Unit:
Κατεύθυνση ΒιοστατιστικήLibrary of the School of Health Sciences
Author:
Αντωνακόπουλος Αθανάσιος
Supervisors info:
Αναπληρώτρια Καθηγήτρια Τουλούμη Παναγιώτα, Επίκουρος Καθηγητής Σιάννης Φώτιος, Διδάκτωρ Βιοαστατιστικής Πανταζής Νικόλαος
Original Title:
Μεροληψία και ακρίβεια εκτιμήσεων σε διαχρονικές μελέτες επαναλαμβανόμενων μετρήσεων με ενδιάμεσες ελλείπουσες τιμές
Summary:
Introduction: HIV type-1 infection is a major problem for public health. With
the recent introduction of Highly Active Anti-Retroviral Treatment (HAART) HIV
infection is now considered a chronic disease. The main categories of HAART are
schemas that contain Boosted Protease Inhibitors (Boosted PIs) and those that
contain Non Nucleoside Reverse Transcriptase Inhibitors (NNRTIs). Nevertheless,
considering the large number of contemporary available antiretroviral regiments
and possible combinations of them and the relative lack of Clinical Trials,
optimal treatment of such patients is still under debate.
Nowadays, due to improved survival of patients with HIV infection, surrogate
markers are used to test the effectiveness of treatment, the most important of
the them being the number of CD4 leukocytes (CD4 count) and viral load. The
existence of missing values of these markers in longitudinal studies is a major
problem, leading to bias or affecting the precision of estimates.
Aim: The aim of this study is to estimate the impact of the two above mentioned
major types of HAART, as well as other factors such as age, sex, prior exposure
to antiretroviral treatment, viral load etc. on longitudinal changes of CD4
counts of patients with HIV infection.
At the same time, the effect of intermittent missingness on bias and precision
of estimates is examined.
Material: Data were obtained from the database of the Greek multicenter AMACS
study. Participants aged more than 15 years and treatment initiation was with
one of the two aforementioned types of HAART after 1/1/1999. The time frame
that is examined for each patient is from HAART initiation until HAART
interruption or change of HAART type that indicates the artificial cencoring of
patient’ s data. Total sample consisted of 1161 cases.
Methods: The average interval between two concecutives CD4 counts was
calculated for each patient. This synoptic statistic was considered to reflect
intermittent missingness as the study design had no common time points for
patients’ follow up. Patients were categorized according to the aforementioned
synoptic statistic. Relationship of average interval between CD4 counts and
follow up time and (by design) censoring of patients’ follow up is examined
using Kaplan Meier survival estimates.
Random effects mixed linear models are used for assessing longitudinal exelixis
of CD4 counts. Root transformation of CD4 count is used, while time is
introduced as a first order linear spline with one breakpoint. The effect of
average interval between two consecutive counts on estimates is examined by
introducing this variable in the model.
Results: Patients with long average interval between CD4 counts had longer
follow up time. Moreover, they were significantly less likely to interrupt
treatment (for any reason) or to interrupt treatment due to ineffectiveness
(p-value<0.001 for both comparisons, log-rank test).
Adjusting for all other variables, Boosted PI HAART leads to faster increase of
CD4 count for the first 6 months (p-value<0.001). Beyond that time point there
is no difference between Boosted PI and NNRTI (p-value:0,578).
Strata of average interval between CD4 counts are significantly different as
far as longitudinal changes of CD4 are considered. Nevertheless, introduction
of this variable in the model has minimal impact on the estimates for other
factors and certainly does not affect the sign nor the significance of
estimates.
Conclusions: Introduction of the synoptic statistic reflecting intermittent
missingness in the model had minimal effect on statistical inference. The
disadvantage of the models used is the strong assumption of data Missing At
Random (MAR), with the possible advantage of better specification of the model.
It should be noted that the variable used to describe intermittent missingness
is also associated with (by design) drop-out missigness.
Keywords:
HIV, HAART, Longitudinal data, Mixed models, Missing values
File:
File access is restricted only to the intranet of UoA.
document.pdf
5 MB
File access is restricted only to the intranet of UoA.
