Accuracy of the Dexcom G6 Continuous Glucose Monitoring System Following Cardiac Surgery

Maintaining tight blood glucose control following surgery is imperative to reducing infections and neurologic dysfunction. This requires frequent blood sampling while in the intensive care unit, leading to increased waste of blood and utilizing time and resources to collect samples and wait for results. Additionally, when on the regular floors, frequent point of care fingersticks for glucose levels are needed for therapeutic intervention.

Continuous glucose monitoring systems (CGMS) are now available for ambulatory use in patients with diabetes. The potential of using such a system in the hospital, including during the postoperative critical care setting and upon transfer to the regular floor, may reduce personnel burden, produce rapid results, and decrease blood waste. However, these systems have not yet been validated for hospital use.

This is a prospective longitudinal study to study the agreement between a novel continuous glucose monitoring system (CGMS) versus current blood glucose monitoring.

Subjects in this study will have their blood glucose measured regularly every 1-3 hours with current methods in the CVICU, and by point of care glucose using the Accuchek Inform II on the regular floors, and the CGMS at the same time will be captured. Subjects will have measurements taken throughout their stay in the CVICU and on the regular floors. Agreement and correlation between systems will be calculated, and error will be classified using the Surveillance Error grid. The proportion of errors of at least moderate risk will be calculated.

Subject characteristics including demographics, comorbidities and baseline disease severity will be summarized using means and standard deviations for continuous measures and frequencies and percentages for categorical factors. To evaluate agreement between methods, concordance correlation coefficients will be calculated, using the method that adjusts for repeated measures as described by King and implemented in software by Carrasco. Analysis will be performed overall, and then stratifying by type of current measure, if the type of measurement varies across settings. Correlations will be calculated using the methods described by Bland and Altman and implemented by Bakdash and Marusich. Differences will also be evaluated for errors according to the Surveillance Error Grid. Rates of moderate or more severe errors will be calculated. For each measure above, 95% confidence intervals will be calculated, accounting for the clustering within subject.. Analyses will be performed using SAS software (version 9.4; Cary, NC) and R software (version 3.5; Vienna Austria).