Development and in-silico assessment of a CGM-based approach to trigger assumption of hypotreatments

The standard protocol for hypoglycemia treatment recommends diabetic patients taking small amounts of carbohydrates, called hypotreatments, as soon as hypoglycemia is revealed. In this work we propose a new CGM-based approach to suggest the assumption of hypotreatments for mitigating/avoiding hypoglycemia. The algorithm exploits the CGM datastream to forecast imminent hypoglycemic events and suggests hypotreatment ingestion. In particular, hypotreatment assumptions are triggered when the dynamic risk (DR) function (Guerra et al. Diabetes Technol. Ther., 2011) predicts a stable glucose level of 70 mg/dl. Early hypotreatment assumptions are also suggested when CGM values are below target glucose with rate-of-change<-1 mg/dl/min. The method is compared with the standard protocol by numerically evaluating the time in hypoglycemia (Thypo) and the post-treatment rebound (PTR) for 100 virtual patients undergoing a single-meal experiment, with forced hypoglycemia, generated by the UVA/Padova T1D simulator. In an ideal, noise-free, scenario the algorithm reduces, on average [5th-95th percentiles], Thypo (from 36 [13-56] to 0 [0-25] min, p<0.0001) without increasing PTR (from 136 [109-178] to 121 [108-141] mg/dl, p<0.0001). Corrupting CGM traces with measurement error, brings to a lower -but still statistical relevant- improvement: Thypo decreases from 41 [0-71] to 25 [0-71] min (p<0.0001), PTR decreases from 176 [117-243] to 137 [109-178] mg/dl (p<0.0001). The (known) sensitivity of DR to noise suggests to improve the performance of the method in a realistic dataset with an enlarged set of merit criteria. Once the robustness of the proposed method is warranted, its applicability could be considered in both insulin pump and multiple-daily-injections therapies