IC-P3-225: Regional atrophy patterns associated with risk of future progression to AD in amnestic MCI using a new time-to-event voxel based method

Abstract

While differences in patterns of atrophy between MCIs who progress to AD within a fixed time period versus non-progressors have been characterized with voxel based morphometry (VBM), additional insight may be gained by identifying regions that are associated with increased risk of future progression to AD. The goal of this study was to use a new time-to-event voxel based analytic method to identify anatomic regions on MRI where atrophy is associated with significantly increased hazard (risk) of future progression to AD in subjects with amnestic mild cognitive impairment (aMCI). This approach differs from traditional VBM by incorporating statistical methods that account for varying lengths of follow-up. Subjects with a diagnosis of aMCI at baseline and a concomitant MRI were followed prospectively until a diagnosis of AD with censoring at last clinical follow-up, death, or a diagnosis of non-AD neurodegenerative disease. Cox proportional hazards models adjusted for age, sex, and education were used to estimate the hazard of progression from aMCI to AD based on voxel level gray matter density (GMD) estimates. Voxel-wise hazard ratios (HRs) are reported comparing subjects with GMD at the 25th percentile to those at the 75th percentile. A total of 95 aMCI subjects (45 women, median age 77 yrs.) were followed for a median of 2 years. Fig. 1 shows a map of voxels with HRs greater than 2 (p<0.05), representing regions where increased atrophy more than doubled risk of progression. The color scale in the Figure indicates the range of significant HRs, with blue corresponding to HR of 2 and red to HR of 6. Not surprisingly, the greatest risk of progression to AD is associated with atrophy of the medial and basal temporal lobes. Other areas of atrophy associated with increased risk of progression are the lateral temporal and parietal neocortex. We have applied time-to-event statistical methods to SPM5 derived GMD estimates in a novel way to identify regions of atrophy that significantly increases aMCI patients’ risk of progression to dementia. In the context of a progressive disease like AD, time-to-event VBM seems more appropriate than traditional two-sample methods.