Abstract:
|
In the analysis of adverse events of therapeutic treatments, often we would like to understand if different types of events tend to occur together within the same patients more or less frequently than random occurrence. In risk and benefit evaluation, we are interested in evaluating if patients are more likely to be responders in terms of both efficacy and safety for certain therapies. In setting of multiple endpoints, treatment benefit can also be evaluated by understanding if patients are more likely to respond to all or some endpoints simultaneously than independently. The current available statistical metrics may not be adequate for such quantification, such as the Venn Diagrams, conditional probabilities, as well as some metrics found in psychometrics. In this paper, we propose simple statistical metrics that can quantify the relative dependency of events among overlapping multiple events and has clear interpretation. We also present the estimation of the confidence intervals and evaluate the properties such as bias, variability and sample size requirement using simulations. Illustration of use in clinical trial data will be provided.
|