Methods to improve glucose variability estimates from censored data in patients with insulin dependent T1DM

• Activity Number: 676 - Analysis and Reporting: Benefit-Risk and Robust Models
• Type: Contributed
• Date/Time: Thursday, August 2, 2018 : 10:30 AM to 12:20 PM
• Sponsor: Section on Medical Devices and Diagnostics
• Author(s): Nicholas Hein* and Christopher Wichman and Lynette Smith and Jennifer Merickel and Andjela Drincic and Matthew Rizzo and Cyrus Desouza
• Companies: University of Nebraska Medical Center and University of Nebraska Medical Center and University of Nebraska Medical Center and University of Nebraska Medical Center and University of Nebraska Medical Center and University of Nebraska Medical Center and University of Nebraska Medical Center
• Keywords: glucose variability; diabetes; continuous glucose monitoring; imputation; censoring

Abstract

High glucose variability (GV) is associated with risk of serious hypo- and hyperglycemic events in patients with insulin-dependent diabetes. Accurate measurement of GV can help guide the best treatment for a patient. Continuous glucose monitoring systems (CGMS) give clinicians and patients real-time access to glucose levels over extended time frames. However, many patients have glucose levels above or below the CGMS sensor thresholds, producing censored values and challenges for accurate estimates of GV. A statistical technique to impute the censored values to calculate accurate GV measures is proposed. A simple replacement strategy (replacement method) and a local non-linear least squares regression method (imputation method) were used to impute censored data in a simulation study. Clinically standard methods (standard deviation (SD), mean amplitude of glucose excursion, and coefficient of variation) were used to calculate GV. Under simulation, the imputation method resulted in a root mean square error (RMSE) of 25.2 and mean bias of 14 when calculating SD. The replacement method had an RMSE 77.3 and mean bias of 65.2. For clinical data, the mean difference in SD was 16.1. We conclude that the imputation method is more accurate than the replacement method.