Applicability of cognitive decline risk scores in a high socioeconomic preventive medicine cohort

Background Validated cognitive decline and Alzheimer's disease risk scores (CAIDE, ANU-ADRI, LIBRA, LIBRA2) were developed in general-population cohorts; how they behave in the high-socioeconomic, health-selected populations typical of preventive-medicine programs and over-represented in ageing cohorts is undescribed. Objective Lacking a cognitive outcome, we did not test predictive validity but described score distributions, inter-score concordance, the components driving variance, and complementary biological data. Methods Four scores were computed for 895 participants aged ≥45 years in the Zoī cohort, a self-referred, self-funded preventive-medicine sample. Concordance was assessed by Spearman correlations and, after dichotomization, by pairwise agreement (percentage agreement, Cohen's κ). Principal component analysis identified components accounting for most between-person variance. Complementary data (DEXA body composition, HbA1c, erythrocyte omega-3) were cross-tabulated with score-derived risk categories. Results Concordance was moderate (ρ = 0.39–0.60), and dichotomized categories were fairly-to-moderately concordant across families (Cohen's κ 0.33–0.52). Protective variables showed ceiling effects (education ≥10 years, 86%; social/cognitive activity, 100%), leaving cardiovascular factors (BMI, hypertension, coronary heart disease) to account for most variance. Despite 91% having normal BMI (≤30 kg/m 2 ), DEXA-measured fat mass showed a dose-response gradient across all scores (CAIDE 4.36–6.26); HbA1c revealed a glycemic gradient invisible to self-reported diabetes. Conclusions In this high-socioeconomic cohort, the scores were moderately concordant and varied almost solely by cardiovascular components, protective factors being saturated. Whether this reflects uniformly low risk or reduced ability to separate individuals cannot be determined without a cognitive outcome. Complementary biological measurements added stratification, supporting augmented risk-assessment tools to be validated against cognitive endpoints.