Exposició GRASS

10:00

Anna Febrer: Graphical and analytical goodness-of-fit methods for parametric survival models with right-censored data.

Resum: Analyzing survival data with parametric models has its advantages and disadvantages. The advantages are that the estimation is easier and estimated survival curves are smoother as they draw information from the whole data. However, the main disadvantage of parametric methods is that they require extra assumptions that may not be appropriate, and the choice of an inappropriate model can lead to incorrect results. Some methods have been developed to assess the goodness of fit of a certain model to the data. But most of them can only be applied to complete data. The aim of this project is to do a bibliographic search and present some graphical and analytical methods to test goodness-of-fit of parametric models for right-censored data.

The introduced graphical tools to validate goodness-of-fit are mostly probability plots; such as the P-P plot, the Q-Q plot, the stabilized probability plot (S-P plot), the empirically rescaled plot (E-R plot) and the cumulative hazard plot. The functions ‘fitdistrcens’ and ‘cdfcompcens’ of the R package ‘fitdistrplus’ are also included in this work since they are useful for estimate the parameters of the model and can plot a comparison between the Kaplan-Meier survival estimator and some parametric survival estimators. Moreover we create an R function that reminds the S 6-plot function. Our function create a 3x3 grid in order to compare the cumulative hazard plots of nine preset distributions. This distributions are the Weibull, Gumbel, Normal, Log-normal, Log-logistic, beta, exponential power and exponentiated Weibull.

About analytical methods, we will explain the idea of how the statistics are constructed and which distribution they follow. This part includes a method to assess goodness-of-fit based on the hazard rates (Hjort, 1989), a method based in the likelihood ratio and only valid for grouped data (Turnbull and Weiss, 1978), a method that consist in embedding the model in a parametric family and using score test (Gray and Pierce, 1983) and a method that construct an statistics based on Hoeffding’s maximum correlation that can be applied only to censored data of Type I or Type II (Grané, 2003).

11:00
Assajos congrés Turquia:

Klaus Langohr: Fhtest: An R Package For The Comparison Of Survival Curves With Censored Data.

Nuria Pérez: Disability-Adjusted Life Years (Daly) For HIV-Infected Patients In Spain: Individual Analysis For The Burden Of Disease.

Moisés Gómez: Selecting The Primary Endpoint In A Randomized Clinical Trial. A Cardiovascular Case Study.