Unravelling the heterogeneity of Mild Cognitive Impairment using Subtype and Stage Inference
Poster No:
146
Submission Type:
Abstract Submission
Authors:
Rodolfo Chiari-Correia1, Carlos Ernesto Garrido Salmon1, Neil Oxtoby2, Alexandra Young2, Francesca Biondo2, James Cole2
Institutions:
1University of Sao Paulo, Ribeirao Preto, Sao Paulo, 2University College London, London, London
First Author:
Co-Author(s):
Introduction:
Individuals diagnosed with mild cognitive impairment (MCI) typically experience initial signs of an abnormal cognitive decline without losing the ability to independently perform basic activities of daily living. This stage is commonly considered to be a prodromal stage of dementia-causing diseases such as Alzheimer's disease (AD). Nonetheless, the most commonly used diagnostic criteria, MCI due to Alzheimer's Disease (AD) [1], have several limitations, resulting in a complex diagnostic process, low accuracy, and a high heterogeneous group of patients [2,3]. One possible way to address this issue, is to classify MCI patients into more biologically specific subgroups - an approach that can be aided by quantitative biomarkers and unsupervised machine learning algorithms. However, clustering algorithms may incorrectly identify subtypes that are merely temporally distinct, that is, essentially only in different stages of a disease progression rather than phenotypically different. In this context, this study aimed to identify MCI subtypes using two different sets of biomarkers and a machine-learning technique named "Subtype and Stage Inference" (SuStaIn) [4]. The Sustain can uncover data-driven disease subtypes but also infer disease progression stages with entirely cross-sectional data.
Methods:
Participants
We analyzed data from 558 MCI subjects (mean age 73.5, 234 Female) and 215 amyloid-negative cognitively normal controls (mean age 72.4, 111 Female) obtained from the Alzheimer's Disease Neuroimaging Initiative (ADNI) database [5].
Biomarker Sets
Neuropsychological: Animals Category Fluency (CATANIMSC), Boston Naming Test (BNTTOTAL), Trail Making Test B (TRABSCOR), Rey Auditory Verbal Learning Test Trail 1 (AVTOT1), RAVLT Immediate (AVTOTAL) [6], RAVLT Learning [6], RAVLT Forgetting [6], Mini Mental State Exam (MMSCORE), and Geriatric Depression Scale (GDTOTAL).
Brain MRI and CSF-derived biomarkers: Volumes of Hippocampus, Cortical, Subcortical, Ventricles, and White Matter Hypointensities. CSF Amyloid-β 42 and Phosphorylated TAU concentration levels.
Biomarkers were selected based on data quality, sample size, variability, and correlation with variables within the same domain. The volume of each brain region was acquired using FastSurfer software [7] and normalized by intracranial volume.
Z-score Sustain
The z-score Sustain version characterizes the progression of a disease through a series of stages, in which each stage is associated with an increase in a biomarker to a new z-score relative to a control population [4].
We analyzed data from 558 MCI subjects (mean age 73.5, 234 Female) and 215 amyloid-negative cognitively normal controls (mean age 72.4, 111 Female) obtained from the Alzheimer's Disease Neuroimaging Initiative (ADNI) database [5].
Biomarker Sets
Neuropsychological: Animals Category Fluency (CATANIMSC), Boston Naming Test (BNTTOTAL), Trail Making Test B (TRABSCOR), Rey Auditory Verbal Learning Test Trail 1 (AVTOT1), RAVLT Immediate (AVTOTAL) [6], RAVLT Learning [6], RAVLT Forgetting [6], Mini Mental State Exam (MMSCORE), and Geriatric Depression Scale (GDTOTAL).
Brain MRI and CSF-derived biomarkers: Volumes of Hippocampus, Cortical, Subcortical, Ventricles, and White Matter Hypointensities. CSF Amyloid-β 42 and Phosphorylated TAU concentration levels.
Biomarkers were selected based on data quality, sample size, variability, and correlation with variables within the same domain. The volume of each brain region was acquired using FastSurfer software [7] and normalized by intracranial volume.
