Abstract
Abstract only
Background:
Carotid endarterectomy (CEA) is associated with improved overall clinical outcomes in patients with acute ischemic stroke (AIS). However, studies on rates and factors associated with readmission following CEA for AIS are scarce. In this study, we used machine learning (ML) methods to identify the factors associated with readmission using a large-scale national database.
Methods:
We used National Surgical Quality Improvement Program (NSQIP) registry (2014-2017) and included patients 18 years or older, who underwent CEA for AIS. AIS and CEA were identified using ICD-9 and ICD-10 diagnosis and CPT procedure codes, respectively. We used Naïve Bayes, Boosted Decision Trees, and Bootstrapped Random Forest classification techniques to explore the predictors of 30-day readmission using demographics, past medical history, and preoperative variables.
Results:
There were a total of 22,373 AIS patients who underwent CEA. Mean (SD) age of the patients was 70.7 (9.4) years, and 61% were men. Majority were non-Hispanic White (80%), followed by non-Hispanic Black (4.6%). During the study period, 1 in 15 AIS patients who underwent CEA experienced 30-day readmission. Bootstrapped Random Forest classification performed best and Naïve Bayes worst with an AUROC of 92% and 59% respectively. The top 5 predictors of 30-day readmission after CEA were Hematocrit, BUN, Creatinine, WBC count, and Platelet count, all collected pre-operatively.
Conclusion:
Our study showed that ML techniques could accurately predict 30-day readmission using pre-operative risk factors. This ML model could be incorporated in EMR as a potential clinical decision support system. Implementing this system could help in early identification of patients who are at high risk for readmission following CEA. This could help physicians to plan and intervene effectively and prevent short-term readmissions; thereby improving quality of care and saving healthcare costs.
