Suicide attempts of community adolescents and young adults : an explanatory and predictive epidemiological approach

Background: Suicide attempts (SA) among community adolescents and young adults represent a major public health burden. SA rates have remained stable for decades. In order to prevent SAs, risk factors need to be identified, along with their potential for SA prevention. Further, new methodological tools for predicting an individual's SA risk have recently been developed. The pros and cons of these tools need to be empirically evaluated.
Method: In the 10-year longitudinal Early Developmental Stages of Psychopathology (EDSP) study, 3021 community adolescents and young adults were interviewed, using the Munich-Composite International Diagnostic Interview DIA-X/M-CIDI. With the aim of identifying risk factors as well as the potential of SA prevention, we evaluated both a wide range of mental disorders and specific traumatic events (TEs). With the aim of evaluating whether Machine Learning (ML) is a better approach to predict an individual's SA risk, compared to a conventional approach, we empirically compared both approaches.
Results: Except for alcohol abuse/dependence, all of the assessed mental disorders are risk factors for the subsequent first lifetime SA. The TEs physical attack, rape/sexual abuse, serious accident, and witnessing somebody else experiencing a TE are risk factors for a future SA. All of the models we used for individual SA prediction showed comparable results.
Discussion: Specific groups should be targeted when planning to conduct a prevention program, e.g. post-traumatic stress disorder patients or victims of rape/childhood sexual abuse. Our study results do not provide evidence in support of the preferential use of ML in predicting an individual's SA risk. Rather, the preferential use of the conventional prediction model is supported by our data, in combination with considerations of interpretability and practicality.