Mathematical modeling for optimal control of BK virus infection in kidney transplant recipients

点击翻译按钮获取中文摘要

BK virus (BKV) infection and nephropathy is a major cause of organ loss following kidney transplantation. There are no effective antivirals for BKV, and standard clinical practice is to reduce immunosuppression, which raises the risk of allograft rejection. There is no consensus on how to safely reduce immunosuppression, and treatment of BKV viremia varies by the attending physician and transplant center. To address this gap in our knowledge for how to optimally manage kidney transplant recipients with BKV viremia, we propose to use mathematical modeling and optimal control theory to develop software-guided management of immunosuppression personalize for individual patients. To translate these models and algorithms to clinical reactive, we need to calibrate and validate them on longitudinal data of kidney transplant recipients with BKV viremia. We propose to utilize NIAID-funded Clinical Trials in Organ Transplantation (CTOT) data sets available in ImmPort, together with pediatric and adult CTOT kidney transplant data not currently in ImmPort, to build the largest longitudinal BKV monitoring data set and use it to calibrate our models. We build this proposed work on our published mathematical model of BKV viremia, which accurately models BKV proliferation and infection of kidney cells, the elicitation of anti-viral and allo-specific cytotoxic T cells that damage kidney cells and reduce graft function, and the non-replenishment of damaged kidney cells resulting in rising creatinine levels. We propose to extend and refine the model to fit the longitudinal clinical data from CTOT kidney transplant recipients with BKV viremia more accurately. To better utilize the extensive patient data from the CTOT studies, we also propose an innovative approach to build a more accurate BKV model by learning the equations describing immune regulation directly from data. Learning will be performed using a neural network emulator, which can approximate arbitrarily complex mathematical functions. This machine learning approach has the advantage of being more flexible and using ancillary data sources that allow a greater degree of personalized model characterization. The CTOT data will be used to inform immune response modeling and to validate the initial model refinements as well as assist us in further model development. We then layer receding horizon control (RHC) algorithms, also known as model predictive control, on top of the mathematical models, to provide adaptive guidance on optimal immunosuppression doses customized to individual patients. The calibration of these models is critical, and a critical component of our proposal is the use advanced statistical methods to estimate model parameters from sparse longitudinal data and to perform sensitivity analysis. Finally, the proposal benefits immensely from the collaboration of three experienced clinicians who manage kidney transplant recipients, who will provide domain expertise, help with the interpretation of data, and guide model and software development to ensure clinical utility. The expected outcome is a useful tool to guide immunosuppressive management that could be tested for its effectiveness in improving patient outcomes and reducing the morbidity of immunosuppressive therapy.