Abstract: SA-PO460
Diagnosis of Renal and Cardiovascular Injuries in Type 1, Type 2, and Prediabetic Patients Using an Ensemble of Machine Learning and Deep Learning Methods
Session Information
- Diabetic Kidney Disease: Clinical - II
November 04, 2023 | Location: Exhibit Hall, Pennsylvania Convention Center
Abstract Time: 10:00 AM - 12:00 PM
Category: Diabetic Kidney Disease
- 702 Diabetic Kidney Disease: Clinical
Authors
- Jazra, Diala, American University of Beirut, Beirut, Lebanon
- Rakka, Mariam, University of California Irvine, Irvine, United States
- Harb, Frederic, University of Balamand, Balamand, North , Lebanon
- Bou Jaoude, Celina Robert, American University of Beirut, Beirut, Lebanon
- Aboujaoude, Adrian Walid, American University of Beirut, Beirut, Lebanon
- Choucair, Mahmoud, American University of Beirut, Beirut, Lebanon
- Eid, Assaad Antoine, University of Michigan, Ann Arbor, United States
- Kanj, Rouwaida, University of Illinois Urbana-Champaign, Urbana, United States
Background
Diabetes is a growing global pandemic with serious complications such as renal and cardiovascular injuries. Diabetic nephropathy, characterized by decreased glomerular filtration rate (GFR) and albuminuria, is a microvascular complication of diabetes. Cardiovascular disease, which can result in injuries to coronary, cerebrovascular, and peripheral arteries, is a macrovascular complication. Studies have shown that renal and cardiovascular diseases often co-occur in individuals with diabetes, but there is currently no machine learning algorithm that was used to diagnose the existence of both complications simultaneously.
Methods
For this study, a dataset of 273 diabetic patients is obtained from the American University of Beirut Medical Center (AUBMC). Chronic kidney disease is defined as a GFR lower than 60 ml/min and/or an albumin to creatinine ratio greater than 30. Cardiovascular disease outcome is measured by the presence of cardiovascular events such as myocardial infarction, heart failure, arrhythmias and valvulopathies. To balance the class distributions, the synthetic minority oversampling technique (SMOT) is used. The Random Forest and Artificial Neural Network models are employed to analyze the data and evaluate the relevance of each feature using feature importance methods. The grid search method is used to fine-tune the hyperparameters of the models.
Results
The Random Forest model achieved an accuracy of 91%. The Extreme gradient Boosting (XGBoost) feature importance method revealed that the most relevant features, in decreasing order of importance, were urine albumin mg/l, serum creatinine, age, LDL, total cholesterol, heart rate, phosphate, systolic blood pressure, carbon dioxide, and potassium. The Neural Network had 75% accuracy with Grid Search. The lower accuracy could be due to the small sample size (273 patients) for deep learning.
Conclusion
The results of the study show that machine learning can be useful in diagnosing renal-cardiovascular complications in diabetic patients.