UC Berkeley

Despite decades-long efforts devoted to enhancing the understanding, prediction, and prevention of death by suicide, suicide rates have continued to rise both in the United States specifically and in many countries worldwide. Although the advent of machine learning techniques has improved our ability to predict suicidal thoughts and behaviors (STBs), few studies have focused on the short-term prediction of these phenomena. Furthermore, the increasing recognition of the individual- and time-varying nature of STBs necessitates the use of individualized predictive models to detect short-term STBs person-by-person with greater precision. In the present study, I used ecological momentary assessment (EMA) data collection methods with idiographic analytic approaches to better describe and predict short-term suicidal ideation and its risk factors on a person-by-person basis. Key factors measured via these EMA methods included variables related to several prominent theories of suicidal behavior, including the interpersonal, hopelessness, and three-step theories of suicide, and key emotions previously demonstrated to predict suicidal thinking. I also employed a series of machine learning techniques to examine whether these approaches could enhance individualized prediction models for short-term suicidal ideation. With a largely female sample of ten (N = 10) individuals that reported at least one suicide attempt in the past year or reported intense ideation for more days than not over the past month. These individuals were also largely Caucasian and identified as sexual minorities. Results demonstrated that short-term suicidal ideation and its risk factors displayed considerable variability over time. Further, results also indicate that individualized models can produce reliable predictions of short-term suicidal ideation—and that these predictions could be further improved by employing machine learning techniques. Furthermore, both auto-regressive and machine learning-based models, on average, outperformed individualized models derived from the interpersonal, hopelessness, and three-step theory of suicide. Taken together, the present findings may represent a first step toward developing a more precise and individualized approach for understanding and preventing death by suicide through better modeling and predicting short-term suicidal ideation and its risk factors. Specifically, these findings suggest that the combination of EMA methodology, idiographic modeling, and machine learning can be used to effectively identify sets of risk factors related to promiment theories of suicide and past research to predict short-term suicidal ideation person-by-person with high precision.