MDV: A Multivariate Data Visualization Tool for Clinical Laboratory Data and Other Time-varying Continuous Measurements

MDV is a software package that provides interactive, multidimensional, dynamic display of data from clinical trials and other pharmaceutical research activities. Originally designed for clinical laboratory data, it can be used for any measured or calculated value repeated over time. Time-varying patterns can be shown in animation, and can be saved as movies. Displays can also include an indication of the extent of abnormality, based on a sufficient population of “normal” data. There are no definite, accepted methods to identify hepatotoxicity based on laboratory data collected during clinical trials. However, hepatotoxicity, like other treatment effects, often creates patterns that can be visually identified. MDV was designed to facilitate the process of finding such patterns in the large volume of laboratory data collected during clinical trials. There are many tools that produce one-, two-, or three-dimensional plots of data, but MDV includes a number of features specifically designed for ease of use and to support the display and exploration of clinical data. A user editable configuration file indicates which inputs define unique subjects, which (e.g., gender or randomized treatment) are attributes for grouping by color or window, which indicate time (e.g., hours or days), and which are analytes to actually be plotted. Measurements can be plotted as raw input, or subjected to a log transformation, which improves the perception of patterns in some analytes, such as the liver enzymes. In the example below, the first two figures show three analytes over the entire course of a clinical trial for placebo and treated subjects, respectively. The wire frame surface represents a 95% probability that the values are within the normal population. Each line connects the measured data for one subject, showing the “trajectory” of that subject’s analytes. Some of the placebo subjects exhibit an excursion from the normal region, but more of the treated subjects do so, and their trajectories cluster in an apparent pattern, suggesting a common effect of the compound. The third figure is similar to the second, but limited to data from a limited period during the trial, giving an indication of the specific temporal nature of the pattern.