Efficiencies of Experimental Designs for an Illness-Death Model

Abstract

Disease development is described by the progressive Markov illness-death model with one disease. The efficiencies of four experimental designs are discussed: the simple survival experiment, serial sacrifice, periodic diagnosis and complete observation. The discussion is based on the 'semiparametric' assumption of piecewise-constant intensities. Simple survival experiments are shown to have very low efficiency, serial sacrifice experiments to be moderately efficient, and periodic diagnosis to be almost as efficient as complete observation. An application to the analysis of the development of leukaemia in laboratory mice is given.