Evaluation of parametric and non-parametric survivalanalysis methods on real data set of cancer patients

Survival analyzes are methods used to reveal the risk of recurrence, death and prognostic factors affecting these risks. Kaplan-Meier and Cox Regression methods are the most preferred nonparametric methods in survival analysis. In this study, it was aimed to evaluate parametric and nonparametric survival analysis methods together in cancer patients. Analyzes were carried out using the R programming language using real data from patients with endometrial cancer, which were obtained from the database of Çukurova University Faculty of Medicine Balcalı Hospital. Firstly, survival curves according to the groups of risk factors (type of operation, age, risk group, comorbidity) were obtained by Kaplan-Meier method and compared through parametric / nonparametric (Logrank, Gehan-Wilcoxon, Tarone Ware, Peto-Peto generalized Wilcoxon and Fleming- Harrington) tests. To use parametric methods, it was determined whether the survival time fit one of the life distributions (Exponential, Weibull, Lognormal, Loglogistic, Generalized gamma) and whether parameters of the determined distribution were similar between the groups were also evaluated. It was found that the lifetimes of all categories of the risk factor groups of interest were different from each other with the parametric / nonparametric tests. Akaike and Bayesian information criteria (AIC, BIC) were utilized in order to select the most suitable model for the data in univariate and multivariate Cox and parametric models. As a result of the evaluation, it was determined that the coefficients obtained from the models were similar to each other. However, it was found that the proportional hazard assumption was not provided in the Cox model and this model has higher AIC and BIC values compared to the parametric models. It was determined that the loglogistic model had the best performance in determining the risk factors of endometrial cancer patients. Nonparametric survival analyzes are frequently used in the literature regardless of the assumptions. However, parametric survival analyzes are more powerful methods, although they are not resistant to hypothetical violations. Therefore, choosing the appropriate method will provide more accurate decision processes in medicine and a statistical perspective to researchers.