Data-driven classification of patients with primary progressive aphasia

Area of Science:

• Neurology
• Linguistics
• Cognitive Science

Background:

• Current diagnostic criteria classify primary progressive aphasia (PPA) into semantic (svPPA), nonfluent (nfvPPA), and logopenic (lvPPA) variants.
• The adequacy of this tripartite classification scheme for PPA is under debate within the scientific community.

Purpose of the Study:

• To investigate the validity of current PPA diagnostic criteria using a data-driven approach.
• To explore alternative groupings of PPA patients based on objective language and cognitive data.

Main Methods:

• Applied k-means clustering to language, semantic, and non-linguistic cognitive data from 43 PPA patients.
• Identified the optimal number of clusters based on data similarity.
• Analyzed the characteristics of each identified patient cluster.

Main Results:

• The k-means algorithm optimally identified three distinct patient groups.
• One group showed selective semantic impairment, largely corresponding to svPPA diagnoses.
• A second group exhibited speech production, repetition, and syntactic deficits, comprising mostly nfvPPA patients and some lvPPA patients.
• A third group presented with more severe, widespread language (speech, repetition, syntax) and cognitive impairments.
• The inclusion or exclusion of non-linguistic cognitive scores minimally impacted the observed clustering patterns.

Conclusions:

• The findings suggest that PPA classification, particularly for non-semantic variants, may be better understood as a spectrum of overall language and cognitive impairment severity.
• The current diagnostic categories might not fully capture the underlying heterogeneity of PPA.
• Further research is warranted to refine PPA diagnostic criteria and understand the progression of language and cognitive decline.