|dc.description.abstract||Longitudinally measured data and time-to-event or survival data are often associated
in some ways, and are traditionally analyzed separately (Asar et al., 2015).
However, separate analyses are not applicable in this case because they may lead to
inefficient or biased results. To remedy this, joint models optimally incorporate all
available information (longitudinal and survival data) simultaneously (Wulfsohn &
Tsiatis, 1997). Furthermore incorporating all sources of data improves the predictive
capability of the joint model and lead to more informative inferences for the purpose
of decision-making (Seyoum & Temesgen, 2017). The primary goal of this analysis
was to determine the effect of repeatedly measured CD4 counts on mortality. The
standard time-to-event models require that the time-dependent covariates of interest
are external; where the value of the covariate at a future time point is not affected by
the occurrence of the event. This requirement would not be fulfilled in this setting,
since the repeatedly measured outcome is directly related to the mortality mechanism.
Hence, a joint modeling approach was required.
We applied the methods developed in this thesis to the CAPRISA AIDS Treatment
program (CAT). We also sought to determine if the patients’ baseline BMI (Body
mass index), baseline age, gender, baseline viral load, baseline CD8 count, baseline
TB status and clinic site, influence the evolution of the CD4 count over time. Various
linear mixed models were fitted to the CD4 count, adjusting for repeated measurements,
as well as including intercept and slope as random effects. Different types of
covariance structures were assessed and the spatial spherical correlation structure
was found to be the best fit. The Cox PH model was employed to model mortality.
Finally the joint model for longitudinal and time-to-event data was fitted.
Out of the 4014 patients, 1457 (36.30%) were male. There were more patients presenting
without TB at ART initiation, 3042 (75.78%) compared to those with prevalent
TB, 972 (24.22%). Results from the multivariable random effects model showed
that the patients gender, age, baseline viral load and baseline CD8 cell count had
statistically significant influences on the rate of change in CD4 cell count over time. The un-adjusted and adjusted hazards regression both found CD4:CD8 ratio, viral
load, gender and age of patients to be significant predictors of mortality. The result
from the joint model in this study indicated that CD4 count change due to HAART
and mortality had been influenced jointly by gender, age, baseline viral load, baseline
CD8 count, time (in years) , CD4:CD8 ratio and by the interaction effects of time
(in years) with TB status, baseline viral load and baseline CD8 cell count. CD4 count
proved to be significantly associated with mortality, after adjusting for age, gender
and other potential confounders
Model diagnostics were performed for validating model assumptions, and our joint
model fitted quite well with fairly good diagnostic attributes. The methods that were
developed in this thesis were applied to the CAPRISA AIDS Treatment program
(CAT) between June 2004 to December 2013.||en_US