Department of Clinical Laboratory, Peking University First Hospital, Beijing, China.
Department of Clinical Laboratory, Shanxi Bethune Hospital, Taiyuan, China.
PeerJ. 2024 Nov 1;12:e18436. doi: 10.7717/peerj.18436. eCollection 2024.
Chronic kidney disease (CKD) is a major public health issue, and accurate prediction of the progression of kidney failure is critical for clinical decision-making and helps improve patient outcomes. As such, we aimed to develop and externally validate a machine-learned model to predict the progression of CKD using common laboratory variables, demographic characteristics, and an electronic health records database.
We developed a predictive model using longitudinal clinical data from a single center for Chinese CKD patients. The cohort included 987 patients who were followed up for more than 24 months. Fifty-three laboratory features were considered for inclusion in the model. The primary outcome in our study was an estimated glomerular filtration rate ≤15 mL/min/1.73 m or kidney failure. Machine learning algorithms were applied to the modeling dataset ( = 296), and an external dataset ( = 71) was used for model validation. We assessed model discrimination area under the curve (AUC) values, accuracy, sensitivity, specificity, positive predictive value, negative predictive value, and F1 score.
Over a median follow-up period of 3.75 years, 148 patients experienced kidney failure. The optimal model was based on stacking different classifier algorithms with six laboratory features, including 24-h urine protein, potassium, glucose, urea, prealbumin and total protein. The model had considerable predictive power, with AUC values of 0.896 and 0.771 in the validation and external datasets, respectively. This model also accurately predicted the progression of renal function in patients over different follow-up periods after their initial assessment.
A prediction model that leverages routinely collected laboratory features in the Chinese population can accurately identify patients with CKD at high risk of progressing to kidney failure. An online version of the model can be easily and quickly applied in clinical management and treatment.
慢性肾脏病（CKD）是一个重大的公共卫生问题，准确预测肾衰竭的进展对于临床决策至关重要，有助于改善患者的预后。因此，我们旨在开发并外部验证一种基于机器学习的模型，利用常见的实验室变量、人口统计学特征和电子健康记录数据库来预测 CKD 的进展。
我们使用单中心的中国 CKD 患者的纵向临床数据开发了一个预测模型。该队列包括 987 名随访时间超过 24 个月的患者。共考虑了 53 项实验室特征纳入模型。我们的研究主要结局是估算肾小球滤过率（eGFR）≤15 mL/min/1.73 m 或肾衰竭。机器学习算法应用于建模数据集（n=296），并使用外部数据集（n=71）进行模型验证。我们评估了模型的区分度（曲线下面积[AUC]值）、准确性、敏感性、特异性、阳性预测值、阴性预测值和 F1 评分。
在中位数为 3.75 年的随访期间，148 名患者发生了肾衰竭。最优模型是基于堆叠不同分类器算法和六个实验室特征建立的，包括 24 小时尿蛋白、钾、葡萄糖、尿素、前白蛋白和总蛋白。该模型具有相当高的预测能力，在验证数据集和外部数据集中的 AUC 值分别为 0.896 和 0.771。该模型还能准确预测患者在初始评估后不同随访时间内肾功能的进展情况。
一种利用中国人群中常规收集的实验室特征的预测模型可以准确识别 CKD 患者中进展为肾衰竭的高危人群。该模型的在线版本可以方便、快速地应用于临床管理和治疗。
