Synopsis

Keywords: Dementia, Aging, Cerebral small vessel disease

Individual brain MRI markers only show at best a modest association with long-term occurrence of dementia. Therefore, it is challenging to accurately identify individuals at increased risk for dementia. We implemented a combined hierarchical clustering analysis based on neurodegenerative and neurovascular brain MRI markers and identified 14 distinct subgroups of individuals with different brain MRI phenotypes. These subgroups had a different long-term risk for dementia; especially the multi-burden brain MRI phenotype showed an increased risk (HR: 13.8 (95%-CI:4.28-44.37)). These findings may in the future be useful to determine patient prognosis and may aid in patient selection for future treatment studies.

Introduction

Most older adults have brain changes on MRI, such as manifestations of neurodegenerative diseases and cerebral small vessel disease (SVD)¹ (for example white matter hyperintensities (WMHs) or lacunes). However, individual brain MRI markers only show at best a modest association with long-term occurrence of dementia². It, therefore, remains challenging to identify individuals who are at increased risk to develop dementia³. Due to heterogenous etiology and mixed pathologies, methods combining different brain MRI markers may likely aid in a more detailed characterization of potential prognostically relevant so called brain MRI phenotypes. We, therefore, aimed to identify different brain MRI phenotypes by combined hierarchical clustering analysis based on neurodegenerative and neurovascular brain MRI markers in community-dwelling individuals. Within each of these brain MRI phenotype subgroups, we determined the long-term dementia risk.

Methods

Participants & study design
The dataset used for the current analysis is part of the AGES Reykjavik study⁴ (n = 3056). At baseline, FLAIR and T1-weighted brain MRI scans were acquired on a 1.5 Tesla Signa Twinspeed system (General Electric Medical Systems, Waukesha, Wisconsin). Dementia outcome was determined as a binary variable (yes/no) by contacting the nursing homes 10.2 ± 2.4 years after the baseline MRI scan. The inclusion and exclusion of participants from the AGES-Reykjavik study for the current study is illustrated in figure 1. Participants who were demented at baseline were excluded from the analyses.
Cluster analysis
The brain MRI markers used to determine the brain MRI phenotypes were brain volumes for the estimation of brain atrophy (brain parenchymal fraction, white matter fraction, grey matter fraction, and lateral ventricle fraction), WMH markers (periventricular/confluent WMH fraction per lobe, deep WMH fraction per lobe, and the WMH shape parameters fractal dimension, solidity, convexity, concavity index and eccentricity⁵), brain infarcts (subcortical, cerebellar and cortical infarcts), microbleeds, and enlarged perivascular spaces (PVS). Brain MRI markers were normalized as z-scores (after multiplication by 100 and natural log transformation when not normally distributed), or otherwise scaled between -2 and 2. Hierarchical clustering was performed applying Ward’s method in R version 4.1.0 (R Core Team, 2021) and packages Nbclust ⁶, factoextra ⁷, cluster ⁸, and dendextend⁹. The dendrogram cut-off was determined using the Dunn index and the heatmap (see figure 2).
Statistical analysis
Cardiovascular risk factors were compared between subgroups with chi-square test for binary variables and one-way ANOVAs for continuous variables. Cox regression was used to estimate the risk of future dementia occurrence within the brain MRI phenotype subgroups (adjusted for age and sex). The reference group was chosen based on having the fewest brain abnormalities. SPSS version 25 (Chicago, IL) was used for the analysis.

Results

The optimal cut-off of the hierarchical clustering model was determined to be at 14 subgroups (figure 2). The main MRI markers per subgroup (S) are illustrated in figure 3. The brain MRI phenotypes of the subgroups ranged from limited burden (S12), mostly atrophy and infarcts (S14), mostly irregularly shaped WMH and atrophy (S3) to a multi-burden subgroup (S2). Baseline characteristics of the study sample per subgroup are shown in Table 1. Subgroup S12 was determined to have the least brain abnormalities and was used as the reference in the survival analysis (Figure 4). Baseline characteristics differed significantly between subgroups (Table 1). Dementia cases at follow-up ranged from 2% to 46% per subgroup. The range of hazard ratios across the subgroups is 4.1-13.8. The multi-burden subgroup S10 showed the largest hazard ratio of 13.8 (95% CI: 4.28-44.37) compared to the reference group (S12).

Discussion

We showed that distinct brain MRI phenotypes can be identified in community-dwelling older adults. Subgroups showed a different risk of developing dementia in the future, with an increased risk especially in individuals in the multi-burden brain pathology subgroup compared to a limited-burden subgroup. Identification of different brain MRI phenotypes can lead to novel insights into the MRI correlates of dementia predisposition. Our results revealed 14 distinct subgroups of individuals with different distributions of brain MRI markers of neurodegenerative and neurovascular disease. These findings may in the future be useful in determining patient prognosis and may aid in patient selection for future treatment studies.

Acknowledgements

This research was funded by an Alzheimer Nederland grant (WE.03-2019-08).