Abstract: FR-PO1174
Analysis of In-Hospital Mortality Risk Factors and Establishment of a Nomogram Prediction Model for Patients with Stage 5 CKD in the Intensive Care Unit (ICU)
Session Information
- CKD: Kidney Function and Extrarenal Complications
October 25, 2024 | Location: Exhibit Hall, Convention Center
Abstract Time: 10:00 AM - 12:00 PM
Category: CKD (Non-Dialysis)
- 2302 CKD (Non-Dialysis): Clinical, Outcomes, and Trials
Author
- Liu, Fanna, The First Affiliated Hospital of Jinan University, Nephrology, Guangzhou, China
Background
Patients with stage 5 chronic kidney disease (CKD5) have a high risk of mortality. ICU patients have lots of abnormal results. Defining the risk factors for in-hospital mortality in ICU CKD5 patients will help doctors pay more attention to the most important factors and intervene in a timely manner. In this study, we analyzed the risk factors related to in-hospital mortality and drew the nomogram to predict in-hospital mortality in CKD5 non-dialysis （CKD 5ND） patients.
Methods
According to the diagnosis code, 495 CKD 5ND patients were selected from the MIMIC-IV database. The differences between the survival group and the death group were analyzed by two independent samples t-test, rank-sum test, or chi-square test. The nomograms to predict the risk of in-hospital mortality in CKD 5ND patients were established respectively. In order to evaluate the nomogram, ROC curve, calibration curve, and DCA curve were used. Boostrap-resampling was used for internal verification. eICU patients were used for external verification. And the predictive performance of the nomogram was compared with commonly scores.
Results
There were 1439 CKD5 patients in the MIMIC-IV database, including 495 non-dialysis patients. There were 425 CKD5 ND patients in the survival group and 70 patients in the death group. Age, SBP, WBC, RDW, and BUN/Scr were the correlation factors for in-hospital mortality in CKD5 ND patients. Internal and external validation showed that the CKD5 ND patients nomogram prediction model had good calibration, clinical benefit, and discrimination (training set AUC=0.832, validation set AUC=0.770). The predictive ability of nomogram was better than the three commonly used scores.
Conclusion
The nomogram prediction model predicting in-hospital mortality of CKD5 ND patients has good discrimination, calibration, and clinical value, which are established by Age, SBP, WBC, RDW, and BUN/Scr.