Developing a Multimodal System for Automated Passive Detection of Suicidal Ideation

In 2020 approximately 1.2 million people in the United States made serious suicide attempts, around 46,000 died. People who are at risk for suicide usually go into moments of crisis when the risk for suicide becomes more severe. The ability to detect when these moments of crisis are occurring is critical to providing timely and effective treatment. A text message or suggestion to call suicide prevention services (such as the lifeline) provided during a time of crisis can save lives. Currently, our ability to detect an acute suicidal crisis relies on a person either self-presenting the crisis or disclosing it in the context of an interview or questionnaire. These methods for identifying crisis are limited because crises can easily happen outside of the context of a specific interview and people are often unmotivated to bring themselves to a hospital or call a provider or crisis services. In this study, we aim to create an objective passive sensing system that can detect a suicidal crisis as it is occurring. One of the key markers of a developing suicidal crisis is an increase in suicidal ideation (wishes to be dead, thoughts about dying, planning one's death, etc.). Our passive sensing system will be designed to detect increasing suicidal ideation using mobile neuroimaging in a group of people who are at high-risk for suicidal crisis. In order to do this, we will develop a neural signature for suicidal ideation in a group of people with major depressive disorder who are currently hospitalized due to a suicidal crisis using simultaneous fMRI/EEG (Aim 1). We will test that this signature is capable of detecting SI as it occurs naturally (meaning not cued by a task) using a mindwandering task in the scanner (Aim 2). Finally, this signature will then be used to track neural activity during the week following discharge from the hospital, which is a well-known time period for suicide risk. At the same time, study participants will also be rating their experienced level of suicidal ideation six times a day. We will use the neural signature to decode neural activity occurring before each self-reported rating of suicidal ideation. We hypothesize that the signature will be able to predict when participants will report greater or lower SI on the basis of the neural activity preceding that report (Aim 3). These findings would help pave the way for a technology that can provide continuous automated passive detection of SI. Future work along this line of research would aim to develop and test a closedloop automated detection and intervention system that could detect SI occurring in high-risk groups using mobile EEG and provide automated interventions via a smartphone app. By doing so, we aim to ensure that people at high-risk are not lost to care because of the unavailability of treatment providers.