The aim of this thesis is to present analyses related to the effect of Highly Active Antiretroviral Therapy (HAART) on the natural history of HIV disease after its introduction in the mid- to late-1990s. Analyses have been carried out predominantly on two large observational studies, the Italian Cohort of patients Naive from Antiretrovirals (I.Co.N.A.) and the UNI-clinic of Frankfurt. Several possible ways of defining virological outcomes have been described. Particular importance has been placed on the analytical problems encountered with missing viral load data. A specific approach modelling the time to therapeutic failure was compared with other commonly used approaches and its advantages and disadvantages have been discussed. A number of these methods were then used to focus on several HAART-related issues. The virological and immunological response to HAART was evaluated in the two cohorts, a general description of these responses was given and relevant predictors were identified. More specific analyses have been carried out to describe the limitations of HAART due to toxicity, to describe the prognostic value of viral load measured at 4 and 8 weeks after starting HAART in predicting the week 24 virological response, and to investigate the prevalence of drug-resistance and its role in predicting virological failure. In addition, analyses have been carried out to describe the frequency and predictive value of low level viral rebounds, to investigate the best time to start HAART, and to compare specific drug regimens. Methodologically, this thesis demonstrates the flexibility and usefulness of survival analysis when used in the context of a "time to therapy failure" approach. All the analyses presented emphasise the value of observational studies in providing data that have important implications in clinical practice and for the treatment of people infected with HIV.