173 - MODELING MEAL TIMES AND CARBOHYDRATES AMOUNTS OF TYPE 1 DIABETES INDIVIDUALS UNDER FREE-LIVING CONDITIONS
Abstract
Background and Aims
While the potential of existing Type 1 diabetes (T1D) simulators in accelerate research in diabetes technology is well-established, current limitations in the capability to mimic some behavioral aspects of patients’ lifestyle impacting on glucose control call for new investigation. Here, in particular, we focus on carbohydrates consumptions and develop a new model for amount and timing of main meals, snacks and hypotreatments.
Methods
The dataset includes 20 European patients studied for 2-month in free-living conditions in a closed-loop trial (Messori et al., Diabetes Technol Ther, 2016). Time and amount of meals are reported for 2759 main meals, 1218 snacks and 958 hypotreatments. Ten different probability density function models are fitted by maximum likelihood to the distributions of carbohydrates amount of each meal category and the time between consecutive meals (Δt). The Kolmogorov-Smirnov test is used to select the most suitable distributions.
Results
Breakfast, lunch and dinner amounts are described by gamma (μ=40.67 g, σ=17.65 g), lognormal (μ=55.70 g, σ=30.36 g) and loglogistic (μ=60.49 g, σ=37.00 g) distributions. Breakfast-lunch and lunch-dinner Δt are modeled by gamma (μ=303.8 min, σ=74.54 min) and generalized-extreme-value (μ=389.6 min, σ=79.08 min) distributions. Snacks amount follows a gamma (μ=24.51 g, σ=17.08 g) distribution, and their Δt from the previous main meal is modeled by a generalized-extreme-value (μ=199.5 min, σ=109.0 min) distribution. Hypotreatments amount has a loglogistic (μ=19.25 g, σ=16.14 g) distribution independently from the hypo-severity.
Conclusions
Once refined using larger datasets, the newly developed models can be incorporated in T1D simulators to allow more realistic in-silico trials.
