Driver gene mutations and clinical features predict bone metastasis risk in NSCLC: a logistic regression model.

Bone metastasis affects approximately 40% of patients with non-small cell lung cancer (NSCLC), significantly impacting patient survival and prognosis. However, the molecular mechanisms linking driver gene mutations to bone metastasis remain poorly understood. This study retrospectively analyzed 362 East Asian population patients with NSCLC to investigate the association and to construct a prediction model based on machine learning. EGFR mutations were the most common genetic alterations (57.7%) and were significantly associated with an increased risk of bone metastasis, with odds ratios of 2.395 for synchronous metastasis and 2.228 for metastasis within one year of diagnosis. Among EGFR subtypes, 19-Del and L858R showed the strongest association with bone metastases. In contrast, other common driver mutations, such as KRAS and ALK, were not significantly associated with bone metastasis. Additionally, advanced T stage (T3-T4), advanced N stage (N2-N3), and adenocarcinoma histology were also significant risk factors for bone metastasis. A logistic regression model, constructed using predictors selected by LASSO regression, showed optimal and stable performance, with validation set AUCs of 0.704 for synchronous bone metastasis and 0.708 for metastasis within one year. SHAP analysis revealed that T stage and EGFR mutation status contributed most to the model's predictions. These findings suggest EGFR mutation is a key molecular driver of early bone metastasis in NSCLC. Integrating molecular markers with clinical features into an interpretable prediction model can improve personalized risk assessment for bone metastasis and facilitate timely intervention.