply receive positive reinforcement; but regardless of response, the system moves on so that all users receive a reminder and motivational message in the afternoon. Thus, we sought to combine the benefits of both non-interactive and interactive approaches.

Finally, within various dialogues, the system delivers hyperlinks to web content types as another means of providing users with diverse content formats (e.g., blog posts, audio content and podcasts, and videos). For example, audio content might be delivered to help guide users through a brief mindfulness exercise, or blog content may provide information on how and why engaging in behavioral activation or positive activities can have an impact on mood. Links can be accessed when users have interest, time, and motivation, but content contained in links is purely supplemental and text message content is intended to be sufficient for users who do not click on links.

2.4. Software infrastructure

All dialogues are implemented using a reusable open-source dialog state machine framework that allows researchers to schedule initial system messages and specify tailored system responses to each possible user reply (i.e., branching logic) via a graphical user interface, creating a rich conversational experience without writing custom software. The software architecture also allows for integration of machine learning algorithms that use data collected via text messages to personalize the types of messages and content delivered to users, as shown in Figure 5. Additionally, the dialog system architecture allows for researchers to program windows in which user responses can be received; these windows close if no response is received, allowing the system to move on and deliver additional content. This approach accommodates a variety of user behavior, including delayed responses and non-responses. Moreover, non-responses or response latency can be stored as variables for use in machine learning algorithms that optimize the timing of message delivery over the course of the 8-week period. For example, it may be that users are delayed or non-responsive because they are habitually busy at certain times of day. Open-source code for components used in the Small Steps SMS system can be accessed via a public code repository (https://github.com/audacious-software/SMS-Dialog-Site-Django).

2.5. Anticipated use

We anticipate that the Small Steps SMS tool can be integrated into mental health information websites where individuals, particularly young adults, are seeking information about their symptoms online. Interested individuals interested in symptom self-management will be able to fill out a brief survey, enter their mobile number, and launch the 8-week series of interactive messages using an entirely automated workflow. Over the 8-weeks of the Small Steps SMS program, users of the system will be able to integrate suggestions, tips, reflective exercises into their everyday lives in order to support symptom self-management. It is expected that the Small Steps SMS system can serve as both a stand-alone tool for individuals who are not interested in formal mental health services, as well as an adjunctive tool that can help reinforce therapeutic principles in a user’s everyday context.

3. Conclusion

The Small Steps SMS system is designed to circumvent many of the barriers young adults face to accessing psychological treatment. The messaging system presented in this work was developed through extensive user feedback, and co-design efforts. The Small Steps SMS system emphasizes content and message diversity and user choice. The specific design elements presented in the paper are intended to support user engagement over time and ensure that those who are not interested in formal treatment are still able to receive evidence-based care to support their own symptom self-management.