Factors associated with mental health literacy among residents of Fengtai District, Beijing: an interpretable machine learning analysis
Year: 2026
Type: preprint
Abstract: Background Mental health literacy (MHL) remains moderate to low in China. Using an interpretable machine learning (ML) approach, we aimed to identify factors associated with MHL attainment among adults in Fengtai District, Beijing, and to quantify their relative contributions. Methods We conducted a secondary analysis of the 2022 Fengtai District Mental Health Literacy Survey, a multi-stage cross-sectional survey. MHL attainment was assessed with the Beijing Resident Mental Health Literacy Monitoring Questionnaire, requiring all three dimension-specific thresholds. Six ML models (logistic regression [LR], random forest [RF], k -nearest neighbors [KNN], decision tree [DT], naive Bayes [NB], and support vector machine [SVM]) were trained on 15 biopsychosocial predictors. RF was prespecified as the primary model. Internal validation used 1000 bootstrap replicates with pooled out-of-bag predictions. Discrimination was assessed primarily by the area under the precision-recall curve (PR-AUC) and secondarily by the area under the ROC curve (AUC). Calibration was assessed by the Brier score and a logistic calibration model. Feature importance was assessed by Gini impurity and SHapley Additive exPlanations (SHAP). Results Of 2004 respondents, 403 (20.11%) attained MHL. All six models exceeded the no-skill PR-AUC baseline of 0.201, with 95% CI lower bounds above it. RF ranked first (PR-AUC 0.364, 95% CI 0.304–0.424), followed by SVM (0.342) and LR (0.325); the others ranged from 0.270 to 0.315, and AUC from 0.604 to 0.684. At the Youden-optimal cut-off, recall ranged from 0.254 to 0.698 and precision was at most 0.354. The RF calibration slope was 1.00 (95% CI 0.84–1.19), the intercept − 0.92, and the Brier score fell from 0.177 to 0.147 after recalibration. Gini impurity ranked age highest (23.03%) and SHAP life satisfaction highest (15.00%); subdistrict, educational attainment, and monthly income were among the top five in both rankings. Life satisfaction, education, income, and prior training contributed positively, whereas older age and insomnia, depressive, or anxiety symptoms contributed negatively. Conclusions Interpretable ML quantified the relative contributions and directions of factors associated with MHL attainment, with educational attainment, life satisfaction, and subdistrict consistently among the leading factors. Given the modest discrimination, the primary model was used for interpretation rather than individual-level screening.
Source: Research Square
Institution: Beijing Fengtai Hospital
Language: English
Subfield: Psychiatry and Mental health
Field: Medicine
Domain: Health Sciences
Open Access status: green
OpenAlex ID: W7225774145