Date Presented 03/26/20
Using big data visual analytics and a total of 86,887 readmitted Medicare beneficiaries living with stroke, a number of biclusters consisting of patient subgroups and their co-occurring multiple chronic comorbidities were determined. This technique informed a multimorbidity self-management program to support patients managing their health after stroke.
Primary Author and Speaker: Ickpyo Hong
Additional Authors and Speakers: Kimberly Hreha, Monique Pappadis, Hashem Shaltoni, Yu-Li Lin, Emmanuel Santillana Fayett, Julianna Dean, Chih-Ying Li, Annalisa Na, Suresh Bhavnani, Timothy Reistetter
PURPOSE: This study aimed to identify the number and boundaries of biclusters consisting of patient subgroups and their multimorbidities using visual analytics. This process intends to provide clinicians (e.g., occupational therapists) and researchers with data-driven patient information to guide the development and implementation of a Multiple Chronic Condition Self-Management program for patients with stroke.
DESIGN: A retrospective secondary data analysis using 100% Medicare Part A claims of patients with stroke.
METHOD: We retrieved Medicare claim records of the elderly hospitalized with stroke in 2013-2014. We identified cases (i.e., stroke patients readmitted within 90 days of discharge) and controls (i.e., not readmitted within 90 days) matched by age, gender, race/ethnicity, and stroke type. This resulted in 43,949 and 42,938 matched pairs in the training (2013) and replication (2014) datasets, respectively. We examined 96 comorbidities extracted from the Centers for Medicare and Medicaid Services Condition Categories (CCs) and identified 72 comorbidities significantly associated with those who were readmitted in the training and replication dataset. We performed a bipartite network analysis utilizing these 72 comorbidities to identify and visualize subgroups of patients, based on co-occurrence patterns of comorbidities.
RESULTS: The analysis revealed five biclusters of patient subgroups and frequently co-occurring multicomorbidities, which had significant clusteredness (Modularity=0.17, z=6.19, p<.001), and significant replication (Rand Index=0.73, z=6.31, p<.001). The five stroke subgroups included: 1) Obesity/Diabetes, 2) Immunity/Infection, Neurological Diseases, Psychiatric/Depression, 3) Brain/Vascular, Neurodegenerative Diseases, 4) Heart, Lungs/Pulmonary Diseases, and 5) Cancer, Psychiatric Disorders.
CONCLUSION: Using 100% Medicare claim data, readmitted patients with stroke had five distinct subgroups of comorbidities. These subgroups of stroke patients informed by the resultant comorbid conditions allow clinicians to tailor self-management programs to improve patient and caregivers knowledge and ability to remain actively engaged in the community. Future studies are necessary to determine specific intervention strategies limiting or prohibiting 90-day readmissions for patients with stroke.
IMPACT STATEMENT: Identifying subgroups of patients with stroke by their multimorbidities informs a Chronic Condition Self-Management program targeted for stroke subgroups at-risk for readmission. Overall this is an important topic because multiple comorbidities are common and yet the science is still being developed on how to best support those seeking interventions.
References
Bhavnani, S. K., Chen, T., Ayyaswamy, A., Visweswaran, S., Bellala, G., & Rohit, D. (2017). Enabling comprehension of patient subgroups and characteristics in large bipartite networks: Implications for precision medicine. AMIA Joint Summits on Translational Science proceedings, 2017, 21–29.2017, 21-29.
Newman, M. (2010). Networks: An introduction: Oxford university press.
