Alternative for the Cox Regression model: using Parametric Models to Analyze the Survival of Cancer Patients

Background: Although the Cox proportional hazard regression is the most popular model for analyzing the prognostic factors on survival of cancer patients, under certain circumstances, parametric models estimate the parameter more efficiently than the Cox model. The aim of this study was to compare the Cox regression model with parametric models in patients with gastric cancer who registered at Taleghani hospital, Tehran, Iran.

Methods: In a retrospective cohort study, 746 patients with gastric cancer were studied from February 2003 through January 2007. Gender, age at diagnosis, distant metastasis, extent of wall penetration, tumor size, histology type, tumor grade, lymph node metastasis and pathologic stage were selected as prognosis , and entered to the models. Lognormal, Exponential, Gompertz, Weibull, Loglogistic and Gamma regression were performed as parametric models ,and Akaike Information Criterion (AIC) were used to compare the efficiency of the models.

Results: Based on AIC, Log logistic is an efficient model. Log logistic analysis indicated that wall penetration and presence of pathologic distant metastasis were potential risks for death in full and final model analyses.

Conclusion: In the multivariate analysis, all the parametric models fit better than Cox with respect to AIC; and the log logistic regression was the best model among them. Therefore, when the proportional hazard assumption does not hold, these models could be used as an alternative and could lead to acceptable conclusions.