Readmission Prediction for Patients with Ischemic Stroke after Discharge

The rate of patient readmissions within a short period after discharge is a significant indicator for the healthcare quality of a hospital. Readmissions may result in an increased cost of a healthcare organization. Design of a model for predicting readmission would benefit on solving the above issues. This study aims to develop a clinical decision support system (CDSS) for predicting readmission of the patient with ischemic stroke (IS) after discharge. The IGS method, which integrates genetic algorithm (GA) and support vector machine (SVM), accompanied with three objective functions, was adopted to develop the CDSS. The data, retrieved from the National Health Insurance Research Database (NHIRD), including 4351 patients (462 with readmission and 3889 without readmission) diagnosed with IS (ICD-9-CM Code 433–435), aged 20 years old and older, treated within 30 days of hospital admission and then discharged to outpatient treatment between Jan. 2007 and Dec. 2009, were used for designing the predictive models. The statistical analysis of demographics (gender and age) and other candidate variables (28) between patients with and without readmission is presented. Twelve of these 30 variables are significantly different (p < 0.05). CDSS models designed using three objective functions achieved predictive performances of accuracy, sensitivity, specificity, and area under ROC curve (AUC) equaling 65.9-66.77%, 58.22-66.66%, 66.88-73.59%, and 0.6773-0.7183, respectively. Future work will focus on improving the predictive performance by including more effective risk factors and comorbidities, as well as integrating GA with more effective AI methods such as deep neural network to increase the predictive performance.