Head-To-Head Comparison of the Accuracy of Different Glucose Sensors with Possible Use in Artificial Pancreas

Background: The Artificial Pancreas is a set system composed by a continuous glucose monitoring (CGM), an insulin pump and a control algorithm responsible for the automatic administration of insulin on the basis of the glucose concentration measured continuously in the interstitial fluid by CGM system. There are different types of CGM system. The accuracy of these devices is a crucial point for the correct functioning, efficacy and safety of the artificial pancreas. Few studies have evaluated their accuracy. Aim: To compare the accuracy of 3 glucose sensors in 2 different studies: first Dexcom G4 Platinum vs FreeStyle Libre, second Dexcom G5 Mobile vs FreeStyle Libre. Methods and results: First study: For 2 weeks, 22 subjects with type 1 diabetes simultaneously wore the FreeStyle Libre (FSL, Abbott, Alameda, CA) and the Dexcom G4 Platinum (DG4P, Dexcom, San Diego, CA). During a hospital phase, patients randomly received the same breakfast with standard or delayed & increased insulin bolus, to induce large glucose swings. Venous glucose was checked every 5-15 min for 6 hours. At home, patients did ≥ 4 reference fingersticks/ day. During home phase, the overall MARD (mean absolute relative difference) in glucose levels was similar for 2 sensors: 12.9 (2.5) % for DG4P vs 13.7(3.6) for FSL (difference not significant [NS]). Accuracy was worse during hypoglycemia for both sensors, without significant difference between sensors. In the euglycemic range, accuracy was better for DG4P [12.0(2.4) % vs 14.0(3.6)%, p 0.026]. In the hospital phase, FSL performed better in the hyperglycemic range. Considering week one, FSLand DG4P had similar accuracy across all glucose ranges, but FSL had a smaller MARD when glucose was changed to >1.5 mg/dl/min. Second study: For 2 weeks 20 subjects with type 1 diabetes simultaneously wore the FreeStyle Libre (FSL, Abbott, Alameda, CA) and Dexcom G5M (DG5M, Dexcom, San Diego, CA). During a hospital phase, patients received the same breakfast with a delayed&increased insulin bolus to induce large glucose swings. Venous glucose was checked every 5-15 min for 6 hours. At home, patients did ≥ 4 reference finger-sticks/day. Twenty type 1 diabetic patients completed the study. During the at-home evaluation, the overall ARD was 12.3% (5.6-21.4) for the FSL and 9.8% (4.7- 18.0) for the G5M (p<0.001). ARD increased during hypoglycemia with both the FSL and G5M sensors and decreased during hyperglycemia. During the hospital phase, G5M performed better than FSL. Considering accuracy during different rates of change, the G5M sensor was more accurate when glycemia was stable and demonstrated better performance than the FSL when glucose increased, both slowly and rapidly. No differences in accuracy were observed when glucose levels decreased rapidly. Conclusions: DG4P performed as well as FSL, both sensors performed less well during hypoglycamia. During glucose swings Libre was more accurate than DG4P. The G5M sensor provides greater accuracy than the FSL sensor.