Z-score Sustain
The z-score Sustain version characterizes the progression of a disease through a series of stages, in which each stage is associated with an increase in a biomarker to a new z-score relative to a control population [4].
Results:
Two subtypes were identified by running SuStaIn on neuropsychological data. Subtype 1 was characterized mainly by changes in z-score in RAVLT forgetting and GDSCORE, while Subtype 2 in the TRABSCOR, as seen in figure 1A. Subtype 2 individuals exhibited worse performance in most neuropsychological tests, as well as lower cortical and hippocampal volumes, and higher ventricular and WM hyperintensities volumes. Additionally, they had a higher conversion rate to AD after 1 and 2 years (figure 2).
Using the MRI and CSF data, the SuStaIn also identified 2 subtypes. Subtype 1 characterization was led by z-score changes in PTAU, while subtype 2 was by changes in ventricular volume and WM hypointensities (figure 1B). Subtype 1 also had a higher percentage of individuals with the ε3ε4 and ε4ε4 APOE allele pair, as well as a higher rate of individuals who converted to AD.
In both analyses, MCI patients in stage 0 were reclassified as subtype 0, where biomarkers abnormalities were sub-threshold (z < 1).
Using the MRI and CSF data, the SuStaIn also identified 2 subtypes. Subtype 1 characterization was led by z-score changes in PTAU, while subtype 2 was by changes in ventricular volume and WM hypointensities (figure 1B). Subtype 1 also had a higher percentage of individuals with the ε3ε4 and ε4ε4 APOE allele pair, as well as a higher rate of individuals who converted to AD.
In both analyses, MCI patients in stage 0 were reclassified as subtype 0, where biomarkers abnormalities were sub-threshold (z < 1).
Conclusions:
The SuStaIn can identify MCI subtypes based on different biomarkers, each with distinct disease progression patterns and neurobiological characteristics. Additionally, it enables the detection of individuals with a higher risk of conversion to AD. (i.e., Neuropsychological subtype 2 and MRI/CSF subtype 1). In our future research, we plan to investigate the ability of the SuStaIn stage to predict conversion to AD using ML models.
Disorders of the Nervous System:
Neurodegenerative/ Late Life (eg. Parkinson’s, Alzheimer’s) 1
Modeling and Analysis Methods:
Classification and Predictive Modeling 2
Keywords:
Degenerative Disease
Machine Learning
MRI
Other - Mild Cognitive Impairment
1|2Indicates the priority used for review
Provide references using author date format
2. Beach, T. G. et al. (2012) ‘Accuracy of the clinical diagnosis of Alzheimer disease at National Institute on Aging Alzheimer Disease Centers, 2005-2010.’, Journal of neuropathology and experimental neurology, 71(4), pp. 266–273. doi: 10.1097/NEN.0b013e31824b211b.
3. Estévez-González, A. et al. (2003) ‘Rey verbal learning test is a useful tool for differential diagnosis in the preclinical phase of Alzheimer’s disease: comparison with mild cognitive impairment and normal aging.’, International journal of geriatric psychiatry, 18(11), pp. 1021–1028. doi: 10.1002/gps.1010.
4. Henschel, L. et al. (2020) ‘FastSurfer - A fast and accurate deep learning based neuroimaging pipeline’, NeuroImage, 219, p. 117012. doi: https://doi.org/10.1016/j.neuroimage.2020.117012.
5. Petersen, R. C. et al. (2010) ‘Alzheimer’s Disease Neuroimaging Initiative (ADNI): clinical characterization.’, Neurology, 74(3), pp. 201–209. doi: 10.1212/WNL.0b013e3181cb3e25.
6. Rabinovici, G. D. et al. (2019) ‘Association of Amyloid Positron Emission Tomography With Subsequent Change in Clinical Management Among Medicare Beneficiaries With Mild Cognitive Impairment or Dementia’, JAMA, 321(13), pp. 1286–1294. doi: 10.1001/jama.2019.2000.
7. Young, A. L. et al. (2018) ‘Uncovering the heterogeneity and temporal complexity of neurodegenerative diseases with Subtype and Stage Inference’, Nature Communications, 9(1), p. 4273. doi: 10.1038/s41467-018-05892-0.