THE EVALUATION OF COX AND WEIBULL PROPORTIONAL HAZARDS MODELS AND THEIR APPLICATIONS TO IDENTIFY FACTORS INFLUENCING SURVIVAL TIME IN ACUTE LEUKEMIA

Introduction: The most important models used in analysis of survival data is proportional hazards models. Applying this model requires establishment of the relevance proportional hazards assumption, otherwise it world lead to incorrect inference. This study aims to evaluate Cox and Weibull models which are used in identification of effective factors on survival time in acute leukemia. Methods: In a retrospective study the information on 197 patients with acute leukemia were collected. The proportional hazards models were used to identify effective factors on survival time. Goodness-of-fit test and two graphical procedures based on the scaled Schoenfeld residuals were used to evaluate the proportional hazards assumption of the Cox and Weibull models. Finally, to assess the accuracy of fitted models, Martingel residuals graphic method was applied for goodness of fit investigation. S-plus statistical package was used for analysis. Results: The results showed that the assumptions of proportional hazards for Cox and Weibull models are correct. Based on the models, variables such as, age, residence place, and WBC, showed significant effect on survival time (P<0.05). Finally, AIC criteria showed that Cox semi- parametric model had better fit on the survival data of the patients with acute leukemia. Conclusion: Assessing the Cox-snell and Martingel residuals plots showed that Cox model is more efficient than Weibull model for the survival time of the patients with acute leukemia. Significance of WBC factor in the model, indicate that control of this factor is effective in increasing patients' survival.