Metformin: We need to either put it in our drinking water or rethink how we study it

STRUCTURED ABSTRACT Objectives To expose the potential impact of residual confounding in common observational study designs investigating metformin using a type 2 diabetes cohort; to propose a more robust study design for future observational studies of metformin. Design Retrospective cohort studies using a prevalent user design conducted in two distinct cohorts: individuals with type 2 diabetes and individuals with prediabetes. Setting Insurance claims database for Medicare Advantage beneficiaries in the United States, 2018-2019. An identical analysis of commercial insurance beneficiaries appears in the supplement. Participants 404,765 individuals with type 2 diabetes, 81,791 individuals with prediabetes. Main outcome measures Total inpatient admission days in 2019, total medical spend (excluding prescription drugs) in 2019. Each of these measures is treated as a binary outcome: greater than zero inpatient days and top 10% medical spend. Results We implement a common observational study design and observe a strong metformin effect estimate associated with reduced inpatient admissions and reduced medical expenditures; we also implement a more robust study design that suggests any estimated effect is attributable to residual confounding related to individuals’ overall health. Conclusions Common observational study designs examining metformin in a type 2 diabetes population are likely impacted by significant residual confounding. By additionally considering numerous negative control outcomes and a complementary prediabetes cohort, the study design proposed here demonstrates efficacy at exposing residual confounding related to overall health, nullifying the claim derived from a standard study design. Trial registration Preregistration available at https://osf.io/qf49p .