A thesis submitted to the Faculty of Science, University of the Witwatersrand, Johannesburg, in fulfilment of the requirements for the degree of Doctor of Philosophy. 21 October, 2014. / Three decades ago, the discovery of the Human Immunodeficiency Virus (HIV) was announced. The subsequent HIV pandemic has continued to devastate the global community, and many countries have set ambitious HIV reduction targets over the years. Reliable methods for measuring incidence, the rate of new infections, are essential for monitoring the virus, allocating resources, and assessing interventions. The estimation of incidence from single cross-sectional surveys using tests that distinguish between ‘recent’ and ‘non-recent’ infection has therefore attracted much interest. The approach provides a promising alternative to traditional estimation methods which often require more complex survey designs, rely on poorly known inputs, and are prone to bias. More specifically, the prevalence of HIV and ‘recent’ HIV infection, as measured in a survey, are used together with relevant test properties to infer incidence. However, there has been a lack of methodological consensus in the field, caused by limited applicability of proposed estimators, inconsistent test characterisation (or estimation of test properties) and uncertain test performance. This work aims to address these key obstacles. A general theoretical framework for incidence estimation is developed, relaxing unrealistic assumptions used in earlier estimators. Completely general definitions of the required test properties emerge from the analysis. The characterisation of tests is then explored: a new approach, that utilises specimens from subjects observed only once after infection, is demonstrated; and currently-used approaches, that require that subjects are followed-up over time after infection, are systematically benchmarked. The first independent and consistent characterisation of multiple candidate tests is presented, and was performed on behalf of the Consortium for the Evaluation and Performance of HIV Incidence Assays (CEPHIA), which was established to provide guidance and foster consensus in the field. Finally, the precision of the incidence estimator is presented as an appropriate metric for evaluating, optimising and comparing tests, and the framework serves to counter existing misconceptions about test performance. The contributions together provide sound theoretical and methodological foundations for the application, characterisation and optimisation of recent infection tests for HIV incidence surveillance, allowing the focus to now shift towards practical application.

The effects of gender and age on the behavioral risk of HIV/AIDS are not clearly understood as previous distinct studies which have been carried out, have given disputable and contradictory outcomes. This study therefore, discusses the statistical methods which can be used to model the influence of age and gender on the behavioral risk factors of HIV/AIDS. In general, generalized linear models are the main methods which can be applied to depict the impact of age and gender on the behavioral risk of becoming infected with HIV/AIDS virus. In this study, the main methods used were logistic regression, log-linear regression and multiple regressions. Behavioral risk was taken as the dependent variable while age, gender, number of sexual partners, religious beliefs and alcohol and drug abuse were fitted as predictor variables. The three statistical methods gave significant results for gender and insignificant results for age. Furthermore, comparisons were made on the three regression methods and the logistic regression gave the best results. It was therefore concluded that gender plays a significant role on the behavioral risk of HIV/AIDS. The results of the study showed that gender of the student and number of sexual partners had a significant effect on the risk behavior of the university students. In future, it may be very important to find out why age is not a significant factor on risk behavior of HIV/AIDS among university students.