Systematic Evaluation and Machine Learning-Based Driver Analysis of Cadmium-Induced Osteotoxicity in Mammals

Cadmium (Cd) is a persistent environmental pollutant with a biological half-life of several decades in animals. However, the osteotoxic effects in the mammalian skeletal system and their underlying mechanisms remain unclear. Using 374 control-observation comparisons from 47 studies, we systematically evaluated Cd-associated skeletal toxicity through meta-analysis and machine learning. The results indicate that Cd exposure significantly impairs overall bone health (13.68%, P < 0.001). TRACP5b (61.52%, P < 0.001) and CTX-I (35.29%, P < 0.01) showed the strongest responses, whereas BMD (10.28%, P < 0.001) and body weight (6.06%, P < 0.001) changed more moderately. TRACP5b responded approximately sixfold more strongly than BMD, supporting bone resorption markers as more sensitive early-warning indicators of Cd-induced skeletal toxicity. The XGBoost model demonstrated the best performance on the test set (R² = 0.752, RMSE = 0.325), highlighting its strong potential for predicting Cd-induced bone toxicity risk. SHAP analysis identified response indicators (9.5%-23.3%) and Cd concentration plus duration (18%) as major contributors to osteotoxic response variation. Biologically, these findings indicate that Cd-associated osteotoxicity is characterized by heterogeneous sensitivity across bone-remodeling endpoints, while exposure concentration and duration modulate the magnitude of this multi-target skeletal response. These findings provide a basis for refining cadmium risk assessment and screening early-warning biomarkers.