Composite risk scores for predicting dementia

Abstract

Purpose of review: A key priority in dementia research is the development of tools to identify individuals at high risk of dementia. This is important to prevent or delay dementia onset and as we move towards personalized medicine.

Recent findings: Numerous models (n > 50) for predicting dementia have been developed. These vary in the number (0 to 20+) and type (e.g. demographics, health, neuropsychological, and genetic) of predictor variables used for risk calculation, follow-up length (1-20 years) and age at screening (mid vs laterlife). Evaluation of the models shows that most have moderate-to-poor predictive accuracy. Few have been externally validated, raising questions about their generalizability outside the cohorts from which they were developed. The results highlight that if additional models are proposed the field will be overwhelmed with many competing risk models, making it difficult to reach consensus on which is best.

Summary: Numerous models for predicting dementia have been proposed but are limited by a lack of external validation and evaluation of economic impact. Innovative methods and data designs may be needed to improve derivation of dementia risk scores. Having a method for predicting dementia risk could transform medical research and allow for earlier testing of intervention strategies.