Affiliations: [a] Institut Universitaire en Santé Mentale de Québec, QC, Canada | [b] Radiology Department, Faculty of Medicine, Université Laval, QC, Canada
Correspondence:
[*]
Correspondence to: Simon Duchesne, Institut Universitaire en Santé Mentale de Québec, 2601 de la Canadiére, Room F-4435, G1J 2G3, QC, Canada. Tel.: (418) 663 5741; ext. 4777; Fax: (418) 663 5971; E-mail: [email protected].
Note: [1] Data used in preparation of this article were obtained from the Alzheimer’s Disease Neuroimaging Initiative (ADNI) database (http://adni.loni.ucla.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.ucla.edu/wp-content/uploads/how_to_apply/ADNI_Acknowledgement_List.pdf
Abstract: Biomarkers, both biological and imaging, are indicators of specific changes that characterize Alzheimer's disease (AD) progression in vivo. Knowing the precise relationship between biomarkers and disease severity would allow for accurate disease staging and possible forecasting of decline. Jack et al. suggested as an initial hypothesis that this relationship be sigmoidal; the objective of this article is to determine, using large-scale population data from ADNI, the precise shape of this association. We considered six different models (linear; quadratic; robust quadratic; local quadratic regression; penalized B-spline; and sigmoid) and used the Akaike Information Criterion to gauge how well these models compare in conforming to the data. We included 576 subjects (229 controls, 193 AD, and 154 mild cognitive impairment subjects who converted to AD) from the ADNI study, for whom baseline data on cerebrospinal fluid amyloid-β (Aβ)42, phosphorylated tau (p-tau), and total-tau (t-tau), hippocampal volumes, and FDG-PET were available. Analysis of this cross-sectional dataset showed that a local quadratic regression model was 42% more likely than a sigmoid to be the best model for Aβ42. This ratio augments to 22% and 73% for Penalized B-Spline in the case of p-tau and t-tau, respectively; to 3500% for the linear model for FDG-PET; and to 6700% for the Penalized B-Spline for hippocampal volumes. Preliminary, cross-sectional evidence therefore indicates that the shape of the association with disease severity is non-linear and differs between biomarkers.
