Comparator Value Pairing Impacts Reported Continuous Glucose Monitoring System Accuracy

In clinical continuous glucose monitoring (CGM) performance studies, the accuracy of a CGM system is assessed by comparing its readings with blood glucose (BG) measurements obtained at the same time. Because CGM readings are stored in fixed intervals (typically five minutes) while comparator measurements are performed manually (typically every 15 minutes), CGM readings and comparator measurements may not be perfectly synchronized. This raises the question of how CGM and BG values should be paired if they were recorded at similar but different timestamps.

A literature review on CGM accuracy studies revealed that multiple different methods have been used in past studies: ¹

1. “Linear interpolation”: Linear interpolation is used to fill the gaps between CGM readings (one-minute intervals) and the CGM reading (interpolated or actual) with the same timestamp as the BG value is utilized (14 studies).

2. “Closest”: The CGM reading that was recorded closest in time to the BG value is paired. This can be simultaneously, before, or after the BG timestamp (30 studies).

3. “CGM after”: The CGM reading recorded simultaneously or after the BG timestamp is always used (18 studies).

4. “CGM before”: The CGM reading recorded simultaneously or before the BG timestamp is always used (four studies).

A difference of up to 1.8% in mean absolute relative difference (MARD) was found between methods using data from a recent study of a CGM system with a five-minute sampling interval (Figure 1). ² Furthermore, these approaches only affect the level of performance observed in CGM performance studies, but not the systems’ clinical usefulness. This considerable variability highlights the need for standardization of the pairing method and also raises the question of the most appropriate method.

Figure 1.

Schematic illustrations of the four pairing methods “linear interpolation” (panel a), “closest” (panel b), “CGM after” (panel c), and “CGM before” (panel d). In this example, CGM recording intervals of five minutes were selected, but the approaches are independent from CGM systems and recording intervals. The mean absolute relative differences (MARD) in the titles were calculated by applying the different pairing methods to data from the same CGM system and comparator measurements collected in a recent CGM performance study. ²

As expected, the “CGM after” method shows the highest (ie, best) accuracy because it can systematically compensate for a CGM system’s time lag and is likely favored by manufacturers for that reason. In contrast, the “CGM before” method leads to the lowest (ie, worst) accuracy as it can exacerbate the time lag. The “linear interpolation” and “closest” methods are a compromise between these two extremes. They are neutral in a technical sense since they do not impact the perceived time lag of the system. However, the “linear interpolation” method can generate CGM readings that are never displayed to the user.

The authors acknowledge that each of the four approaches has its own strengths and limitations, which may be more or less relevant depending on specific goals of CGM performance studies. However, in the interest of making results from different studies more comparable through standardization, the authors recommend using the “closest” pairing method in all future CGM performance evaluations. This recommendation is based on the relative popularity of this method in the literature as well as the fact that the “closest” pairing method is neutral regarding time lag while only pairing actually recorded CGM readings. In case that two CGM readings were recorded equidistant to the BG timestamp, which can occur if the CGM sampling interval is an even number, it is recommended to pair the earlier CGM reading.