Examining heterogeneity in dementia using data-driven unsupervised clustering of cognitive profiles

Area of Science:

• Neurology
• Gerontology
• Data Science

Background:

• Dementia presents as a heterogeneous condition with varying symptoms and progression rates.
• Existing dementia subtyping research often focuses narrowly on Alzheimer's disease, uses limited baseline data, and relies on expensive imaging biomarkers.
• There is a need for data-driven approaches to understand dementia heterogeneity across the spectrum of cognitive decline.

Purpose of the Study:

• To identify distinct dementia subtypes using a data-driven, unsupervised clustering approach.
• To overcome limitations of previous studies by analyzing longitudinal cognitive data and avoiding reliance on expensive imaging.
• To explore the relationship between identified dementia subtypes and disease progression over time.

Main Methods:

• Employed a data-driven unsupervised clustering algorithm (SillyPutty) combined with hierarchical clustering.
• Utilized longitudinal cognitive assessment scores from a real-world clinical dementia cohort.
• Analyzed temporal relationships between subtypes and disease progression, moving beyond cross-sectional data.

Main Results:

• Identified distinct dementia subtypes within the clinical cohort.
• Subtypes in the very mild or mild dementia stages exhibited significant heterogeneity in cognitive profiles.
• These early-stage subtypes also showed varied risks for disease progression.

Conclusions:

• Data-driven subtyping using longitudinal cognitive data can reveal significant heterogeneity in dementia, particularly in early stages.
• This approach offers a more comprehensive understanding of dementia heterogeneity compared to studies limited to Alzheimer's disease or cross-sectional data.
• Understanding subtype-specific progression is crucial for personalized dementia care and future research.