Supervisors info:
Χρήστος Θωμαδάκης, Διδάκτωρ, Ιατρική Σχολή, ΕΚΠΑ
Γιώτα Τουλούμη, Καθηγήτρια. Ιατρική Σχολή, ΕΚΠΑ
Λουκία Μελιγκοτσίδου, Αναπληρώτρια Καθηγήτρια, Τμήμα Μαθηματικών, ΕΚΠΑ
Summary:
Through this thesis, we attempted to model the rate of clinical visits for people with HIV in Greece under combination antiretroviral therapy. For the statistical analysis of data, we relied on the notion and fit of calendar and waiting time models, which form the two main methodologies for handling of recurrent events. At a first stage, we described the necessary clinical background for the examined disease, focusing on all the basic concepts and parameters which were finally utilised in the context of the conducted analysis. After that, we utterly developed the theory of the used methodologies, starting from simple assumptions of the corresponding modeling and reaching to extensions of those, which permit the application of these specific approaches for the description of quite complex problems. More specifically, the methodology of calendar time, was established based on the Poisson counting processes, having as its main goal the modeling of the rate of events. On the other hand, the methodology of waiting times was based upon the theory of renewal counting processes, with the corresponding modeling focusing on the hazard of a new event (e.g clinical visit). For both approaches, the basic regression equations were given, derived from the semi-parametric Cox model, while some generalisations were also examined such as the use of stratified analysis or the insertion of subject-specific random effects. The utilised data were taken from the electronic database of the multi-centre AMACS study, which includes registered information for people with HIV from large hospital units in Greece. The studied outcome was the realisation of clinical visits without any content limitation, the times of which were examined according to the calendar scale (time distance from the first individual therapy) as well as the one coming from the formed gaps (time elapsed since the previous visit). For the better evaluation of the studied events' frequency, time-fixed and time-varying factors were included, with their choice being based on the expected association and the available literature. Through the application of various models and the conduction of diagnostic tests, we concluded that the frailty ones are the most suitable for the description of the complex visit processes. The above, lead to basing our final inference according to the results of this specific modeling. From the explanatory variables chosen, the CD4 cell count and the calendar time of the first treatment showed statistically important negative association with the studied outcome. In the opposite direction was the association of the clinical AIDS diagnosis, while the individuals that were infected through drug injection showed reduced visit frequency compared to those that were infected through sexual intercourse. Moreover, the existence of high viral load resulted in conducting clinical visits more frequently than having low or non detected one, while the same was observed for older individuals (at ART initiation) compared to younger ones. Lastly, we compared the two used methodologies based on their fit to the data and concluded that the waiting time approach greatly surpasses that of the calendar time. The assignment was completed through the summary of the main results and their comparison with corresponding ones from other studies, the reference of analysis strong points and weaknesses and the commenting on the study's external validity.
