A cautionary note on appropriate statistical methods to compare dose responses of methionine sources.

Two sources of methionine (Met) activity are frequently used in commercial feed formulation: DL-2-hydroxy-4-(methylthio) butanoic acid (HMTBA), most commonly available as an 88% solution with 12% water; and DL-methionine (DLM, 99% powder). Despite the fact that both compounds have been in commercial use for over 50 yr, controversy and confusion remain with respect to their relative bioefficacy (RBE). This paper presents a review of the use of a nonlinear common plateau asymptotic regression technique (NLCPAR) that has been used to compare the 2 Met sources with particular emphasis on the validity of the basic assumptions of that model. The thesis of this paper is that the controversy is due, at least in part, to the misapplication of this regression technique to estimate the RBE of HMTBA and DLM. The NLCPAR model is a bioassay with the key dependent assumptions that HMTBA is a dilution of DLM, and that each follows dose-response curves of the same form and approach a common plateau. Because both provide Met activity, it may be considered reasonable to accept these assumptions; however, specifically testing them demonstrated that the assumption of a common dose-response is not supported by data. The common plateau assumption was tested with an alternative approach of fitting nonlinear separate plateaus asymptotic regression (NLSPAR) to a set of 13 published broiler studies in which the NLCPAR model had been used to estimate RBE of HMTBA and DLM. The hypothesis of a common plateau was rejected (P < 0.01), meaning that the conclusion that HMTBA had lower bioefficacy than DLM based on the NLCPAR methodology was not valid. An example using published data demonstrated that the NLSPAR model was a significantly better fit than the NLCPAR model, and showed that HMTBA and DLM followed different dose responses. Consequently, there was no single value for RBE for the entire dose range; rather, the RBE of the 2 compounds varied with use level. The evidence presented here indicates that separate plateau models should be used when comparing these 2 products. These more valid models can then be used for predictions of differences between HMTBA and DLM at levels of expected use.