Keele Research Repository

Prognostic models are used to predict an individual’s future health outcomes, including the risk of disease progression and the development of further complications. The statistical methodology used to develop these models is often naïve to the presence of competing events, these are events which prevent or alter the probability of an outcome of interest from occurring. Not appropriately accounting for competing events is known to produce inflated absolute risk predictions for time-to-event outcomes, this bias is known as competing risks bias. However, there has been relatively little research about competing events in prognostic model research, for which absolute risk predictions are a key outcome.
This thesis investigates the presence and impact of competing events on prognostic model research. To begin, two reviews were conducted to determine the presence, reporting, and management of competing events in current prediction model literature. Then competing risks statistical regression methods were applied to develop and internally validate a prognostic model using existing study data. These models were compared to models developed using standard time-to-event analysis techniques, naïve to competing events, with an external validation study. Finally, a simulation study was performed to identify the circumstances for which competing risks bias affects the predictive ability and calibration of prognostic models, with an overall aim to provide guidance for the optimal approaches to incorporate competing risks in prognostic model research.