Abstract
BACKGROUND: The built environment is a key intervenable target for public health, yet its nonlinear and threshold relationships with schizophrenia incidence remain poorly understood. METHODS: We analyzed township-level schizophrenia incidence (2019-2023) in Anhui, China, using data from the National Severe Mental Disorder Registration System. Built environment features were derived from multi-source geographic data. We characterized spatiotemporal patterns and modeled associations using Ordinary Least Squares (OLS), Geographically Weighted Regression (GWR), and an interpretable XGBoost model explained by SHAP values. RESULTS: The mean annual incidence was 13.46 per 100,000, with significant spatial clustering (Moran's I = 0.47, P < 0.001). The SHAP-XGBoost model outperformed both OLS and GWR. Key built environment predictors included population density, NDVI, distance to blue space, street connectivity, and blue space area. These factors exhibited complex nonlinear relationships with schizophrenia risk; for example, population density showed a U-shaped association with a risk threshold around 15,000 persons/km(2). Interaction effects between multiple features were also identified. CONCLUSION: This study provides robust evidence that the built environment is significantly and nonlinearly linked to schizophrenia incidence. The identified thresholds and interactions offer concrete, actionable guidance for urban planning aimed at mental health promotion.