Operational definitions of type 2 diabetes influence study population characteristics and outcome associations: Empirical evidence from the UK Biobank.

AIMS: Operational definitions of type 2 diabetes (T2D) in multi-source cohorts may identify heterogeneous case sets, affecting epidemiological findings. We examined whether alternative T2D definitions altered phenotype composition and associations with clinical outcomes. METHODS: We implemented five T2D definition families in 501,936 UK Biobank participants, distinguishing diagnosed T2D from HbA1c-defined undiagnosed diabetes. We compared case composition, diagnosis timing, and polygenic and clinical profiles. Associations with mortality, myocardial infarction (MI), stroke, and dementia were estimated using Cox proportional hazards models, with hazard ratios (HRs) and 95 % confidence intervals (CIs) calculated. RESULTS: Diagnosed case counts varied across definitions (13,873-22,394), whereas undiagnosed diabetes showed high cross-family agreement (3,207). Undiagnosed diabetes had a more adverse cardiometabolic profile but weaker outcome associations than diagnosed components. The definition relying mainly on linked diagnosis records identified the narrowest case set and produced the highest diagnosed-only HRs across the four outcomes, especially for MI (3.82; 95 % CI 3.61-4.04) and dementia (3.79; 95 % CI 3.52-4.09). Including undiagnosed diabetes to the case sets attenuated HRs by 1.2-13.8 %, with the largest attenuation for dementia. CONCLUSIONS: Alternative T2D definitions changed case composition and outcome associations. Inclusion of undiagnosed diabetes shows substantial cardiometabolic burden but less accumulated disease history. Record-dependent definitions preferentially capture clinically documented cases disease, while careful adjudication of self-reported diabetes information with this provides wider population coverage. Definition choice should align with the intended research question and exposure contrast.