Serious mental illnesses (SMI), such as schizophrenia and bipolar disorder, are disabling and intrusive disorders that can impact educational attainment, work productivity, social functioning, and life expectancy [1]. Psychosocial and psychopharmacological interventions can improve symptoms and health, employment, and education [2]. For individuals with schizophrenia, maintenance antipsychotic treatment is the primary approach for treatment; however, on average, 42% of patients with schizophrenia have been found to be nonadherent to medication treatment, often resulting in worse symptomatology [3,4]. Worsening symptoms can lead to decreased social interaction and make patients less likely to seek help [5,6]. Thus, patient symptoms can worsen or relapse while presenting limited opportunities for mental health clinician awareness, and little opportunity to intervene.

Although warning signs for symptom exacerbation or relapse differ among individuals, they are often consistent within an individual over time. Exacerbations start as mild and increase as symptoms worsen, with 75% of families reporting observable changes between 2-4 weeks before relapse [7]. A study found that the interval between worsening symptoms and relapse was at least one week in 50% of patients [8]. Given a window of opportunity to intervene, interventions were developed and tested to prevent relapse or hospitalization through intermittent drug techniques; however, these interventions showed modest to no effectiveness when implemented [4]. There was noted difficulty getting individuals to reliably self-monitor and report their warning signs early enough for medication changes to prevent relapse, suggesting a need for additional patient support and more reliable monitoring.

Mobile apps and passive mobile sensing technologies could be an approach to help people with SMI self-monitor [9]. Passive mobile sensing can use sensor and phone data from mobile devices to detect health-related behaviors (eg, exercise, social interaction, sleep), which may allow one to detect changes in symptoms and functioning across various behavioral domains like activity, sociability, and sleep [10,11]. These associations have been found in patients with SMI. For example, behavioral indicators for psychiatric symptoms of mood and anxiety disorders monitored through mobile sensing were found to predict clinically assessed symptoms of depression and posttraumatic stress disorder [12]. In patients with bipolar disorder, depressive and manic symptoms were associated with changes in activity and phone communication, while other studies found activity and location to be correlated with mood states [13-16]. In patients with schizophrenia, self-reported stress, depression, and psychotic experiences were also associated with mobile sensor data related to sleep, activity, and communication, signaling the potential for developing predictive models from passive mobile sensing data to aid in identifying changes in symptoms and functioning [17].

Approximately 75% of individuals in treatment for SMI report being willing to participate in digital interventions for stress, health, and mental health [18]. Additionally, patients with SMI find computerized self-assessments acceptable provided they have a clinical application and can help prevent serious adverse events for themselves or others. In a previous study of patients with schizophrenia, passive mobile sensing was found feasible when participation was voluntary [19]. This suggests at least a degree of acceptability among patients with SMI to use passive mobile sensing for managing SMI symptomatology.

Although a few previous mobile sensing studies have centered on individuals with SMI, very few have focused on US veterans—a population with unique physical and mental health needs that may also benefit from this technology. Approximately 4% of veterans have SMI and an estimated 44% of veterans with SMI have moderately severe or severe symptoms [20]. According to a review of health care utilization by veterans, it was found that veterans with SMI were the highest users of inpatient services, with 22% and 41% of patients requiring hospitalization or an emergency department visit, respectively [21]. There are significant effects on patients with SMI and their families associated with worsening symptoms that can lead to diminishing quality of life and medical health. Relative to patients with other mental illnesses, patients with SMI had the greatest percentage of deaths during a 1-year follow-up period [21]. The human costs associated with relapse in patients with SMI are significant, and there is a need to address the difficulty with getting patients to reliably self-monitor and promptly report their warning signs.

A 2014 Veterans Affairs study found that 60% of 249 patients with SMI had a smartphone, with 64% of smartphone users also using apps on their phones [22,23]. Veterans with SMI used similar smartphone functions as the general population and typically kept their phones with them most or all of the time, suggesting that passive mobile sensing may be feasible for this population [22]. Additionally, as time is required for the detection of worsening symptoms and intervention, the period between warning signs and potential relapse is noteworthy. Veterans with SMI on average experienced early warning signs of worsening symptoms 208 days prior to relapse, similar to patients in the general population with SMI [24]. Considering smartphone usage and period of time prior to relapse, passive mobile sensing may be a viable means for patients with SMI to improve outcomes through the development of self-monitoring tools that can help detect or predict worsening symptoms, allowing for earlier assessment, intervention, and improved outcomes.

This study aims to determine the feasibility, safety, and acceptability of passive mobile sensing in patients in treatment for SMI, and to develop and study predictive analytic models to identify associations between symptoms and behavioral domains measured.

The objectives are as follows:

The study consists of two phases: a user-centered design phase and a mobile sensing clinical trial phase (Figure 1). The study uses a passive mobile sensing app that displays data gathered in the form of interactive patient dashboards. The user-centered design phase (Phase I) focuses on usability of and perceptions regarding the app. The mobile sensing phase (Phase II) studies the safety, feasibility, and acceptability of passive mobile sensing, and uses data to develop predictive models for behavioral domains associated with SMI symptoms.

