Assessment of dementia risk scores in predicting mild cognitive impairment: A comparison of CogDrisk, CAIDE, LIBRA, and ANU-ADRI.

Abstract

Background: Given the lack of widely accessible dementia treatments, identifying individuals at high risk of dementia is vital for prevention. No prior study has compared multiple validated dementia risk tools for predicting mild cognitive impairment (MCI) across multiple datasets. We assess the performance of the CogDrisk, ANU-ADRI, CAIDE, and LIBRA in predicting MCI.

Method: Data were obtained from the ARIC, Whitehall II, and PATH Through Life cohorts. Participants without dementia or MCI at baseline were included. Risk scores were computed using available risk factors and analysed using logistic regression, with Area Under the Curve (AUC) estimates. Multiple imputation was used to evaluate the impact of missing data.

Results: The ARIC (n = 5778), Whitehall II (n = 6387), and PATH (n = 2115) cohorts had mean baseline ages of 51.9, 55.8, and 62.5 years, with follow-ups of 28.2, 15.7, and 11.2 years, respectively. AUCs for MCI prediction were generally similar across tools and datasets. Dementia prevalence following MCI was highest in ARIC (23.6%), followed by Whitehall II (14.1%) and PATH (7.0%). In ARIC, CogDrisk showed slightly better AUCs for predicting MCI cases that progressed to dementia. Whitehall II and PATH showed mixed results, with wider confidence intervals for progressing MCI cases, and higher AUCs for non-progressing MCI cases using CogDrisk and ANU-ADRI. All tools performed consistently when predicting dementia without prior MCI.

Discussion: Dementia risk scores demonstrated comparable performance of MCI prediction and are more sensitive for identifying cases that progress to dementia, supporting their greater utility for informing risk reduction strategies.