Early detection of Alzheimer's disease subtypes in Amsterdam Dementia Cohort

Abstract

AbstractBackgroundHeterogeneity in the atrophy profile and clinical presentation of Alzheimer’s disease (AD) is well documented in literature. Defining tangible data‐driven subtypes and detecting these subtypes early in an individual is paramount to devising precise intervention strategies. In this study, we estimated distinct data‐driven AD subtypes and their corresponding atrophy timelines from cross‐sectional data and used it for predicting the subtypes of preclinical and prodromal patients.Method1725 AD (age=66.8±8.4), 856 mild cognitively impaired (MCI, age=67.1±8.0), 672 subjective cognitive decline (SCD, age=61.4±10.2) with unknown or positive amyloid status, and 756 controls or SCD with negative amyloid status (CN, age=58.8±9.0) were selected from the Amsterdam Dementia Cohort. Baseline T1w MRIs were processed using Freesurfer and the volumes were selected after automatic quality check (Rosen, NeuroImage, 2018). AD patients in the cohort were used for identifying distinct subtypes using non‐negative matrix factorization (ten Kate, Brain, 2018). Sequences of regional abnormalities were modelled using discriminative event‐based model (Venkatraghavan, NeuroImage, 2021) to identify early atrophy regions in each subtype. The trained model was used to predict the subtypes of MCI and SCD patients. Lastly, we estimated disease severity scores for each individual in the subtype they belong in, based on their position along the respective atrophy timelines.ResultWe identified four AD subtypes: typical subtype (NAD=453(26.3%)), early frontal cortical atrophy (FCA) (NAD=475(27.5%)), early posterior cortical atrophy (PCA) (NAD=349(20.2%)), and early subcortical atrophy (SA) (NAD=362(21.0%)). 86(5.0%) AD patients did not belong in any of these subtypes. The atrophy timelines of each subtype are shown in Figure 1. The trained model predicted the subtypes in 653 MCI (NTypical=202, NFCA=188, NPCA=86, and NSA=177) and 380 SCD patients (NTypical=83, NFCA=158, NPCA=52, and NSA=87). 203 MCI and 292 SCD patients did not belong to any subtype. Individual severity score increased for worse disease stages in each subtype, as seen in Figure 1, validating both the predicted subtypes and the estimated timelines.ConclusionWe identified four AD subtypes, their atrophy timelines, and used it for early detection of subtypes in MCI and SCD patients. Future work should focus on devising precision intervention strategy for patients in each subtype.