Machine Learning Clustering Algorithm Analysis of Dementia and Acute Ischemic Stroke in 53,000 Patients
Aryan Malhotra1, Adam Kiss1, Hao Yu1, Rafay Khan1, Mateusz Faltyn2, Fawaz Al-Mufti3
1New York Medical College, 2Trilemma Capital, 3Westchester Medical Center
Objective:
We aim to identify patient groups with unique comorbidity profiles in the 2015-2019 National Inpatient Sample (NIS), and assess the contributions of these profiles to mortality, discharge disposition, and length-of-stay in the sample of patients with dementia and acute ischemic stroke (AIS).
Background:
Not Applicable
Design/Methods:
The 2015-2019 NIS was assessed using ICD-10 PCS/CM codes to identify patients with dementia and concomitant AIS. A machine learning clustering analysis evaluated the population based on 50 comorbidities, complications and clinical covariates. Optimal clustering was determined using the Davies-Bouldin Index (DBI) and Calinski-Harabasz Index (CHI). Between-cluster multivariate regression analysis was performed to assess risk of mortality and non-routine discharge. Kruskal-Wallis H-Testing was performed to assess variance in length-of-stay between clusters. Statistical analysis was performed using Python.
Results:
53,411 patients were clustered into 4 groups ranging from 1096 to 41,557 patients. Mortality ranged from 6.01% in Cluster 1 to 19.80% in Cluster 4. Clusters 2-4 each displayed significantly higher rates of mortality [OR Range 1.32-3.86, p<0.001] relative to Cluster 1. These results are visualized in a heatmap. Cluster 1 had the greatest prevalence of dyslipidemia amongst groups. Cluster 4 had the greatest prevalence of sepsis, aspiration pneumonia, and arrhythmia. Risk of non-routine discharge was lowest in Cluster 2 [OR 1.21, p<0.001] and highest in Cluster 4 [OR 4.77, p<0.001] when compared to Cluster 1. Kruskal-Wallis H-testing and post-hoc pairwise comparison of length-of-stay variances showed significant (p<0.001) differences between all clusters with the greatest test statistics occurring when comparing group 4 to all others.
Conclusions:
This analysis distinguishes four groups with unique comorbidity profiles. This clustering approach enables the assessment of different comorbidity presentations of patients with varying mortality, discharge disposition, and length-of-stay, offering a foundation for further analysis to inform clinical decision-making.
Disclaimer: Abstracts were not reviewed by Neurology® and do not reflect the views of Neurology® editors or staff.