Survey-Weighted Explainable Gradient Boosting for Regional Disparities in PHQ-9-Defined Depression in Korea: 2024 Korea Community Health Survey
Year: 2026
Type: article
Abstract: PURPOSE: To analyze regional disparities in depression defined by the Patient Health Questionnaire-9 (PHQ-9) among Korean adults using spatial statistics and survey-weighted prediction models based on the 2024 Korea Community Health Survey (KCHS), and to visualize region-specific contributing factors using SHapley Additive exPlanations (SHAP) METHODS: This cross-sectional secondary analysis evaluated 231,469 respondents. Survey weights were applied to logistic regression, XGBoost, and LightGBM models to classify concurrent depression status, operationally defined as a PHQ-9 score of 10 or higher. Spatial autocorrelation was assessed using global Moran's I and Local Indicators of Spatial Association. SHAP values quantified global feature importance and the spatial heterogeneity of dominant positive mean contributors across 250 harmonized municipal units RESULTS: The national weighted prevalence of PHQ-9-defined depression was 4.1 percent, with weak positive global spatial autocorrelation and regional variation ranging from 0.9 percent to 8.2 percent. In the independent test set, LightGBM achieved an area under the receiver operating characteristic curve of 0.92, an area under the precision-recall curve of 0.43, and a Brier score of 0.03. No local association remained significant after false discovery rate correction. SHAP analysis identified perceived stress, subjective happiness, and subjective health status as the most influential global predictors. Mapping revealed spatial heterogeneity: perceived stress was the most frequent dominant positive mean contributor, whereas other municipal units showed subjective health status, subjective happiness, smoking status, sleep duration, income, and other contributors. Regional contributor rank retention varied across bootstrap samples (median, 72.5%). CONCLUSION: Psychosocial and structural characteristics contributed to concurrent classification of PHQ-9-defined depression, with exploratory geographic variation in aggregated SHAP contributions. Combining explainable machine learning with spatial statistics may inform further local assessment. IRB APPROVAL/EXEMPTION: This study was exempted from review by the Institutional Review Board of Kangbuk Samsung Hospital (KBSMC IRB No. 2026-02-008).
Source: Asian Nursing Research
Authors: SeongKwang Kim, SukJae Park, YoSeop Ryu
Institutions: Kangwon National University, Kangbuk Samsung Hospital
Language: English
Subfield: Psychiatry and Mental health
Field: Medicine
Domain: Health Sciences
Open Access status: gold
Estimated APC paid: $1,500
Funder: Samsung
OpenAlex ID: W7225098703
