Predicting long term survival in myxofibrosarcoma: Development and evaluation of Machine learning models for 2- and...

Myxofibrosarcoma is a rare soft tissue sarcoma with complex growth patterns and variable survival outcomes, making prognosis challenging. Machine learning (ML) offers a promising approach to enhance survival predictions. This study explored key predictive features and assessed ML model performance and generalizability.

We included 3400 patients from the National Cancer Database for model training, and 112 patients from our institutional registry for external validation. We employed a Random Forest algorithm to identify predictive features. These features were used to train four models: Random Forest, single Neural Network, Neural Network Ensemble, and Weighted Ensemble combining all approaches. Model performance was assessed using area under the curve (AUC), F1-scores, and Brier scores.

Tumor size, age, and metastatic status were identified as key predictors of survival across models. In internal validation, multiple models showed promise, but the Neural Network Ensemble demonstrated balanced performance for both 2- and 5-year overall survival. In external validation, the Neural Network Ensemble outperformed the Random Forest model, and achieved AUC of 0.965 and 0.934 for 2- and 5-year overall survival, respectively.

We successfully developed and validated ML algorithms that accurately predicted survival outcomes in patients with myxofibrosarcoma, with insights for model selection.