Stage vs subtype hypothesis: A longitudinal MRI study investigating the heterogeneity in AD

Abstract Background Longitudinal measurements of brain atrophy have revolutionized our understanding of how Alzheimer’s disease (AD) pathophysiology evolves during the disease course. The distribution of neurofibrillary tangles (NFTs) in AD, varies between patients and correlates in vivo atrophy measures and specific profiles of cognitive decline. Longitudinal progression of atrophy markers in AD will help understanding the course of the disease. We wanted to investigate if atrophy differences between patients with AD dementia already exist at the time of the clinical onset and estimate their atrophy patterns over time. Method Our dataset consist of Freesurfer preprocessed T1‐weighted structural MRIs(2 to 5 longitudinal MRI visits) for 322 amyloid positive AD diagnosed patients and 304 amyloid negative cognitively unimpaired (CU) subjects. We applied an in‐house longitudinal clustering pipeline on the AD sample after controlling for normal ageing effect (CU sample), to explore the heterogeneous patterns of AD over time, after the AD clinical onset. Result The model revealed heterogeneous patterns of atrophy in the AD subjects, predominantly in the baseline data, but also in their trajectories over time. Three distinct patterns of regional brain atrophy at AD onset and decline over time were identified. Approximately 55% of the AD patients presented the typical AD atrophy pattern, including some subjects who initially showed atrophy mainly in the temporal lobes but progressed similarly to typical AD. Atypical patterns of atrophy were also found in our dataset: i) Minimal atrophy AD with very few atrophy at baseline and subtle atrophy over time (10%), ii) hippocampal sparing early onset (HSEO) with posterior cortical atrophy (15%) and hippocampal sparing late onset (HSLO) with milder posterior atrophy and later AD onset (10%). The HSEO cluster has steeper atrophy trajectories and faster decline in all cognitive domains compared to HSLO. A 10% of the sample had no clear classification in any of the groups. Regional atrophy between subtypes was associated with longitudinal cognitive performance. Conclusion Our results show not only that AD is heterogeneous which is known from cross sectional data, but also that atrophy subtypes exist most probably due to different diseases and not due to different stages of the disease.