Soft-Voting colorectal cancer risk prediction based on EHLI components

Nutrition and lifestyle factors are closely related to the incidence of Colorectal Cancer (CRC). Indeed, the adoption of a healthy lifestyle can potentially minimize the risk of CRC in the long term. Identifying people at risk of developing CRC will guide them to screening and motivate behavior change. We therefore developed and validated a model to predict CRC risk based on age and Extended Healthy Lifestyle Index (EHLI) components in a Moroccan population. The CCR Nutrition database was used to drive and validate the proposed model. It is a database from a multicenter case-control study of the Moroccan population with 1496 pairs. In a comparative study, thirteen machine learning models, ten simple and three combined, were developed to find the best one. For a better applicability in daily medical practice, a web application was developed. For the training and testing, 5-fold cross-validation was used. The Soft-Voting classifier based on CatBoost, LightGBM and Gradient-Boosting models provided an increase performance with an average accuracy equal to 0.6583 ± 0.054. Soft-Voting CRC risk prediction based on age and EHLI components would identify individuals at risk of colorectal cancer in the Moroccan population and could contribute to improve prevention by encouraging the adoption of a healthy lifestyle.