Predicting skeletal fluorosis severity using machine learning across diverse fluoride-exposed populations in China

Area of Science:

• Environmental Health
• Bone Metabolism
• Machine Learning in Medicine

Background:

• Skeletal fluorosis (SF) is a debilitating bone disease from chronic fluoride overexposure, impacting millions globally.
• Current diagnostic methods, primarily radiography, often identify SF at advanced stages, hindering early intervention.
• There is a critical need for advanced tools to predict SF risk and severity for timely public health action.

Purpose of the Study:

• To develop and validate an interpretable machine learning model for assessing skeletal fluorosis severity.
• To identify key predictors of SF severity across diverse fluoride exposure routes.
• To provide a data-driven tool for early risk screening in high-risk populations.

Main Methods:

• A predictive model was built using demographic, environmental, and biomonitoring data from 1,309 individuals in Chinese fluoride-endemic regions.
• Variable selection was performed using LASSO regression, followed by training and validation of five machine learning algorithms.
• Model performance was assessed using AUC, and SHAP analysis was employed for interpretability.

Main Results:

• The Random Forest model demonstrated strong predictive performance with an AUC of 0.875 (training) and 0.832 (test).
• Key predictors of SF severity included pain score, joint function, age, and urinary fluoride (UF) concentration.
• The model effectively captured regional variations in fluoride exposure and SF severity.

Conclusions:

• An interpretable machine learning framework offers a robust method for early skeletal fluorosis risk screening and severity stratification.
• This approach facilitates targeted public health interventions by enabling timely identification of at-risk individuals.
• Data-driven methodologies show significant utility in large-scale environmental health surveillance for bone diseases.