Developing a Clinician-friendly Online Tool for Survival Prediction in Colon Cancer Patients: A Bayesian Model Averaging for Risk Factor Selection and Dynamic Nomogram

Translational statistics merges biostatistics and clinical research to communicate research findings effectively. Nomograms, graphical representations integrating independent prognostic factors, are valuable tools in colorectal cancer (CRC) research. Bayesian models for variable selection in survival outcome prediction offer advantages through Bayesian model averaging (BMA). This study aimed to utilise BMA for variable selection and develop a clinician-friendly online dynamic nomogram for survival prediction.

A retrospective study utilised the Cabrini Monash colorectal neoplasia database, including colon cancer patients who underwent surgery. Data on demographics, perioperative risks, treatment details, mortality, morbidity, and survival were collected. BMA was employed for Bayesian variable selection to identify effective risk factors for survival prediction. Sensitivity analyses using Cox-LASSO and imputation of missing data were performed. Prognostic online dynamic nomograms were constructed using selected risk factors and the R-package DynNom.