Fig. 4

LASSO linear regression models of baseline and change VAT area and proportion. The x-axis displays the variables selected by the LASSO model, and the y-axis represents the estimated β-unstandardized coefficients. The magnitude and direction by which each variable affects the baseline VAT area (A) and proportion (B) and the 18-month relative changes in VAT area (C) and proportion (D) are represented by the color (blue for positive associations and red for negative associations) and length of the bars. Baseline VAT area (A) model was trained on a set of 227 participants, tested on 55 participants and validated on 259 participants. Baseline VAT proportion (B) model was trained on a set of n = 218 participants, tested on n = 53 participants and validated on n = 143 participants. The VAT area change (C) model was trained on a set of n = 180 participants, tested on n = 46 participants and validated on n = 207 participants. The VAT proportion change (D) model was trained on a set of 172 participants, tested on 43 participants and validated on 212 participants. Abbreviations: LASSO, least absolute shrinkage and selection operator; VAT, visceral adipose tissue; WC, waist circumference; MAP, mean arterial pressure; TG, triglycerides; HDLc, high-density lipoprotein cholesterol; HOMA-IR, homeostatic model assessment of insulin resistance; GGT, gamma-glutamyl transferase; ALKP, alkaline phosphatase; hsCRP, high-sensitivity c-reactive protein; ALT, alanine transaminase