Dementia risk prediction: A comparative analysis of the ANU-ADRI, CAIDE, CogDrisk, LIBRA, and LIBRA2 indices in the HUNT study.

Abstract

Background/objective: Dementia is a major global health concern, necessitating effective risk assessment tools and early intervention. This study compared the performance of five modifiable dementia risk indices - ANU-ADRI, CAIDE, CogDrisk, LIBRA, and LIBRA2 and a "demographics-only" (age, education) model.

Methods: We analyzed data from 5247 Norwegian participants in the Trøndelag Health Study (HUNT4 70+, 2017-2019) and dementia risk indices from baseline data in HUNT3 (2006-2008). Logistic regression models assessed associations between standardized index scores and all-cause dementia and Alzheimer's disease (AD) across age group (<65 vs. ≥65 years), sex, and APOE4 status.

Results: During the mean follow-up of 10.6 (9.3-12.3) years (SD=0.74), all indices significantly predicted dementia and AD, though none outperformed the demographics-only model. CogDrisk showed significantly better discriminative ability than all other indices (0.76, 95 % CI:0.74-0.78; DeLong p < 0.05), followed by LIBRA (0.75, 95 % CI:0.72-0.77) and ANU-ADRI (0.74, 95 % CI:0.72-0.76). LIBRA2 (0.69, 95 % CI:0.66-0.71) and CAIDE (0.59, 95 % CI:0.56-0.61) had significantly lower accuracy (DeLong p < 0.001). Removing demographics maintained rank order but reduced accuracy across all indices. Stratified analyses showed stronger performance in those ≥65 years and females at HUNT3, while APOE4 status did not affect performance.

Conclusion: All indices were associated with dementia risk, with CogDrisk performing best across all conditions, and LIBRA2 and CAIDE performing weakest. No index outperformed a model including age and education only. Future research should refine risk indices for age- and sex-specific applications and assess whether simpler demographic models may suffice in some contexts.