Background: Randomized trials have shown that cotrimoxazole prophylaxis significantly reduces mortality in untreated HIV infection in resource-limited settings, but no trial has assessed whether benefits also occur on ART. Addressing this question outside of a trial requires causal models as use of cotrimoxazole prophylaxis after ART initiation is likely to influence and be influenced by underlying prognosis.

Methods: DART is a randomized trial of management strategies in symptomatic ART-naive adults with CD4 <200 cells/mm³ initiating triple drug ART in 4 centers (3 in Uganda including 1 satellite, 1 in Zimbabwe). In each center, cotrimoxazole prophylaxis was prescribed at the treating physician's discretion. Marginal structural models with stabilized time-dependent inverse probability treatment weights were used to estimate the causal effect of cotrimoxazole prophylaxis (excluding 137 patients in a pilot sexually transmitted infection study) splitting follow-up into 4-week periods (stratifying by center and randomized monitoring strategy to compute weights).

Results: By March 2007, 3179 patients contributed 9214 years follow-up (median 3 years) and 267 deaths. Cotrimoxazole use differed by center (12%, 77%, 71%, and 72% of follow-up). Time-dependent predictors of cotrimoxazole use included lower CD4 and hemoglobin (last and last-but-1), WHO stage 3/4 event in the previous 4 weeks or earlier, and sexually transmitted infection randomization (not randomized, randomized to continuous ART or sexually transmitted infections). In the first 12 weeks after randomization when use depended only on baseline factors, cotrimoxazole prophylaxis was associated with a significant reduction in mortality (OR = 0.51, 95%CI 0.32 to 0.83; p = 0.007), but less evidence of reductions in malaria (0.82, 0.63 to 1.05; p = 0.12) and WHO 4 events (0.84, 0.54 to 1.20; p = 0.34). After 12 weeks cotrimoxazole was associated with a smaller reduction in mortality (0.72, 0.51 to 1.02; p = 0.07), a similar reduction in malaria (0.76, 0.62 to 0.94; p = 0.01) and no effect on WHO 4 events (1.03, 0.73 to 1.46; p = 0.87).

Conclusions: Marginal structural models applied to observational data where cotrimoxazole prophylaxis was not used continuously and depended on time-varying confounders (which could be influenced by previous use) suggest that current cotrimoxazole prophylaxis reduces risk of mortality and malaria, but not WHO 4 events, in patients receiving ART. Mortality benefits appear to be greater during the first 12 weeks, which could be due to greater effects at ART initiation or in sicker patients.