| Abstract: |
Background: Cardiovascular diseases (CVD) affect approximately 80 million Americans and the heart disease is the leading cause of death in the U.S.. Researchers have been working on ways to improve disease management and treatment strategies through developing better risk prediction tools. However, one major challenge is that even when patients with CVD have similar clinical characteristics, many may experience adverse events (e.g., death) sooner than others. Electronic health records (EHR) that contain a wealth of patient data provides an opportunity to apply deep machine learning and artificial intelligence to identify variables that predict patient outcomes.
Purpose: Using artificial intelligence to identify important features that predict 5-year mortality among 40 variables from 1000 patients selected from the Emory Cardiovascular Biobank (Atlanta, Georgia, USA). The long-term goal is to develop algorithms that ingest patient data and provide predicted patient prognosis in real time to facilitate effective disease risk management, ultimately leading to better clinical care.
Results: Out of these 1000 patients, 145 experienced death in 5 years of follow-up. The features we considered included but not limited to demographics, vitals, comorbidity, lab tests, kidney function, and 7 inflammatory biomarkers. IBM’s Watson AutoAI CGB classifier as the core algorithm against the target variable was used. Six features were identified as important predictors of death: B-type natriuretic peptide (BNP), soluble urokinase-type plasminogen activator receptor (suPAR), sex, newfeature_1, history of heart failure, white blood cell.
Discussion: BNP is known a significant biomarker that associated with severity of the disease and has been widely used for diuretic therapy treating for heart failure patients. However, BNP value can also be elevated in patients with chronic renal disease. The plasma soluble urokinase-type plasminogen activator receptor (suPAR) is a biomarker for renal focal segmental glomerulosclerosis (FSGS) and has yet to be decided for its significance in predicting its role in severity of heart disease in a mix of renal and heart disease. Our findings shed lights on its combined value on predicting mortality in the near future and could be adopted in clinical practice in the future. |