Unit:
Postgraduate Programme Biostatistics & Health Science DataLibrary of the School of Health Sciences
Author:
Oikonomidis Alkibiadis
Supervisors info:
Γ. Τουλούμη, Καθηγήτρια, Ιατρική, ΕΚΠΑ
Φ.Σιάννης, Επικ. Καθηγητής, Μαθηματικό,ΕΚΠΑ
Β.Σύψα,Επικ. Καθηγήτρια, Ιατρική, ΕΚΠΑ
Original Title:
The use of different timescales in multiple event survival analysis with application to care adherence in HIV-positive patients of the greek cohort study AMACS
Translated title:
The use of different timescales in multiple events survival analysis with application to care adherence in HIV-positive patients of the greek cohort study AMACS
Summary:
The current thesis has three aims. One is to provide the chance of getting familiar
with the management of large databases the second is to determine the effect of the
timescale that is used in correlated event data and the third is to link factors to patient
attrition from HIV monitoring and treatment. For that reason, two different timescales; gap
and calendar timescale and a handful of models are implemented, in order to determine
which timescale is more appropriate.
Data may be correlated due to multiple events of the same subject. In the case
where the study's subjects are experiencing the same type of event multiple times, we
refer to the data as recurrent event data. In the analysis of such data, correlation among
events and heterogeneity of subjects may present simultaneously in many situations.
Therefore we need to find a way to model the association within the observations of a
cluster (subject) as well the variance between different subjects (clusters). Variance
corrected and shared frailty models can be used for that reason. We also consider the
conditional shared frailty model in order to model both correlation and heterogeneity with
the use of event-based baseline hazards and random effect. All models were fitted to a
dataset of HIV-positive patients in Greece provided by AMACS.
Our simulations showed that the simple frailty models performed slightly better than
all other models that were fitted as the conditional frailty models were quite unstable.
However the fittings on the empirical data pointed to the opposite direction, as the
conditional frailty model under the calendar timescale seemed to be the most appropriate
way to assess subject heterogeneity as well as event dependence.
Main subject category:
Health Sciences
Keywords:
Unobserved heterogeneity, Frailty, Variance corrected, Correlated survival data, Calendar timescale, Gap timescale, Event dependence, Repeated events, Recurrent events
