Figure 3.

(A) Assessment of linearity in associations of BMI and all‐cause mortality in the UK Biobank sample of White British ancestry using BMI. Observational associations between BMI and all‐cause mortality obtained using conventional Cox regression adjusted for secular trends (date of birth), current occupation, qualifications, smoking status, alcohol intake, and physical activity. (B) Assessment of linearity in associations of BMI and all‐cause mortality in the UK Biobank sample of White British ancestry using instrument‐free BMI. Approximate analogue using MR stratified by categories of the instrument‐free exposure (divided at the 5th, 10th, 25th, 50th, 75th, and 85th percentile) adjusted for secular trends (date of birth) and first 10 genetic principal components. Localized average causal effects were then joined together and plotted against the corresponding percentiles of the original exposure. Linearity tests were conducted after removing data below or above the 1st or 99th percentile, respectively, because of the scarcity of data toward the tails of the BMI distribution. Hazard ratios (HRs) were calculated relative to the mean BMI (27 kg/m²), with 1,000 bootstrap resamples to obtain 95% confidence intervals (CIs). The black lines represent the fitted HRs from cubic spline models (with mean BMI as the reference).