Journal of School of Public Health and

  Background and Aim: In many diagnostic studies, including surveying the survival of patients with gastric cancer where each individual after surgery can experience more than one type of event, and the occurrence of one type of event hinders the occurrence of other types of events, the question of competing risk is raised. For checking the effect of each covariate on the occurrence of any event and estimating the hazard function, Cox and Fine and Gray models are used. In the event that the assumptions of two models do not hold, using them will be an incorrect course of action. One way to overcome this problem is to use models that have higher flexibility.

  Materials and Methods: In this study, the demographic, clinical and therapeutic characteristics of 330 patients with gastric cancer who referred from January 1996 to April 2000 to the Cancer Institute of Iran Imam Khomeini Hospital and underwent surgery, including their type, and the time of occurrence of the first event (locoreginal replace/death) for each patient from medical records were collected and evaluated. Using this information, the cumulative hazard function of relapse of disease was plotted by means of three models Cox, Fine and Gray and the flexible one, and was checked against the observed cumulative incidence function of recurrence of disease and, finally, their performance was evaluated.

  Results: Nearly, for each event, the proportionality assumption holds for all the variables . According to the graph of cumulative incidence function for the event of interest (recurrence), it can be seen that the Cox model, has overestimated the cumulative incidence function and the curves of two other models are very similar and also similar to the observed curve. However, the cumulative incidence function of the flexible model is smoother than the others.

  Conclusion: In the competing risk framework, Cox model is not very useful in practice while it seems that the flexible model is not only a good alternative to the Fine and Gray model but will also be superior to it when the assumption of proportionality does not hold.