All patients installed the Mindstrong Health app and provided informed consent for the use of data in research and product development before enrolling for services. This analysis was conducted under a secondary data analysis protocol to identify clinically relevant associations in active and passive data collection at Mindstrong.
As part of routine clinical care, patients were asked to report their mental illness symptoms every 60 days through the mobile app via the Diagnostic and Statistical Manual of Mental Disorders Fifth Edition (DSM-5) Self-Rated Level 1 Cross-Cutting Symptom Measure-Adult Survey (DSM L1). Due to possible patient burden given the serious mental illness population, the DSM L1 was chosen as a routine clinical screener to comprehensively assess a wide range of clinical domains while ensuring assessment brevity. Specifically, the DSM L1 consists of 23 questions that assess 13 clinical domains, including depression, anger, mania, anxiety, somatic symptoms, suicidal ideation, psychosis, sleep problems, memory, repetitive thoughts and behaviors, dissociation, personality functioning, and substance use (on average, 1-3 items are present per clinical domain) [25 (link)]. The depression domain in the DSM L1 has 2 items, which assess loss of interest in pleasurable activities and depressed mood (feeling sad). These 2 items are comparable to the short-form Patient Health Questionnaire (PHQ2) [26 (link)]. The DSM L1 questionnaire was completed by patients every 60 days via the Mindstrong app. Screenshots of the DSM L1 survey in the Mindstrong app are shown inFigure 1 . A total of 984 assessments of depressive symptoms over a 1-year period were included in this analysis.
In this retrospective observational study, patients were selected based on high compliance with the DSM L1. We intentionally selected individuals with higher compliance because higher compliance (or more repeated measures within the same person) allows examination of the within-person association between self-reports of depressive symptoms and the behavioral measures computed from human-smartphone interactions.
Because the DSM L1 prompted participants to report their depressive symptoms that occurred during the past 2 weeks (or 14 days), we included 14 days of smartphone interaction data prior to each depressive symptom measurement date (symptom-date) in the analysis. The temporal alignment between the DSM L1 survey and smartphone metadata is shown inFigure 2 A.
The app collects smartphone metadata and contains information about interactions with smartphones in an unobtrusive manner. These metadata of device usage and its touchscreen are collected unobtrusively by proprietary software on the Android operating system. The metadata include various touchscreen behaviors (eg, clicking and scrolling), device-level behaviors (eg, turning the smartphone screen on and turning the smartphone screen off), masked keyboard behaviors (eg, typing characters from the left or right side of the smartphone’s keyboard and not the exact character), and change of foreground apps (eg, text messaging apps and entertainment apps). The starting time of each instance of usage of the smartphone device and its touchscreen was recorded with a timestamp at the millisecond level. These metadata are collected locally on the patient’s smartphone and are then transmitted with encryption to a Health Insurance Portability and Accountability Act (HIPAA)– and ISO 27001–compliant cloud storage service (Amazon Web Services). All personnel who can access patients’ metadata and assessment data (including diagnosis, demographics, and self-reported survey data) complete annual HIPAA training. Data were deidentified prior to analysis.
As part of routine clinical care, patients were asked to report their mental illness symptoms every 60 days through the mobile app via the Diagnostic and Statistical Manual of Mental Disorders Fifth Edition (DSM-5) Self-Rated Level 1 Cross-Cutting Symptom Measure-Adult Survey (DSM L1). Due to possible patient burden given the serious mental illness population, the DSM L1 was chosen as a routine clinical screener to comprehensively assess a wide range of clinical domains while ensuring assessment brevity. Specifically, the DSM L1 consists of 23 questions that assess 13 clinical domains, including depression, anger, mania, anxiety, somatic symptoms, suicidal ideation, psychosis, sleep problems, memory, repetitive thoughts and behaviors, dissociation, personality functioning, and substance use (on average, 1-3 items are present per clinical domain) [25 (link)]. The depression domain in the DSM L1 has 2 items, which assess loss of interest in pleasurable activities and depressed mood (feeling sad). These 2 items are comparable to the short-form Patient Health Questionnaire (PHQ2) [26 (link)]. The DSM L1 questionnaire was completed by patients every 60 days via the Mindstrong app. Screenshots of the DSM L1 survey in the Mindstrong app are shown in
In this retrospective observational study, patients were selected based on high compliance with the DSM L1. We intentionally selected individuals with higher compliance because higher compliance (or more repeated measures within the same person) allows examination of the within-person association between self-reports of depressive symptoms and the behavioral measures computed from human-smartphone interactions.
Because the DSM L1 prompted participants to report their depressive symptoms that occurred during the past 2 weeks (or 14 days), we included 14 days of smartphone interaction data prior to each depressive symptom measurement date (symptom-date) in the analysis. The temporal alignment between the DSM L1 survey and smartphone metadata is shown in
The app collects smartphone metadata and contains information about interactions with smartphones in an unobtrusive manner. These metadata of device usage and its touchscreen are collected unobtrusively by proprietary software on the Android operating system. The metadata include various touchscreen behaviors (eg, clicking and scrolling), device-level behaviors (eg, turning the smartphone screen on and turning the smartphone screen off), masked keyboard behaviors (eg, typing characters from the left or right side of the smartphone’s keyboard and not the exact character), and change of foreground apps (eg, text messaging apps and entertainment apps). The starting time of each instance of usage of the smartphone device and its touchscreen was recorded with a timestamp at the millisecond level. These metadata are collected locally on the patient’s smartphone and are then transmitted with encryption to a Health Insurance Portability and Accountability Act (HIPAA)– and ISO 27001–compliant cloud storage service (Amazon Web Services). All personnel who can access patients’ metadata and assessment data (including diagnosis, demographics, and self-reported survey data) complete annual HIPAA training. Data were deidentified prior to analysis.
Free full text: Click here
