Analysis of depression in El Salvador using machine learning models

Abstract

Introduction. Depression is a frequent mental disorder and one of the leading causes of disability worldwide. It has a multifactorial origin, resulting from the interaction between biological, psychological, social, and structural factors. Objective. Analyze the factors associated with depression in adults and older adults in El Salvador. Methodology. A crosssectional analytical study with a predictive approach was conducted on 7249 participants. A logistic regression model based on machine learning models was applied, trained on 80 % of the data and evaluated on 20 %, optimized through cross-validation and Monte Carlo simulations. The risk profile was categorized using clustering analysis. Results. Depression was associated with anxiety OR 10.385; (95% CI 8.760–12.310), post-traumatic stress disorder OR 4471 (IC 95 % 3257-6138), COVID-19-related stress OR 2.42; (95% CI 1.437–4.092), suicidal ideation OR 1.968; (95% CI 3.257– 6.13), recent discrimination OR 1.338; (95% CI 1.090–1.643), being female OR 1.291; (95% CI 1.072–1.55), unmet basic needs OR 1.192; (95% CI 1.0161.399), and functional disability OR 1.044; (95% CI 1.038–1.051), p < 0.05. The average AUC was 0.836. Clustering analysis identified three groups: high, medium, and low risk. The high-risk group had low social integration and high functional and emotional impairment. The departments of Morazán and Chalatenango had the highest proportion of high risk. Conclusion. Depression is influenced by a complex interaction of emotional, social, and structural factors, with a higher prevalence in women and differences in the geographical distribution of risk, which requires comprehensive and targeted interventions