The study was reviewed and approved by Northwestern University’s institutional review board. Users were recruited via fliers placed at university-affiliated and private outpatient mental health practices. Eligible users were 18-65 years old and had a DSM-IV (Diagnostic and Statistical Manual of Mental Disorders–fourth version) diagnosis of bipolar disorder with a minimum of 2 acute mood episodes within 2 years of enrollment. Individuals were excluded if they (1) were not in current psychiatric care; (2) met criteria for a substance-use disorder within the last 6 months; (3) met criteria for another psychiatric diagnosis, or had symptoms for which participation in the study was either inappropriate or dangerous (including current severe suicidal ideation or a serious suicide attempt in the last 12 months); (4) were pregnant or planned to become pregnant; (5) had a visual, hearing, voice, or motor impairment that would prevent completion of the study procedures or limit smartphone use; or (6) were unable to speak or read English. As the intervention primarily targets mood episode relapse prevention, a current mood episode at the baseline assessment was an additional exclusion criterion for the pilot study.

Before an initial telephone screening, users completed informed consent by telephone or online. Users completed written consent before a face-to-face (F2F) clinic visit. Initial screening to establish a bipolar disorder diagnosis was conducted via telephone using the Mini–International Neuropsychiatric Interview [46]. If eligible, users completed an F2F interview with a study clinician (psychiatrist or psychologist) using an abbreviated version of the Affective Disorders Evaluation and the Clinical Monitoring Form [47,48]. Individuals with a confirmed diagnosis at the clinic visit enrolled in the field trial, design interview, and/or usability testing and could additionally continue to the next phase if they chose. For the pilot study, users also completed a baseline telephone assessment using the Clinical Monitoring Form to assess current clinical status. All individuals who participated in earlier phases of intervention development were offered an opportunity to participate in subsequent development phases. Users were compensated for their time and travel costs: US $10 was given toward travel costs and the telephone assessment; US $15 was given for participation in the clinical assessment, baseline and monthly telephone assessment, exit interview, and app training; and US $25 was given for the design and usability interviews.

At study enrollment, 12 users were included and were 21-62 (mean 38, SD 14) years old. Of these 12 users, 4 were male and 8 were female; 12 were non-Hispanic white; 3 were married or living as married, 3 were divorced, and 6 were never married; 5 had completed some college, 2 had completed college, and 5 had completed more than college; 2 were students, 6 were employed, 2 were unemployed, and 2 were on disability.

Similar to existing, empirically supported F2F psychotherapy interventions for bipolar disorder, the LiveWell intervention aims to decrease episode relapse, reduce symptom burden, and improve quality of life [3,4,10-15,58-60]. The intervention seeks to achieve these outcomes by assisting individuals with managing behaviors proposed to underlie the impact of existing therapies [7,11-13,15,18,19,31,61,62]. LiveWell thus emphasizes identifying early warning signs and symptoms of relapse, developing plans and monitoring for relapse, and then enacting and adjusting plans as needed. In addition, LiveWell engages users in a similar process to support taking medications as prescribed, obtaining adequate sleep duration, and maintaining regular routines. LiveWell also addresses strengthening social support, managing stressors, and engaging in healthy habits regarding diet, exercise, and substance use.

LiveWell consists of technological and human support components, including a smartphone app, a secure server, a website, and coaching support (Figure 1). The smartphone app consists of 5 components: Foundations, Toolbox, Wellness Plan, Daily Check-in, and Daily Review. It provides foundational information on bipolar disorder self-management (Foundations) and a toolbox with self-assessment surveys and skills practice (Toolbox). The Foundations and Toolbox components support developing a personalized Wellness Plan to reduce relapse risk and manage signs and symptoms. The core of the intervention is a Daily Check-in, in which users monitor medication adherence, sleep duration, routine, and wellness. Based on Daily Check-in data, the Daily Review provides tailored feedback that directs users to relevant psychoeducation content (eg, using lifestyle skills to reduce relapse risk or coping skills for managing early warning signs and symptoms).

LiveWell also utilizes human support, a coach to improve app use adherence, self-management, and communication with mental health providers [63]. The app provides data summaries and alerts via a secure server and website to help the coach provide support and facilitate communication with mental health care providers. Personnel without professional mental health training provide coaching support to reduce costs and increase access [64,65]. A clear division of labor between the technology and the coach ensures that coaches operate within the scope of nonclinical practice. The technology functions as the psychotherapeutic strategy expert that provides status summaries and alerts to the coach, who uses structured scripts and flowsheets to serve as a technology use concierge.

Monitoring is a major determinant of behavior change [41] and an essential strategy of empirically supported bipolar disorder psychotherapies [3,11,25,60]. Individuals with bipolar disorder are interested in using self-monitoring tools [30,31]. Thus, the initial development of LiveWell focused on creating a smartphone-based self-monitoring tool for tracking moods, thoughts, sleep, and routine. In particular, the ability to distinguish between different levels of wellness (eg, doing well or responding as expected to events versus experiencing early warning signs of an episode) may be essential to staying well [33].

