CAMPUS-P Clinical Trial
Early detection of suicidal risk in students promises to be an important step toward suicide prevention. We present the promises and challenges of using an informatics platform interfaced through a smartphone app for early detection in an at-risk population of students ages 8-21. The use of AI has been shown to be effective in identifying suicide ideation risk through machine learning models in previously published clinical trials (Pestian et al., 2016).
Hypothesis
The MHSAFE app, a smartphone app with a specific interview process called the MHSAFE probes, can be accepted and seamlessly integrated into the workflow of a school-based therapy session, with the app recording acoustical information of sufficient quality for text translation and analysis. The overall goal of the study is to investigate the feasibility of using technology during a therapy session to collect conversations and pertinent Mental Health data. Ultimately these conversations will be analyzed with artificial intelligence (AI) to identify the risk of aggression, self-harm, and suicide.
Methods
In this IRB approved study, conducted in collaboration with The Children’s Home of Cincinnati (TCH), a large behavioral health services group, multiple mental health interviews were conducted with students ages 8-21 at 8 schools utilizing 10 licensed mental health therapists. Therapists conducted weekly therapy sessions with ongoing student mental health clients which were recorded using the MHSAFE app on a variety of smartphones. The interviews included the MHSAFE probes: open ended questions focused on Hope, Anger, Secrets, Fear, and Emotional Pain. The PHQ-9 and GAD-7 were administered at every interview and the C-SSRS or C-SSRS Since Last Visit were used at the therapists' discretion. Therapists used the MHSAFE app to provide concurrent critical assessment of the student’s mental health status, rating each student’s likelihood for imminent suicidality, aggression, or self-harm. Actionable reports were
managed using a standardized safety plan. The recordings were transcribed using various software, then compared to manual transcription text for accuracy. Data captured will be used to train a machine learning model for early detection of language suggestive of suicidality, violence or self-harm. Therapists also rated the app for ease of use and acceptability in a school based therapeutic setting.
Results
In this 3-month study, 67 students and their guardian(s) provided informed consent and assent. The study enrolled 60 students who generated 267 interviews, of which 240 were evaluable (90% of sessions). Non-evaluable cases reflect the approximate 3 case learning curve for mental health therapists new to the use of this technology. The study safety plan was activated when 3 students expressed suicidal ideation.
Feedback obtained from the mental health therapists in the form of a survey and focus groups on the app’s usability, experience and value was overwhelmingly positive.
Conclusion
The MHSAFE app can be successfully incorporated in a school setting as a point-of-care tool and can collect audio data of high enough quality to be further analyzed with AI. Therapist acceptance of using new technological advances in school-based therapy session was strong. Future advances will provide additional opportunities for app deployment and AI model development to support prevention through early detection of aggression, self-harm, and suicide risk.