This project will develop specific statistical methods to evaluate, and to reduce, bias due to the potential for early stopping, in estimation of both primary and secondary parameters, and will develop appropriate p-values for secondary hypotheses. Secondary analyses are defined as inference regarding parameters other than the primary parameter or inference regarding the primary outcome parameter in subset analyses. The basis for secondary inference is the Whitehead model, where two measures (primary and secondary) are available from every patient in a survival setting without censoring. Extensions beyond the Whitehead model are also to be investigated. An additional area to be addressed is the issue of incorporating late data (i.e., data accumulated after a trial has been stopped, say for safety reasons, before formal stopping criteria have been met). Outcomes are shown to be Brownian motion processes. The project will require mathematical analysis and approximation, algorithmic and software development and simulation studies. It is also planned to reanalyze some published trials of cardiovascular treatments to assess bias due to the sequential nature of those trials, and to determine how developed methods would have affected the bias.
