| dc.description.abstract | Multivariate or cluster failure time data are common in survival analysis and
nding an appropriate method to model the correlation among the observations
is a very important issue for valid and reliable statistical inference. The primary
objective of this project was to review various models for clustered survival data
with focus on frailty models and their properties. Semi parametric Cox marginal
and frailty models were used to analyze observed right censored data from a
multicenter clinical trial. A simulation study was conducted to assess the impact
of frailty distribution mis-speci cation on parameters estimates. Di erent settings
in terms of the number of centers and true heterogeneity parameter were
considered.
From the observed data, the estimated heterogeneity parameters were small
yielding insigni cant center e ect. From the simulation study, the regression coe
cient was less a ected by mis-speci cation of the frailty distribution and initial
simulation settings compared to the heterogeneity parameter. In conclusion, in
the absence of center e ect, event times were homogenous between and within
the centers. From simulation study, gamma frailty model would be a practical
choice in real data analysis with time to event endpoint when the regression parameters
are of primary interest and when the choice of frailty distribution is not
straightforward.
Key | en_US |
