The Need for Statistically Informed Analyses
Despite the encouraging preliminary findings using variations on cluster analysis, the developmental protocol for a reliable therapeutic agent or diagnostic test needs a rigorous statistical treatment. In particular, we believe that models developed for clinical applications must be able to quantify false positive errors as well as false negative errors (corresponding to the traditional type I and type II error rates).
Clustering methods and ad hoc filtering schemes have the capacity to direct our attention in a qualitative sense to genes that may be of interest to researchers, but they do not provide a clear framework for assessing both error rates. Without using standardized statistical tools designed to measure the frequencies of false positive errors and false negative errors it is difficult to assess either the reproducibility or the generalizability of the results of pilot studies.
When human life is at risk and limited health-care research dollars are being spent on developing new diagnostic tests and therapeutic agents, the importance of using statistically informed data analysis tools is quite clear.