Real-Time Change-Point Detection Algorithm with an Application to Glycemic Control for Diabetic Pregnant Women

Abstract

Glycemic control in pregnancies of diabetic women is still suboptimal; birth defects and late miscarriages (i.e., second trimester miscarriages) are much more common in diabetic pregnancies than in the general population. This paper presents a pilot study for real-time detection of dangerous changes in glucose level, namely such that are associated with birth defects or miscarriage during the first trimester of pregnancy. Its main goals are to present an algorithm and to verify that it has practical potential to notify early enough of an increased risk of adverse outcomes in diabetic pregnancies. The study included eight women with type 1 diabetes who wore a Continuous Glucose Monitor (CGM; a device that reads and transmits the glucose level every five minutes) during the entire first trimester. Nonparametric change-point detection methods were applied on CGM data; results show evidence that an increase in glucose variability is associated with heightened risk for late miscarriage, and that this change could have been detected early enough to reduce fluctuations. By contrast, standard indicators for glycemic control in pregnancy failed to identify this peril.