Clinical StudyStructural neuroimaging correlates of cognitive status in older adults: A person-oriented approach
Introduction
Analytical approaches to human inquiry can be classified as either variable or person oriented. The aim of variable-oriented research is to identify relationships between variables, such as the relationship between a clinical variable (for example, diagnostic status) and a neuroimaging marker (for example, grey matter volume). In this approach, the individual is viewed as a “data carrier”, and the interest is in the inter-relations between the variables at the aggregate level [1]. This approach is typical for the analysis of neuroimaging data, where techniques such as group comparisons, correlations, and the general linear model predominate. In contrast, the aim of person-oriented research is to identify patterns or groupings of individuals under the assumption that the study sample might be drawn from multiple populations. Statistical techniques such as cluster analysis are typical for this approach to research.
Person-oriented approaches are yet to be employed extensively in neuroimaging research. One reason for this involves the high-dimensional nature of neuroimaging data. Cortical thickness data, for example, typically contains >100 datapoints for each participant [2]. While there are now techniques readily available to analyse such data in a person-oriented framework, they have yet to permeate into the neuroimaging community. The aim of this paper is to introduce the concept of person-oriented research in neuroimaging, and demonstrate its application to high-dimensional structural imaging data. Readily available software packages are used [3], [4], which make the uptake of this approach relatively straightforward.
Von Eye and Bogat [5] describe three criteria for person-oriented research. First, an underlying assumption of the analysis is that the sample was drawn from more than one population. Statistical techniques are used to identify different groups within the data that are characterised by different patterns of variables. The second criterion involves the validation of the different groupings. To avoid circularity, it is essential to use variables that were not used to form the groupings. In the case of neuroimaging data, phenotypic data are often suitable candidates. Finally, the groupings must be interpreted in the context of a theory. In the case of neuroimaging in neurodegenerative disease, the known pathological substrate provides this context.
This study used data from a large Alzheimer’s disease (AD) study containing participants with varying degrees of cognitive function. The advantage of using this population is that there are likely to be structural changes in some participants, driving cluster formation [6]. The known pathological substrate of the disease provides a framework for interpreting the findings, and the associated measures of cognitive impairment provide a means of validating the cluster solution. As described in detail below, this study used a stepwise variable selection approach to reduce the number of variables used for clustering. Cluster analysis was then performed using a validated model-based approach [7]. These approaches overcome the challenges of high-dimensional data and facilitate the objective selection of the number of clusters to extract [8]. This approach has been successfully applied in the analysis of high-dimensional genetic data [9]. Given the known distribution of pathology in AD, it was predicted that clustering information will be encoded in the mesial temporal, parietal, and posterior cingulate structures [10], [11], [12]. Groups with greater atrophy were expected to have lower scores on cognitive measures, and would be more likely to carry a clinical diagnosis of cognitive impairment.
Section snippets
Participants
The data used in this study were obtained from the Alzheimer’s Disease Neuroimaging Initiative (ADNI) database. The ADNI was launched in 2003 by the USA National Institute on Aging, the National Institute of Biomedical Imaging and Bioengineering, the Food and Drug Administration, private pharmaceutical companies and non-profit organisations, as a US$60 million, 5 year public–private partnership. The primary goal of ADNI has been to test whether serial MRI, positron emission tomography, other
Cluster solution
The variable selection procedure identified two variables that contributed significantly to cluster membership. These were the left and right entorhinal cortices. Model-based clustering was then performed on these two variables. As shown in Fig. 1, the best fitting clustering solution contained two groups (BIC = −745.18).
