Dynamic Prediction of Renal Failure Using Longitudinal Prognostic Information among Patients with Chronic Kidney Disease and Kidney Transplant

点击翻译按钮获取中文摘要

Project Summary/Abstract Patients with chronic kidney disease (CKD) and patients receiving kidney transplantation (KTx) are at risk of kidney/graft failure. Accurate estimation of the time of these adverse clinical events is of great importance for patient counseling and for the timing of interventions. In clinical practice, these patients are often monitored at recurrent clinical visits for the progression of the disease. It is desirable to have tools that can make personalized, real-time prediction of the risk of kidney/graft failure at each clinical visit, adapting to the time- varying patient conditions. Currently, the published risk prediction equation for CKD and KTx are usually developed by relating risk factors measured at an earlier time point, such as baseline, to the time of subsequent adverse event in a regression model. This approach cannot incorporate the longitudinal data from all the clinical visits, and may generate suboptimal or biased risk estimation and are not suitable for real-time prediction. Building upon on recent advancement in dynamic prediction (DP) methodology from the statistical literature, we aim to develop personalized, time-adapted risk prediction equations for CKD and KTx respectively. The proposed works include developing novel DP methods for kidney/graft failure with adjustment for the competing risk by death, external validation and re-calibration, and creating software for routine use in clinical practice. For CKD, the prediction model of kidney failure will be developed from the Chronic Renal Insufficiency Cohort Study (CRIC) data, and validated using the electronic health records of Veterans Health Administration. For KTx, the prediction model of graft failure will be developed from the Wisconsin Allograft Recipient Database (WisARD), and validated using the Scientific Registry of Transplant Recipients (SRTR). The statistical methodology and software can be used in other medical specialties beyond nephrology to develop risk prediction models for adverse clinical events from longitudinal data.