Related Experiment Video
Updated: Jun 13, 2026
Comparison of Predictive Performance of Three Lymph Node Staging Systems in Colorectal Signet Ring Cell Carcinoma Based on Machine Learning Model
Published on: April 18, 2025
Integrating laboratory tests with FIGO staging using machine learning improves long-term prognostic assessment in
Jing Wang1, Yongkang Yang2, Zhichao Chen3
1Department of Obstetrics and Gynaecology, Second Affiliated Hospital of Shantou University Medical College, Shantou, China.
Background:
Limited access to robust biomarkers and regional disparities in standardized pathology and molecular profiling contribute to heterogeneous long-term outcomes among endometrial cancer (EC) survivors. Routine laboratory tests are inexpensive, widely available, and may offer complementary prognostic information beyond tumour staging.
Methods:
A single-centre retrospective longitudinal cohort study of 760 women with EC who underwent surgery between 2010 and 2020 was conducted. Forty-eight routine laboratory tests and cancer features were collected. Ten machine-learning models integrating laboratory measures and tumour stage were developed to predict 5- and 10-year overall survival (OS) and disease-free survival (DFS). To further assess the causal relevance of genetically predicted circulating biomarkers for risk of EC, the analysis was supplemented with bidirectional two-sample Mendelian randomization (MR) and colocalization analysis.
Results:
Using selected features, including seven laboratory measures, the K-nearest neighbour model achieved an area under the receiver operating characteristic curve (AUC) of 0.876 [95 % confidence interval (CI) 0.769-0.956] for 10-year OS in the test set, and outperformed International Federation of Gynaecology and Obstetrics (FIGO) stage and age alone. For 10-year DFS, the support vector machine model showed acceptable discrimination (AUC 0.801, 95 % CI 0.617-0.942), but did not surpass FIGO stage. In additional 5-year analyses, the artificial neural network model achieved an AUC of 0.860 for OS. In contrast, the extra trees model achieved an AUC of 0.897 for DFS, with a clearer gain for OS than for DFS compared with established clinical variables. As a supplementary analysis, MR did not identify a robust causal association between circulating biomarkers and risk of EC.
Conclusion:
Integrating routine laboratory tests with FIGO stage improves long-term mortality prediction in EC survivors. This low-cost and accessible approach may enhance prognostic assessment, particularly in settings where advanced molecular profiling is not available routinely.