Résumé
Current definitions of sensor detected hypoglycemia (SDH) give high levels of asymptomatic hypoglycemia on continuous glucose monitoring (CGM). We defined individually optimised thresholds and durations for sensor detected hypoglycemia (SDH) that better identify symptomatic hypoglycemia experienced by people with insulin treated diabetes. We analysed 10 weeks of blinded CGM (Libre 2) and FitBit data from 435 participants [217 type 1, 218 type 2]. They self reported symptomatic hypoglycemia on the Hypo-METRICS smartphone app. We used particle Markov chain Monte Carlo optimization to generate the threshold and duration of SDH that maximizes detection of symptomatic hypoglycemia for each individual (and by Fitbit sleep status). When using individual definitions of SDH, precision for detection of symptomatic hypoglycemia increased by 23% with 2% loss of sensitivity vs the consensus definition of Level 1 SDH (70mg/dl > 15 mins). Precision and sensitivity increased by 41% and 20% respectively vs level 2 SDH (54 mg/dl > 15 mins). Increased precision, with minimal change of sensitivity, were greatest during sleep, and irrespective of diabetes types and sleep status (Table). Optimizing SDH definitions at the individual level and by sleep status could minimize false alarms without increasing missed clinically important hypoglycemia episodes.