Estimating Likelihood of Dementia in the Absence of Diagnostic Data: A Latent Dementia Index in 10 Genetically...

Epidemiological research on dementia is hampered by differences across studies in how dementia is classified, especially where clinical diagnoses of dementia may not be available.

We apply structural equation modeling to estimate dementia likelihood across heterogeneous samples within a multi-study consortium and use the twin design of the sample to validate the results.

Using 10 twin studies, we implement a latent variable approach that aligns different tests available in each study to assess cognitive, memory, and functional ability. The model separates general cognitive ability from components indicative of dementia. We examine the validity of this continuous latent dementia index (LDI). We then identify cut-off points along the LDI distributions in each study and align them across studies to distinguish individuals with and without probable dementia. Finally, we validate the LDI by determining its heritability and estimating genetic and environmental correlations between the LDI and clinically diagnosed dementia where available.

Results indicate that coordinated estimation of LDI across 10 studies has validity against clinically diagnosed dementia. The LDI can be fit to heterogeneous sets of memory, other cognitive, and functional ability variables to extract a score reflective of likelihood of dementia that can be interpreted similarly across studies despite diverse study designs and sampling characteristics. Finally, the same genetic sources of variance strongly contribute to both the LDI and clinical diagnosis.

This latent dementia indicator approach may serve as a model for other research consortia confronted with similar data integration challenges.