Models de Competing Risks
Seminari GRASS - 8 de novembre 2006
DATE: 8 de novembre 2006 a les 11 hores
SPEAKER: Núria Porta
IDIOM: Català
PLACE: Aula C5016, Edifici C5, Campus Nord, UPC
SUMMARY:
In life testing experiments, the failure of an individual may be classified into one of k>1 mutually exclusive causes. Competing Risks models are used to model such data, in which a failure time T>0 and a cause of failure
C$\in${1,?,k} can be observed for an individual. In this presentation, an overview of different regression models for competing risks will be outlined.
Two big frameworks are found when it comes to these models. Firstly, models such as Proportional Hazards or Aalen?s Additive Hazards models, in which inference is based on cause-specific hazards. Secondly, sub-distribution modelling or pseudo-values regression modelling, among others, in which inference is based on cause-specific cumulative incidence functions. In a Competing Risks setting, cumulative incidence functions are more informative to explore relationship between distinct modes of failure, since they take into account the overall survival of the patient. Moreover, they directly estimate the magnitude of the proportion of patients suffering each cause-specific hazard, and thus often are the primary efficacy measures of interest. Modelling of the cumulative incidence function arises from the need to estimate the effect of covariates directly on them.
To complement the theoretical presentation, models will be fitted to data from 893 bladder cancer patients which are followed up to some failure occur. Failure may be due to recurrence of the tumour, progression of the tumour or death due to cancer. It is aimed to study the relationship with gender, age, multiplicity of the primary tumour at diagnose, and grade of the primary tumour.
SPEAKER: Núria Porta
IDIOM: Català
PLACE: Aula C5016, Edifici C5, Campus Nord, UPC
SUMMARY:
In life testing experiments, the failure of an individual may be classified into one of k>1 mutually exclusive causes. Competing Risks models are used to model such data, in which a failure time T>0 and a cause of failure
C$\in${1,?,k} can be observed for an individual. In this presentation, an overview of different regression models for competing risks will be outlined.
Two big frameworks are found when it comes to these models. Firstly, models such as Proportional Hazards or Aalen?s Additive Hazards models, in which inference is based on cause-specific hazards. Secondly, sub-distribution modelling or pseudo-values regression modelling, among others, in which inference is based on cause-specific cumulative incidence functions. In a Competing Risks setting, cumulative incidence functions are more informative to explore relationship between distinct modes of failure, since they take into account the overall survival of the patient. Moreover, they directly estimate the magnitude of the proportion of patients suffering each cause-specific hazard, and thus often are the primary efficacy measures of interest. Modelling of the cumulative incidence function arises from the need to estimate the effect of covariates directly on them.
To complement the theoretical presentation, models will be fitted to data from 893 bladder cancer patients which are followed up to some failure occur. Failure may be due to recurrence of the tumour, progression of the tumour or death due to cancer. It is aimed to study the relationship with gender, age, multiplicity of the primary tumour at diagnose, and grade of the primary tumour.
Share: