Researchers have developed a benchmark called FairGlucose to evaluate the fairness of continuous glucose monitoring (CGM) systems. They created a dataset with 300 patients from different demographics and tested 33 AI models on it. The results show that population-level validation can mask disparities between subgroups, such as type 1 and type 2 diabetes patients. This suggests that relying solely on aggregate metrics may not be sufficient to ensure fairness in digital health AI. The study recommends subgroup-disaggregated reporting as a standard practice. --- Why it matters: This matters because it highlights the need for more nuanced evaluation methods in digital health AI, particularly when it comes to fairness and equity. Engineers and researchers working on CGM systems should consider the potential disparities between subgroups and strive to develop models that perform well across different demographics. Source: https://arxiv.org/abs/2608.18296