AKI prediction and recovery in hospitalized patients with COVID-19

ABSTRACT

Background: AKI is a complication in patients hospitalized with COVID-19 and is associated with poor outcomes. We aimed to develop predictive models for AKI development and recovery in patients hospitalized with COVID-19. Methods: Patients with a positive SARS-CoV2 PCR admitted to 19 Texas hospitals from 3/13/2020-1/1/2021 were included. AKI presence and stages were determined using KDIGO guidelines. Individuals with AKI present on admission (POA) were excluded for predictive models. Patients were followed for 90 days to evaluate for renal recovery (serum creatinine ≤1.1 times baseline). Nested models for AKI were built using logistic regression Model 1 included age, sex, race, smoking status, presence of hypertension (HTN), diabetes (DM), chronic kidney disease (CKD), coronary artery disease (CAD), and chronic heart failure (CHF), and use of ACEI/ARB;Model 2, added admission WBC, hs-CRP, and hemoglobin;Model 3, added ferritin and D-Dimer to Model 2 to assess for accuracy improvements. 10-fold stratified cross validation was done to evaluate model performance. Results: Of 8392 patients, 2702 (32%) had AKI, of which 2281 (84%) recovered by 90 days 92% of stage 1, 75% of stage 2, and 40% of stage 3 AKI, p for trend <0.001. After excluding AKI present on admission, 776 of 5671 developed AKI during the hospitalization. Percentages of AKI stages 1, 2 and 3 were 67%, 8%, and 25%. Overall, 152 (20%) of 776 required RRT. Patients with AKI were older, more likely to be male, black, and have hypertension, diabetes, coronary artery disease, congestive heart failure, and CKD. The interval improvement of each AKI predictive model was statistically significant, with last model AUC of 78.1 (95% CI 76.3%-79.9%) and all p<0.001. The final model had improvement in all metrics when compared to Models 1 and 2, with a sensitivity of 69%, specificity 76%, positive predictive value 32%, negative predictive value 94%, positive likelihood ratio 3.02, and negative likelihood ratio 0.40. Conclusions: AKI is common among patients hospitalized with COVID-19, but a large proportion recover renal function by 90 days. Recovery rate is lower based on stepwise higher stages of AKI. Addition of inflammatory biomarkers to demographics and medical comorbidities can improve prediction of AKI in this patient population.