Developing a Clinician-Friendly Online Dynamic Nomogram for Survival Prediction in Colon Cancer Patients

This study used Bayesian model averaging to develop an online dynamic nomogram for predicting colon cancer survival, showcasing its value in integrating key prognostic factors for clinical use.

This retrospective study analysed surgical colon cancer cases from the Cabrini Monash colorectal neoplasia database using Bayesian model averaging to identify survival risk factors. Model performance was validated with data from Alfred Hospital, and dynamic online nomograms were developed using the DynNom R package.

The study analysed 2475 colon cancer patients (2010-2021), reporting an overall mortality rate of 6.4 per 100 (95% CI: 5.9-7.1) and post-relapse mortality of 7.0 per 100 (95% CI: 6.4-7.6), with 5-year overall and relapse-free survival probabilities of 0.75 and 0.74, respectively. Bayesian model averaging identified key predictors with posterior inclusion probabilities greater than 0.3, including age, ASA score, stage, chemotherapy, lymph node ratio and cancer side. The model showed strong performance (C-index: 0.84 training, 0.80 validation; AUCs: 0.88-0.91 training, 0.84-0.88 validation), good calibration and clinical utility across thresholds. A dynamic nomogram incorporating these factors was developed using the entire dataset as an accessible online tool to support personalised survival prediction in clinical practice.

This study showcases the robust capabilities of Bayesian model averaging in uncovering key prognostic factors for colon cancer survival. By integrating these predictors into a dynamic online nomogram, it delivers a powerful, clinician-friendly tool that significantly enhances prognostic accuracy and enables personalised, data-driven decision-making in oncology care.