Predicting suicidal thoughts and behavior among adolescents using the risk and protective factor framework: A large-scale machine learning approach

Abstract

Introduction Addressing the problem of suicidal thoughts and behavior (STB) in adolescents requires understanding the associated risk factors. While previous research has identified individual risk and protective factors associated with many adolescent social morbidities, modern machine learning approaches can help identify risk and protective factors that interact (group) to provide predictive power for STB. This study aims to develop a prediction algorithm for STB among adolescents using the risk and protective factor framework and social determinants of health. Methods The sample population consisted of more than 179,000 high school students living in Utah and participating in the Communities That Care (CTC) Youth Survey from 2011-2017. The dataset includes responses to 300+ questions from the CTC and 8000+ demographic factors from the American Census Survey for a total of 1.2 billion values. Machine learning techniques were employed to extract the survey questions that were best able to predict answers indicative of STB, using recent work in interpretable machine learning. Results Analysis showed strong predictive power, with the ability to predict individuals with STB with 91% accuracy. After extracting the top ten questions that most affected model predictions, questions fell into four main categories: familial life, drug consumption, demographics, and peer acceptance at school. Conclusions Modern machine learning approaches provide new methods for understanding the interaction between root causes and outcomes, such as STB. The model developed in this study showed significant improvement in predictive accuracy compared to previous research. Results indicate that certain risk and protective factors, such as adolescents being threatened or harassed through digital media or bullied at school, and exposure or involvement in serious arguments and yelling at home are the leading predictors of STB and can help narrow and reaffirm priority prevention programming and areas of focused policymaking.

The use of machine learning on administrative and survey data to predict suicidal thoughts and behaviors: a systematic review

“…Adolescents . Eleven studies assessed predictive risk factors of suicidal thoughts and behaviors in children or adolescents ( 38 , 40 , 49 , 50 , 58 , 63 , 66 , 67 , 75 , 76 ). Czyz et al.…”

COVID-19 Stress, Religious Affiliation, and Mental Health Outcomes Among Adolescents

“…While the current study is unique in its exploration of specific COVID-19 stressors and adolescent mental health challenges, findings are consistent with previous studies linking stressful life events and mental health challenges. For example, previous research demonstrates that stressful life events, such as family conflict [ 26 ], problems in school [ 27 ], and family fighting, [ 28 ] are risk factors for adolescent mental health challenges. When considering life challenges related to COVID-19, evidence suggests that COVID-19-related stressors have strained families and negatively influenced adolescent mental health [ 29 ].…”

Mapping of machine learning approaches for description, prediction, and causal inference in the social and health sciences

“…Studies with this aim have tested, for example, candidate modifiable factors associated with childhood cognitive performance ( 85 ). In another application, gradient boosting and SHAP plots were used to identify the top 10 risk factors of suicidal thoughts and behavior in adolescents, all related to sociodemographic factors and family and peer relationships ( 86 ).…”