Comparison of Survival analysis of Gastrointestinal Cancer patients using parametric and Cox models

The most common cancer in Iran is digestive system cancer, the highest incidence of which is reported from the Caspian Sea littoral. This study aimed to determine factors affecting the survival of patients with gastrointestinal cancer using the Cox and parametric models; the 2 models were compared. This survey was a prospective study conducted between 1990 and 1991. Data were collected through the Cancer Registry Center in Babol, which functions under supervision of the School of Public Health and Institute of Public Health Research, Tehran University of Medical Sciences. We tried to identify cases of gastrointestinal cancers. The individual characteristics of 484 patients, namely, age, sex, family history, marital status, smoking status, occupation, ethnicity, medication status, education, residence (urban, rural), and type of cancer were recorded. The patients were followed up for a period of 15 years, i.e., until 2006 year. To determine the effective factors on survival of patients, the Cox model and parametric models such as exponential, weibull, log-normal, log-logistic, and the AIC criteria and residuals were used to compare the effectiveness of the models. The SAS and STATA software were used for data analysis, with a significant level of 0.05. Sixty-six percent of the patients (total n=484) were males and 34% females, with a mean age of 59 and 55 years, respectively. Their distribution according to type of cancer was as follows: esophageal cancer, 359 (74.2%); stomach cancer, 110 (22.7%); colorectal cancer, 15(3.1%). Estimated one-, three-, and five-year survival rates were 24%, 16% and 15%, respectively. The results of this study reveal that gender and family history can be strong risk factors for GI cancer. Log-normal and log-logistic models in multivariate and univariate analyses gave almost similar results. However, based on AIC criteria and residuals analysis, the log-logistic model gives the best fit as compared to other parametric models and can be used instead of the Cox model for determining factors affecting survival of patients with gastrointestinal tract cancer.