COVID-19 and Acute Kidney Injury in the ICU: Magnitude, Timing, and the Role of Admission Severity in a Large Overlap-Weighted Cohort

Abstract Background: Acute kidney injury (AKI) is a frequent complication in intensive care unit (ICU) patients and is associated with increased mortality and long-term kidney dysfunction. During the coronavirus disease 2019 (COVID-19) pandemic, high AKI rates have been reported in critically ill patients, but the magnitude and timing of any association in ICU populations remain uncertain-particularly when accounting for baseline severity of illness and missing pre-admission serum creatinine values. Methods: We analyzed 28,612 adult ICU stays (25,923 patients) from the Northwest Intensive Care Unit (NWICU) database between March 2020 and December 2022, excluding patients with end-stage kidney disease (ESKD). COVID-19 status was determined based on laboratory confirmation of SARS-CoV-2 infection when available, or otherwise by ICD-10 diagnosis codes, consistent with clinical practice during the study period. AKI was identified in two post-admission windows-0-48 hours and 0-7 days after ICU admission-using Kidney Disease: Improving Global Outcomes (KDIGO) creatinine-based criteria. Missing covariates were handled using multiple imputation by chained equations (MICE). We estimated the effect of COVID-19 on AKI risk using overlap-weighted logistic regression to balance pre-admission covariates, and repeated analyses adjusting for the Baseline Physiology Index (BPI), a composite severity score derived from the first 12 hours of ICU data. Results: Baseline creatinine was available in 57.7% of ICU stays. In the overlap-weighted cohort (effective sample size ≈ 11,600; maximum absolute standardized difference ≈ 0.01), the standardized 7-day AKI risk was 18.9% (95% confidence interval CI 18.5-19.4) in COVID-negative patients and 23.2% (95% CI 22.7-23.7) in COVID-positive patients, corresponding to a risk difference of +4.3 percentage points (pp) (95% CI 3.6-4.9) and an odds ratio (OR) of 1.29 (95% CI 1.24-1.35). Machine-learning-based propensity scores yielded similar results. At 48 hours, no meaningful difference was observed (+0.2 pp; OR 1.02, 95% CI 0.87-1.19). Adjusting for BPI attenuated the 7-day risk difference by approximately 19% in the full cohort and approximately 41% in the BPI-observed subset, but a substantial association persisted. Conclusions: In a large, ICU-admission-anchored cohort balanced on pre-admission traits, COVID-19 was linked to 4-5 extra AKI cases per 100 patients in the first week (not the first 48h). Adjustment for baseline physiologic severity attenuated this risk, indicating partial mediation by illness at presentation, while residual effects suggest other pathogenic pathways. The integrated approach-anchoring at admission, dual-rule KDIGO ascertainment, overlap weighting, and severity-attenuation-offers a robust framework for future COVID-AKI research.