Content area
Abstract
Each year millions of people develop suicide plans. These plans are consistently cited as critical indicators of suicide risk by clinicians, researchers, and suicide prevention agencies. Having the ability to accurately predict suicide plans, even without a full understanding of the causal risk mechanisms underlying them, would allow us to better identify individuals who require intervention–a critical step toward suicide prevention. The present study sought to describe the nature of suicide plans in the short term and predict suicide plans and related outcomes. Secondary data analysis was performed on an existing longitudinal dataset composed of 1021 participants. Participants were asked to complete suicide-related measures at baseline and 3-, 14-, and 28-days post-baseline. Analyses focused on three key outcomes: (a) suicide plans, (b) suicide preparatory behaviors, and (c) suicidal intent. To describe the nature of suicide plans in the short term, cross-sectional associations were calculated between these outcomes and relevant psychological constructs. To examine our ability to predict these outcomes, univariate logistic regression, multivariate logistic regression, and machine learning models were implemented.
Results indicated that nearly all participants in the study sample endorsed a lifetime history of suicide plans. While most participants endorsed thinking of their suicide plan over the 28-day follow-up period, intent to act on these plans was only moderate on average. All suicide plan features (i.e., method, time, place) generally demonstrated consistent, yet small, cross-sectional associations with related constructs. Univariate predictors performed at approximately chance level prediction accuracy, and in general, predictive model performance was poor for traditional multivariate and LASSO models and excellent for Random Forest models. Overall, predictive results supported a complex conceptualization of risk for suicide plans, preparation to act on plans, and increased suicidal intent. The study limitations and implications for research and clinical practice are discussed.
Online supplementary materials accessible through ProQuest are a full model correlation matrix and LASSO feature selection table.
Details
