External Validation of the Augmented Renal Clearance Predictor in Critically Ill COVID-19 Patients

Abstract

The ARC predictor is a prediction model for augmented renal clearance (ARC) on the next intensive care unit (ICU) day that showed good performance in a general ICU setting. In this study, we performed a retrospective external validation of the ARC predictor in critically ill coronavirus disease 19 (COVID-19) patients admitted to the ICU of the University Hospitals Leuven from February 2020 to January 2021. All patient-days that had serum creatinine levels available and measured creatinine clearance on the next ICU day were enrolled. The performance of the ARC predictor was evaluated using discrimination, calibration, and decision curves. A total of 120 patients (1064 patient-days) were included, and ARC was found in 57 (47.5%) patients, corresponding to 246 (23.1%) patient-days. The ARC predictor demonstrated good discrimination and calibration (AUROC of 0.86, calibration slope of 1.18, and calibration-in-the-large of 0.14) and a wide clinical-usefulness range. At the default classification threshold of 20% in the original study, the sensitivity and specificity were 72% and 81%, respectively. The ARC predictor is able to accurately predict ARC in critically ill COVID-19 patients. These results support the potential of the ARC predictor to optimize renally cleared drug dosages in this specific ICU population. Investigation of dosing regimen improvement was not included in this study and remains a challenge for future studies.

Keywords

• renal elimination
• COVID-19
• intensive care units
• machine learning
• external validation
• augmented renal clearance