A comparison of classification tree and linear regression analysis for the assessment of vaccine quality

LÓPEZ, MARÍA VIRGINIA; MARANGUNICH, LAURA; BARTOLONI, NORBERTO; PARREÑO, VIVIANA; RODRIGUEZ, DANIELA; VENA, MARÍA MARTA; IZUEL, MERCEDES

Lisboa, Portugal

Congreso; 56th Session of the International Statistical Institute; 2007

International Statistical Institute, Portugal

Frequently, the assessment of an event by quantitative measures leads the researcher to decision making on the basis of certain values that allow to classify cases into categories. When the variable can be measured by two methods, these must be concordant in the final result. In general, in these cases the researcher is interested in the most convenient method in what concerns costs and simplicity. In this sense, an example in livestock production is the assessment of the immunogenic quality of a vaccine against a pathogenic virus infection of importance in animal health. The high costs and the difficulty in finding seronegative bovines pose the need for developing standardized tests in laboratory animals to ensure the presence of reliable products in market. The use of a smaller animal species, naturally seronegative for the virus, could represent a convenient alternative to the use of bovines. In this paper we compare both techniques for the classification of vaccines in accordance of their antigenic level based in the antibody response of bovines and a smaller laboratory animal species. In this study both techniques had similar performance. With classification tree the interval between limits was wider for the intermediate class, this could be convenient for the laboratory. An additional advantage in using classification tree is the possibility of adding another variable, i.e. difference with baseline antibody titre, without loosing interpretability of the results. This variable could be useful in some viral infections when it is difficult to find serological negative bovines.