Modelling survival in HIV cohorts with applications to data from Zomba, Malawi

The Human Immunodeficiency Virus (HIV) pandemic still remains a major public health concern worldwide. The World Health Organization (WHO) estimates that approximately over 70% of people living with HIV in the world are in sub-Saharan region. Malawi is one of the worst affected countries in sub-Saharan Africa with prevalence reaching up to 16% in some areas. Recent study reports, largely in Africa, comparing outcomes for HIV patients with Kaposi’s sarcoma (HIV/KS) and HIV patients without KS indicate poor prognosis and poor health outcomes amongst HIV patients with KS. While efforts are being made to improve the management and care for the HIV/KS patient group, there is also need for continued efforts to better understand the survival patterns in this patients. The work presented in this thesis attempts to investigate the survival patterns in different patient subgroups in HIV cohorts in Malawi by using advanced and novel statistical techniques with an ultimate aim of informing targeted patient treatment and management practices. In this thesis, we aim to address the following four objectives; (1) to identify risk factors for mortality among HIV patients diagnosed with Kaposi’s sarcoma during routine initiation of ART, (2) to model the survival pattern among HIV patients diagnosed with KS, (3) to model local geographical variations in survival among HIV patients on ART, (4) to quantify transition dynamics in HIV and TB co-infection using multi-state modelling. For the first two objectives, we considered extended Cox models and parametric models. We also used a novel approach of accounting for high attrition in cohorts in which we used a ’gold-standard’ data to compare survival in our cohort. Sensitivity analyses indicated consistencies in our approach providing an insight into how model results change when using this comparison approach. Overall We noted an early mortality with most patients dying in the first five months after starting HIV treatment. Patients with TB and the patients who started in the early era of ART were significantly at risk of dying. The model diagnostics indicated that (i) a random effects Cox/Log-Gaussian frailty model and (ii) a flexible parametric proportional hazards model, describe the risk of mortality in the HIV/KS patients well. For the third objective, spatial survival models were considered. The study showed existence of possible residual spatial variation in survival after adjusting for age, sex, KS status, TB status and unobserved individual frailties. To further aid our understanding, we used the choropleth maps to indicate areas with substantially high probability of mortality risk at different cut-off values. These results highlight the local geographical variations in survival in HIV populations, an element more often ignored in most studies on HIV data. For the last objective, we considered the homogeneous continuous time multistate Markov models. In this study we found that patients in TB free status had a relatively higher probability of transitioning to being diagnosed with TB compared to dying while in TB free status. However, the cumulative transition hazards for the ’TB free ! death’ transitions compared to the "TB free ! TB infection" transitions were only higher during the early days of HIV treatment. This result emphasize how early periods after starting HIV treatment is crucial to ensure better prognosis. We also noted significant gender differences in the ’TB-free ! death’ transitions. It is anticipated that the findings in this thesis will help to inform treatment and management practices of HIV patients. The findings provide clear outcome pathways taken by HIV/TB patients before experiencing a terminal outcome. More importantly, the findings could help inform policies aimed at improving overall survival in HIV cohorts by establishing targeted patient management and treatment strategies and also formulating a more efficient triage system for care and treatment of particular group of patients