Development of a soil cadmium bioaccessibility prediction model for human health risk assessment

Soil cadmium (Cd) contamination threatens human health, especially in industrial regions. Conventional risk assessments rely on total metal concentrations while neglecting bioaccessibility and exposure parameter uncertainty, potentially biasing estimates. This study compiled 195 datasets of soil properties and Cd bioaccessibility across China. Correlation analysis and random forest modeling were employed to identify key factors and develop predictive models for bioaccessible Cd content. Total Cd content was the dominant determinant, with soil pH, sand fraction, and iron content exerting additional effects. The random forest model exhibited excellent performance (test set coefficient of determination [R²] = 0.87). Nationwide predictions revealed a pronounced north-south contrast in Cd bioaccessibility (1.38%-55.60%). Incorporating predicted values into Monte Carlo-based probabilistic risk assessment reduced health risk estimates by 3.30-7.61 times, highlighting bioaccessibility as a major source of uncertainty, particularly for children. This model provides an effective tool for rapid prediction of soil Cd bioaccessibility and supports refined health risk assessment and management.