Response to Sensitivity of the Glycemia Risk Index to Effects of Automated Insulin Delivery Initiation

Donaldson et al ¹ summarized glycemic results for a real-world population initiated on automated insulin delivery (AID) systems and previously reported in greater detail. ² At 12 months, the Glycemia Risk Index (GRI) improved by 8.7 percentiles, and time in range (TIR^{70-180 mg/dL}) improved by 8.1%. The GRI is calculated as the quality of glycemia based on a hypoglycemia component and a hyperglycemia component. In Donaldson’s population, following AID use, the improvement in the GRI was limited to a reduced hyperglycemia component (reflecting less time in hyperglycemia), whereas the hypoglycemia component remained unchanged (reflecting no change in time in hypoglycemia). ¹

This was an unusual patient cohort with a low average rate of time below range (TBR^{<70 mg/dL)}) of only 1.3% at onset of AID treatment. ² This low incidence of hypoglycemia did not change from the intervention. By contrast, in the Wireless Innovation for Seniors with Diabetes Mellitus Randomized Controlled Trial (WISDM RCT) of people with type 1 diabetes (PWT1D), the mean pre-intervention TBR^{70-180 mg/dL} was 5%. ³ In the Diabeloop study on PWT1D, an AID system resulted in TBR reduction from 2.4% to 1.3%, ⁴ which is the same level as the pretreatment TBR^{70-180 mg/dL} of the Donaldson cohort. ² Donaldson’s real-world cohort had a baseline %CV of 34%, ² which is consistent with “stable glucose levels” (defined as <36%) in the 2019 international consensus statement on continuous glucose monitoring (CGM). ⁵

As Donaldson’s team demonstrated, in a population with a low frequency of hypoglycemia and stable glucose levels, an intervention not affecting the population’s minimal baseline hypoglycemia is unlikely to benefit GRI much more than TIR. They reported that the improvement in GRI exceeded the improvement in TIR at both time periods where these investigators compared the two metrics (delta GRI in percentiles vs delta TIR in percent at 13¹ vs 11² at 1 month and 8.7¹ vs 8.1¹ at 12 months). The effect of treatment (AID initiation in the Donaldson series) depends not only on the treatment but on the baseline characteristics of the group being treated.

The GRI was developed to reflect both hypoglycemia and hyperglycemia. If there is no overall change in hypoglycemia, then the overall hypoglycemia component of the GRI should not change. Reducing severe (Level 2) hyperglycemia (glucose > 250 mg/dL) can reduce the GRI substantially, but if improvements are limited to reducing mild (Level 1) hyperglycemia (glucose > 180-250 mg/dL), then the hyperglycemia component of the GRI should decrease less. ⁶ Donaldson’s report did not separate time above range into time in mild hyperglycemia and time in severe hyperglycemia.^1,2 In a study population with a higher frequency of hypoglycemia or greater glycemic variability than Donaldson’s, the difference between improvement in GRI and TIR would be more impressive.

We agree with Donaldson’s team’s conclusion that evaluating GRI in cohorts with higher baseline rates of hypoglycemia following initiation of diabetes interventions would further assist with evaluating the clinical utility of GRI. When an intervention is administered to a population with larger baseline risks for hypoglycemia or greater baseline glycemic variability than Donaldson’s, GRI would be expected to show greater sensitivity than TIR^{70-180 mg/dL}.