How well do dementia risk scores predict dementia? A systematic review and meta-analysis

In our systematic review and meta-analysis, we aimed to assess the predictive accuracy and clinical applicability of ten dementia risk assessment scores for incident dementia in cohort studies. Three of 10 risk scores had sufficient, comparable data for a meta-analysis. We analyzed the Cardiovascular Risk Factors Ageing and Dementia (CAIDE), Lifestyle for brain health (LIBRA), and Alzheimer's Disease Risk Index (ANU-ADRI) in middle- and older-aged baseline populations with short, medium, and long follow-up durations, starting free from dementia, with or without other health conditions. A total of 29 studies met our inclusion criteria; 13 were included in the meta-analyses. CAIDE performed best, particularly among middle-aged participants at baseline, with longer follow-up periods, yielding pooled areas under the curves (AUCs) of 0.74-0.76. ANU-ADRI showed stable, moderate predictive accuracy with AUCs of approximately 0.65-0.71. The original LIBRA showed the lowest discrimination, with AUCs ranging from 0.51 to 0.60, whereas the modified LIBRA performed better, with AUCs of 0.65-0.67 at short follow-up and a pooled AUC of 0.70 (95% CI 0.66-0.74) at medium follow-up. Dementia risk assessment scales show variable predictive performance across scores, baseline age, and follow-up duration, and may be useful for dementia risk stratification in clinical practice. As a substantial proportion of dementia risk is linked to potentially modifiable factors, identifying high-risk individuals may help guide timely risk-reduction strategies. Further prospective validation is needed across different populations and clinical settings.