Dementia risk prediction in the general population: external validation of a prediction model in the population-based LifeLines Cohort Study.

Background: Models for dementia prediction in primary care are necessary to identify individuals at risk for developing dementia, but their implementation in clinical practice is partly limited due to lack of external validation or use of high-cost variables. We externally validated the predictive performance of a simple yet promising dementia risk prediction model.

Methods: We assessed discriminative ability with a c-statistic with 95 % confidence interval, using age, history of stroke, subjective memory complaints and need for assistance with a relatively complex task as predictors. This was done on 10,007 individuals that participated in the Lifelines-cohort study. Assessment of dementia in the Lifelines Cohort Study is self-reported in the follow-up questionnaires.

Results: Mean follow-up at LifeLines timepoint 1b was 1.5 years, mean follow-up at LifeLines timepoint 2a was 3.3 years and mean follow-up at LifeLines timepoint 3a was 9.1 years. Overall, 36 participants self-reported dementia development. Discriminative ability of the model overall dementia development yielded a c-statistic of 0.62 [95 % CI=0.48-0.70], and performed slightly better at follow-up 2a 0.67 [95 % CI=0.57-0.78]. However, calibration of the model in this external validation cohort was poor, with systematic overestimation of the predicted risk.

Conclusion: In this study the basic dementia risk prediction model overestimated the risk of dementia, but had reasonable discriminative ability in the Lifelines cohort. Within this validation cohort the potential of the model is underestimated due to low incidence of reported dementia. Further validation is required to determine the true value of the model. Studies assessing its implementation feasibility in primary care should also be conducted.