A prediction model of pulmonary metastasis risk in pediatric patients with stage IIB osteosarcoma in the long bone of extremities.

INTRODUCTION

While epiphyseal plates may resist osteosarcoma invasion, the correlation between epiphyseal involvement (EI) and pulmonary metastasis (PM) or prognosis remains unclear, and no PM prediction models specifically target pediatric patients.

MATERIALS AND

METHODS

This study enrolled 221 patients (≤14 years) with stage IIB osteosarcoma in the long bone of extremities. Using LASSO and multivariate Cox regression analyses, we identified significant risk factors for PM and prognosis, integrating them into a nomogram and nine machine learning (ML) models. After comprehensive performance evaluation, the optimal model was selected to predict 2-year PM risk, stratifying patients into high- and low-risk groups by median risk score.

RESULTS

EI significantly correlated with increased PM risk and poorer prognosis; however, when tumors did not cross the epiphyseal plate, metastasis incidence and prognosis remained comparable irrespective of the tumor-epiphyseal distance. Key risk factors included EI, elevated alkaline phosphatase (ALP), decreased lactate dehydrogenase (LDH), poor chemotherapy response, and elevated LDH ratio. The Random Forest (RF) model showed optimal predictive performance for risk stratification.

CONCLUSION

This study establishes the first pediatric-specific PM risk prediction model for osteosarcoma, enabling personalized management, precise prognosis assessment, and optimized resource allocation, thereby demonstrating artificial intelligence's value in biomedical research.