ATTD 2015 E-POSTER PRESENTATIONS CL-NoIP CL + IP 256 Mean Plasma Glucose (mg/dL) 129 – 13 123 – 13 % Glucose Values 70–180 mg/dL # Hypoglycemic # Hyperglycemic Episodes ( < 60 mg/dL) Episodes ( > 250 mg/dL) 81 77 6 4 9 2 A-113 AUC Glucose after Meals 245 – 176 239 – 162 Prandial Glucose Excursion (mg/dL) 100 – 45 99 – 48 EFFECT OF AN INSULIN INFUSION SITE WARMING DEVICE ON PERFORMANCE OF A CLOSED-LOOP SYSTEM S. Weinzimer1, G. Bitton2, B. Grosman3, J. Sherr1, N. Patel1, C. Michaud4, E. Tichy1, L. Carria1, W. Tamborlane1, E. Cengiz1 1Pediatrics, Yale University, New Haven, USA 2R + D, Insuline Medical, Petach Tikvah, Israel 3Closed-Loop, Medtronic Diabetes, Northridge, USA 4Pediatrics, Children’s Hospital of Philadelphia, Philadelphia, USA The effectiveness of closed-loop (CL) systems in mitigating post-prandial hyperglycemia is limited by slow absorption of subcutaneously delivered insulin and prolongation of the ‘meal bolus’ over several hours by the controller algorithm. Warming the skin at the site of insulin infusion with the InsuPatch (IP) accelerates insulin absorption following a standard subcutaneous bolus in an open-loop setting, but whether skin warming im- proves the performance of closed-loop systems has not been established. Nine T1D subjects (age 20 – 7 y, duration 11 – 9 y, A1c 7.4 – 0.7%) were studied during two 24 hours of CL control using Medtronic ePID + IFB controller: without (CL-NoIP) and with skin warming (CL + IP) to 40°C. Meals were identical for both conditions and were accompanied by small manual insulin bo- luses (0.08 units/kg). Mean blood glucose was slightly lower with CL + IP vs CL-NoIP (p = 0.01); but time within target, prandial glucose excursion, and AUC post-prandial glucose were similar (Table). There were fewer instances of hypoglycemia (BG < 60) and hyperglycemia (BG > 250) in CL + IP condition vs CL-NoIP. Skin warming with the InsuPatch resulted in lower mean glucose levels and fewer instances of hyperglycemia and hypo- glycemia, but did not significantly change glycemic responses to meals. To best utilize site warming technology, larger subcuta- neous depots of insulin given as manual pre-meal manual doses may be required for optimal prandial glucose control. 257 HYPOGLYCEMIC ALERT BASED ON ONLINE GLUCOSE PREDICTION USING MODEL MIGRATION METHOD FOR SUBJECTS WITH TYPE I DIABETES C.X. Yu1, C.H. Zhao1, Y.J. Fu2 1State Key Laboratory of Industrial Control Technology Department of Control Science and Engineering, Zhejiang University, Hangzhou, China 2Research Triangle Park, Becton Dickinson Technologies, North Carolina, USA Avoiding hypoglycemia episodes are major challenges for diabetes mellitus. A timely alert of hypoglycemia episodes be- fore they occur can allow enough time for the patient to take necessary actions to avoid hypoglycemia. In our previous work, a rapid model development strategy using the idea of model migration, which is developed by ad- justing the model parameters of inputs based on only a few samples from new subjects, was successfully used for online glucose prediction using simulation data. Here, more focus was put on hypoglycemic alert using model migration method and evaluation based on clinical data provided by BD. Twenty-four unidentified clinical subjects are used to test the model migration method (MM) where only 24 samples (two hour) are required to adjust parameters. Also, it is compared with subject-dependent standard modeling method (SD) where about 288 samples were used for model development. The results in Table 1 indicate that the prediction accuracy of MM is comparable to that of SD. Also, it reveals similar time lag for hypoglycemic alert. The sensitivity and specificity of MM are comparable or even better than SD. Therefore, the proposed MM can be regarded as an effective and economic modeling method instead of repetitive SD for hypoglycemic alert. Table 1. Hypoglycemic Alert Using Two Different Methods for 30 min Ahead Glucose Prediction Method RMSE (mg/dL) Time Lag for hypo alert (samples) Sensitivity for hypo alert (%) Specificity for hypo alert (%) 19.55 – 7.03 19.49 – 6.60 4.10 – 2.85 5.00 – 2.07 66.97 64.71 99.33 99.68 MM SD 258 AN AUTOMATIC DETECTION METHOD FOR VARIABILITY OF SIGNAL-TO-NOISE RATIO IN CONTINUOUS GLUCOSE MONITORING C.H. Zhao1, H. Zhao1 1State Key Laboratory of Industrial Control Technology Department of Control Science and Engineering, Zhejiang University, Hangzhou, China One major difficulty of denoising methods for continuous glucose monitoring (CGM) is that the noise level can vary with time which may cause unreliable filtering performance. Here, to make sure that the filter parameters can be tuned in time, an automatic noise variability detection method is proposed for Kalman filter (KF) by checking previous filtering residuals. A confidence interval [x-2r, x + 2r] is enclosed where r is the standard deviation of residuals between filtered and measured