Predictive modelling for high-risk stage II colon cancer using auto-artificial intelligence

Background Recently, stratification of high-risk stage II colon cancer (CC) and the need for adjuvant chemotherapy have been the focus of attention. The aim of this retrospective study was to define high-risk factors for recurrent stage II CC using Prediction One auto-artificial intelligence (AI) software and develop a new predictive model for high-risk stage II CC. Methods The study included 259 consecutive pathological stage II CC patients undergoing curative resection at our institution between January 2000 and December 2016. Prediction One software with five-fold cross-validation was used to create a predictive model and receiver operating characteristic (ROC) curve. Predictive accuracy of AI was evaluated using the area under the ROC curve (AUC). We also evaluated the importance of variables (IOV) using a method based on permutation feature importance (IOV > 0.01 defined high-risk factors) to evaluate disease-free survival (DFS). Results The median observation period was 6.1 (range = 0.3-15.8) years. Thirty-seven patients had recurrence (14.3%); the AUC of the AI model was 0.775. Preoperative carcinoembryonic antigen > 5.0 ng/mL (IOV = 0.047), venous invasion (IOV = 0.014), and obstruction (IOV = 0.012) were high-risk factors contributing to cancer recurrence. Patients with 2-3 high-risk factors had lower 5-year DFS than those with 0-1 factor (87.4% vs 62.7%, p < 0.001). Conclusions We developed a new predictive model that could predict recurrent high-risk stage II CC with high probability using auto-AI Prediction One software. Patients with >= 2 of the aforementioned factors are considered to have high risks for recurrent stage II CC and may benefit from adjuvant chemotherapy.