Diabetes stems from different metabolic disruptions which makes it a complex disease whose burden is continuously increasing. The objective of the research carried out was to combine model-based and feature-based data-driven approaches to develop innovative digital health technologies aimed at improving diabetes prevention and optimizing its diagnosis and therapy. Different approaches were here developed starting from two sources of metabolic data, i.e., challenge (either from intravenous or oral glucose tolerance tests) and continuous glucose monitoring (CGM) data. As for challenge data, the three developed model-based approaches allow to derive parameters to quantitatively assess physiological processes (i.e., insulin clearance, alpha-cell insulin sensitivity) involved in disease development and progression. Then, the possibility of combining modeling and data-driven methodologies was explored to develop a digital prevention framework in the field of gestational diabetes to identify women at greater risk of type 2 diabetes progression; an important result was unraveling which features contribute to determine this increased risk. As for CGM data, given the lack of guidelines in its characterization, a great effort was put in standardizing the calculation of the multiplicity of metrics. Then, the potential of such metrics was explored through the formulation of feature-based machine learning approaches laying the foundation for an integrated diabetes prevention/management multipurpose tool for: i) innovative diagnosis and early identification of complications (i.e., diabetic retinopathy) ii) prediction of hypoglycemic events due to different causes such as hemodialysis or physical exercise; in this latter case a dedicated metric called HIKE was formulated. Regarding pharmacological therapy, this thesis addressed the open problem of quantifying the exogenous insulin that reaches the circulation (bioavailability) through an in-silico approach based on dynamic modeling.
Il diabete è una malattia complessa il cui peso è in continuo aumento. L’obiettivo della ricerca è stato quello di combinare approcci basati su modello e/o su caratteristiche e guidati dai dati, al fine di sviluppare tecnologie digitali innovative volte a migliorare la prevenzione della malattia e a ottimizzarne diagnosi e terapia. Si sono sviluppati diversi approcci a partire da due fonti di dati metabolici: test di tolleranza al glucosio (endovenoso o orale) e dati di monitoraggio continuo del glucosio (CGM). Nel primo caso sono stati sviluppati tre approcci basati su modello per la valutazione quantitativa di processi fisiologici (clearance dell'insulina e sensibilità all'insulina delle cellule alpha) coinvolti nello sviluppo e nella progressione della malattia. Si è poi esplorata la combinazione di modelli e metodologie basate su dati per sviluppare un sistema di prevenzione digitale nel campo del diabete gestazionale volto a identificare le donne a maggior rischio di progressione verso il diabete di tipo 2; un risultato importante è stato individuare quali caratteristiche hanno un ruolo rilevante in tale progressione. Circa i dati CGM, vista la mancanza di linee guida nella loro caratterizzazione, si è svolto un lavoro mirato a standardizzare il calcolo di molteplici metriche. In seguito, il potenziale di tali metriche è stato esplorato nella formulazione di approcci di apprendimento automatico basati su caratteristiche, gettando le basi per uno strumento di prevenzione e gestione integrato per: i) diagnosi innovativa e identificazione precoce di complicanze (esempio, retinopatia diabetica) ii) previsione di eventi ipoglicemici dovuti a varie cause come emodialisi o esercizio fisico; in quest'ultimo caso è stata formulata una metrica dedicata denominata HIKE. Nell’ambito della terapia farmacologica, si è affrontato il problema della quantificazione dell'insulina esogena che arriva in circolo attraverso un approccio in-silico basato sulla modellizzazione dinamica.
Digital health technologies to improve diabetes prevention and optimize therapy: from model-based approaches to feature-based machine learning / Piersanti, Agnese. - (2024).
Digital health technologies to improve diabetes prevention and optimize therapy: from model-based approaches to feature-based machine learning
PIERSANTI, AGNESE
2024-01-01
Abstract
Diabetes stems from different metabolic disruptions which makes it a complex disease whose burden is continuously increasing. The objective of the research carried out was to combine model-based and feature-based data-driven approaches to develop innovative digital health technologies aimed at improving diabetes prevention and optimizing its diagnosis and therapy. Different approaches were here developed starting from two sources of metabolic data, i.e., challenge (either from intravenous or oral glucose tolerance tests) and continuous glucose monitoring (CGM) data. As for challenge data, the three developed model-based approaches allow to derive parameters to quantitatively assess physiological processes (i.e., insulin clearance, alpha-cell insulin sensitivity) involved in disease development and progression. Then, the possibility of combining modeling and data-driven methodologies was explored to develop a digital prevention framework in the field of gestational diabetes to identify women at greater risk of type 2 diabetes progression; an important result was unraveling which features contribute to determine this increased risk. As for CGM data, given the lack of guidelines in its characterization, a great effort was put in standardizing the calculation of the multiplicity of metrics. Then, the potential of such metrics was explored through the formulation of feature-based machine learning approaches laying the foundation for an integrated diabetes prevention/management multipurpose tool for: i) innovative diagnosis and early identification of complications (i.e., diabetic retinopathy) ii) prediction of hypoglycemic events due to different causes such as hemodialysis or physical exercise; in this latter case a dedicated metric called HIKE was formulated. Regarding pharmacological therapy, this thesis addressed the open problem of quantifying the exogenous insulin that reaches the circulation (bioavailability) through an in-silico approach based on dynamic modeling.I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
