The proposed research focuses on the continued development of statistical methods for interim analyses of longitudinal studies. Previous research efforts developed a flexible group sequential procedure through an alpha spending function and applied the method to linear mixed effects models and a general class of linear rank statistics which are often used in the analysis of repeated measurement studies. We specifically want to investigate further the application of interim analyses for linear mixed effects models and for repeated categorical methods. The alpha spending function requires an estimate of the information fraction, as do all group sequential methods. An appealing estimate is the fraction of total Fisher information. For the linear mixed effects model, this estimate of information fraction is more complex than for comparison of proportions of means. We have completed some preliminary work (See Lee and DeMets, 1993) but more work is required before this aspect is totally resolved. In addition to the information fraction, we also want to evaluate the bias in the estimate of rate of change for trials which terminate early. This issue has been addressed for proportions and means but not for this method. This issue is especially of concern for extrapolating results to health care policy resulting from such trials. . We also have proposed developing methods for specific repeated categorical measures and considering issues of robustness, information fraction, and estimation bias, paralleling our efforts for the linear mixed effects model. Finally, we propose some enhancements for our software package for the flexible alpha spending function approach. We also plan to draft a clinicians handbook as a nontechnical introduction to data monitoring, including discussion of the alpha spending function, as well as many examples from completed trials.
Showing the most recent 10 out of 17 publications
