Head-to-head comparison of diffusion imaging markers for cognitive decline and risk of dementia: The population-based Rotterdam Study.

Abstract

Diffusion tensor imaging (DTI) is a promising tool for early detection of white matter deterioration, but the optimal DTI marker for assessing risk of cognitive decline and dementia is undetermined. The aim of this study was to compare the predictive value of multiple diffusion measures for cognitive decline and dementia risk in a population-based study. We measured peak-width skeletonized mean diffusivity (PSMD), normal-appearing white matter mean diffusivity (MD), and fractional anisotropy (FA) in participants aged 45 years and older who underwent 1.5T MRI between 2006-2015 as part of the ongoing prospective population-based Rotterdam Study. In a head-to-head comparison between DTI markers, we determined risk of cognitive decline and dementia during follow-up, using linear regression and Cox proportional hazard models, respectively, that also include age, sex, education, and presence of lacunes or microbleeds. Of 5230 participants (mean age: 65.0 years +/-10.9 [standard deviation], 55.8% women), 286 developed dementia during a median follow-up of 9.8 years. For dementia risk, PSMD and global MD were superior regarding model fit, with the highest hazard ratio for global MD (HR: 1.74 [95%CI: 1.53-1.98]). Similarly, PSMD and MD had the highest discriminatory value (C-statistic, for PSMD: 0.886, and for MD: 0.882), albeit differences with other measures were small. Of all participants, 4433 (84.3%) had complete data on all cognitive tests at baseline, of whom 2800 (63.2%) completed their follow-up visit after on average 5.6 years. For global cognition both at time of MRI and at follow-up, PSMD was superior in terms of model fit and showed the highest regression coefficients, followed by tract-specific measures of MD and FA. In conclusion, when directly comparing structural connectivity measures, PSMD is superior to MD and FA for prediction of cognition and cognitive decline, while PSMD and MD perform best with respect to dementia risk.