This Merit Award resubmission in response to the RFA CX-18-023 addresses one of the top VA priorities, suicide prevention. Recognizing those at the highest risk of suicidal behavior with an imminent need for acute medical intervention remains a fallible subjective decision based on known risk and protective factors. Unfortunately, the contribution of each of these risk factors is small. Thus, there is an urgent need to develop adequate algorithms to predict imminent suicide risk. The overall objective of this application is to test the value of intrinsic brain activity as a marker of acute suicidal behavior and examine potential clinical correlates. Our central hypothesis is that a neural pattern classifier based on resting state functional connectivity will identify acute risk for suicidal behavior, by discriminating recent suicide attempters from current suicidal ideators, in a reproducible and specific fashion. This application is the progression of our pilot work that used machine learning to show that neural pattern classification of resting state-fMRI data allowed a specific differentiation of recent suicidal attempters (within three days of the attempt) from patients currently endorsing suicidal ideation with 79% accuracy. We plan to test our central hypothesis by using resting state functional connectivity to discriminate depressed Veterans who recently attempted suicide (n=80) from depressed Veterans with suicidal ideation (n=80), and non-suicidal stress controls (n=40). We will build on our previous work, replicating the same strategy that resulted in a trained classifier in a larger independent and more heterogeneous sample, and test whether the addition of demographic, clinical, cognitive and biological variables associated with suicide may improve the classifier accuracy (AIM 1). We will examine the temporal specificity of our classifier testing its ability to discriminate: a) clinically stable suicide attempters: attempters rescanned 5-8 days later when symptom severity had subsided, from suicidal ideators, and b) depressed patients with and without lifetime history of suicide attempts. We will also scan a stress-control cohort of age-, sex-matched non-suicidal controls hospitalized in medical-surgical units and attempt to distinguish them from suicidal ideators (AIM 2). Exploratory AIM1 will be a step towards translation, we will examine resting state functional connectivity obtained in 1.5T and 3T scanners. In exploratory AIM2 we will attempt to identify a responsible mechanism by using regression analysis between the most discriminating connectivity pathways between recent attempters and ideators and suicide attempt intent and lethality.
We aim to test the reproducibility and specificity of a neural pattern classifier to discriminate recent suicide attempters from current suicidal ideators as a proxy measure of acute suicide risk. This neural pattern classifier, directly based on the function of the ultimate agent of human behavior, has the potential to significantly inform on suicide risk assessment, using an already widely available technology.
Approximately 22 Veterans die by suicide each year in the United States. This proposed study will recruit 160 18-65 year-old depressed patients who had recently attempted suicide or are currently experiencing suicidal thoughts, as well as 40 patients currently hospitalized for medical complaints with no suicidal thoughts. In order to maximize our recruitment, we will recruit Veterans from the Northport VAMC, NY, and civilian patients from Stony Brook University Hospital. At least half of the recruited subjects will be Veterans. These patients will undergo an interview and psychological assessments, blood draw, and either or one or two sessions of functional magnetic resonance imaging. We aim to replicate and refine our previous finding of determination of acute risk for suicide with a pattern classifier based on brain resting activity.
