Incomplete multiresponse designs and surrogate endpoints in clinical trials (Q1338364)

The primary objective of the current study is to focus on the design and analysis aspects of a general class of incomplete multiresponse designs, adoptable in clinical trials, which incorporate an appropriate surrogate variate to enhance statistical information without escalating (much) the cost of experimentation. The conventional surrogate model (incorporating a validation sample) considered by \textit{M. S. Pepe} [Biometrika 79, 355- 365 (1992; Zbl 0751.62049)] can be characterized as a special case of an incomplete multiresponse design (IMD) wherein one has only one primary variate and a surrogate variate; moreover, only a completely randomized design (treatmentwise) is envisaged in her work. In Section 2, we formulate an IMD incorporating a surrogate variate in a more general setup, and study various aspects of such a design. In spite of the isomorphism of design aspects, there is a basic difference between a conventional linear model and a so-called regression model in clinical trials. This is apparent from the respective formulation of statistical functions on which inference is to be drawn in a valid and yet reasonably efficient manner. This may generally limit the scope of applicability of standard parametric procedures in such IMD with surrogate variates. In that respect, semiparametric models fare well, and in Section 3 we provide an outline of both such procedures. Our main concern is to develop appropriate nonparametric procedures which may not need the basic (and somewhat stringent) regularity conditions inherent in a parallel semiparametric model. We propose to defer most of these technicalities to a subsequent communication. However, for the sake of illustration, in Section 4, we consider a simple IMD (with one primary variate and a surrogate variate), and discuss the related nonparametrics in some detail.