Short Course: Methods of Risk Prediction for Patients with Competing Events

In medicine, the risk prediction model is often based on the estimated probability of an event (e.g., disease incidence) derived from a survival regression model. If a person is at risk of multiple mutually exclusive events, i.e., competing events, the occurrence of one event may prevent the occurrence of any other competing event from happening, or may alter the probability of the occurrence of other event. If competing events are independent to each other, standard survival regression models can be employed in risk prediction. However, when competing events are not independent to each other, treating competing events as usual noninformative censoring in standard survival analysis modeling may result in bias in risk prediction. Several statistical methods have been developed to estimate the probability of occurrence of an event when people are subject to competing risks. Proper selection of the method of risk prediction is crucial to producing accurate result.

The objectives of the course are to disseminate the concepts, methods, and the recent statistical tools for risk prediction with data involving competing risks, and to enrich a network of researchers who use data science and analytic methods in medicine. At the end of the course, attendees will understand why standard statistical regression models are not appropriate for analyzing data of competing risks, and will be able to identify two different types of competing risks in practice, and understand the importance of diagnostics in risk prediction modeling.

The course will be led by Joyce Chang, PhD and Jonathan Yabes, PhD, Departments of Medicine and Biostatistics, University of Pittsburgh and held in the Graduate School of Public Health Auditorium (G23) on Saturday, April 21 from 9AM-4:30PM.