The general objective of randomized clinical trials is to assess if the treatment effect on a given population is clinically meaningful. In this way, one obtains an average estimate of the treatment effect over the trial population. However, a growing need for medical practitioners is to be able to predict with sufficient precision the efficiency of a given treatment for a given patient. There is little information in the literature about this issue. We have previously proposed a treatment-startified Cox model including interaction between treatment and patient's covariates, to identify and predict the responders to a therapy. In this paper, we focus on the assessment of the predictive power of the model. The performance of the predictive model for a population and for an individual was statistically validated internally and externally from several aspects. The prediction correlates well with the observation. Thus, we suggest that this approach would be useful in identifying and predicting the responders to a therapy, subject to an appropriate and more extensive validation process in real setting.