Adamo, Sarah (2024) The importance of identifying predictive outcomes in patients affected by chronic and disabling diseases: a Machine Learning approach. [Tesi di dottorato]

Documento PDF ITEE_Thesis_37cycle_SarahAdamo.pdf Visibile a [TBR] Amministratori dell'archivio Download (9MB) | Richiedi una copia

Abstract

In the last few years the scenario of clinical research and diseases study has completely changed due to the introduction of Artificial Intelligence (AI) in healthcare. Through the AI application, and particularly through the involvement of Machine Learning (ML) techniques, a new way of ap- proaching the analysis of pathologies and diseases has been possible. The management of patients affected by chronic and disabling disease represents nowaday a challenge for clinical research, since those patients are more frequently exposed to the occurrence of acute events and the wors- ening of clinical conditions and of the quality of life too. Moreover, there is still a lack of knowledge in several diseases, particularly for what may concern the early detection and the choice of the best treatment. For this reason, three main clinical case studies were involved in this research: long- term COVID-19 disease, asthma disease and Parkinson’s Disease (PD). Both supervised and unsupervised ML models were implemented in order to identify the best parameters that were able to predict an early diagnosis or a clinical outcome and to identify phenotypes with similar clinical characteristics. Starting from the acquisition of data, the definition of an effective outcome was fundamental for a better training of models. Analyses gained good results in terms of evaluation metrics and iden- tified parameters with a high predictive value: the 6-Minute-Walking-Test (6MWT) for the estimation of rehabilitation improvements in long-term COVID19, a new cut-off value of the eosinophils count for the identification of clinical clusters with specific characteristics impacting on the prediction of further exacerbations, the effectiveness of gait analysis variables in the prediction of a pre-dementia stage in PD. In conclusion, this studies demonstrated to provide powerful methods for the identification of fundamental parameters in the management of chronic and disabling disease, thus supporting the clinical decision making.

Downloads

Actions (login required)

Modifica documento