Evaluation of parametric models by the prediction error in colorectal cancer survival analysis

Aim: The aim of this study is to determine the factors influencing predicted survival time for patients with colorectal cancer (CRC) using parametric models and select the best model by predicting error’s technique. Background: Survival models are statistical techniques to estimate or predict the overall time up to specific events. Prediction is important in medical science and the accuracy of prediction is determined by a measurement, generally based on loss functions, called prediction error. Patients and methods: A total of 600 colorectal cancer patients who admitted to the Cancer Registry Center of Gastroenterology and Liver Disease Research Center, Taleghani Hospital, Tehran, were followed at least for 5 years and have completed selected information for this study. Body Mass Index (BMI), Sex, family history of CRC, tumor site, stage of disease and histology of tumor included in the analysis. The survival time was compared by the Log-rank test and multivariate analysis was carried out using parametric models including Log normal, Weibull and Log logistic regression. For selecting the best model, the prediction error by apparent loss was used. Results: Log rank test showed a better survival for females, BMI more than 25, patients with early stage at diagnosis and patients with colon tumor site. Prediction error by apparent loss was estimated and indicated that Weibull model was the best one for multivariate analysis. BMI and Stage were independent prognostic factors, according to Weibull model. Conclusion: In this study, according to prediction error Weibull regression showed a better fit. Prediction error would be a criterion to select the best model with the ability to make predictions of prognostic factors in survival analysis.