Assessing the Sensitivity of Brain-Age to Alzheimer's Disease in different Ethnic Groups

197 

Abstract Submission 

Zeena Shawa¹, Aghogho Onojuvbevbo², Sophie Martin¹, Neil Oxtoby¹, James Cole¹

¹University College London, London, England, ²Sheffield Hallam University, Sheffield, England

Zeena Shawa, MRes  
University College London
London, England

Aghogho Onojuvbevbo  
Sheffield Hallam University
Sheffield, England

Sophie Martin, MRes  
University College London
London, England

Neil Oxtoby, PhD  
University College London
London, England

James Cole, PhD  
University College London
London, England

Alzheimer's Disease is the most common neurodegenerative disease and cause of dementia [1,2]. The global burden of dementia is growing, with the number of people living with dementia projected to increase to 152 million by 2050. This growth is estimated to rise particularly in low and middle-income countries [2]. Although there have been advances regarding predicting dementia onset and progression, it is important that the performance of these research outputs are verified in different populations. Additionally, there is a lack of literature examining the potential impact of ethnic and racial factors [3, 4]. Brain-age is an index of the brain's biological age derived from structural imaging. It correlates with an increased risk of dementia in memory clinic patients and has the potential to aid in early dementia diagnosis [5]. However, a significant portion of the brain-age literature uses less diverse cohorts [5, 2]. Thus, this research aims to investigate the sensitivity and generalizability of brain-age in non-white individuals.

We analysed data across 23 sites from the National Alzheimer's Coordinating Centre (NACC) database [6]. 389 cognitively normal (CN) individuals (68.7±8.5 years; 278 female) and 189 patients living with Alzheimer's Disease (AD) (73.0±9.9 years; 118 female) were included, after some filtering to age and sex-match the groups (non-White and White for CN and AD each). Fig. 1 contains the demographics across all groups. The brainageR model [7] was used to estimate individuals' brain-age. This model was trained on n=3377 predominantly White healthy individuals from seven public datasets. To assess brainageR's generalisability, we compared the brain-predicted age difference (brain-PAD) between CN and AD groups within the non-White and White populations in the NACC dataset. T-tests and Cohen's d effect sizes were compared between the groups examined.

Fig. 2(a) shows the brain age against chronological age for individuals in each group and associated R² values for the line of best fit. Both CN groups have an R² value above 0.5, indicating that there is a moderate and similar amount of variance explained in these groups. This provides no evidence of ethnic differences in model fit in CN samples. Mean Absolute Error (MAE) and 95% confidence intervals of brain predicted age compared to chronological age for each group was: 6.66 ± 1.37 for Non-White AD, 5.93 ± 0.84 for Non-White CN, 5.86 ± 1.19 for White AD individuals, and 5.36 ± 0.76 for White CN. Consequently, the MAE of individuals with AD for both ethnic groups are within the confidence intervals of the respective CN groups.

A Welch Two Sample t-test showed significant differences in the mean Brain-PAD values between the CN and AD groups (p=0.000 for both White and Non-White, Fig. 2(b)), reflecting the impact of AD on brain structure. The Cohen's d effect sizes when comparing the White AD and White CN subgroups were d=0.83 ± 0.22, while for the Non-White CN and Non-White AD subgroups was d=0.55 ± 0.25. The effect size of AD on brain-PAD is weaker in the Non-White group, though still significantly greater than 0 (based on 95% CIs).

A brain-age model trained on a less diverse dataset generalised similarly to white and non-white CN groups and was partially robust to AD when tested in a sample of different ethnicity. Thus, brain-age could be used in under-represented groups to aid patient prognosis, clinical trial stratification, disease staging, and more [8, 9]. However, more work is required to demonstrate and improve generalisability in further demographic groups.

Neurodegenerative/ Late Life (eg. Parkinson’s, Alzheimer’s) ¹

Aging

Classification and Predictive Modeling ²

Other Methods

Aging

Data analysis

Degenerative Disease

Modeling

MRI

Neurological

Other - Brain-Age