2022年 · 第102卷第46期

Urinary neutrophil gelatinase-associated lipocalin (NGAL), a biomarker of acute kidney injury, has been used in urine strip tests. However, there are limited data on its use in low-and middle-income countries, where diagnosing acute kidney injury remains a challenge. The study prospectively enrolled 250 children aged 2 to 18 years with sickle cell anemia, including 185 children hospitalized for a painful (vasoocclusive) crisis and 65 reference children followed up in usual care at a sickle cell anemia clinic. The definition of kidney injury employs a range of serum creatinine measurements and a modified definition of sickle cell anemia by the Global Organization for Improving Kidney Disease Prognosis (KDIGO). The average age of the children enrolled in school was 8.9 years, of which 42.8% were female children. Of the hospitalized children, 36.2% developed kidney injury and 3.2% died. There was a strong correlation between urine NGAL levels determined with test strips and standard enzyme-linked immunosorbent assay (0.71 in hospitalized children and 0.88 in the usual care reference population). NGAL levels were elevated during renal injury and significantly increased at all stages of injury. Adjusted for age and sex, the relative risk of kidney injury in hospitalized children undergoing high-risk strip tests (300 ng/ml and above) was 2.47, 95%CI: 1.68 to 3.61, the risk of death increased by 7.28 times, 95%CI：1.10~26.81。 In children with sickle cell anemia and acute kidney injury, urinary NGAL levels are significantly elevated and may predict the risk of death.

Acute kidney injury (AKI) is the most common and serious complication of sepsis, with high mortality and disease burden. Early prediction of AKI is key to timely intervention and ultimately improved prognosis. Prediction of AKI in patients with sepsis has been a hot topic in critical care medicine research. In recent years, due to the development of statistical theory and computer technology, machine learning has attracted the attention and recognition of clinicians. This study aims to establish and validate a predictive model based on a novel machine learning algorithm for predicting AKI occurrence in critically ill sepsis patients. This study extracted sepsis patient data from the Intensive Care Unit Medical Information Set III (MIMIC-III) database, used Boruta algorithm for feature selection, and used machine learning algorithms such as logistic regression, K-nearest neighbor algorithm (KNN), support vector machine (SVM), decision tree, random forest, extreme gradient boost (XGBoost) and artificial neural network (ANN) and ten-fold cross-validation methods to construct the model. The performance of these models was evaluated in terms of identification, calibration, and clinical application, respectively. In addition, the discrimination ability of the machine learning-based model was compared with that of the Sequential Organ Failure Assessment (SOFA) model and the customized Simplified Acute Physiological Function Score II (SAPS II) model. A total of 3 176 critically ill patients with sepsis were included for analysis, of which 2 397 (75.5%) developed AKI during hospitalization. A total of 36 variables were selected for model construction, and models such as logistic regression, KNN, SVM, decision tree, random forest, ANN, XGBoost, SOFA and SAPSⅡ score were established. The areas under the operating characteristic curve of subjects were 0.736, 0.664, 0.735, 0.749, 0.779, 0.755, 0.821, 0.646 and 0.702, respectively. Among all models, the XGBoost model has the best predictive performance in terms of identification, calibration, and clinical application. Therefore, the study considers the machine learning model to be a reliable tool to predict AKI in sepsis patients. Among them, the XGBoost model has the best predictive performance and can be used to assist clinicians in identifying high-risk patients and implementing early interventions to reduce mortality.