Premium
Towards harmonizing subtyping methods for PET and MRI studies of Alzheimer’s disease
Author(s) -
Mohanty Rosaleena,
Mårtensson Gustav,
Poulakis Konstantinos,
RodriguezVieitez Elena,
Grothe Michel,
Nordberg Agneta K,
Ferreira Daniel,
Westman Eric
Publication year - 2020
Publication title -
alzheimer's and dementia
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 6.713
H-Index - 118
eISSN - 1552-5279
pISSN - 1552-5260
DOI - 10.1002/alz.042807
Subject(s) - subtyping , alzheimer's disease neuroimaging initiative , dementia , positron emission tomography , concordance , neuroimaging , psychology , oncology , medicine , atrophy , disease , pathology , neuroscience , computer science , programming language
Background Identification of biological subtypes in Alzheimer’s disease (AD) has become a growing field for understanding the heterogeneity of the disease. Subtypes, originally identified by postmortem markers (Murray et al., 2011), have been translated to utilize in vivo biomarkers such as those based on structural magnetic resonance imaging (sMRI) and positron emission tomography (PET). While there exists methodological variability across studies, comparable characteristics are reported at group‐level. However, the question that remains to be explored and the motivation for this study is: do group‐level similarities translate to individual‐level agreement across subtyping methods? Method We compared five subtyping methods reported in previous studies in a two‐part approach: (a) we validated the methods in 89 AD dementia patients (reference group: 70 healthy individuals; HC) from ADNI using sMRI only; (b) we extended and applied the methods to 30 AD dementia and 53 prodromal AD patients (reference group: 200 HC) from ADNI with concurrent sMRI and tau PET. All AD patients were Aβ positive while all HC were Aβ negative. Subtyping methods included four hypothesis‐driven (Byun et al, 2015; Ferreira et al., 2017; Murray et al., 2011; Risacher et al., 2017) and a data‐driven (Poulakis et al., 2018) algorithms and were implemented as outlined in each original study. Up to four subtypes were identified, namely: typical AD (TAD), limbic predominant (LP), hippocampal sparing (HS) and minimal atrophy (MA). Group‐level and individual‐level comparisons across methods were performed. Result In cohort (a), similar frequencies of subtypes were observed between reproduced and published values, thus, establishing proof‐of‐concept. However, comparison of individual‐level subtypes showed disagreement across methods. In the extended cohort (b), large disagreements were observed at the individual level. This is reflected by the low values of Cohen’s kappa, measuring the agreement between methods (Figure 1). Conclusion Although characteristics of subtypes are comparable at group‐level, there exist large disparities in individual‐level agreement across subtyping methods. Study of subtypes is growing rapidly by incorporation of multiple biomarkers and application to AD continuum. Thus, there is an urgent need for consensus and harmonization across methods to push the field forward.