Longitudinal and Bayesian modelling of polyomavirus nephropathy and rejection after kidney transplantation

Since the beginning of clinical kidney transplantation, more than 60 years ago, immunosuppression aimed to reduce the incidence of acute rejection. Despite the current use of powerful immunosuppressive agents, primarily targeting T-cell activation, and despite improvement on the short term, graft outcome on the long term has improved only marginally. The reason for graft failure after kidney transplantation is primarily progressive destruction of renal parenchyma, in large part due to antibody-mediated rejection, late T-cell mediated rejection, recurrent or de novo glomerular disease, polyomavirus nephropathy or non-specific progressive chronic transplant injury. In many cases, the cause of graft failure is multifactorial, a combination of these different histological disease processes. In addition to the heterogeneity in the histopathological presentation of graft injury, there is also a substantial heterogeneity in its clinical presentation. Rejection can present in many different scenarios, as delayed graft function (DGF; i.e. initial not functioning of the graft after transplantation), slow recovery of graft function after transplantation, acute increases in serum creatinine levels from baseline, subacute increases in serum creatinine, slow increases in serum creatinine over time, stable but poor kidney function, occurrence of proteinuria etc. Rejection even occurs relatively frequently (between 8-20%) subclinically, i.e. in patients with good and stable graft function, only detected by performing protocol-specified biopsies at regular time points after transplantation. There is currently no validated algorithm that integrates the histopathological phenotypes of rejection to the pretest probability for rejection calculated from the clinical presentation and graft functional changes. Moreover, based on the pathophysiology of rejection and polyomavirus nephropathy, different donor-recipient risk parameters are also relevant in the risk prediction, such as the HLA antigen and HLA molecular mismatch between donors and recipients, NK cell missing self, donor-specific HLA antibodies (DSA) and polyomavirus replication in urine and blood. Although these factors can be calculated from the routinely collected clinical data, none of these factors has been included in such probabilistic models. In this thesis, we will use advanced mixed/joint modelling and Bayesian statistical models to create and evaluate probabilistic models for these graft injury processes. The models will be cross-validated and calibrated internally, and also externally. In addition, decision curve analysis will enable estimating the actual clinical benefit of the models that will be developed.