Predicting Adolescent Suicide Risk From Cellphone Usage Data and Self-Report Assessments

Abstract

As suicide is a leading cause of adolescent death, innovative evaluation of imminent suicide risk factors is needed. This study followed high-risk adolescents who presented with recent suicidal thoughts and behaviors (STB) for six months. They were digitally monitored and periodically observed during in-clinic visits. We aimed to classify their STB levels and identify severe cases based on two types of digital monitoring: (1) weekly self-reported questionnaires by patients and (2) and continuously collected cellphone use data. We present a novel approach for utilizing the immense amounts of unlabeled cellular logs in a supervised classification problem. Satisfying prediction results from both data types showed the feasibility of using digital monitoring for STB prediction. Such a capability may enrich periodic clinical assessments with frequent digital follow-ups and raise awareness whenever necessary.