Objectives: We aimed to evaluate the effectiveness of a unique chemotherapy regimen, identify factors influencing overall survival (OS), and compare the predictive performance of six machine learning models in Chinese osteosarcoma patients. Methods: A retrospective analysis was conducted on 390 patients with osteosarcoma who were treated between 2009 and 2019. All patients received standardized neoadjuvant chemotherapy (ifosfamide + methotrexate + adriamycin or ifosfamide + adriamycin + cisplatin, depending on age) and subsequent surgery. Clinical and pathological data were collected. Survival analysis was performed using Kaplan–Meier curves and log-rank tests. Multivariate analysis and survival prediction were conducted using Cox proportional hazards models and six machine learning algorithms [random forest (RF), AdaBoost, CatBoost, Extra Trees, XGBoost, and LightGBM) validated via five-fold cross-validation. Clinical net benefit was assessed using decision curve analysis (DCA). Results: The cohort had a mean age of 19 years, with 62.47% male participants and 88.82% diagnosed at stage II. The 3-year and 5-year survival rates were 76.00% (95% CI: 71.60%–80.40%) and 65.00% (95% CI: 60.20%–69.80%), respectively. Multiple factors—including tumor type, surgical method, recurrence/metastasis, tumor necrosis rate, and serum biomarkers (lactate dehydrogenase (LDH), alkaline phosphatase (ALP), platelet count (PLT), white blood cell count (WBC), and red blood cell count (RBC))—were significantly associated with OS. Among the machine learning models, RF and Extra Trees demonstrated the highest predictive accuracy (AUC = 0.960), followed by CatBoost (0.942), AdaBoost (0.897), LightGBM (0.879), and XGBoost (0.853). Calibration curves showed excellent agreement between predicted and observed survival probabilities. DCA confirmed that RF and Extra Trees provided superior net benefit across a wide range of threshold probabilities. Conclusion: The unique chemotherapy regimen showed superior survival outcomes. Prognostic evaluation should integrate multiple clinical and pathological indicators. Machine learning models, particularly RF and Extra Trees, offer powerful tools for individualized survival prediction and treatment planning in osteosarcoma.

Introduction: Osteosarcoma (OS) is the most common bone tumor, characterized by a high incidence, rapid progression, and frequent metastases. The implementation of chemotherapy has made important progress, while the necrosis rate is limited and the survival rates remain unsatisfactory, therefore novel approaches are needed.Methods: We used proteomic analysis to characterize the molecular landscape of patients exhibiting different levels of chemotherapy-induced necrosis.Results: Patients with low necrosis rate (≤70%) showed distinct expression patterns, with significant upregulation of proteins involved in DNA replication, metabolism, and mitochondrial pathway. The Runx1-related signaling pathway was also identified as potentially involved in disease progression. Remarkably, Mitochondrial Ribosomal Protein L4 (MRPL4) and Macrophage Erythroblast Attacher, E3 Ubiquitin Ligase (MEMA) were identified as hub proteins in MEGENA analysis and the public database. By integrating with immunohistochemistry, the higher expression level was verified in samples of OS patients compared to those of healthy people. Discussion: Overall, our project improves the knowledge of the expression pattern with different necrosis rates of OS samples, and the findings of MRPL4 and MAEA indicate the potential role in chemoresistance and provide new targets for the therapeutic strategy for OS patients with a low necrosis rate.