Fig. 3

The significant causalities in the bidirectional two-sample Mendelian randomization analysis. These results all met the following criteria: the P-value derived from IVW method was < 3.40 × 10^–5; all estimates from four Mendelian randomization methods showed consistent directions of association (positive or negative); no heterogeneity was identified by the Cochran’s Q test after removing the underlying outliers by MRPRESSO; P-value of MR-Egger intercept and MR-PRESSO global tests were both > 0.05; the estimate was not biased by single single-nucleotide polymorphism in leave-one-out analysis. SA, surface area; IVW, inverse-variance weighted; MR-PRESSO, MR Pleiotropy RESidual Sum and Outlier; IVs, instrumental variables