Can Kidney Function Be Used to Predict Survival of COVID-19 in Hospitals? Predictive Modelling in a Retrospective Cohort Study

Abstract

Introduction/Background and aims: Blood biomarkers have been included in several mortality prediction models in the literature [1,2] where it has been noted that mortality of coronavirus disease 2019 (COVID-19) has been linked to changes in kidney function. Therefore, the role of the kidneys is of interest in understanding COVID-19 severity and outcome in patients. The study aimed to investigate the link between kidney blood biomarkers and the survival from severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) infection on admittance to hospital in a retrospective cohort study. Methods/summary of work: De-identified and pseudo-anonymised patient data included in this study were approved by the ethics committee as part of the existing Cardiac MRI Database NHS REC IRAS Ref: 222349 and University of Brighton REC (8011). In a retrospective cohort study using data extracted from the Laboratory Information Management System, the data were collected at a large NHS Foundation Trust hospital in the Yorkshire and Humber regions, UK. The extracted dataset included 400 patients between the age of 18 and 102 years who were admitted to Accident and Emergency unit between 01/03/2020 and 31/08/2020. All patients had a confirmed reverse transcription polymerase chain reaction (RT-PCR) test for SARS-CoV-2, a recorded survival/non-survival outcome within 55 days, and kidney function blood tests conducted within 36 hours of SARS-CoV-2 PCR test. Results/Discussion: Following candidate selection, the models were finalised with the predictors age, sodium, potassium, urea, and creatinine. Logistic regression showed good discrimination for in-hospital death within the development cohort with an AUROC of 0.770 (95% CI: 0.663-0.846). Random forest had an AUROC of 0.710 (95% CI: 0.620-0.780). The specificity for the logistic regression was high with 0.907, whereas the sensitivity was 0.538. Similar values were calculated for the random forest model specificity (0.889) and lower sensitivity (0.423). The stratified 5-fold cross-validation included the training split of 80% of all included patients (n = 320), and each fold was used as test set once (n = 64). Discrimination of the logistic regression in the validation cohort was 0.780 (95% CI: 0.720-0.838) and higher than in the development cohort. The AUROC for the random forest model also improved compared to the development cohort (0.766, 95% CI: 0.730-0.810). Conclusion: The kidney prediction model can be used to obtain accurate predictions of survival of patients with SARS-CoV-2 in patients within 55 days of admittance to hospital. Further external validation is planned with larger datasets from NCCID and Biobank.

Rights

This content is not covered by the Open Government Licence. Please see source record or item for information on rights and permissions.

Cite as

Downloadable citations

Identifiers

Repository URI

http://eprints.gla.ac.uk/273788/