Application of machine learning in depression risk prediction for connective tissue diseases

Abstract

Abstract This study retrospectively collected clinical data from 480 patients with connective tissue diseases (CTDs) at Nanjing First Hospital between August 2019 and December 2023 to develop and validate a multi-classification machine learning (ML) model for assessing depression risk. Addressing the limitations of traditional assessment tools, six ML models were constructed using univariate analysis and the LASSO algorithm, with the categorical boosting (Catboost) model emerging as the best performer, demonstrating strong predictive ability across different depression severity levels (none_F1 = 0.879, mild_F1 = 0.627, moderate and severe_F1 = 0.588). Additionally, the study provided an interpretation of the best-performing model using SHAP and developed a user-friendly R Shiny application ( https://macnomogram.shinyapps.io/Catboost/ ) to facilitate clinical use. The findings suggest that the Catboost model represents a significant advancement in assessing depression risk among CTD patients, highlighting the potential of ML in enhancing mental health management for this patient population.

Keywords

• Connective tissue disease
• Depression
• Machine learning
• Multi-classification algorithms
• Catboost