Evaluating the correspondence of neuropsychological profiles with clinical diagnostic criteria for frontotemporal lobar degeneration and Alzheimer's disease: insights from k-means clustering.

INTRODUCTION: Diagnosis of frontotemporal lobar degeneration (FTLD) is complicated by high heterogeneity within diagnostic groups and overlap between them. Using an unsupervised machine learning technique this study aimed to increase understanding of neuropsychological profiles in the spectrum of FTLD syndromes and to assess the consistency of these neuropsychological profiles with diagnostic criteria. METHODS: Data from the German FTLD Consortium (FTLD-c) including behavioral variant of frontotemporal dementia (bvFTD), primary progressive aphasia (PPA) syndromes, i.e., semantic variant (svPPA), and non-fluent variant (nfvPPA), healthy controls, and amnestic Alzheimer's disease (AD), and logopenic variant (lvPPA) was used. K-means clustering was applied on neuropsychological and behavioral data. Analysis was repeated for varying cluster numbers (k = 2-9). Resulting clusters were inspected on the correspondence with current diagnostic groups and neuropsychological profiles of the clusters were explored. RESULTS: The results indicated that, based on neuropsychological profiles, particularly homogeneous clusters emerged for healthy controls, svPPA and bvFTD. In contrast, amnestic AD did not separate well from other diagnostic groups, and nfvPPA and lvPPA clustered together. Inspecting the neuropsychological profiles of the clusters showed that segregation of bvFTD from other patient groups was most supported by companion-rated questionnaires focused on behavior. Patients with svPPA showed impairment in both language and verbal memory tasks. LvPPA and nfvPPA showed impairment in processing speed as well as short term and working memory. DISCUSSION: Using clustering this study provides insight into data-driven grouping of participants with FTLD based on neuropsychological deep-phenotyping. Whereas our approach confirmed svPPA and bvFTD to have comparatively homogeneous neuropsychological profiles, results suggest that the non-fluent PPA syndromes-lvPPA and nfvPPA-are difficult to segregate clinically, at least with the tests applied in our cohort. Our findings may suggest the need for further specification of clinical diagnostic criteria for these diseases.