Introducing frailty models as a random effect model for a pooled trial analysis

Abstract

Cox proportional hazard models are the most popular way to analyze survival data. Heterogeneity in survival outcomes of cancer patients in a dataset influence the shape of mortality rate observed. Frailty models provide a way to investigate and to describe this variation.

The main aim of this study was to compare different extended Cox models that try to capture the heterogeneity in a pooled dataset and to assess the robustness of the models comparing their estimates, confidence intervals and p-value and their contribution in explaining the heterogeneity.

Secondly, we wanted to investigate whether the heterogeneity can be captured using different frailties terms. Unobserved heterogeneity might be derived from the patient population, treatment protocols etc. and not as such from the differences in cancer site alone.