Accuracy and limitations of continuous glucose monitoring using spectroscopy in critically ill patients

OptiScanner devices, continuous glucose monitoring devices that perform automated blood draws via a central venous catheter and create plasma through centrifugation, measure plasma glucose levels through mid-infrared spectroscopy at the bedside. The objective of this study was to determine accuracy and practicality of the devices in critically ill patients attempting glycemic control.

The plasma glucose level was measured by the devices and in comparative plasma samples using Yellow Springs Instrument (YSI) plasma analyzers. After adding several previously unrecognized interferences in the interference library, we reanalyzed the mid-infrared signals and compared the resulting plasma glucose level with the reference value. Results are presented in Clarke error grids, glucose prediction errors and Bland-Altman plots and expressed as correlation coefficients.

We analyzed 463 comparative samples from 71 patients (median 6 (4 to 9) samples per patient). After calibrating the system, a Clarke error grid showed 100% of the values in zones A or B. The glucose predictor error demonstrated that 86% of the glucose values < 75 mg/dL were within ± 15 mg/dL of the YSI results and 95% ≥ 75 mg/dL were within 20% of the comparative YSI results. Bland-Altman plot showed a bias of -0.6 with limit of agreement of -24.6 to 23.3. The Pearson correlation coefficient was 0.93 and R2 was 0.87. In one third of the patients the devices had to be disconnected prematurely (that is before planned disconnection) because of repeated occlusion alarms suggesting blood draw errors.

The devices needed calibration for several previously unrecognized interferences. Thereafter, accuracy of the device to measure plasma glucose levels in 'our cohort' of critically ill patients improved, but external validation is highly recommended. The automated blood draw system of the devices needs further improvement to make this device of value for clinical use (trial registration (Netherlands Trial Register): NTR2864).