Survival analysis and prognosis: A comparison of three approaches (for the analysis of nasopharix data)

In the present paper we consider the problem of estimating the lifetime distribution S, on the basis of data related to patients affected by tumor disease. Each element of the sample is then characterized by a dependent random variable Y (lifetime) and a set of random variables X = {X1, X2,..., XN} related to age, tumor morphology, amount of radiotherapy and so on. The aim of the analysis is finding an estimate of S(Y{plus 45 degree rule}X) which allow forecasting on the basis of the information given by X. We consider in depth an application to a set of nasopharinx cancer data, collected by the Istituto Tumori "Regina Elena" of Rome (Italy) and report the results obtained by different nonparametric and semiparametric models (Cox model, Nelson-Aalen estimation, Bayes dynamic estimation). A comparison of the results, obtained by the different methods, is considered to find out the peculiarities of each one. This allows using them as a joint procedure of forecasting, in contrast with the usual interpretation as competing, not complementary, methods. © 1990.