Work

Type 1 diabetes patient decision-making modeling for the in silico assessment of insulin treatment scenarios

Year: 2017
Type: dissertation
Abstract: In type 1 diabetes (T1D) exogenous insulin is administered to compensate for the absence of endogenous insulin production by pancreas beta-cells. T1D subjects must finely tune insulin doses to maintain blood glucose (BG) concentration within the normal range (70-180 mg/dl). For such a purpose, every day, T1D subjects need to frequently monitor their BG concentration and make several treatment decisions, e.g. the calculation of insulin and carbohydrate (CHO) doses to counterbalance, respectively, high and low BG values. The safety and effectiveness of T1D insulin therapies are normally assessed by clinical trials, which unfortunately are usually time-demanding, expensive and often present constraints of low numerosity and short duration, with consequently low probability of observing rare but risky situations, like severe hypoglycemia. These limitations can be overcome by the use of in silico clinical trials, based on computer simulations, that allow to test medical device-based treatments in a large number of subjects, over a long period, under reproducible conditions, at limited costs, and without implicating any risk for real subjects. A popular powerful tool to perform in silico clinical trials in T1D is the UVA/Padova T1D simulator, i.e. a model of glucose, insulin and glucagon dynamics in T1D subjects. However, to test insulin therapies in a real-life scenario, the UVA/Padova T1D simulator alone is not sufficient because a mathematical description of other fundamental components, like the device used for glucose monitoring and the patient's behavior in making treatment decisions, is required. The aim of this thesis is to design a mathematical model of T1D patients making treatment decisions fully usable for the comprehensive in silico assessment of insulin treatment scenarios. In particular, in the first part of the thesis we develop three submodels that the UVA/Padova T1D simulator requires (as complement) to pursue this scope. Specifically, we design a model of self-monitoring of blood glucose (SMBG) device, a model of minimally-invasive sensor for continuous glucose monitoring (CGM), and a model of the patient’s behavior in tuning CHO intakes and insulin doses according to SMBG and/or CGM measurements. The parameters of these models are either fitted on real data or derived from literature studies. The overall model, in the following called T1D decision-making (T1D-DM) model, can be used for several in silico experiments. To demonstrate its usefulness, in the second part of this thesis we apply the T1D-DM model to assess safety and effectiveness of nonadjunctive CGM use, i.e. the use of CGM measurements to make treatment decisions without requiring confirmatory SMBG measurements collected by fingerstick. This specific application is currently of great scientific and industrial interest for the diabetes technology research community because, until clinical evidence of its safety is provided, nonadjunctive CGM use cannot be approved by U.S. regulatory agencies, like the Food and Drug Administration. The thesis is organized in six chapters. In Chapter 1, after introducing T1D therapy, the importance of in silico clinical trials is discussed, both in general and specifically for the assessment of nonadjunctive CGM use. Then, some state-of-art simulation techniques are briefly introduced discussing their open problems. The aim of the thesis is illustrated at the end of the chapter. In Chapter 2, we analyse more in depth the limitations of the approaches currently available in the literature for the assessment of insulin treatments. In particular, we demonstrate that a recently proposed simulation method to "replay" in silico real-life treatment scenarios has domain of validity limited to small adjustments of basal insulin, calling for the development of more sophisticated techniques like that proposed in this thesis. In Chapter 3, our simulation method based on the T1D-DM model is presented. This model allows to simulate, in a real-life scenario, the glucose profiles of T1D subjects using SMBG and/or CGM to make treatment decisions. The T1D-DM model is composed of four components: A) the UVA/Padova T1D simulator, B) a model of glucose monitoring devices, C) a model of patient's behavior and treatment decisions and D) a model of the insulin pump. In particular, as far as B) is concerned, two different SMBG error models are derived by data collected with two popular SMBG devices (One Touch Ultra 2 and Bayer Contour Next USB). Using a recently published methodology which takes into account the main sensor error components, a CGM model is derived from data collected by a state-of-art CGM sensor (Dexcom G5 Mobile). Regarding C), a model of the patient's behavior in making treatment decisions based on SMBG and/or CGM, such as administration of insulin boluses and hypotreatments, is designed to simulate treatments based on i) SMBG, ii) adjunctive CGM, or iii) nonadjunctive CGM. In order to reproduce a real-life scenario, the model includes components describing the mistakes real subjects commonly make, such as miscalculation of meal CHO content and early/delayed insulin administrations. In Chapter 4 and Chapter 5, two in silico trials based on the T1D-DM model are designed to assess nonadjunctive CGM use. In the first trial, nonadjunctive CGM is compared to SMBG and adjunctive CGM over a two-week period in 100 virtual subjects. Results show that the use of CGM (both adjunctive and nonadjunctive) significantly improves glycemic control compared to SMBG, while no significant change is observed between adjunctive CGM and nonadjunctive CGM. This suggests that CGM is ready to substitute SMBG for T1D treatment. In the second trial, the impact of thresholds used for CGM hypo/hyperglycemic alerts on the performance of nonadjunctive CGM use is assessed. Results show that time in hypoglycemia is reduced by nonadjunctive CGM use with any alert setting, while time in hyperglycemia is significantly worsen by nonadjunctive CGM use, compared to SMBG, when the high alert threshold is set to 350 mg/dl or higher. Finally, the major findings of the work carried out in this thesis, its possible applications and margin of improvements are summarized in Chapter 6.
Language: English
Field: Medicine
Open Access status: green
OpenAlex ID: W2783729690
Cites:
Cited by:
Related to: 10