Detecting multimorbidity patterns in Alzheimer's disease using unsupervised machine learning: A nationwide emergency...

Area of Science:

• Geriatric Medicine
• Neurology
• Health Services Research

Background:

• Alzheimer's disease (AD) patients present complex comorbidities to emergency departments (EDs).
• Understanding temporal shifts in multimorbidity patterns among AD patients is crucial for effective care.
• Existing research lacks insight into evolving comorbidity phenotypes in AD-related ED visits.

Purpose of the Study:

• To identify temporal shifts in comorbidity-based phenotypes among older adults with AD visiting EDs.
• To analyze changes in multimorbidity patterns between 2007 and 2022.
• To apply unsupervised clustering to understand evolving AD patient profiles.

Main Methods:

• Analysis of Nationwide Emergency Department Sample (NEDS) data for adults aged ≥60 with AD diagnoses (2007, 2012, 2017, 2022).
• Mapping of ICD-9/10 codes to 30 clinically relevant comorbidities annually.
• Application of k-means clustering to identify comorbidity subgroups and heatmaps for temporal composition analysis.

Main Results:

• Four stable but evolving comorbidity clusters were identified across the study period.
• Early cohorts (2007-2012) showed dominance of cardiovascular and respiratory conditions.
• Recent cohorts (2017-2022) exhibited increased prevalence of nonspecific, frailty-related symptoms (e.g., fatigue, GERD).
• The proportion of ED visits documenting AD declined, possibly due to coding shifts or diagnostic overshadowing.

Conclusions:

• The comorbidity landscape for AD-related ED visits is evolving, shifting towards vaguer syndromes and complex multimorbidity.
• Findings highlight the need for dementia-aware triage and dynamic phenotyping in emergency care.
• Improved strategies are essential for managing cognitively impaired older adults in ED settings.