医学
逻辑回归
列线图
回顾性队列研究
队列
接收机工作特性
肾移植
随机森林
队列研究
推导
肾移植
试验预测值
内科学
置信区间
递归分区
移植
数据集
回归分析
回归
超声波
肾脏疾病
外科
集合(抽象数据类型)
血流动力学
肾移植
时间点
肾
并发症
预测建模
肾病科
作者
ZiHao Ni,Tianying Xing,WeiHong Hou,Xinyu Zhao,Yunlu Tao,Fubo Zhou,Yingqi Xing
摘要
OBJECTIVES: This study aimed to identify kidney transplant recipients with functioning grafts at the 3-month post-transplant who were at high risk of subsequent adverse graft outcomes. We developed predictive models and nomograms integrating ultrasound hemodynamic, clinical, and laboratory parameters. METHODS: We analyzed 376 kidney transplant recipients enrolled from January 2020 to April 2023. The 2020-2022 cohort was randomly divided into a training set (n = 200) and an internal random holdout set (validation set 1; n = 131). The subsequently enrolled 2023 cohort was retained as the temporal validation set (validation set 2; n = 45). All ultrasound predictors were assessed at the approximately 3-month post-transplant. Risk factors were selected using least absolute shrinkage and selection operator (LASSO) and logistic regression. Predictive models were constructed using logistic regression and 5 machine-learning algorithms. Model performance was evaluated in terms of discrimination, calibration, and clinical utility, and a nomogram was developed for visualization. SHapley Additive exPlanations (SHAP) were used to interpret the machine-learning models. RESULTS: The final logistic regression model included graft renal segmental artery resistance index (Grsa RI), preoperative albumin, and postoperative day 5 platelet-to-lymphocyte ratio (Po5 PLR). Its AUCs were 0.893 in the internal random holdout set (validation set 1) and 0.859 in the temporal validation set (validation set 2). The random forest model, selected on the basis of the highest training-set AUC among the machine-learning models, achieved AUCs of 0.916 and 0.821 in validation sets 1 and 2, respectively. DeLong tests showed no statistically significant difference in AUC between the logistic regression and random forest models in either validation cohort (p = .958 and p = .561, respectively). Preoperative albumin, Po5 PLR, and Grsa RI were independently associated with adverse graft outcomes. CONCLUSIONS: The prediction models effectively identified high-risk recipients using pre- and postoperative variables, providing an actionable strategy for early intervention to improve transplant outcomes by targeting modifiable factors.
科研通智能强力驱动
Strongly Powered by AbleSci AI