Over 20% of people with psychosis continue to exhibit suicide ideation despite receiving gold-standard interventions. This incomplete success, while problematic, is unsurprising. Interventions on suicide ideation are often chosen because they are effective at the group level, leaving open the possibility of a mismatch between the targeted mechanisms and those driving a specific patient’s thoughts of suicide. Personalized treatments aimed at what causes more severe suicide ideation in each specific person with psychosis should yield better results. But such personalization is not feasible without an efficient, accurate method of determining individual-level causes of changes in suicide ideation’s severity. Our project’s primary purpose is to determine whether a computational approach pioneered by our team can identify individual-level causes of fluctuation in suicide ideation severity.
To begin, we will use Ecological Momentary Assessment (EMA) to capture suicide ideation and its putative determinants in the daily lives of 50 people with psychosis. Potential determinants will include psychosis symptoms and variables implicated in widely-accepted theories of suicide (ex: hopelessness, perceived burdensomeness). We will analyze these EMA data with causal discovery algorithms, which use machine learning to infer causal relations between variables in observational datasets. This procedure will yield network diagrams that describe the causes of more severe suicide ideation within each individual participant as well as on average, across participants (Aim 2). We will explore each diagram’s consistency with widely-accepted theories of suicide.
To maximize confidence in these diagrams’ accuracy, we will develop a novel causal discovery algorithm that accounts for the nested structure of EMA data (Aim 1). We will conduct simulation studies that quantify this algorithm’s ability to recover known causal relations from synthetic EMA data, and benchmark it against algorithms used in our past EMA work. To further gauge confidence in the accuracy of network diagrams, we will assess whether they capture the inter-individual heterogeneity in the causes of suicide ideation that past research suggests is present among people with psychosis.
Our project will have numerous immediate effects on suicidology. By assessing whether causal discovery algorithms can produce diagrams describing what causes more severe suicide ideation in specific individuals, and gauging confidence in these diagrams via multiple complementary strategies, our project will illuminate whether causal discovery approaches could inform personalized treatment for suicide ideation in people with psychosis. By ascertaining whether and how positive symptoms impact suicide ideation’s severity, our project will enable tailoring of widely-accepted theories of suicide to people with psychosis. The causal discovery algorithm we develop will enable suicidologists to more effectively dissect out within- and between-persons causal pathways that culminate in suicide ideation; improved knowledge of these pathways will support development/selection of better individual and population-level interventions.
After the grant period, we will examine whether our network diagrams predict treatment-related changes in suicidality among patients with psychosis or treatment-resistant depression. If so, we will conduct a large-scale RCT pitting treatments for suicidality selected via therapist intuition against those selected based on our network diagrams.