The cluster solution is shown in Fig. 2 with reference to left and right entorhinal cortex thickness. Participants in the first cluster, labelled the “typical” subgroup, had
Discussion
This study demonstrated the application of model-based clustering with variable selection in a high-dimensional structural neuroimaging dataset. In a large group of aged participants, two regions were found to discriminate between subgroups, namely the left and right entorhinal cortices. Model-based clustering performed on these variables revealed two subgroups. The first, named the “typical” group, was relatively homogeneous with entorhinal cortical volumes in the upper-range of sample. This
Conclusions
This study demonstrates the successful application of person-oriented methods to structural neuroimaging data. The combination of variable selection and model-based clustering identified two clusters in a large sample of participants from an AD study. The two regions that survived variable selection were consistent with the known neuropathology of AD, and the clusters differed on key phenotypic variables that were not used to form the groups. Taken together, model-based clustering of structural
Conflicts of Interest/Disclosures
Data used in preparation of this article were obtained from the Alzheimer’s Disease Neuroimaging Initiative (ADNI) database (adni.loni.usc.edu). As such, the investigators within the ADNI contributed to the design and implementation of ADNI and/or provided data but did not participate in analysis or writing of this report. A complete listing of ADNI investigators can be found at: http://adni.loni.usc.edu/wp-cont. The author declares that they have no financial or other conflicts of interest in
Acknowledgements
Data collection and sharing for this project was funded by the Alzheimer’s Disease Neuroimaging Initiative (ADNI; National Institutes of Health Grant U01 AG024904) and Department of Defense ADNI (Department of Defense award number W81XWH-12-2-0012). ADNI is funded by the USA National Institute on Aging, the National Institute of Biomedical Imaging and Bioengineering, and through generous contributions from the following: AbbVie, Alzheimer’s Association; Alzheimer’s Drug Discovery Foundation;
References (37)
FreeSurfer
Neuroimage
(2012)- et al.
An automated labeling system for subdividing the human cerebral cortex on MRI scans into gyral based regions of interest
Neuroimage
(2006) - et al.
“Mini-mental state”: a practical method for grading the cognitive state of patients for the clinician
J Psychiatr Res
(1975) - et al.
Variable selection in model-based clustering: a general variable role modeling
Comput Stat Data Anal
(2009) - et al.
Volumes of the entorhinal and perirhinal cortices in Alzheimer’s disease
Neurobiol Aging
(1998) - et al.
Volume reduction of the entorhinal cortex in subjective memory impairment
Neurobiol Aging
(2006) - et al.
MRI correlates of general intelligence in neurotypical adults
J Clin Neurosci.
(2016) - et al.
The person-oriented versus the variable-oriented approach: are they complementary, opposites, or exploring different worlds?
Merrill-Palmer Quarterly
(2006) - et al.
Measuring the thickness of the human cerebral cortex from magnetic resonance images
Proc Natl Acad Sci USA
(2000) - Scrucca L, Raftery A. Clustvarsel: A package implementing variable selection for model-based clustering in R. 2014....
Person-Oriented and variable-oriented research: concepts, results, and development
Merrill-Palmer Quarterly
Neuropathologic alterations in mild cognitive impairment: a review
J Alzheimer’s Dis
Model-based Gaussian and non-Gaussian clustering
Biometrics
Model-based clustering, discriminant analysis, and density estimation
J Am Statist Assoc
Model-based clustering and data transformation for gene expression data
Bioinformatics
The cortical signature of prodromal AD Regional thinning predicts mild AD dementia
Neurology
The cortical signature of Alzheimer’s disease: regionally specific cortical thinning relates to symptom severity in very mild to mild AD dementia and is detectable in asymptomatic amyloid-positive individuals
Cereb Cortex
Cited by (3)
Data-Driven Analyses of Longitudinal Hippocampal Imaging Trajectories: Discrimination and Biomarker Prediction of Change Classes
2022, Journal of Alzheimer's DiseaseIdentifying and evaluating clinical subtypes of Alzheimer’s disease in care electronic health records using unsupervised machine learning
2021, BMC Medical Informatics and Decision Making
- 1
Data used in preparation of this article were obtained from the Alzheimer’s Disease Neuroimaging Initiative (ADNI) database (adni.loni.usc.edu). As such, the investigators within the ADNI contributed to the design and implementation of ADNI and/or provided data but did not participate in analysis or writing of this report. A complete listing of ADNI investigators can be found at: http://adni.loni.usc.edu/wp-cont.