Reasoning in time-based medical data: combining visual analytics, data mining, and artificial intelligence methods

This thesis presents an in-depth exploration of visual analytics, data mining, and artificial intelligence methods to improve clinical outcomes with time-based data. We address the hypothesis that key patterns and relationships can be identified in clinical temporal data to significantly improve the prevention, diagnosis, and treatment of various health conditions. Each use case in this thesis addresses this hypothesis in a different manner: the impact of daily habits on older adults, disease path progression, and heart failure decompensation detection.
The first use case focuses on a tool developed for caregivers to manage virtual coaching for older adults, assessing the effectiveness of personalized coaching plans. The tool groups participants using semi-supervised classification methods and then utilizes evolution charts for insight extraction. It has been applied in a European cohort and its usability has been proven through usability tests.
The second use case focuses on creating a methodology to analyse disease trajectories. To achieve this, we propose an impact measure between diseases to estimate trajectory impact, along with a graph-based interactive visualization for easy comprehension. We applied this method to a cohort of 71,849 patients, and we identified significant disease progression patterns, such as the link between dementia following transient ischaemic attacks and strokes. Due to the positive results, the approach has been extended to another hospital, demonstrating its broader applicability.
The third use case aims to predict cardiac decompensation events (CDE) in patients with chronic heart failure using telemonitored data and advanced computational models. The proposed model provides stable and reliable predictions, enhancing the triage process for patient care. Additionally, the precedent model has been validated in three external cohorts.