American Statistical Association New York City Metropolitan Area Chapter Mailman School of Public Health Columbia University Department of Biostatistics Colloquium

Competing risks are classically summarized by cause specific hazards and the cumulative incidence function. To obtain a full understanding of the data, these identifiable quantities should be viewed simultaneously for all possible events. Another available quantity is the conditional probability of a competing risk, which is the probability of failure from a particular cause given that no other competing event has occurred. When one event is of a particular interest, this quantity provides useful information, as it displays a probability adjusted for the other events. However, the use of the conditional probability has been limited due to the lack of a regression modelling strategy. I will present such regression model taking a functional view of the competing risks data, which treats such data as high dimensional temporal data. Insight gained from this methodology is illustrated using a dataset on patients suffering from monoclonal gammopathty of unknown significance. The focus throughout the talk will be the clear definition of competing risks endpoints and the proper interpretation of results from the multiple analyses, which may appear discrepant and difficult to synthesize.