Integrating laboratory tests with FIGO staging using machine learning improves long-term prognostic assessment in endometrial cancer.
Summary
This retrospective study developed machine learning models to improve long-term prognostic assessment in endometrial cancer. By integrating routine laboratory tests with FIGO staging, researchers demonstrated a significant enhancement in predicting 5- and 10-year overall and disease-free survival. The K-nearest neighbour model achieved an AUC of 0.876 for 10-year overall survival, outperforming FIGO staging and age alone. This low-cost and accessible approach could optimize prognostic assessment, especially in settings where advanced molecular profiling is limited.
Analysis
CLINIQUE: This study proposes an innovative prognostic approach for endometrial cancer, integrating easily accessible routine biomarkers. The K-nearest neighbour model demonstrated superior predictive capability for 10-year overall survival (AUC of 0.876) compared to standard methods, making it a promising prognostic biomarker. This finding could justify prospective validation trials to assess its impact on risk stratification and therapeutic decision-making. Clinical impact could be seen in the medium term (3-5 years), particularly in resource-limited settings, by offering a more accurate prognostic assessment without requiring expensive molecular profiling. BIOINFO: The study developed ten machine learning models, including K-nearest neighbour, SVM, ANN, and Extra Trees, to predict overall and disease-free survival. Integrating 7 selected laboratory measures with FIGO staging yielded robust performance, with AUCs up to 0.897 for 5-year disease-free survival. While the model architectures are not detailed as "novel," the application of these algorithms to routine laboratory data to improve prognosis is relevant. Validation on an internal test set is a good start, but external validation would be necessary for reproducibility. If deployed clinically, this model could complement or replace less accurate prognostic tools, helping to prioritize patients for intensive treatments or closer follow-up, especially in centers without access to advanced profiling technologies.