You are here:
THE ANALYSIS OF MIXED DISCRETE AND CONTINUOUS OUTCOMES USING DESIRABILITY FUNCTIONS.
Citation:
MOSER, V. C., T. COFFEY, AND C. GENNINGS. THE ANALYSIS OF MIXED DISCRETE AND CONTINUOUS OUTCOMES USING DESIRABILITY FUNCTIONS. Presented at Society of Toxicology, San Diego, CA, March 05 - 09, 2006.
Description:
Multiple types of outcomes are sometimes measured on each animal in toxicology dose-response experiments, and multiple analyses may increase the overall type I error. One approach to analyzing these outcomes in an integrated way is through the use of a composite score. We introduce the use of desirability functions as a way of deriving an overall score that uses information from each of the outcomes. This methodology is commonly used in the quality control engineering literature but has not been applied in toxicology. Desirability functions transform observed responses of any type to a 0-to-1 unitless scale, where smaller scores indicate less desirable responses (i.e. increased toxicity). The geometric mean is used to combine the scores and then a univariate statistical analysis can be performed. Compared to other composite scores discussed in the toxicology literature, this approach has two advantages. First, because of the mathematical nature of the geometric mean, the overall score is more sensitive to deviations from the typical response at each dose group. As a consequence, the estimate of the threshold is more sensitive to evidence of toxicity. Second, weights may be incorporated into the desirability function, making it possible to prioritize the importance of each endpoint. Using data from five outcomes (motor activity, brain and blood cholinesterase activity (ChE), gait score and tail pinch score: a combination of ordinal, count, and continuous responses) from a neurotoxicity experiment, we demonstrate the use of desirability functions to derive a composite score. We analyze the overall score using a nonlinear exponential threshold model. Using the composite score the lowest dose tested exceeded the 95% confidence limits of the dose threshold. Further analyses indicated the composite score was sensitive to slight toxicity, mainly due to ChE inhibition. This research was partially supported by NIEHS training grant #T32 ES007334 and does not reflect U.S. EPA policy.