落花, 看到这个有关CGM MARD和你分享。虽然libre 3 MARD 数值比 G7好，但是很多人实际体验没有G7准

At ATTD, details around Roche CGM came out at a plenary session; the product is 'pre-CE Mark' but Roche noted on 2.1.24, it was 'very close' to launching. From what we can see, the product is promoted as having 14-day wear, a MARD of 9.2%, Apple Watch connectivity, a 2-hour glucose prediction curve and 30 min and night low prediction. The form factor looks larger than either ABT or DXCM and compares to DXCM G7 and ABT L3 MARDs of ~8.2% and 7.8% and 10- and 14-day wear, respectively (and DXCM's Stelo and aspirations for the G-Series to reach 15 days).

MARD alone is not the best accuracy marker and should not be used as the sole parameter to characterize the accuracy of CGM systems. MARD uses only a small portion of CGM data in its calculation, doesn't tell us much about comparative performance at low glucose levels (where some CGMs have struggled) and does not account for additional information like trend and rate of change given by CGMs. On that point, investors have been asking about whether Roche's longer-timeframe prediction algos could be an advantage. We note MDT has promoted similar features in the past with its low SG forecast (hypo prediction up to 4 hours in advance) and other predictive alerts. DXCM has been relatively dismissive of this approach, noting a lot can change in the real world, making longer-term predictions a moving target.