Abstract
BACKGROUND: With the increasing reliance on big data analytics in poststroke dysphagia research (which drives related health care decisions), accurate classification of large, complex data sets is crucial. Although medical coding is commonly used to define patient cohorts, there is no administrative-level instrument for classifying poststroke dysphagia severity. Thus, we aimed to develop a novel classification framework for poststroke dysphagia severity. METHOD: We used data from 445 patients diagnosed with acute ischemic stroke and dysphagia from the 2017 Medicare 5% Limited Data Set to develop our poststroke dysphagia severity classification framework. For our exploratory analysis, we used unsupervised k-means clustering to categorize patients based on dysphagia indicators constructed from International Classification of Diseases, Tenth Revision, codes. The resultant clustering solution was applied to three random 60% samples of the data set to verify the stability of the dysphagia severity clusters. RESULTS: Cluster analysis resulted in a three-cluster algorithm characterizing mild, moderate, and severe dysphagia severity. Inspection of the clusters revealed that dysphagia severity categories were not analogous to stroke severity categories. CONCLUSIONS: We developed a novel framework to classify poststroke dysphagia severity using administrative data. We also found discordance between stroke severity and dysphagia severity, which has implications for classification methodologies in administrative poststroke dysphagia research. Future studies are needed to validate this classification framework.