Martina Vettoretti, Italy

University of Padova Department of Information Engineering

Presenter Of 1 Presentation

ASSESSMENT AND COMPARISON OF THE MEASUREMENT ERROR COMPONENTS OF DEXCOM G5 MOBILE AND EVERSENSE CONTINUOUS GLUCOSE MONITORING SYSTEMS

Session Type
ORAL PRESENTATION SESSION
Date
22.02.2020, Saturday
Session Time
10:30 - 12:00
Channel
La Paz
Lecture Time
10:30 - 10:40

Abstract

Background and Aims

The measurement error of continuous glucose monitoring sensors results from the combination of different error sources, including plasma-interstitium kinetics, calibration error and random noise. The aim of this work is to compare the measurement error components of two popular systems: Dexcom G5 Mobile (DG5M) and Eversense.

Methods

Eleven subjects were monitored in parallel using DG5M and Eversense. In the middle of sensors’ lifetime, subjects attended 6-hour clinical sessions where reference YSI measurements were collected every 15 min. Data were analysed using the methodology proposed by Facchinetti et al. (IEEE TBME 2014), which allows dissecting and quantifying the three main error components: the physiologic delay due to the plasma-interstitium kinetics (modeled as first-order linear dynamic system), the systematic error due to imperfect calibration (linear regression model), and the residual random noise (autoregressive model).

Results

Plasma-interstitium time-constant was 12.31 [6.13-17.41] min for Eversense and 5.96 [2.64-9.16] min for DG5M. For increasing glucose concentration (range: 50-270 mg/dl), the average sensor calibration error ranged linearly from +4.92 to +1.44 mg/dl for Eversense and from -5.58 to –10.00 mg/dl for DG5M. Random noise standard deviation was 7.79 [5.25-11-86] mg/dl for Eversense and 4.01 [3.56-5.52] mg/dl for DG5M.

Conclusions

Eversense showed larger physiologic delay and random noise compared to DG5M. The calibration error was mostly positive (overestimation) for Eversense and mostly negative (underestimation) for DG5M over the entire glucose range. These different characteristics should be taken into account when comparing results obtained with the two sensors, e.g. for proper tuning of decision-making strategies.

Hide