A Neurocomputational Approach to Discriminate Between Suicide Ideators and Attempters

Many people think about suicide, but only a fraction actually attempt it. Understanding why some individuals only contemplate suicide while others attempt suicide is critical for identifying risk factors specific to suicide attempt (rather than ideation) that can be used to predict any one person's future likelihood of attempting suicide. In this study, we take a critical first step toward developing a suicide attempt “risk algorithm” using machine learning based on a novel combination of clinical, behavioral and neuroimaging variables. For the first time, we will study behavioral and neuroimaging variables associated with decision-making and cognitive control to determine how accurately these measures, in conjunction with known clinical risk factors, discriminate between suicide ideators and attempters. Adopting an “ideation-to-action” conceptual framework for research on suicide, we anticipate that a combination of behavioral deficits in decision-making and cognitive control, lower volumes of associated brain structures, and altered patterns of brain activation on a probabilistic learning neuroimaging task, will most accurately classify suicide ideator versus attempters. We will recruit 100 adults with suicidal ideation, half of whom will have a past suicide attempt, and a comparison group of 50 people without a history of suicidal ideation or a suicide attempt. To identify individuals for this study who are at a high risk for suicidal ideation and suicide attempt, all participants, including the comparison group, will carry a diagnosis of major depressive disorder or borderline personality disorder. Intensity of suicidal ideation and number and medical lethality of suicide attempts will be comprehensively characterized. Known clinical risk factors for suicide attempt (e.g., anxiety, posttraumatic stress, substance use problems, and sexual abuse history) will be measured using a self-report questionnaires and structured clinical interviewing. Sources of information will be individual participants, collateral medical records, and corroborating reports from people well-acquainted with the participants. Participants will complete behavioral tests of decision-making and cognitive control, a high-resolution structural brain scan, and a probabilistic learning task while functional brain activation is measured. These variables will be submitted to a machine learning classification procedure that will identify the optimal combination of clinical, behavioral and structural/functional neuroimaging variables that most accurately differentiates the suicide ideators versus attempters. Overall, our study will make new strides toward identifying a novel combination of behavioral and neuroimaging risk factors, in conjunction with known clinical risk factors, that distinctly reflect the risk for suicide attempt beyond those associated with suicidal ideation. Importantly, our research addresses a critical challenge in suicide prevention research by developing an individual-person-level “risk algorithm” for suicide attempt based on a diverse and comprehensive set of multi-modal risk factors. Ultimately, this line of research will help to direct suicide prevention efforts toward people who have contemplated suicide and are identified through this algorithm as being at the highest risk of attempting suicide, thereby taking crucial steps toward tackling the suicide rate in the United States, Canada and around the globe.