Abstract
Cognitive frailty (CF) is dynamic, but its longitudinal relationship with diabetes mellitus (DM) is unclear. We examined CF transitions and incident DM in three prospective cohorts. We included 23,458 adults aged ≥ 45 years without baseline DM from CHARLS, HRS and ELSA. Repeated assessments defined CF states and transitions. Generalized estimating equations (GEE) estimated population-averaged associations with subsequent DM. Longitudinal targeted maximum likelihood estimation (LTMLE; n = 13,616) estimated cumulative DM risks under hypothetical sustained patterns. In pooled adjusted GEE models, favourable dynamics were associated with lower next-wave DM odds than unfavourable dynamics (OR 0.66, 95% CI 0.59-0.74). In the three-category analysis, recovery had lower odds than persistent vulnerability/worsening (OR 0.79, 95% CI 0.66-0.95), but higher odds than persistent normality in the four-category analysis (OR 1.30, 95% CI 1.06-1.58). LTMLE estimated risks of 11.8% versus 18.1% under sustained favourable versus unfavourable dynamics (RD -6.2 percentage points, 95% CI -8.2 to -4.3; RR 0.66, 95% CI 0.56-0.76). Fine-Gray results treating death as a competing event were concordant. Repeated CF assessment may identify adults with persistent or worsening vulnerability who warrant intensified DM prevention. Associations of favourable dynamics of CF with lower DM risk require cautious interpretation and intervention studies.