Data-driven subtyping of Alzheimer's emergency presentations using unsupervised machine learning.

BACKGROUND

Older adults with Alzheimer's disease (AD) frequently present to emergency departments (EDs) with multiple coexisting conditions. However, age-specific patterns of multimorbidity in AD-related ED encounters remain incompletely characterized.

OBJECTIVE

To identify and describe age-specific multimorbidity subtypes among AD-associated ED encounters using unsupervised machine learning and heatmap-based diagnostic profiling.

METHODS

We conducted a cross-sectional analysis of the 2022 Nationwide Emergency Department Sample. AD-associated encounters were identified by an ICD-10-CM G30.x diagnosis code in any diagnosis field. The cohort included 125,461 ED encounters and was stratified into four age groups: 60-64, 65-74, 75-84, and ≥ 85 years. For each age group, the 30 most frequent co-occurring diagnoses were converted into binary indicators. KMeans clustering with eight clusters per age group was used for heatmap-based subtyping, while Uniform Manifold Approximation and Projection and Hierarchical Density-Based Spatial Clustering of Applications with Noise were used to visualize diagnostic structure.

RESULTS

Distinct multimorbidity profiles were identified across all age groups. Among adults aged 60-74 years, clusters commonly included psychiatric and metabolic conditions, such as depression, anxiety, diabetes, chronic kidney disease, substance-use diagnoses, and dehydration. Among adults aged ≥ 75 years, clusters more frequently included cardiorenal disease, urinary tract infection, metabolic encephalopathy, respiratory failure, do-not-resuscitate status, and palliative care. The cohort had a mean age of 81.7 ± 7.2 years, and 62.6% of encounters involved women.

CONCLUSION

Patterns of coexisting conditions in AD-related ED visits differed substantially by age. Adults aged 60-74 years more often showed psychiatric and metabolic combinations, whereas those aged ≥ 75 years more often showed infection, cardiorenal disease, respiratory failure, and end-of-life care markers. Recognizing these age-related patterns may help clinicians anticipate common care needs when evaluating patients with AD in the ED.