This paper is concerned with methods for choosing optimal treatment fo
r particular patients; the central idea is that information contained
in covariates may be useful for this kind of decision making. For each
patient estimated conditional survivor functions for each mode of tre
atment may be compared and the optimal treatment may be defined as the
one providing higher predicted survival. Regression survival models f
orm the basis for a pertinent statistical decision rule, and a procedu
re for retrospective validation of such a rule is given. The usefulnes
s of the proposed approach is illustrated with an application to the t
reatment of patients with cancer of cervix uteri. The results of data
analysis suggest that the choice of optimal treatment based on covaria
te information is of considerable promise for practical use.