Evaluating allograft risk models in organ transplantation: Understanding and balancing model discrimination and calibration.

Abstract

In the field of organ transplantation, the accurate assessment of donor organ quality is necessary for efficient organ allocation and informed consent for recipients. A common approach to organ quality assessment is the development of statistical models that accurately predict posttransplant survival by integrating multiple characteristics of the donor and allograft. Despite the proliferation of predictive models across many domains of medicine, many physicians may have limited familiarity with how these models are built, the assessment of how well models function in their population, and the risks of a poorly performing model. Our goal in this perspective is to offer advice to transplant professionals about how to evaluate a prediction model, focusing on the key aspects of discrimination and calibration. We use liver allograft assessment as a paradigm example, but the lessons pertain to other scenarios too.