Abstract: SA-PO368
Machine-Learning Approaches for Major Adverse Cardiovascular Events (MACE) Prediction in Patients on Hemodialysis
Session Information
- Hypertension, CVD, and the Kidneys: Clinical Research
October 26, 2024 | Location: Exhibit Hall, Convention Center
Abstract Time: 10:00 AM - 12:00 PM
Category: Hypertension and CVD
- 1602 Hypertension and CVD: Clinical
Authors
- Chen, Cheng-Hsu, Division of Nephrology, Taichung Veterans General Hospital, Taichung City, Taiwan
- Hou, Shun Fang, Division of Nephrology, Taichung Veterans General Hospital, Taichung City, Taiwan
- Wang, Min-Shian, Smart Healthcare Committee, Taichung Veterans General Hospital, Taichung, Taiwan
Background
Major Adverse Cardiovascular Events (MACE) are common complications of hemodialysis (HD) patients that include myocardial infarction (MI), stroke, cardiac arrhythmia and heart failure (HF). The objective of the current study was to predict MACE among our HD patients.
Methods
HD patients above 18 years old were recruited 29356 HD sessions for the study between 2014 to 2023 from our hospital database of the TSN-KiDiT (kidney, dialysis, and transplantation integrated software), which is integrated operation management system and quality control for Taiwan Society of Nephrology. Different Machine learning algorithms: including RandomForest (RF), XGBoost, logistic regression (LR), and KNN( K Nearest Neighbor) were employed. Clinical attributes, electrolytes, dialysis adequacy and blood flow (BF), cardiothoracic ratio (CT ratio) and biomarkers were explored in predicting MACE. The feature importance was determined using mean decrease accuracy.
Results
Overall, 28788 HD sessions were included in the analyses, there were 3791 events of MACE within 12-month. The XGBoost Model demonstrated a prediction accuracy of 88.92% with the area under the receiver operating characteristic curve (AUROC) 94.42%, which is higher as compared to the RF 84.54% [AUROC 94.95%], the LR model 65.23% [AUROC 65.23%], however, the KNN has the best accuracy 92.45% [AUROC 93.28%] with less sensitivity 59.47%, respectively. The classification accuracy of the models for cardiac arrhythmia was 89.01%, which was higher than prediction accuracy for AMI (83.67%), and heart failure (HF: 82.84%). Age, CT ratio, glucose, transferrin saturation, albumin, ferritin were the major predictors of MACE.
Conclusion
The ML models had shown acceptable performance in predicting MACE in HD patients. Age, CT ratio, glucose and other biomarkers were important predictors of MACE, which is consistent between the individual components of MACE, such as cardiac arrhythmia, MI, and HF. These parameters can be calibrated as prognostic parameters of MACE events in HD patients.