Supplementary file 1_Insulin resistance–related metabolic burden, white matter microstructure, and cognitive change: the HABS-HD study.docx

Selection stability was assessed using 1,000 participant-level bootstrap samples. All candidates underwent longitudinal mixed-effects analysis, with indirect estimates calculated as baseline exposure–white matter coefficients multiplied by white matter-by-time coefficients. Intervals used 2,000 participant-level cluster bootstrap samples.

Sensitivity analyses adjusted for white matter hyperintensity (WMH) volume and covariate-by-time interactions.</p>Results<p>Of 2,606 participants, 788 had two cognitive assessments a median of 2.17 years apart. Fourteen candidate pathways were selected, all involving RD and TyG or TyG-WHtR; 11 had 95% conditional Monte Carlo uncertainty intervals excluding zero. Two of 36 models met the supplementary model-level FDR criterion.

Five longitudinal pathways had negative indirect estimates with 95% bootstrap intervals excluding zero: four involved right cingulum RD and processing speed or executive function, and one involved right superior corona radiata RD and processing speed. Four also met the FDR criterion for RD-by-time interactions. The TyG-WHtR–right cingulum RD–executive function indirect estimate was -0.0054 per year (95% bootstrap interval, -0.0110 to -0.0009).

Selection frequencies ranged from 6.7% to 42.0%. After WMH and WMH-by-time adjustment, intervals excluded zero for two of these five pathways, decreasing to one with additional core covariate-by-time interactions.</p>Conclusion<p>Triglyceride–glucose index and TyG-WHtR showed statistical indirect associations with cognition through white matter RD, with some pathways associated with subsequent cognitive change. Limited selection stability and sensitivity to WMH and covariate-by-time adjustment warrant evaluation in independent cohorts.</p>