Comparing Cox regression and parametric models for survival of patients with gastric carcinoma.

ABSTRACT

BACKGROUND: Researchers in medical sciences often tend to prefer Cox semi-parametric instead of parametric models for survival analysis because of fewer assumptions but under certain circumstances, parametric models give more precise estimates. The objective of this study was to compare two survival regression methods - Cox regression and parametric models - in patients with gastric adenocarcinomas who registered at Taleghani hospital, Tehran. METHODS: We retrospectively studied 746 cases from February 2003 through January 2007. Gender, age at diagnosis, family history of cancer, tumor size and pathologic distant of metastasis were selected as potential prognostic factors and entered into the parametric and semi parametric models. Weibull, exponential and lognormal regression were performed as parametric models with the Akaike Information Criterion (AIC) and standardized of parameter estimates to compare the efficiency of models. RESULTS: The survival results from both Cox and Parametric models showed that patients who were older than 45 years at diagnosis had an increased risk for death, followed by greater tumor size and presence of pathologic distant metastasis. CONCLUSION: In multivariate analysis Cox and Exponential are similar. Although it seems that there may not be a single model that is substantially better than others, in univariate analysis the data strongly supported the log normal regression among parametric models and it can be lead to more precise results as an alternative to Cox.