Evaluation of parametric models with estimating the prediction error by apparent loss method in analyzing survival of colorectal patients

Colorectal cancer is the most common gastrointestinal cancer. Investigating the factors that predict survival time for these patients is important. The purpose of this study was comparison of parametric models by estimating the prediction error and also identifying the effective factors on predicted survival time of patients with colorectal cancer. This cohort study was conducted with 600 patients who were suffered from colorectal cancer in Taleghani Hospital of Tehran between 2001 to 2005 and they were followed up for at least 5 years. For identifying the effective factors on survival time، of the patients we analyzed the data by some parametric models such as Weibull، Exponential and Log logistic and compared these models with the estimation of prediction error by apparent loss method. Among 600 patients there was 344 men (57. 3%) and 256 wemon (47. 7%). Of total، 151 patients were died that 62. 3% of them were men. Univariate analysis showed that the effect of BMI، sex، staging of tumor، tumor site were significant but in multivariate model staging of tumor and BMI were significant. By the estimation of prediction error، the best model was Log logistic. With respect to the importance of survival time prediction، we found that we can use the prediction error to compare the parametric models. In addition، because of effectiveness of tumor stages and BMI in the patients’ survival time، survival time could be increased by an on-time diagnosis and an appropriate controled diet.