Related Experiment Video
Updated: May 22, 2026
A Mouse Model of Incompletely Resected Soft Tissue Sarcoma for Testing (Neo)adjuvant Therapies
Published on: July 28, 2020
Development and temporal validation of a machine learning-based model to predict postoperative recurrence and guide
Zhenguo Zhao1, Xinyu Li1, Xinfeng Wang2
1Department of Orthopedics, National Cancer Center/National Clinical Research Center for Cancer/Cancer Hospital, Chinese Academy of Medical Sciences and Peking Union Medical College Beijing 100021, China.
Background:
Soft tissue sarcoma (STS) is highly heterogeneous and has a high risk of recurrence so that the accurate prognosis of postoperative recurrence and the value of radiotherapy are critically important. We developed a machine learning model to predict postoperative recurrence in soft tissue sarcoma patients, providing a data-driven tool to optimize adjuvant radiotherapy decisions.
Methods:
We retrospectively analyzed 642 STS patients who underwent radical surgery at the China National Cancer Center from 2010 to 2025. In order to determine the essential predictors and build strong models, we used a machine learning approach based on wrapper methods. The performance of models was rigorously evaluated using temporal validation, where the concordance index (C-index), time-dependent receiver operating characteristic (ROC), and decision curve analysis (DCA) were used to identify the most effective predictive architecture.
Results:
There are a total of 11 feature subsets are identified, which are combined with 11 machine learning algorithms in a combinatorial manner, resulting in 121 predictive models. Among these models, the Cox proportional hazards model combined with Random Survival Forests (COXPH+RSF, CRM) demonstrates the best predictive performance. The C-index for CRM is 0.923 (95% CI 0.878-0.935) in the training cohort, 0.867 (95% CI 0.850-0.875) in the cross-validated training cohort, and 0.807 (95% CI 0.765-0.819) in the temporal validation cohort. Time-dependent calibration curves, time-dependent ROC curves and DCA evaluation confirms that the CRM achieves high predictive precision and clinical utility. We also release an open-access online platform to host our model and to support its practical application. And staging system based on the CRM provides a new clinical reference for determining postoperative adjuvant radiotherapy strategies in this patient cohort.
Conclusions:
The CRM demonstrates the best predictive capacity concerning recurrence after surgery in STS patients, which could be of immense potential to assist clinicians in assessing disease severity, guiding patient follow-up, and informing adjuvant treatment strategies.
