Development of a Prediction Score for In-Hospital Mortality in COVID-19 Patients with Acute Kidney Injury: A Machine Learning Approach

Abstract

Introduction: Acute kidney injury (AKI) is frequently associated to COVID-19, adding severity and increasing mortality risk. \n \nObjective: The aim of the study was to develop and validate a prognostic score at hospital admission for predicting in-hospital mortality in COVID-19 patients with AKI (AKI-COV score).Design: Cross-sectional multicenter prospective cohort study. \n \nSetting: The Latin America AKI COVID-19 Registry has been conducted in 57 cities in 12 countries from Latin America. Model training was performed on a cohort of patients admitted from May\xa01 to December 31, 2020. \n \nParticipants: Eight hundred and seventy COVID-19 patients with AKI defined according KDIGO serum creatinine criteria were included between 01 March to 31 December 2020. \n \nMaterial and Methods: We evaluated four categories of predictor variables available at the time of AKI diagnosis: (1) demographic data; (2) comorbidities and condition at admission; (3) laboratory exams at admission; (4) characteristics and causes of AKI. We used a machine learning approach to fit models in the training set using 10-fold-cross validation. Predictors with more than 30% missing were removed. We select the best model and confirm the accuracy in a validation cohort using the area under the receiver operating characteristic curve (AUC-ROC). \n \nMain Outcome Measured: In-hospital mortality. \n \nResults: There were 544 (62.5%) in-hospital deaths. Increasing age, mechanical ventilation, use of vasopressors, leukocytosis, hypertension, severe condition at admission, AKI ethiology, and need kidney replacement therapies (KRT) were associated with increased risk of death. Longer time from symptoms to hospitalization or to AKI diagnosis, and higher urine output were associated with reduced risk of death. The coefficients of the best model (Elastic Net) were used to build the predictive ImAgeS score. The final model has an AUC-ROC of 0.823 [95% CI 0.761 – 0.885] in the validation cohort. \n \nConclusion: We developed a predictive model using only demographic data, comorbidities, hospital admission condition, laboratory variables and causes of AKI that shows good accuracy and is easily applicable. The use of AKI-COV score may assist health-care workers in identifying hospitalized COVID-19 patients with AKI that may require more intensive monitoring and can be used for resource allocation. \n \nFunding Statement: This study was partially funded by Latin American Society of Nephrology and Hypertension (SLANH). \n \nDeclaration of Interests: The authors declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article. \n \nEthics Approval Statement: The Institutional Review Board of the Clinica Los Olivos, Cochabamba, Bolivia approved the study.