Use of an Artificial Neural Network to Determine Prognostic Factors in Colorectal Cancer Patients

Background & Objectives: The aim of this study was to determine the prognostic factors of Iranian colorectal cancer (CRC) patients and their importance using an artificial neural network (ANN) model. Methods: This study was a historical cohort study and the data gathered from 1,219 registered CRC patients between January 2002 and October 2007 at the Research Center for Gastroenterology and Liver Disease of Shahid Beheshti University of Medical Sciences, Tehran, Iran. For determining the risk factors and survival prediction of patients, neural network (NN) and Cox regression models were used, utilizing R 2.12.0 software. Results: One, three and five-year estimated survival probability in colon patients were 0.92, 0.71, and 0.48 and for rectum patients were 0.86, 0.71, and 0.42, respectively. By the ANN model, pathologic distant metastasis, pathologic regional lymph nodes, tumor grade, high risk behavior, pathologic primary tumor, familial history and tumor size variables were determined as ordered important factors for colon cancer. Tumor grade, pathologic stage, age at diagnosis, tumor size, high risk behavior, pathologic distant metastasis and first treatment variables were ordered important factors for rectum cancer. The ANN model lead to more accurate predictions compared to the Cox model (true prediction of 89.0% vs. 78.6% for colon and 82.7% vs. 70.7% for rectum cancer patients). Conclusion: This study showed that ANN model is a more powerful tool in survival prediction and influential factors of the CRC patients compared to the Cox regression model. Therefore, this model is recommended for predicting and determining of risk factors of these patients.