Fig. 4: Four-way decomposition and structural equations modeling of the association between Poor Oral Health Problems (POHP) and brain white matter microstructural integrity metrics (FA, MD, ICVF, ISOVF, OD, z-scored) by the single principal component of selected plasma proteomic biomarkers (k = 2: GDF15 and WFDC2), overall (N = 4663): UK Biobank 2006–2021.

A Four-way decomposition. B Structural Equations Model. DE_EST Direct effect estimate, DE_P Direct effect p value, DE_SE Direct effect standard error, ereri_cde excess relative risk due to neither mediation nor interaction or controlled direct effect, ereri_intmed excess relative risk due to mediated interaction or mediated interaction, ereri_intref excess relative risk due, to interaction only or interaction referent, ereri_pie excess relative risk due to mediation only or pure indirect effect, FA fractional anisotropy, ICVF Intracellular Volume Fraction, IE_EST Indirect effect estimate, IE_P Indirect effect p value, IE_SE Indirect effect standard error, ISOVF Isotropic Volume Fraction, MD mean diffusivity, OD Orientation Diffusion, Pd Periodontal disease, POHP Poor Oral Health Problems, pct_cde percent of total effect that is controlled direct effect, pct_intmed percent of total effect that is mediated interaction, pct_intref percent of total effect that is interaction referent, pct_pie percent of total effect that is pure indirect effect, TE_EST Total effect estimate, TE_P Total effect p value, TE_SE Total effect standard error, tereri Total excess relative risk, UK United Kingdom. See supplementary Table 5 for protein abbreviations. Note: Full model 1 included age, sex, race (Non-White vs. White), SES, household size, LE8 total score as exogenous variables; Full model 2 is Full model 1 with additional exogenous variables being fasting time (hours), season at baseline assessment, and time between baseline assessment and neuroimaging visit (days). *P < 0.05; **P < 0.010; ***P < 0.001.