Abstract:
Although remarkable decrease in the number of HIV related morbidity and mortality was recorded since the start of ART, exploring the effects of different factors determining survival of patients plays principal role So as to enhance ART. In such cases, survival analysis is the most appropriate method and its parametric version yields more powerful estimates than Cox PH provided that, distributional assumptions satisfied. Hence, the performances of Cox PH model and different parametric Models in both metric (PH and AFT) were compared on 3042 eligible subjects followed for at least six years using secondary data obtained from University of Gondar Hospital, ART database. Model specific stepwise variable selection procedures were performed on 11 Covariates that appeared to be significant by Non-Parametric and independent analysis to fit models with the corresponding set of covariates resulting possibly minimum loss of information. Based on the statistical information criteria, the Generalized Gamma AFT Model provided best fit to the dataset as compared to Cox PH, Exponential, Weibull, Gompertz, Log-Logistic and Log-Normal Models. Accordingly, advanced WHO clinical stages (III and IV), lower CD4 percent (12-15%), TB co-infection, being bedridden or ambulatory functional status, being relatively old in age, having relatively lower weight, the presence of Opportunistic Infections and Risky Behaviors were strongly related to relatively minimum median survival time (accelerated death time) under the Generalized Gamma AFT Model. Gender, Household Size and Educational Level were not selected to fit final Generalized Gamma AFT Model, although they were significant by independent analysis unlike Occupational Status, Marital Status and Cotrimoxazol that were insignificant at all. Statistical software, SPSS and STATA were used for analysis.