Rayhan Lal, United States of America
Stanford University Division of Endocrinology, Department of Pediatrics & MedicinePresenter of 2 Presentations
Insulin Pumps
OPTIMIZING FORMULAS FOR BASAL, CARB RATIO AND SENSITIVITY FACTOR FOR PREDICTIVE CONTROLLERS: LESSONS LEARNED FROM LOOP
Abstract
Background and Aims
Rather than utilizing user-entered settings, prediction-based commercial automated insulin delivery (AID) systems calculate internal representations of basal, carbohydrate-to-insulin (CIR) ratio and/or insulin sensitivity factor (ISF) to generate estimates of future glucose. Frequently simple formulas are used that incorporate a proportionality constant and total daily dose (TDD), total daily basal or weight. In contrast, the open-source AID system Loop relies on the settings provided by users and healthcare providers. We attempt to optimize the formulas used for basal, CIR and ISF based on data from Loop users.
Methods
Utilizing data from the Loop observational study we define an “aspirational” cohort of individuals who provide complete data (³90% CGM availability), meeting the standard international consensus on Time in Range clinical targets, no time below 40mg/dL and normal BMI. We then perform fitting to the traditional equation forms utilizing TDD and test additional forms that also include age, BMI and average daily carbohydrate consumption to optimize fitting.
Results
219/743 (29%) of Loop observational study participants who had settings data met inclusion criteria for the “aspirational” cohort. When fitted to TDD alone, basal and CIR tended to be more aggressive while ISF tended to be less aggressive than the standard equations. Incorporating BMI, daily carbohydrate consumption and total daily dose provided a significantly better fit than did formulas using TDD alone.
Conclusions
The data provided in the Loop observational study allows the creation of new formulas for basal, CIR and ISF that can benefit all users of prediction-based AID.