Supplementary Material for: Incorporation of Urinary Neutrophil Gelatinase-Associated Lipocalin and Computed Tomography Quantification to Predict Acute Kidney Injury and In-Hospital Death in COVID-19 Patients

Background: The prevalence of acute kidney injury (AKI) in COVID-19 patients is high, with poor prognosis. Early identification of COVID-19 patients who are at risk for AKI and may develop critical illness and death is of great importance. Objective: The aim of this study was to develop and validate a prognostic model of AKI and in-hospital death in patients with COVID-19, incorporating the new tubular injury biomarker urinary neutrophil gelatinase-associated lipocalin (u-NGAL) and artificial intelligence (AI)-based chest computed tomography (CT) analysis. Methods: A single-center cohort of patients with COVID-19from Wuhan Leishenshan Hospital were included in this study. Demographic characteristics, laboratory findings, and AI-assisted chest CT imaging variables identified on hospital admission were screened using least absolute shrinkage and selection operator (LASSO) and logistic regression to develop a model for predicting the AKI risk. The accuracy of the AKI prediction model was measured using the concordance index (C-index), and the internal validity of the model was assessed by bootstrap resampling. A multivariate Cox regression model and Kaplan-Meier curves were analyzed for survival analysis in COVID-19 patients. Results: One hundred seventy-four patients were included. The median (±SD) age of the patients was 63.59 ± 13.79 years, and 83 (47.7%) were men.u-NGAL, serum creatinine, serum uric acid, and CT ground-glass opacity (GGO) volume were independent predictors of AKI, and all were selected in the nomogram. The prediction model was validated by internal bootstrapping resampling, showing results similar to those obtained from the original samples (i.e., 0.958; 95% CI 0.9097–0.9864). The C-index for predicting AKI was 0.955 (95% CI 0.916–0.995). Multivariate Cox proportional hazards regression confirmed that a high u-NGAL level, an increased GGO volume, and lymphopenia are strong predictors of a poor prognosis and a high risk of in-hospital death. Conclusions: This model provides a useful individualized risk estimate of AKI in patients with COVID-19. Measurement of u-NGAL and AI-based chest CT quantification are worthy of application and may help clinicians to identify patients with a poor prognosis in COVID-19 at an early stage.

研究背景：新型冠状病毒肺炎（COVID-19）患者中急性肾损伤（acute kidney injury, AKI）的患病率较高，且预后不佳。尽早识别存在AKI风险、可能进展为重症乃至死亡的COVID-19患者，具有重要临床意义。
研究目的：本研究旨在构建并验证一款针对COVID-19患者AKI与院内死亡的预后模型，纳入新型肾小管损伤生物标志物尿中性粒细胞明胶酶相关脂质运载蛋白（urinary neutrophil gelatinase-associated lipocalin, u-NGAL）以及基于人工智能（artificial intelligence, AI）的胸部计算机断层扫描（chest computed tomography, CT）分析指标。
研究方法：本研究纳入武汉雷神山医院的单中心COVID-19队列患者。收集患者入院时的人口学特征、实验室检查结果及AI辅助胸部CT影像学指标，经最小绝对收缩和选择算子（least absolute shrinkage and selection operator, LASSO）与Logistic回归筛选后，构建AKI风险预测模型。采用一致性指数（concordance index, C-index）评估模型的预测准确率，并通过自助重采样法评估模型的内部有效性。针对COVID-19患者的生存分析，采用多因素Cox回归模型与Kaplan-Meier曲线完成。
研究结果：本研究共纳入174例患者，患者中位年龄（±标准差）为63.59±13.79岁，其中男性83例，占比47.7%。u-NGAL、血清肌酐、血尿酸及CT磨玻璃影（ground-glass opacity, GGO）体积均为AKI的独立预测因子，且全部被纳入列线图。通过内部自助重采样对模型进行验证，结果与原始样本分析结果一致（C-index为0.958；95%置信区间：0.9097~0.9864）。AKI预测的C-index为0.955（95%置信区间：0.916~0.995）。多因素Cox比例风险回归分析证实，高u-NGAL水平、增大的GGO体积及淋巴细胞减少均为不良预后与高院内死亡风险的强预测因子。
研究结论：本模型可为COVID-19患者的AKI风险提供有效的个体化评估。u-NGAL检测与基于AI的胸部CT定量分析具备临床应用价值，可帮助临床医师尽早识别预后不良的COVID-19患者。