New Approach to a Parametric Regression in Survival Analysis
| Authors | |
|---|---|
| Year of publication | 2013 |
| Type | Conference abstract |
| MU Faculty or unit | |
| Citation | |
| Description | Parametric methods are not used so often due to a necessity to estimate the baseline hazard function. It is required to specify the probability distribution of the survival times when dealing with the methods. Unlike the Cox semiparametric model the assumption of proportional hazards need not to be fulfilled. That makes the parametric methods useful in some special cases. Exponential, Weibull, gamma and lognormal distributions are the most frequently used in the methods. All of this distributions are defined to infinity that causes overestimation of the probability of survival in longer survival times. We decided to transform standardly used distributions to be defined in a finite interval. The newly obtained distributions were formated and applied to the data of patients with breast cancer in the fourth stage of the disease. The maximum likelihood method was used to estimate the distributions parameters. Various models were compared visually and on the basis of Akaike information criterion. |
| Related projects: |