Predicting venous thromboembolic events in patients with cancer using a new machine learning paradigm

Abstract

[en] The rise of machine learning in the last decade has facilitated great advances in fields such as medicine, where very powerful models have been developed, capable of predicting certain medical conditions with an accuracy never seen before. The present work is focused on predicting one of the leading causes of death among patients with cancer: venous thromboembolic events (VTE). Over the years, several statistical models based on clinical/genetic data have been developed, and have made it possible to create some risk assessment tools, like the Khorana score [2]. However, none of them are based on machine learning. In this way, we propose a new model that uses advanced machine learning techniques and is able to outperform all models currently available. Furthermore, the model is based on a very recent and promising learning paradigm that has barely been tested, hence it is a great opportunity for us to explore and evaluate it. This breakthrough ultimately has an impact on the patient’s quality of life, improving the ability to detect patients at high risk of developing a VTE, who would benefit from preventive treatment.