Due to the potential importance of this ability, the field-trial Daily Check-in included 7-point scales to monitor mood and thoughts (Figure 2, Table 1). During the app training session, coaches helped users establish mood and thought anchors (ie, words describing their prior experiences at different wellness levels) to make these scales more personally relevant and useful. However, coaches noticed that users' mood and thought anchors often overlapped (eg, “planning for my future” as a mood anchor; Table 1), and some anchors appeared to reflect behaviors, physical symptoms, and changes in motivation (eg, “sluggish” as a thought anchor; Table 1). In addition, users reported that the 7-point range was too restrictive (eg, “If people are more depressed, it would probably be useful to have a wider scale to describe it”). As a result, the mood and thought rating scales were collapsed into a single 9-point wellness rating scale (Figure 2).

Users were also given the option to complete the Daily Check-in multiple times a day, but coaches noted that allowing multiple check-ins did not appear to elicit the reflection needed to identify different wellness levels. Instead, the Daily Check-in seemed to be capturing momentary reactions to daily hassles and uplifts [66]. To encourage users to engage in reflective monitoring rather than in-the-moment rating, the Daily Check-in was restricted to allow only one check-in per day, and the coaching scripts were adjusted to encourage reflection on wellness status using the personalized anchors.

Content and tools for the app and coaching support were developed based on information from empirically supported psychotherapies for bipolar disorder [3,12,13,15], health psychology behavior change theories [39,40,67-74], and chronic disease self-management models [75-83]. Design sessions and usability testing were then conducted to obtain user feedback on the overall app design, Daily Check-in, Daily Review, Foundations lessons, and the F2F coaching app training session.

During the design sessions, the Daily Check-in had separate early warning sign buttons to emphasize the importance of early warning signs (Figure 2). User feedback indicated these buttons were unnecessary and confusing with the transition to the 9-point rating scale. Thus, the Daily Check-in was simplified by removing these buttons, and early warning signs were clearly defined as +2 and -2 on the wellness rating scale (Table 2). Additionally, the Foundations content and coaching scripts were updated to emphasize the use of the wellness rating scale for early warning signs recognition.

During usability testing, user feedback suggested that exercise and diet were not adequately addressed. Therefore, Foundations and Toolbox content was added to cover the importance of exercise and diet [84-86]. User feedback also suggested that finding the right doctor and establishing a therapeutic alliance was not adequately discussed and that the rationale and approach to identifying a hospital in the case of a severe episode were unclear. As a result, the team lesson about working with a psychiatrist was expanded, the rationale for learning about a hospital for inpatient care was clarified, and information about using mental health directives was added.

In their posttask questionnaire responses, users indicated that the app's overall design and organization were straightforward, easy to use and understand, reasonable in terms of time commitment, and met their expectations (Table 3, Multimedia Appendix 5). Users also indicated that both the Foundations and Daily Review were interesting, relevant, and taught them something new.

The pilot study tested the complete intervention, including the technology and human support. At the F2F session, coaches helped users identify personalized wellness anchors (Table 2) and set mutually agreed-upon target goals (medication adherence, sleep duration, routine bedtime and rise time, and wellness rating range) for daily monitoring (Figure 1). The coaches encouraged users to set goals known to facilitate staying well [11-13,87], such as taking their medications 100% of the time, sleeping the recommended amount each day (7-9 hours, with 6-10 hours reasonable for some), going to bed and starting their day within a 1.5-hour window, and keeping their wellness ratings within a balanced range (-1/+1; Table 2).

Users found differentiating between maintaining a regular routine and getting the right amount of sleep confusing. Therefore, a separate button for recording sleep duration was added to the Daily Check-in (Figure 1), and coaching scripts were updated to clarify the behavioral targets and goal setting process (Multimedia Appendix 6). Users also expressed difficulties understanding the Daily Review feedback regarding goal achievement. The first page of Daily Review feedback uses bar graphs to display the percent of days over the last 7 days that the user met their target goal (Figure 3). A title was added to the Daily Review bar graphs to indicate that the percentage bars correspond to goal achievement over the last 7 days. A hover bar feature was also added so that users could view their personalized goals for each behavioral target (Figure 3). Some users found the Daily Review feedback repetitive when target goals were consistently not met. To allow users to sample additional Daily Review content under these conditions, a Review Something Else component was added and made available after completing the suggested Daily Review feedback. Coaches found that users did not always follow up on Daily Review feedback to contact their providers. To address this, when feedback prompting communication with providers was given, a pop-up message with a link to one’s psychiatrist’s phone number was added to the Daily Review. Email alerts to coaches were also added, and coaching scripts were developed to help the coaches contact the user and psychiatrist under these conditions [63].

