Back to search

National and Kapodistrian University of Athens

Statistical modeling of CD4 cell count evolution in HIV-positive individuals before and after antiretroviral treatment initiation under missing data due to various dropout mechanisms

Abstract

dc:description

In longitudinal epidemiological studies, the evolution of markers related todisease progression is often of great interest. However, longitudinal markerdata often suffer from the presence of missing data, frequently in the formof dropout. Missing data though can ultimately undermine inferences under certain cases. It has been shown that likelihood-based methods, suchas linear mixed models (LMMs) could provide unbiased results even afterignoring the dropout mechanism, given that the dropout probabilities depend on the observed marker values (MAR). When missingness dependson unobserved quantities (MNAR), joint modeling of the marker evolutionand the dropout process is generally required for unbiased estimates. In thisthesis, we address several issues arising due to incomplete data, motivatedby the epidemiology of HIV, focusing mainly on longitudinal modeling ofCD4 counts before/after treatment initiation.When modeling CD4 count trajectories during untreated HIV infection,CD4 counts are mainly censored due to treatment initiation, with the natureof this mechanism remaining debatable (MAR or MNAR). Several shared random effects models (SREMs), a specific subclass of MNAR joint models,have been fitted to such data. Motivated by this example, we analyticallyshow that specific SREMs, when fitted to data subject to certain MARmechanisms, can produce seriously biased marker rate of change estimates.In addition, we propose a more robust alternative SREM model that workswell under specific MAR and MNAR dropout mechanisms.Under MAR, CD4 cell counts during untreated HIV infection are usually modeled through LMMs with random intercept and random slope. We analytically show that using a random intercept and slope structure when thetrue covariance structure is more complex can lead to seriously biased estimates, with the degree of bias depending on the magnitude of the MARdrop-out. Under misspecified covariance structure, we compare, in terms ofinduced bias, the approach of adding a fractional Brownian motion (BM)process on top of random intercepts and slopes with the approach of using splines for the random effects. Moreover, to discriminate between theexamined approaches in real-data applications, we adopt Bayesian modelcomparison criterion based on the posterior model probabilities.When modeling CD4 counts after treatment initiation, dropout can occurdue to death in care or disengagement from care, which are competingevents. Death is usually considered to correspond to MNAR dropout thoughthe nature of the mechanism of disengagement from care is less clear. Thissetting calls for joint modeling. We propose a flexible class of SREMsto jointly model the marker evolution and multiple causes of failure usingcumulative incidence functions (CIFs) in the survival submodels, with theCIFs depending on the "true" marker value over time. The fact that theall-cause CIF should be bounded by 1 is formally considered. The proposedmodels are extended to account for potential failure cause misclassificationthrough double sampling. We also provide a multistate representation ofthe whole population by defining mutually exclusive discrete states basedon the "true" marker values and the competing risks. Based solely on theassumed joint model, we derive fully Bayesian inference for state occupationand transition probabilities.

Degree

thesis:*
Grantor dc:publisher
National and Kapodistrian University of Athens
Year dc:date
2022

Author and committee

dc:creator, dc:contributor.*
Authors dc:creator
  • Thomadakis, Christos
  • Θωμαδάκης, Χρήστος

Subjects

dc:subject × 22

Rights

Language dc:language
eng

Identifiers

dc:identifier.*
Identifier
10.12681/eadd/51791

Chain of custody

source
Harvested from
Greek National Archive of PhD Theses
Base URL
phdtheses.ekt.gr/eadd_oai/request
Last updated
2026-11-18
Source record
OAI-PMH GetRecord
citation

Thomadakis, Christos; Θωμαδάκης, Χρήστος. Statistical modeling of CD4 cell count evolution in HIV-positive individuals before and after antiretroviral treatment initiation under missing data due to various dropout mechanisms. National and Kapodistrian University of Athens, 2022. http://hdl.handle.net/10442/hedi/51791