Distribución del valor p para la prueba t en estudios de no-inferioridad con muestras paralelas

In randomized controlled clinical trials, treatment efficacy is considered as its capability to produce beneficial effects. Efficacy of an experimental treatment is usually established comparing its effectiveness respect to a placebo or comparing with the effect of an active control treatment. The first stage to determine effectiveness of a new treatment is placebo-controlled trial, and, as a second stage, a superiority trial will compare new treatment with a treatment uniformly accepted as a standard one. Sometimes the use of a placebo control is considered unethical; in such a case, a non-inferiority trial may be appropriate. In non-inferiority the new treatment dont need to be superior to a control, it is enough that it doesn't be unacceptably worse and additionally to be better than placebo. This assert rules the construction of non-inferiority hypothesis, which is based in a non-inferiority margin; in its determination is necessary to consider both statistical reasoning and clinical judgement. The possibility to claim incorrectly non-inferiority is one of the methodological flaws in non-inferiority studies and is related with the determination of non-inferiority margin. Two samples t-test is the statistical criteria to analyze non-inferiority hypothesis when mean value is the parameter to measure treatments efficacy. The objective of present report was to obtain and to characterize the p-value probability distribution for t-test in non-inferiority designs and to establish the relationship between this distribution and test's power function, as a result a characterization of the effect of sample size and non-inferiority margin over test sensibility was obtained, in addition an application of p-value distribution to meta-analysis was done introducing a criteria to analyze the effect of a particular study in results integration.