Background
Dementia, including Alzheimer’s disease (AD), is a critical and growing global health challenge, presenting a significant burden to individuals, families, and healthcare systems worldwide. Early identification of individuals at elevated risk is essential for timely intervention, prevention, and resource allocation. Multiple risk assessment tools incorporating modifiable risk factors have been developed to predict dementia risk. However, their comparative performance and clinical utility remain unclear. This article reviews a recent comparative analysis by Stubs et al. that evaluated five dementia risk indices—ANU-ADRI, CAIDE, CogDrisk, LIBRA, and LIBRA2—within the context of the large-scale, population-based Trøndelag Health Study (HUNT) in Norway.
Study Design and Methods
The study cohort comprised 5,247 Norwegian participants from HUNT4 70+ (2017-2019), with baseline risk factor data drawn from HUNT3 (2006-2008). The authors evaluated five modifiable risk indices—ANU-ADRI, CAIDE, CogDrisk, LIBRA, and LIBRA2—standardizing the scores for analytic comparability. To contrast these multi-factor indices with simpler demographic models, a “demographics-only” model including age and education was also employed.
Logistic regression analyses assessed associations between each index and incident all-cause dementia and AD over a mean follow-up of 10.6 years. Stratified analyses examined predictive performance by age (<65 vs. ≥65 years), sex, and APOE ε4 carrier status. Model discrimination was quantified using the area under the receiver operating characteristic curve (AUC), with statistical comparisons made via DeLong’s test.
Key Findings
All five indices were significantly associated with dementia risk, corroborating their relevance in risk stratification. However, no index surpassed the discrimination achieved by the demographics-only model incorporating age and education.
Among the risk indices, CogDrisk demonstrated the highest discriminative ability (AUC=0.76, 95% CI: 0.74-0.78), significantly outperforming the other indices (p<0.05). LIBRA (AUC=0.75, 95% CI: 0.72-0.77) and ANU-ADRI (AUC=0.74, 95% CI: 0.72-0.76) followed closely. By contrast, LIBRA2 (AUC=0.69, 95% CI: 0.66-0.71) and CAIDE (AUC=0.59, 95% CI: 0.56-0.61) showed markedly lower accuracy (p<0.001), with CAIDE’s performance particularly limited in this cohort.
Excluding demographic variables from the indices decreased predictive accuracy across all models but retained the relative ranking among indices, underscoring the dominant prognostic value of age and education.
Stratified analyses revealed that risk prediction was more robust among participants aged 65 years and older and females at baseline, suggesting possible age- and sex-related variations in risk factor impact or index calibration. APOE ε4 status did not significantly influence the indices’ predictive performance.
Expert Commentary
The findings underscore several important clinical and methodological points. First, while composite modifiable risk indices have utility, simple demographic factors—age and education—remain powerful predictors of dementia risk. This raises questions about the incremental clinical value and complexity of using multifactorial indices versus simpler models, especially in resource-limited settings.
CogDrisk’s superior performance may reflect its comprehensive inclusion of behavioral and lifestyle factors, reinforcing the multifactorial and modifiable nature of dementia risk. However, the relatively poor performance of CAIDE—a widely known index—suggests that index validation in diverse populations and settings is critical prior to widespread clinical implementation.
Age- and sex-specific differences highlight the necessity of tailored risk prediction tools that adjust for population heterogeneity. The lack of effect modification by APOE ε4 status may indicate that these indices capture risk pathways partially independent of genetic susceptibility, or that APOE effects are adequately accounted for through other variables.
Limitations include reliance on baseline risk factors measured approximately a decade prior to dementia diagnosis, which may not reflect changes over time. Also, the study population’s homogeneity (Norwegian adults) may limit generalizability to other ethnic groups or geographic locations.
Conclusions and Future Directions
This comparative analysis confirms that several established dementia risk indices significantly predict incident dementia, yet none clearly outperforms a simple demographic risk model based on age and education. Among indices, CogDrisk shows the best accuracy and may offer added value in clinical and research settings, but further refinement to enhance age- and sex-specific prediction is warranted.
Future research should explore integrating dynamic longitudinal risk factor data, genetic and biomarker information, and consideration of sociodemographic diversity. Furthermore, pragmatic studies are needed to evaluate whether simpler demographic models might suffice for screening purposes in specific clinical or public health contexts, thus balancing predictive accuracy with feasibility.
In clinical practice, practitioners should be aware of the strengths and limitations of various dementia risk assessment tools and consider employing them alongside traditional demographic indicators to inform individualized prevention strategies.
References
Stubs J, Langballe EM, Livingston G, Anstey KJ, Deckers K, Mathews FE, Kivimäki M, Strand BH, Rokstad AM, Krokstad S, Selbæk G. Dementia risk prediction: A comparative analysis of the ANU-ADRI, CAIDE, CogDrisk, LIBRA, and LIBRA2 indices in the HUNT study. J Prev Alzheimers Dis. 2025 Nov;12(9):100326. doi: 10.1016/j.tjpad.2025.100326. Epub 2025 Aug 18. PMID: 40829975; PMCID: PMC12501341.
