Predicting outcomes in patients with kidney disease: methodology and clinical applications

Abstract

In part 1 of the thesis Predicting Outcomes in Patients with Kidney Disease, key differences between etiological and prediction research are explored and it is shown that observational research often conflates etiology and prediction which leads to incorrect causal conclusions. A framework for the external validation of prognostic models is provided and it is shown how competing events can be dealt with when externally validating a time-to-event prognostic model. These results are applicable to many clinical research fields, including nephrology as exemplified in part 2. Within the six applied chapters in part 2, prediction models for various adverse outcomes in patients with advanced kidney disease are identified, validated and developed. The thesis provides a broad overview of prognostic model applications in patients with chronic kidney disease, including comprehensive external validation studies for kidney failure prediction models, mortality prediction models and graft failure prediction models. Models to predict mortality on conservative care and dialysis and models to predict adverse outcomes after kidney transplantation were developed and validated. These results may improve shared decision-making processes and individualized medicine for patients with kidney disease.