Predicción probabilística de estados glucémicos para pacientes con diabetes tipo 1 mediante el análisis de datos composicionales
Year: 2023
Type: dissertation
Abstract: Diabetes mellitus (DM) is a chronic disease that affects a large number of people worldwide, making it a global pandemic. Type 1 diabetes (T1D) accounts for approximately 10 % of all cases. It is characterized by autoimmune destruction of the pancreatic β cells, which are responsible for producing insulin. This loss of β cells leads to a permanent lack of insulin, resulting in an abnormal state of blood glucose (BG) homeostasis known as hyperglycemia. Subsequently, it can cause both chronic microvascular (retinopathy, neuropathy, and nephropathy) and macrovascular (cardiovascular and cerebrovascular diseases) complications, as well as other acute complications. In the treatment of T1D, exogenous insulin is necessary to reduce BG levels to normoglycemia (70-180 mg/dL), which has been established as the control target. Normalization of BG is the main task of diabetes treatment, attempting to minimize hypoglycemia and hyperglycemia events. The main mechanisms of intervention in glycemic control in people with T1D are insulin administration and dietary adjustment. In both cases, the effectiveness of the treatment is influenced by the delay in insulin absorption and action and food intake. Therefore, the ability to predict the future glycemic profile is essential to helping the patient make decisions and avoid risky situations. The availability of continuous glucose monitoring (CGM) systems has allowed for the systematic collection of glucose measurements at short intervals (5 or 15 min), generating a large amount of real-time data. This increase in data availability has led to the development of new mathematical prediction models that, along with improved measurement accuracy, allow for more reliable and long-term predictions despite the uncertainty and variability inherent in glucose measurements. This thesis presents a simulation tool for T1D, where cohorts of virtual patients (VPs) are generated, incorporating models of long-acting insulins to evaluate multiple daily injection (MDI) and continuous subcutaneous insulin infusion (CSII) therapies in challenging and realistic scenarios. A mathematical methodology based on compositional data (CoDa) is proposed to validate a probabilistic model of transition between different categories of periods, providing a novel metric that could be used in any process that needs validation with compositional data. Finally, an individualized model is presented to predict the mean and coefficient of variation (CV) of glucose for the following 2 and 4 hours. From these predictions, the estimated minimum and maximum BG values are calculated. An information system called "traffic light" has been implemented and validated, which updates people on their glycemic status, risks related to hyperglycemia, hypoglycemia, and CV for the next hours. The incorporation of long-acting insulin was evaluated in the MiceLab diabetes simulator, where simulation results were compared with those obtained in clinical trials. The validation and proposal of prediction algorithms were evaluated using sets of measurements from individuals with T1D who use CGM devices. The results are promising and suggest that these models could improve the accuracy of BG prediction, thus contributing to technological advancements and the optimization of therapies to improve the quality of life for individuals with T1D
Source: DUGiDocs (Universitat de Girona)
Institution: Universitat de Girona
Language: English
Subfield: Endocrinology, Diabetes and Metabolism
Field: Medicine
Domain: Health Sciences
Keywords: type 1 diabetes, blood glucose prediction, compositional data analysis, continuous glucose monitoring, virtual patient simulation, insulin pump therapy, long-acting insulin, probabilistic prediction model, coefficient of variation, glycemic control, hypoglycemia, hyperglycemia, model validation
Open Access status: green
OpenAlex ID: W7033327670
