How do explicit, implicit, and sociodemographic measures relate to concurrent suicidal ideation? A comparative machine learning approach

Introduction
Suicide is a leading cause of death, and decades of research have identified a range of risk factors, including demographics, past self-injury and suicide attempts, and explicit suicide cognitions. More recently, implicit self-harm and suicide cognitions have been proposed as risk factors for the prospective prediction of suicidal behavior. However, most studies have examined these implicit and explicit risk factors in isolation, and little is known about their combined effects and interactions in the prediction of concurrent suicidal ideation.
Methods
In an online community sample of 6855 participants, we used different machine learning techniques to evaluate the utility of measuring implicit self-harm and suicide cognitions to predict concurrent desire to self-harm or die.
Results
Desire to self-harm was best predicted using gradient boosting, achieving 83% accuracy. However, the most important predictors were mood, explicit associations, and past suicidal thoughts and behaviors; implicit measures provided little to no gain in predictive accuracy.
Conclusion
Considering our focus on the concurrent prediction of explicit suicidal ideation, we discuss the need for future studies to assess the utility of implicit suicide cognitions in the prospective prediction of suicidal behavior using machine learning approaches.