Unit:
Κατεύθυνση ΒιοστατιστικήLibrary of the School of Health Sciences
Author:
Apostolidis Apostolos
Supervisors info:
Ιωάννης Ντζούφρας, Καθηγητής, Τμήμα Στατιστικής, Οικονομικού Πανεπιστημίου Αθηνών
Κωνσταντίνος Γιαννούτσος, Καθηγητής, Department of Biostatistics and Health Data Science, Indiana University Indianapolis
Γιώργος Μπακογιάννης, Αναπλ. Καθηγητής, Department of Biostatistics and Health Data Science, Indiana University Indianapolis
Original Title:
A simulation study of the Bayesian Weibull competing risk model with missing cause of failure
Translated title:
A simulation study of the Bayesian Weibull competing risk model with missing cause of failure
Summary:
In survival studies, more than one cause of failure is a frequent phenomenon and
sometimes the cause of failure is missing under MAR assumption. Several
approaches are suggested some of them are Bayesian or frequentist, parametric
or semi-parametric methodologies. The Bayesian Weibull competing risk with
missing cause of failure is a model that has not been adequately described in the
bibliography. As a consequence, this master thesis first tries to efficiently describe
this model and second, compare it with an existing Maximum pseudo-partial
estimation method. This Bayesian Weibull competing risk with missing event type
model uses the Bayesian methodology to impute the missing cause of failure and
estimate the desirable coefficients-parameters. The imputation of the missing
cause of failure is conducted by treating those missing observations as parameters.
The form of the coefficients - parameters of the model are derived from a Weibull
competing risk model and they are estimated via a Bayesian methodology. The
Maximum pseudo-partial estimation method is a computationally efficient
method in which its coefficients are estimated by a weighted-probability Cox
regression model. The simulated data are derived from the statistics of the EA-
IeDEA HIV study which in this study a heavy under-reporting issue of the event
type is observed. The results of the simulation study indicate that both
methodologies effectively handle this issue when the assumptions of the models
are applied. Both the proportionality and the Weibull assumption significantly
impacts the validity of the results in the Bayesian Weibull model, but in case they
are not true, one can use accelerated time failure models instead of a Weibull one
to tackle this issue.
Main subject category:
Health Sciences
Keywords:
Bayesian, Competing, Survival, Risk, Missing, Simulations
