Objective Among adults with type 2 diabetes (T2D), several (but not all) studies show that being overweight (body mass index (BMI): 25.0–29.9 kg/m²) or obese I (BMI: 30.0–34.9 kg/m²) near the time of diagnosis, is unexpectedly associated with reduced all-cause mortality compared with normal weight—the obesity paradox. We addressed whether this observation is causal (eg, a true protective effect); due to confounding (including effect modification); or due to selection (‘collider’) bias.

Research design and methods We performed a matched population-level cohort study using primary care records from Salford, UK (1995–2012) in 10 464 patients with incident T2D paired (1:3) with 31 020 individuals who never developed T2D. We estimated HRs for associations of BMI with all-cause mortality using Cox models, stratified by smoking status.

Results Median follow-up was 8.7 years. For never smokers, the hazard of all-cause mortality increased from 25 kg/m², in a linear manner, with increasing BMI in the T2D cohort (HR per 5 kg/m²: 1.23, p_trend<0.001) and in the non-diabetes cohort (HR per 5 kg/m²: 1.34, p_trend<0.001). In contrast, among ever smokers, BMI-mortality relationships were U-shaped in the T2D and non-diabetes cohorts. Evidence of the obesity paradox in ever smokers, with and without T2D, argued against a selection bias, but supported a contribution of effect modification by smoking (p_interaction=0.009). Results were stable to various sensitivity analyses.

Conclusions In this cohort, the obesity paradox is mainly explained by smoking as an effect modifier. These findings indicate that the obesity paradox does not challenge standard weight management recommendations among T2D patients.