Concurrent and Prospective Prediction of Suicidal Ideation in Adolescents Using Multimethod Data and Machine...

Despite an increased focus on prevention, rates of suicidal thoughts and behaviors have failed to decline and robust predictors have yet to be identified, highlighting a critical need to integrate multiple risk processes across methodological approaches in predictive models.

The current pilot study leveraged machine learning with multimethod data to predict concurrent and prospective suicidal ideation (SI) in a sample of adolescents 14 to 17 years of age (N = 165) oversampled for depression and depression risk. Predictors included clinical diagnoses and comorbidity load, interviewer-rated chronic stress, self-reported internalizing symptoms, daily experiences of positive and negative affect, and neural measures of emotion processing and reward responsiveness. The presence or absence of SI was measured at baseline using self-report and a clinician interview. SI was self-reported again at a 6-month follow-up and was reassessed via a clinician interview approximately 1 year after baseline. Random forest classification models with a synthetic minority oversampling technique were implemented and cross-validated.

Random forest classification outperformed logistic regression, predicting SI with high precision and recall both concurrently and prospectively (F1s = 0.81-0.85). Predictor importance analyses showed that cognitive symptoms of depression and average positive affect were prominent predictors across models. Surprisingly, chronic stress and neural measures demonstrated limited predictive utility.

Findings from this pilot study support the potential of machine learning algorithms with multimethod data in the prediction of SI in adolescence, although replication and extension are needed.

We worked to ensure that the study questionnaires were prepared in an inclusive way. We worked to ensure sex and gender balance, as well as race, ethnic, and/or other types of diversity in the recruitment of human participants. While citing references scientifically relevant for this work, we also actively worked to promote sex and gender balance in our reference list. The author list of this paper includes contributors from the location and/or community where the research was conducted who participated in the data collection, design, analysis, and/or interpretation of the work. One or more of the authors of this paper self-identifies as a member of one or more historically underrepresented racial and/or ethnic groups in science. One or more of the authors of this paper self-identifies as a member of one or more historically underrepresented sex and/or gender groups in science.