Predicting Post-Traumatic Stress Disorder (PTSD) symptoms in women suffering from breast cancer using machine learning

Abstract

Over the course of the last few years, the phenomenon of breast cancer is constantly increasing in frequency, but the mortality of the disease is decreasing, thanks to the continuous advance of modern medicine and technological tools. A new reality dawns upon us, in which we need to help the surviving women BOUNCE back psychologically and reintegrate them smoothly in our society and workforce. Amidst these unprecedented circumstances, this diploma thesis, part of the European Union – funded BOUNCE project, researches the potential of using machine learning algorithms to try and accurately predict Post Traumatic Stress Disorder (PTSD) symptoms in women suffering from early breast cancer and ultimately aims to create the optimal model and associated methodology. The data used were gathered at the four oncology centers – pilots: IEO (Milan, Italy), HUS (Helsinki, Finland), HUJI (Jerusalem, Israel) and Champalimaud (Lisbon, Portugal). The preprocessing methods used took into account the expected heavy imbalance of our medical data, the limited number of samples and the high number of features to consider, as well as the machine learning classifiers to be used. The model training leveraged repeated cross-validation in order to tune their hyper-parameters and the best models were evaluated on a separatelyheld test set to simulate unknown real-world data. The experiments conducted were part of an ablation study that tried to identify the important aspects of our preprocessing and modelling procedure, and also pinpoint the important features that indicate high probability of developing PTSD symptoms and therefore greatly impact our models, leading them to better predictions. The resulting procedure was tested when being used on data from completely different hospitals to check its geographical generalizability. The outcome of this study demonstrates considerable promise and highlights the potential of machine learning in the field of medicine and more specifically in predicting diseases and psychological disorders.