The use of sum statistics for multiple endpoints in clinical and preclinical trials. Violeta Labarta Centro de Ingeniería Genética y Biotecnología, División de Ensayos Clínicos. Ave 31 e/ 158 y 190 Playa, Apartado postal 6162, Ciudad de la Habana CP 10600, Cuba. Havana, Cuba e-mail: clintr@cigbdec.cigb.edu.cu 1. Introduction The efficacy of two or more treatments frequently are measured for more than one response variable. Although univariate methods for individual assessing of each characteristic are useful in this setting, there is the additional need for a single, overall, objective probability statement that addresses the question of whether or not the experimental therapy is efficacious (Obrien, 1984). The standard analysis for the comparison of two multivariates samples, which is based on Hotelling’s T2 statistic, addresses somewhat the wrong question and consequently has very poor power for the alternatives of primary interest. A second approach would be to assign per-experiment error rates to each of the univariate test by using Bonferroni’s inequality. This approach may lack power for alternatives in which most or all measures of efficacy are improved. This will be of particular concern when the number of endpoints studied is large relative to sample size. This problem will be exacerbated when the measures of efficacy are highly correlated, as is often the case. In the last 15 years have been reported tests that carry out a global analysis. O’Brien’s Ordinary Least Squares test for testing the multivariate one-sided hypothesis when the covariance matrix is unknown. Exact tests, P{decision for H1 µx=µy} = α, for any ∑, where α is the level of the test: test of stochastic order of Wei-Lachin (Wei L.J., Lachin J.M. ,1984), test of Follman (Follmann D., 1996) and Standardized Sum and Principal Components tests published recently (Läuter J., 1996; Läuter J., Glimm E., Kropf S., 1996). Standardized Sum and Principal Components tests can be applied in differences between two or more populations have to be tested and a factorial structure of the means and covariances can be supposed. He proposed statistics which permit exact t and F tests keeping the error of first kind at the prescribed level. We did a MatLab program v.5.2. to calculate the Standardized Sum and Principal Components tests. These tests were applied in four preclinical and clinical studies with the objective of comparing their results with standard analyses. We proposed that methods for studies of scaring and the studies of lesions of the acne. 2. Materials and Methods 7DEOH8VHGVWXGLHV Study Treatments Disease Subjects n 1 8 ulcer rats 58 2 5 burn picks 74 3 4 acne humans 226 4 2 acne humans 75 Response variables in the studies of scaring (1 y 2): wound area, not wound area, perimeter, middle radio, circularity, eccentricity and lineal growth. We select for multivariate analyses: wound area, perimeter and middle radio. Response variables in the studies of lesions of the acne (3 y 4): comedones, papule, pustule, cyst, nodules and abscess. We select for multivariate analyses all of then. 3. Results and Discussion 7DEOH5HVXOWVRIWKHDSSOLFDWLRQRIWKHSURFHGXUHVLQHDFKVWXG\ Study Standard analysis Standardized Sum Principal Components 1 RRao=0.95 FSS= 0.97 FPC= 0.99 (d.f.=21 y 138) (d.f.=3 y 54) (d.f.=3 y 54) PR=.5224 PSS=0.416 PPC=0.403 2 RRao=0.95 FSS= 1.06 FPC= 1.08 (d.f.=12 y 185) (d.f.=3 y 70) (d.f.=3 y 70) PR=.4910 PSS=0.372 PPC=0.365 3 RRao=0.39 FSS= 0.44 FPC= 0.45 (d.f.=15 y 602) (d.f.=6 y 219) (d.f.=6 y 219) PR=0.9807 PSS=0.851 PPC=0.843 4 FT2=0.77 tSS=0.75 tPC=0.76 (d.f.=6 y 68) (d.f.=73) (d.f.=73) PT2=0.5929 PSS=0.457 PPC=0.448 The p values in Standardized Sum and Principal Components tests were lower than in the standard analyses and the power were better too. A definitive response to the clinical variable measured in those studies were possible by used Standardized Sum or Principal Components test but not when we used classical analyses. These tests are useful for preclinical and clinical studies particularly when the sample size is small for standard analyses. Standardized Sum and Components Principal are methods of choise for studies of scaring and the studies of lesions of the acne, because the power is better than in classical analyses with the same sample size. REFERENCES Follmann, D. (1996). A simple multivariate test for one-sided alternatives. J. Amer. Statist. Assoc. 91, 854-861 Läuter J. (1996). Exact t and F Tests for Analyzing Studies with Multiple Endpoints. Biometrics 52, 964-970. Läuter J., Glimm, E. Kropf, S. (1996). New multivariate test for data with an inherent structure. Biom. Journ. 38, 5-23. O’Brien, P.C. (1984). Procedures for comparing samples with multiple endpoints. Biometrics 40, 1079-1087. Wei, L.J., Lachin, J.M. (1984). Two-sample asymptotically distribution-free tests for incomplete multivariate observations. J. Amer. Statist. Assoc. 79, 653-661. FRENCH RÉSUMÉ Nous avons fait un MatLab programme v.5.2. pour calculer la Somme Standardisée et les épreuves des Composants Principaux. Ces épreuves ont été appliquées dans quatre études preclinique et clinique avec l'objectif de comparer leurs résultats avec les analyses standardes. Nous avons proposé ces méthodes pour les études d'effrayer et les études de lésions de l'acné. Une réponse définitive au variable clinique mesuré dans ces études était possible par l’usage de la Somme Standardisée ou des Composants Principaux mais pas quand nous avons utilisé des analyses classiques. La Somme Standardisée et les Composants Principaux sont méthodes de choisir pour les études d'effrayer et les études de lésions de l'acné, parce que le pouvoir est meilleur que dans les analyses classiques avec la même dimension de l'échantillon.