Study Design
Challenges in Evaluating the Impact of CGMs on Racial Disparities: Design Considerations for a Target Trial Emulation using All of Us Data Chloe R. Bennett* Chloe Bennett Robert Cavanaugh Louisa H. Smith
The burden of type 2 diabetes (T2D) disproportionately affects Black people, perpetuating health inequities. Continuous glucose monitors (CGM) improve glycemic control and reduce complications, potentially mitigating racial disparities in T2D outcomes.
This study uses the target trial framework to estimate the effect of CGM initiation on racial disparities in hospitalization. We specified a target trial in which non-Hispanic Black and white individuals with T2D and no prior CGM use are assigned to either initiate CGM use or continue standard monitoring.
We emulated the trial using data from the NIH All of Us Program, which includes electronic health record (EHR) and survey data. 17,507 participants had self-reported T2D, of which 13,360 had EHR data, and 83% of those had a T2D-related code appear in their EHR at any time. To emulate the target trial, eligible participants were 18+ with A1C measurement within 30 days and no CGM use within the past 6 months. 9298 participants met eligibility criteria at least once, 72% of whom were eligible for more than one iteration of the sequential emulated trial. We assessed time to hospitalization after meeting eligibility criteria (time zero).
In naïve analyses, both unadjusted and adjusted for sociodemographic characteristics, white participants consistently had higher hospitalization rates compared to Black participants, and the prescription of a CGM was strongly associated with hospitalization. Adjusting for diabetes-related characteristics and comorbidities reduced the effect estimate by approximately half, but a positive association with CGMs remained. We discuss possible origins of the presumed remaining bias and assess the effects of design choices. Specifically, we discuss definitions of eligibility criteria in survey and EHR data, confounding by indication, differences in monitoring by indication, as well as choices of time zero, a grace period for initiating treatment, and of outcome and censoring definition.