Roberto Visentin, Italy
University of Padova Department of Information EngineeringPresenter of 1 Presentation
USE OF AN AUTOMATED SMARTPHONE-BASED CARBOHYDRATE ESTIMATOR IN TYPE 1 DIABETES THERAPY: CLINICAL IMPACT ASSESSED BY COMPUTER SIMULATION
Abstract
Background and Aims
Carbohydrate (CHO) counting is a burdensome task for people with type 1 diabetes (T1D). Wrong estimation of meal CHO content increases risks of post-prandial hypo-/hyper-glycemia. An automated CHO estimator (ACE) using food pictures taken from a smartphone can potentially help subjects with T1D improving their glycemic control while reducing the burden of diabetes self-management. Here, we aim to in silico compare the clinical impact of the ACE vs. patient CHO estimation (PCE) in T1D therapy.
Methods
One hundred and twenty-eight food items, for which true CHO amounts were available, were scanned with an early ACE prototype. For each item, the Relative Signed Error (RSE) of CHO estimation was calculated and the resulting distribution was fitted to a Student's t probability function model. This model was incorporated into the UVA/Padova T1D simulator, already equipped with a statistical model describing PCE error (Vettoretti et al., 2018). Thus, a 3-day 3-meal/day in silico head-to-head trial was performed to compare the effects of ACE vs. PCE in terms of relevant glucose control metrics.
Results
Our data showed that ACE slightly overestimated true CHO amount (median [25th,75th] percentile RSEACE=0.9% [-7.5%,19.3%]), while patients usually tended to underestimate it (RSEPCE=-4.9% [-21.3%,7.5%]). Simulated glucose control metrics are reported in Table 1.
Conclusions
Despite statistically significant, the differences in outcome metrics are marginal from a clinical point of view. We conclude that, to date, the use of the ACE prototype likely would not improve glucose control but might ease T1D management. These results warrant confirmation in a head-to-head clinical trial.