Time-dependent estimates of recurrence and survival in colon cancer: clinical decision support system tool development for adjuvant therapy and oncological outcome assessment.

Abstract

e14500 Background: Unanswered questions remain regarding treatment efficacy in colon cancer (CC), especially those determining high-risk node-negative cohorts that may benefit from adjuvant therapy. We sought to evaluate the use of machine learning and classification modeling to estimate survival and recurrence in CC. Methods: We used the Department of Defense Automated Central Tumor Registry (ACTUR) to identify primary CC patients treated between January 1993 and December 2004. Cases with events or follow-up that passed quality control were stratified into one-, two-, three-, and five-year survival cohorts. ml-BBNs were trained using machine-learning algorithms and k-fold cross-validation, and receiver operating characteristic (ROC) curve analysis used for validation. Results: There were 5,301 cases stratified into cohorts. Survival cohort Areas-Under-the-Curve (AUCs) ranged from 0.85–0.90, positive-predictive-values (PPVs) for recurrence and mortality ranged from 78-84% and negative-predictive-values (NPVs) from 74-90%. Cross-validation showed that the ml-BBNs produce robust individual estimates of recurrence (p<0.001) and mortality (p<0.001) based on readily available clinical-pathological information in the context of adjuvant chemotherapy. Conclusions: Tumor registry data and machine-learned Bayesian Belief Networks produce robust classifiers. These Clinical Decision Support System tools yield clinically relevant estimates of outcomes that may assist clinicians in treatment planning.