心力衰竭患者AKI发生的临床预测模型构建

摘要 目的：构建心力衰竭患者AKI（acute kidney injury）发生的临床预测模型，对早期高危患者识别提供依据。方法：回顾性分析新疆医科大学第一附属医院2018年1月至2020年12月明确诊断心力衰竭患者350例，其中AKI患者104名（29.7%），非AKI患者246名（70.3%），将其按7:3 比例随机分为建模队列(n=245)和验证队列(n=105)。构建 LASSO回归分析建模队列，基于 logistic 回归结果构建HF-AKI（heart failure-acute kidney injury）患者的诺顿图，同时对模型进行校准，同时验证模型效益。结果：单因素分析得到25个差异变量，LASSO回归、多因素逐步logistics 回归，最终得到5个差异变量：年龄、住院天数、入院肌酐、射血分数、是否使用抗生素。构建HF-AKI 患者的临床预测模型并绘制成诺顿图。构建训练组和验证组诺顿图的 ROC曲线 AUC大小分别为 0.730和 0.794，通过Hosmer-Lemeshow检验，验证组虽然没有训练组的拟合优度优异，但P＞0.05，表明该诺顿图模型同样具有良好的校准度。结论：本研究成功构建了HF-AKI的临床预测模型，经过系列验证提示该模型的训练组和验证组均具有净收益范围，具有一定的临床价值。

Construction of Clinical Prediction Model for AKI in Patients with Heart Failure

ABSTRACT Objective: To establish a clinical prediction model for AKI occurrence in patients with heart failure, and to provide evidence for early identification of high-risk patients. Methods: A retrospective analysis was performed on 350 patients with heart failure confirmed by the First Affiliated Hospital of Xinjiang Medical University from January 2018 to December 2020, including 104 patients (29.7%) with AKI and 246 patients (70.3%) without AKI. They were randomly divided into a modeling queue (n=245) and a validation queue (n=105) in a 7:3 ratio. The LASSO regression analysis modeling cohort was constructed, and the Norton diagram of HF-AKI patients was constructed based on logistic regression results. Meanwhile, the model was calibrated and the benefits of the model were verified. Results: 24 differential variables were obtained by univariate analysis, LASSO regression, multi-factor stepstep logistics regression, and finally 5 differential variables were obtained: age, length of stay, admission creatinine, ejection fraction, antibiotic use or not. The clinical prediction model of HF-AKI patients was constructed and plotted into Norton diagram. The ROC curve AUC of the Norton diagram of the constructed training group and the verification group were 0.730 and 0.794, respectively. Hosmer-lemeshow test showed that the goodness of fit of the verification group was not as good as that of the training group, but P>0.05. It shows that the Norton diagram model also has good calibration degree. Conclusion: The clinical prediction model of HF-AKI was successfully constructed in this study. After a series of validation, it was suggested that both the training group and the validation group of this model had a net income range and had certain clinical value.