The role of the independent variable to glucose sensor calibration

In vivo subcutaneous glucose sensor accuracy depends on the calibration method. Sensor accuracy was assessed during standard oral glucose tolerance tests in six non-diabetic subjects each wearing six subcutaneous glucose sensors (Medtronic MiniMed). Paired blood glucose (B(G)) and sensor current readings were used for retrospective sensor calibration using either B(G) or sensor current as the independent variable. Sensor accuracy after calibration was assessed using three criteria: linear regression between B(G) and sensor glucose (S(G)); correlation; and mean absolute difference (MAD), defined as 100 x |B(G) - S(G)|/B(G). Calibration with B(G) as the independent variable resulted in unbiased estimates of regression slope (1.02, not different than 1, p< 0.01) and y-intercept (-1.06 mg/dL, not different than 0, p< 0.01). In contrast, calibration with sensor current as the independent variable resulted in biased estimates of slope (0.76, different than 1, p< 0.01) and y-intercept (31.25 mg/dL, different than 0, p< 0.01). However, with sensor current as the independent variable, the MAD was lower than the corresponding value for calibration with B(G) at the x-axis (15.00 +/- 0.47% vs. 18.35 +/- 0.63%, p< 0.01). The Pearson correlation coefficient between B(G) and S(G) was higher when using sensor current as the independent variable (R = 0.82 vs. R = 0.79 when using glucose on the x-axis). We suggest that despite the fact that calibration with sensor current as the independent variable leads to a bias in the estimate of B(G), it is a more appropriate calibration method when the primary concern is minimization of the MAD between S(G) and B(G).