These factors were to be included in a final logistic regression model if their inclusion changed the unadjusted odds ratio for adherence and outcomes by 15% or more. We chose to perform logistic regression rather than methods for survival analysis because our primary aim was to relate occurrence of early adverse outcomes to early adherence during the initial period of ART, and was not to compare time-to-event by adherence. A challenge inherent in determining the relationship between early adherence and early outcomes is that some patients initiate ART and never return, in which case their adherence is DCMU unknown. Requiring patients to return for at least one ART refill in order to be included in the study would eliminate such patients from the analysis but could introduce potentially severe selection bias due to the relationship between loss to follow-up and death. We therefore chose to include these individuals, and in the primary analysis the adherence of patients who initiated but never returned for ART was imputed as,0.95, which is a ‘‘missing = failure’’ approach taken by others. To evaluate the effects of this assumption on the results we then performed a sensitivity analysis repeating the primary analysis after recategorizing these patients’ adherence as $0.95. Because all patients who initiated and never returned were counted as outcomes, the primary analysis counting these patients as suboptimally adherent was considered representative of the strongest likely relationship between early adherence and early adverse outcomes in the cohort. Most studies relating adherence to outcomes have expressed this relationship in terms of odds ratios,Thapsigargin relative risks or relative hazards. These measures of association express the etiologic relationship between exposure and outcome but do not convey the overall public health importance of a risk factor to occurrence of an outcome at the population level. This study, based on prospectively collected data from routine clinical practice in Botswana, provides evidence that suboptimal early ART adherence as measured by pharmacy refill data and pill counts increases the risk of early adverse outcomes.