A critical review and classification of dementia risk assessment tools to inform dementia risk reduction.

Addressing modifiable dementia risk factors requires reliable risk assessment methods. We aimed to synthesise knowledge on risk scores for all cause dementia, Alzheimer's disease (AD) and vascular dementia, classify them according to target population, evaluate their content, cost, appropriateness of validation studies, and suitability for implementing risk reduction guidelines. A systematic search was conducted of PubMed, Cochrane Collaboration, ProQuest, Scopus, Embase, and PsycINFO databases using a pre-registered protocol. Data on risk factors, target population, predictive validity, cost, and alignment with WHO guidelines were extracted. Random-effects meta-analysis was performed. Of 45 risk scores identified, 29 were for all-cause dementia, including 11 based on late-life cohorts, 6 on midlife, and 7 covering mid to late-life. The pooled C-statistic across development and validation studies of dementia risk scores was 0.69 (95 % CI: 0.67, 0.71). Development study AUCs were higher than validation study AUCs and dropped from 0.74 to 0.66 for risk scores developed for clinical samples and from 0.79 to 0.71 for AD specific scores (which include functional indicators non-independent of disease). There were no validated risk scores for vascular dementia. Dem-NCD, CogDrisk, ANU-ADRI and LIBRA risk scores incorporated most WHO-recommended risk factors and demonstrated accuracy comparable to the overall pooled C-statistic. We conclude that across the field, there are methodological limitations relating to validation, and inappropriate comparison of tools designed for different purposes or target populations. However, there are now several validated, risk scores for all-cause dementia and AD that assess modifiable factors and offer cost-effective dementia risk assessment and risk reduction advice.