Bayesian Risk Prediction Model For Colorectal Cancer Mortality Through Integration Of Clinicopathologic And Genomic Data

This multivariable study on colorectal cancer survival prediction used the BART model to demonstrate comparable model performance compared to other nonlinear learning models and LASSO linear regression. Seven significant and stable variables were identified as predictors for 5-year colorectal cancer-specific survival in stage II-III: positive lymph node count, negative node count, depth tumor of invasion, MSI status, tumor site, age, and extent of extraglandular necrosis. A risk prediction calculator was developed to categorize patients into low-, intermediate-, and high-risk groups.