Abstract: TH-PO323
Machine Learning for Identification of Near-Term All-Cause and Cardiovascular Death Among Patients Undergoing Peritoneal Dialysis
Session Information
- Home Dialysis - I
November 02, 2023 | Location: Exhibit Hall, Pennsylvania Convention Center
Abstract Time: 10:00 AM - 12:00 PM
Category: Dialysis
- 802 Dialysis: Home Dialysis and Peritoneal Dialysis
Authors
- Xu, Xiao, Peking University First Hospital Department of Nephrology, Beijing, Beijing, China
- Dong, Jie, Peking University First Hospital Department of Nephrology, Beijing, Beijing, China
Background
Although more and more cardiovascular risk factors have been verified in peritoneal dialysis (PD) populations in different countries and regions, it is still difficult for clinicians to accurately predict who will have cardiovascular events and when the cardiovascular events will occur and cause death. We developed and validated machine learning-based models to predict near-term all-cause and cardiovascular death.
Methods
Machine learning models were developed on the Peritoneal Dialysis Telemedicine-assisted Platform Cohort (PDTAP) of 7539 PD patients enrolled between June 2016 and April 2019, which was randomly divided into a training set and an internal test set by 5 random shuffles of 5-fold cross-validation, to predict of 3-month cardiovascular death and all-cause death. We chose objectively collected markers such as patient demographics, clinical characteristics, and laboratory, and dialysis-related variables to inform the models and assessed the predictive performance using a range of common performance metrics, such as precision, accuracy, and area under the curve (AUC).
Results
CVDformer model was used to predict 3-month cardiovascular death of PD patients. In the test set, CVDformer model had a precision of 87.96% - 89.45% and an accuracy of 86.10% - 88.74%. best than the LSTM model had 78.78% - 80.45% for precision and 76.53% - 79.12% for accuracy. The AUC was 0.88-0.90 in identifying the presence of near-term all-cause death and cardiovascular death using the CVDformer model.
Conclusion
We developed and used a novel combination of machine learning methods to assess the all-cause mortality risk and cardiovascular death risk in 3 months. The ability to identify the potential risks of all-cause and cardiovascular death with an inexpensive, widely available, and automatic procedure has important practical implications, particularly for the management of dialysis patients.