At week 4, users were asked to complete reading the Foundation lessons, and coaches assisted users in developing a lifestyle plan to reduce relapse risk (Table 4) and a coping plan for managing signs and symptoms (Table 5). Originally, generic suggestions for the plans were included. However, when coaches asked users to personalize their plans, some wanted to keep the generic plan. Generic plans were removed for the diet, exercise, substances (Attend), stress management (Tranquil), and build-support (Social) sections of the Reduce Risk plan to encourage users to think of relevant personal plans (Multimedia Appendix 7). Users' personalized goals for sleep duration and bedtime/rise time windows, and a generic recommendation to take all of their psychiatric medications daily, were retained. Coaching scripts were adapted to encourage the use of implementation intentions for planning [88-90], and users were given examples of implementation intentions in the generic risk-reduction plans (eg, “If you miss medications for more than 3-4 days in a row, then discuss with your psychiatrist”).

Users reported confusion and difficulties with setting and achieving target goals and did not consistently follow up with Daily Review recommendations to contact providers; thus, the coach role was expanded from app use support [65] to more active support for self-management and clinical care communication. Clear coach email alerts and dashboard summaries, scripts, and flowsheets were developed to support these additional roles [63]. The coaching scripts were modified to emphasize creating target behavioral goals, assisting users with developing plans, reinforcing success, and making adjustments when goals were not met (Multimedia Appendix 6). This increased self-management support included developing flowsheets and tip sheets to help coaches guide the user to appropriate self-management content [63]. Questions from coaches about how to address worsening symptoms resulted in additional structured protocols to assess suicidality and functional impairment, including clear instructions on when to call for real-time clinical support. Details of the rationale and implementation of the coaching support are described elsewhere [63].

When users finished the 8-week pilot study, they were asked to complete a usability questionnaire (Multimedia Appendix 9) [91,92]. Responses are summarized in Table 6 for questions in which ≥7/11 users rated an item as “strongly agree” or “agree,” or ≥3/11 users rated an item as “strongly disagree” or “disagree.” All users found the Daily Check-in easy to use, and most indicated that it made them more aware of their early warning signs and sleep duration. Users expressed that the Foundations were easy to understand, about the right length, and relevant. They reported that they liked being able to personalize the Wellness Plan and found it relevant for their continued use of the app. Finally, users emphasized the importance of coaching support. They found the coach supportive and the calls convenient in terms of their schedule and appropriate length. Most users indicated that they were motivated by the coach to review intervention content and change their behaviors after coach calls.

Thematic analysis of exit interviews identified 6 usability themes: ease of use, ease of learning, usefulness, barriers, suggestions, and technical limitations (Multimedia Appendix 10). Users indicated that they found the app easy to use.

Some users reported that having self-management resources on their phones was especially convenient when they experienced symptoms. “It’s sort of like having a therapist in your phone” (2041).

Users found the Foundation lessons and Wellness Plan relevant and useful. They stated that they liked being able to personalize their wellness anchors and plans and that this process made their app use more meaningful.

Users also stated that personalizing their information inspired self-awareness.

Moreover, users felt that the coaching support was motivating, useful, and helped them get more out of their app use.

Users offered many suggestions about changes or additions to data viewing, navigation, personalization, and monitoring, such as adding tracking for energy level, alcohol use, diet, and exercise. “There should be more visual things . . . like video clips” (2063).

Users noted the app’s navigation could be complicated, which sometimes led to not using parts of the app; however, coaching support helped resolve navigation difficulties. “Navigating was a little hard to get used to. . . . I’m not sure I ever looked at [the toolbox]” (2041).

Users still found that the Daily Review feedback could be repetitive if they were consistently experiencing problems with one specific behavioral target (eg, routine). In some cases, this repetitive feedback led to a discontinuation of use. “I got fed up after a couple weeks ‘cause [the Daily Review] was the same thing” (2063).

Some users felt that technical issues impeded their use of the app. Specifically, users experienced problems with the reminders to check in, which contributed to irritation and reduced reliance on the reminders.

Other users expressed difficulties with study equipment, such as frustration with poor phone battery life. “I had some difficulty with the phone. [It] would die all the time” (2016).

However, most users felt that the study team was helpful and responsive in resolving technical issues.

A thematic analysis of users’ mood, thoughts, and wellness rating scale anchors was completed to explore what types of signs and symptoms are relevant for monitoring wellness. The anchors for the mood, thought, and wellness rating scales were coded into subtypes and types (Multimedia Appendix 11). When the field trial anchors were formally coded, only 46% of the mood scale’s anchors were coded as “mood,” while the remainder were coded as “cognition,” “physical,” “behavior,” or “motivation.” For the thought scale, 69% of the anchors were coded as “cognition,” with the remainder coded mostly as “mood” and “behavior.” In contrast, when users identified personal anchors for the pilot study wellness scale (Table 7), the anchor types were coded as “behavior” (38%), followed by “cognition” (28%), “mood” (17%), “physical,” (10%) and “motivation” (7%). In terms of subtypes, anchors for thought content (17%), sleep (12%), thought process (10%), negative mood (10%), energy (9%), and social interactions (8%) accounted for two-thirds of the subtypes.