Graphical analysis of multivariate pain data in analgesic trials

In order to compare analgesic treatments effectively one must measure pain over time. In a single-dose clinical analgesic trial one typically obtains repeated pain measurements from each patient during a relatively short period (4-6 hours). Such measurements constitute multivariate data, which are usually reduced by simple addition to a single derived pain measure for analysis of between-treatment differences. In this article we consider graphical procedures for the display of multivariate, clinical analgesic data. The first of these are the traditional time-effect curves, with added "standard error bars." We also examine a multivariate display, the biplot, which is based on principal component analysis. This technique provides a representation of the raw pain scores at each time of measurement by a vector originating from the origin of a two-dimensional graph. The multivariate pain scores of individual patients are summarized by points on the graph. Differences in pain scores over time (or between treatment groups) may be examined by relating points (or ellipses representing treatment groups) to different vectors. Using actual data from a postsurgical analgesic trial, we compare the two approaches and show that the multivariate approach is a useful addition to the standard technique for display of such data.