Use of Many Covariates in Clinical Trials

Publisher Website
Google Scholar
Abstract
Clinical trials are almost always limited in number of patients, and when end points are qualitative, even more limited in the number of the rarer end point (often death). Yet we often feel we should make use of many covariates, more than should be used in any single regression. Composites formed by weighting the potential covariates are a natural choice. Weights based on the degree of significance of individual separate regressions involving compartments-combined data seem theoretically plausible and proved practically effective in the Anturane Reinfarction Trial. (It often pays to tailor the complexity of fitting to the extent of data available. In some circumstances, fitting an imprecise composite can help while fitting all its components separately loses.) In more complicated and data-richer situations, like the National Halothane Study, a smear-and-sweep procedure blending one classification after another into a composite classification may be helpful.

This publication has 4 references indexed in Scilit: