Machine learning algorithms and web-based prognostic tool for different histological subtypes of osteosarcoma: a retrospective cohort

Background: Osteosarcoma (OSC) is a rare but aggressive bone cancer and predicting survival outcomes remains a critical challenge in clinical practice. This study aims to evaluate the performance of machine learning models in predicting survival outcomes for OSC patients using regression and classification approaches. Methods: We analyzed data from 1,471 OSC patients, with 55.7% male and a mean age of 26.13 years. Descriptive and survival analyses were performed using Python. Regression models (Linear Regression, LightGBM, XGBoost, and Random Forest) were deployed to predict survival as a continuous outcome and evaluated using Mean Absolute Error, Mean Squared Error, and R-squared (R 2 ). Classification models (Logistic Regression, Decision Tree, Support Vector Machine, and Random Forest) were used to predict 1-year, 3-year, and 5-year survival status, evaluated using accuracy, precision, recall, and F1 score. Results: Among regression models, XGBoost performed best, achieving the lowest MAE (3.715) and MSE (20.788) and the highest R 2 (0.213). For classification tasks, Logistic Regression outperformed other models, achieving the highest accuracy for 1-year (0.829), 3-year (0.749), and 5-year (0.719) survival predictions. The Random Forest and Decision Tree models showed competitive performance, while the Support Vector Machine struggled with long-term survival predictions. Conclusion: Machine learning models, particularly XGBoost for regression and Logistic Regression for classification, demonstrate strong potential for predicting survival outcomes in OSC patients. These findings underscore the utility of machine learning in enhancing clinical decision-making and suggest avenues for future research, including incorporating additional clinical variables and advanced modeling techniques to improve long-term survival predictions.