Circulating osteocalcin and incident dementia in the UK Biobank: a prospective cohort and exploratory multi-omics analysis

Background We examined baseline relative circulating osteocalcin (OCN) in relation to incident dementia in UK Biobank. Methods The longitudinal cohort comprised 43,942 participants; models used available complete records. Cox models used baseline OCN standardized per SD and follow-up censored at diagnosis, death, or November 27, 2024. Cognitive, MRI, and baseline molecular analyses were exploratory. Results In the fully adjusted model, OCN was not clearly associated with all-cause dementia (HR 1.052; 95% CI, 0.988–1.120; p = 0.117; N = 40,253; 1,095 events). The vascular dementia estimate was nominally positive (HR 1.159; 95% CI, 1.010–1.328; p = 0.035), but did not survive Holm correction across three secondary outcomes (adjusted p = 0.105) and weakened after excluding baseline stroke. Adding OCN changed the mean fivefold cross-validated C index from 0.84628 to 0.84642 (ΔC = 0.00014). No baseline cognitive association survived Holm correction across four tests. Nominal hippocampal and WMH associations in the larger MRI sample were not evident in matched extensively adjusted samples. Repeated-visit cognitive models did not yield adequately sized multiple classes for interpretable class-membership regressions. Baseline measured-only molecular analyses identified OCN-related correlations, but age- and sex-adjusted BCAA correlations did not survive panel-wise Bonferroni correction. Conclusion Baseline relative OCN did not show a clear fully adjusted all-cause dementia association or useful incremental discrimination. Subtype, cognitive, imaging, and molecular findings remain exploratory. These data do not establish a compensatory mechanism or clinical predictive utility.