All participants are recruited from the VA Greater Los Angeles Healthcare System (GLA VA), with clinicians included if they are providing care for veterans with SMI. The following inclusion criteria are used: (1) clinical diagnosis of SMI, defined as schizophrenia, schizoaffective disorder, or bipolar disorder, (2) risk for symptoms based on having had, during the past year, psychiatric hospitalization, psychiatric emergency care, lived at a crisis program, or more than 6 outpatient visits, and (3) ownership of a phone running Android OS with a data plan. The following exclusion criteria are used: (1) under the age of 18 years and (2) has a conservator/legally authorized representative.

A list of potentially eligible participants was identified using Veterans Affairs data systems. In addition, flyers were posted at clinics and announcements were made at group meetings to invite participation. Individuals interested in participating had a phone screening verifying diagnosis, age, and last treatment. After confirming potential eligibility, prospective participants were scheduled for an in-person or virtual screening visit at which further information was provided and eligibility assessments were re-verified. Additionally, individuals were asked if they own a smartphone running Android OS with a data plan. If an individual was eligible, full informed consent was obtained, and they were enrolled.

The study has been approved by the Institutional Review Board of the GLA VA (project number IIR 19-392).

Following consent and enrollment, participants underwent a baseline assessment. This consisted of an interview regarding demographics, symptoms, functioning, treatment utilization, mobile phone usage, housing status, substance abuse, and cognitive deficits (Table 1). Additionally, throughout the sensing phase of the study, weekly assessments were conducted by phone with participants to assess their sleep, sociability, activities, symptoms, and safety. The rate of serious adverse events during participation will be monitored.

At the baseline assessment visit, the app was downloaded onto the participant’s phone and activated. Every time the app is activated on a device, an individualized ID is generated by the server. Following activation, the app runs in the background and collects data passively without further action from the user. The server generates an automated report daily that includes the IDs of participants who have not recently transmitted any data, and research staff contact participants to investigate and troubleshoot.

The app collects passive behavioral data in three domains that have been previously shown to be measurable with these data: activities, sleep, and sociability (Figure 3). The app uses the AWARE application programming interface (API) that manages the querying and data packaging, transmission, and storage. The input data are drawn from phone utilization, sensors, software on Android phones that process sensor data, and Google APIs (Table 2). The Google Fused Location API and Google Activity Recognition API are used for postprocessing of sensor data. Although most recent phones contain most or all of these data sources, some older phones may lack certain sensors; the app uses all available sensor data.

The data collected from the app are encrypted and cached locally on the device, and transmitted to the platform’s remote server when Wi-Fi is available. If Wi-Fi is not available, data are cached locally until a connection is established. Data are transmitted through an encrypted channel to a web application on a server where it is stored in a secure, encrypted format. Following confirmation of successful transmission, data are deleted from the phone.

A random subsample of participants was selected to partake in 3- and 9-month evaluation assessments, which are semistructured interviews that are recorded and transcribed until data saturation is achieved. Interviews assess for acceptability of passive sensing, usefulness, ease of use, attitudes, and behavioral intention toward use. Participants also provide additional feedback regarding the app. There is a 9-month follow-up and final visit for all participants, in which measures from baseline are readministered.

During informed consent, participants are informed about how the system works; how data are collected, transmitted, and stored; and who has access to the data. The mobile app communicates with a secure server where sensor data are housed. Data are encrypted when at rest on the mobile device, in transit to the server, and on the server.

This study will determine the feasibility and acceptability of passive mobile sensing, and estimate the extent to which behavioral data predicts behaviors and psychiatric symptoms in patients with SMI. We anticipate that passive mobile sensing will be feasible and acceptable in most patients with SMI. We hypothesize that passive mobile data will be associated with behaviors and psychiatric symptoms, and that these associations will vary among individuals.

Passive mobile sensing could be an innovative method to improve self-monitoring of behaviors associated with worsening symptoms of SMI. There is increasing support for the validity of behavioral indicators for mobile sensing platforms as predictive of psychiatric symptoms [12,41]. Additionally, the feasibility and acceptability of passive mobile sensing in patients with SMI has been explored with encouraging results [41,42].

Some challenges are expected in recruitment, enrollment, and retention in the patient population. To overcome these difficulties, strategies like monitoring and tracking patient flow, fair and appropriate compensation, and flexible assessment scheduling are used. Due to the COVID-19 pandemic, it has been difficult to conduct in-person screening visits; consequently, the study is adopting virtual screening, consents, and assessments to recruit and enroll participants.

Additionally, as the data are collected from smartphones, we expect missing data points. Various approaches will be considered to adjust for missing data when creating activity profiles. One such approach is the similarity assumption that the proportion of time that a participant spends at a given activity level during observed times is like that during unobserved times, which is similar to the missing at random assumption in longitudinal data analysis. Another approach is to use the planned machine learning approaches to evaluate missing data patterns.

If found to be feasible, acceptable, and safe, passive mobile sensing can become a clinical tool for patients with SMI, allowing access to dashboards of activity, sociability, and sleep that can be used as a tool for monitoring symptoms and behaviors. The adoption of such apps may also be viable and its integration into clinical care may further improve clinical outcomes for patients with SMI. The findings of this study hope to guide the development of passive sensing and modeling that could dynamically assess mental health and identify the risk for worsening illness. It hopes to eventually guide the development of a platform that is acceptable and desirable to patients, helping patients and their clinicians to monitor their clinical status, identify the risk for relapse, and allow for early intervention.