Reproducibility of the Continuous Glucose Monitoring System Matches Previous Reports and the Intended Use of the Product

A recent article by Metzger et al. (1) questioned the reproducibility of the continuous glucose monitoring system (CGMS; Medtronic MiniMed). We were surprised by the authors’ conclusions, as their results were consistent with previously published reports. We believe the authors’ expectations for the CGMS did not coincide with its intended use. We were also puzzled by their use of a subjective assessment of clinical decision making when a validated tool, i.e., the Clarke error grid, is available (2). Finally, Metzger et al. dismissed the impact of an update to the CGMS software (Solutions 3.0) that does in fact improve the reproducibility evident in their data.

Although Metzger et al. reported results based on a limited number of subjects (n = 11), the results are quite similar to those reported in a large postmarketing study (n = 235) (3). Correlation was 0.93 mg/dl (vs. 0.91), and bias was 0.0 mg/dl (vs. −3.91). Metzger et al. also reported that 69% of sensor-sensor pairs had >10% difference, which compares closely with the previously reported median difference of 12.6% between sensor-meter pairs. Even the meters used to calibrate the CGMS provide only 56 to 69% of self-monitored blood glucose values within 10% of corresponding laboratory results (4).

As stated in the instructions for use, information provided by the CGMS “is intended to supplement, not replace, blood glucose information using standard home glucose monitoring devices” by providing glucose pattern and trend information for 24–72 h. Metzger et al. stated, “Clinical decisions should not be made on the sole basis of glucose sensor data” (1). When used with self-monitored blood glucose and HbA_1c values, clinical decisions based on CGMS profiles are designed to help the diabetes team optimize management.

Metzger et al. included several figures depicting discrepancies between sensor-sensor profiles, the most obvious being patient K. The authors failed to note that the sensor software labeled the depressed tracing as not meeting optimal accuracy criteria and that it therefore should have been further investigated by the clinician before use. The authors further report a 35% discrepancy in the clinical interpretation of sensor-sensor profiles. This method of assessment is subjective and has not been validated. Moreover, based on a Clarke error grid analysis, >93% of the CGMS readings are clinically accurate or acceptable.

Metzger et al. report technical problems in 18% of the profiles generated by Solutions 2.0 software. The upgraded Solutions 3.0 software resolves many of the technical reasons for which data were discarded in their analysis, corrects the abrupt shift in value at midnight, improves the accuracy and reproducibility of the data downloads, and improves the agreement between sensor and meter values as measured by mean absolute percent difference (18.4 vs. 16.1%) and correlation (0.91 vs. 0.92) (5).

An established body of literature, including a supplement to the journal Diabetes Technology & Therapeutics (6), supports the performance and utility of the CGMS to guide therapy adjustments based on glycemic excursions and to improve and predict HbA_1c. The results reported by Metzger et al. should be weighed against the encouraging results and conclusions of both evolving and previously reported research.