Abstract
Background: Chronic kidney disease (CKD) patients with coronavirus disease 2019 (COVID-19) are at significant risk of death. However, clinical identification of high-risk individuals remains suboptimal despite the recognition of many pathophysiological and comorbidity-related risk factors. We aim to develop a clinically simple machine learning (ML)-based score to predict acute COVID-19 mortality among CKD patients. Methods: CKD inpatients with COVID-19 were prospectively enrolled from December 2022 to January 2023 with a three-month follow-up. Feature selection from clinical and laboratory results was performed through least absolute shrinkage and selection operator and stepwise selection. Logistic regression, support vector machine (SVM), random forest, and extreme gradient boosting were applied for ML model development. A predictive score for mortality was constructed using logistic regression. We compared predictive ability between the proposed score and other published scores. Results: 219 CKD patients were included and had a high mortality rate of 25.1%. The SVM model exhibited the best performance, with the validation area under the receiver operating characteristic curve (AUC) being 0.946 (95% CI 0.918, 0.974). The COVID-19 vaccination status, age, monocyte percentage, prothrombin activity, cardiac troponin T, and total bilirubin ("VAMPCT") were the most relevant factors and utilized to develop the scoring system with an AUC of 0.960 (95% CI 0.935, 0.985). Conclusion: ML models predicting three-month mortality had favorable performance for CKD patients with COVID-19. The VAMPCT mortality score provided a user-friendly approach.