As HAART scale-up has progressed, major challenges inherent in following large numbers of patients in resource-limited settings have emerged. One problem relates to keeping patients in care, and published rates of loss to follow-up after HAART initiation in antiretroviral therapy programs in sub- Saharan Africa have varied widely, from approximately 0 to 39%. Furthermore, while patient tracing is ideal it is not always feasible and passive follow-up, where patients who miss visits are not traced, is common. From a public health perspective, the most important measure of an antiretroviral therapy program��s effectiveness is survival after HAART. However, losses to follow-up may threaten the validity of analyses of survival if the proportion of patients dying after HAART is high and only known deaths are counted as events. Furthermore, censoring, or not counting, patients lost to follow-up as having died, while standard, could lead not only to inaccurate estimates of survival but to biased estimates of risk factors for death as well. The latter issue, called informative censoring, may occur in survival analyses when patients who are lost to follow-up are both censored and at high risk of having died. Since many cohort studies of survival after HAART have had substantial rates of losses to follow-up after HAART, and since both monitoring antiretroviral therapy scale-up efforts and improving these efforts depends on accurate reporting of survival rates and risk factors for death, evaluating and quantifying these issues is of global health importance. To investigate this further, we analyzed how outcomes and risk factors for death after HAART initiation in a large public treatment GW786034 molecular weight program in sub-Saharan Africa would be reported in two scenarios: one before and one after patient tracing. Date of HAART initiation and date of outcome were used as the start and end point of follow-up time, respectively. Patients who were alive and in care and patients who were lost to follow-up were censored as of the date of their last clinic visit. Kaplan-Meier plots were used to present 1-year survival estimates before and after patient tracing. Risk factors for death before and after patient tracing were evaluated using Cox proportional Tubulin Acetylation Inducer purchase hazards models after testing the validity of the proportional hazards assumption. Factors where the 95% confidence interval of the point estimate for the unadjusted relative hazard did not cross 1 on unadjusted analysis were retained and evaluated in a multivariable model. However, in order to evaluate if the inclusion of possible confounders not meeting this criterion in the analysis affected the study��s findings, we also evaluated the results after forcing variables plausibly associated with survival in the multivariable model. Collinearity between potential risk factors was assessed by examining the standard errors for the hazard ratios when the multivariable Cox regression model was fitted.