Research on the Prediction of Coal Workers' Pneumoconiosis Based on Easily Detectable Clinical Data: Machine Learning...

Area of Science:

• Occupational Medicine
• Pulmonary Medicine
• Data Science & Machine Learning

Background:

• Coal workers' pneumoconiosis (CWP) is a significant occupational lung disease causing irreversible damage.
• Early CWP prediction is crucial for halting pulmonary fibrosis progression.
• Current prediction methods using imaging and biomarkers are costly and inconvenient.

Purpose of the Study:

• To develop a machine learning (ML) model for CWP prediction using easily obtainable clinical data.
• To explore the utility of occupational history, lung function, and blood indicators for CWP risk assessment.

Main Methods:

• A prediction framework utilized a dataset with multidimensional clinical features.
• Six ML algorithms were trained and validated using cross-validation and a test set.
• Hyperparameter optimization and model interpretability (Shapley Additive Explanation) were performed.

Main Results:

• All ML models demonstrated high predictive performance, with small differences on the test set.
• Light gradient boosting and categorical boosting achieved high AUC (0.974-0.975).
• Key predictors identified include age, forced expiratory volume/forced vital capacity, and platelet count.

Conclusions:

• Combining clinical features with ML algorithms shows promise for CWP prediction.
• This approach offers a convenient and effective strategy for early CWP diagnosis and intervention.
• Clinical biomarkers possess independent predictive value for CWP risk.