Dexcom

AIMS:Continuous glucose monitoring (CGM) remains underutilized in low- and middle-income countries (LMICs). We extend earlier observations on the feasibility and impact of CGM among people living with type 1 diabetes (T1D) in Rwanda in a real-world continuation phase study. METHODS:This was a 1-year continuation phase of a single-arm, mixed-methods, prospective observational study conducted in Kigali, Rwanda, from August 2022 to September 2024. Completers of the 12-month Phase I were transitioned to a current-generation CGM device and, in months 19-24, reduced frequency of clinic visits reflecting routine care. The primary outcomes were change in haemoglobin A1c (HbA1c) and the CGM-based metrics time in range (TIR, 3.9-10 mmol/L), and time below range (TBR, <3.9 mmol/L). Secondary outcomes included self-reported hospitalizations, incidences of severe hypoglycaemia and diabetic ketoacidosis, as well as diabetes-related questionnaire results. RESULTS:At the end of Phase I, 40 of the original 50 participants entered Phase II with mean HbA1c, TIR and TBR of 44 mmol/mol (6.2%), 44.9% and 5.6%, respectively, all metrics being significantly different from study start. These did not change significantly by the end of the study (48 mmol/mol [6.6%], 46.5% and 7.9%, respectively). No hospitalizations, three episodes of severe hypoglycaemia and two episodes of diabetic ketoacidosis were reported. Most respondents were satisfied with the device, used it consistently and trusted the values provided. CONCLUSIONS:CGM-related improvements in HbA1c and TIR among Rwandans living with T1D were maintained for at least 24 months. CGM should be considered as an important tool to improve diabetes self-management in LMICs.

BACKGROUND:Rebound hyperglycemia (RHyper), rebound hypoglycemia (RHypo), extended hyperglycemia (EHyper), and extended hypoglycemia (EHypo) are newly defined continuous glucose monitoring (CGM) metrics. Here, we investigated the characteristics of these new metrics and the relationship between new CGM metrics and standard metrics. MATERIALS AND METHODS:In this retrospective cohort study, 30,000 CGM users with at least 90 days of CGM data were randomly selected from Dexcom Clarity database. Standard and new CGM metrics were calculated for each user. Four different cutoffs were used to define RHyper and RHypo, and two cutoffs were used to define EHyper and EHypo events. The number of RHyper, RHypo, EHyper, and EHypo events per week, mean duration of events, and mean area under the curve of events were calculated. For rebound events, the rate of change (ROC) was calculated. Pearson correlation and simple linear regression were used to analyze the data. RESULTS:Mean time in 70-180 mg/dL was 61.8 ± 20.7%, mean glucose was 173 ± 37.1 mg/dL, and coefficient of variation (CV) was 32.1 ± 7.2%. RHyper, RHypo, and EHyper were more frequent during daytime and increased throughout the day. EHypo mostly occurred during nighttime. CV correlated strongly with RHyper (70-180 mg/dL) events/week (r = 0.67) and RHypo (180 to 70 mg/dL) events/week (r = 0.64). Time in range had the strongest correlation with EHyper events/week (r = -0.88) among new metrics. RHyper events and RHypo events were strongly correlated with each other (r = 0.92). RHyper and RHypo ROC have a stronger correlation with CV than the correlation between CV and time below range (TBR) metrics. CONCLUSIONS:For rebound and extended metrics, the most important metric was the number of events/week. RHyper and RHypo had a stronger correlation with CV and hypoglycemia metrics (TBR) than the correlation between CV and TBR. Thus, rebound events have the potential to detect hypoglycemia events caused by glycemic variability.

Safe Reinforcement Learning (RL) algorithms are typically evaluated under fixed training conditions. We investigate whether training-time safety guarantees transfer to deployment under distribution shift, using diabetes management as a safety-critical testbed. We benchmark safe RL algorithms on a unified clinical simulator and reveal a safety generalization gap: policies satisfying constraints during training frequently violate safety requirements on unseen patients. We demonstrate that test-time shielding, which filters unsafe actions using learned dynamics models, effectively restores safety across algorithms and patient populations. Across eight safe RL algorithms, three diabetes types, and three age groups, shielding achieves Time-in-Range gains of 13-14\% for strong baselines such as PPO-Lag and CPO while reducing clinical risk index and glucose variability. Our simulator and benchmark provide a platform for studying safety under distribution shift in safety-critical control domains.