Identifying Factors Affecting Depressive Symptoms and Incidence of Mental Health Diagnosis within 1 Year among 2SLGBTQ+ Youth During COVID-19 Using Machine Learning Methods

Abstract

Abstract Purpose: There is a lack of longitudinal studies that examine changes in mental health among 2SLGBTQ+ youth during the COVID-19 pandemic. Hence, using a Canadian cohort of young 2SLGBTQ+ (16-29 years old), weidentified the factors that predicted increases in self-reported depressive symptoms scores (n = 882) and those that predicted incidence in diagnosis of mental health problems within 1 year among a subset of the cohort who were never diagnosed with mental health problems at baseline (n = 344) using machine learning techniques. Methods: Random Forest regression and classifier were used to identify factors associated with the outcomes. Data were split into training and test sets; Root Mean Square Error (RMSE) and area under the curve (AUC) were used as the criteria to evaluate model performance on the test set. Results: The top ten predictors of each outcome were identified, nine of them were the same for both outcomes. These shared nine variables were: self-rated mental health (SRMH), adverse childhood experiences (ACE), depressive symptoms, stressful life, internalized homophobia, outness, community connectedness, enacted stigma, and perceived sexual stigma at baseline. Some predictors had a unique non-linear relationship with the outcomes. Conclusion: These analyses suggest that 2SLGBTQ+ specific factors (such as outness) and one’s psychological well-being were the most important factors in predicting one’s future mental health. Social identities such as gender or sexuality appeared to be less significant in affecting one’s mental health. More studies with larger samples are needed to better understand some of the complex non-linear associations.