Background:  Observational studies have demonstrated that CD4 T cell counts often remain stable in presence of drug-resistant viremia. These observations are confounded by the possibility that the outcome of interest (CD4 T cell counts) may influence decisions as to when to modify therapy, leading to a potential bias. Marginal structural models control for such time-dependent confounders.

Methods:  We identified all subjects in a prospective clinic-based cohort who failed to achieve or maintain undetectable HIV RNA levels on a new regimen. The influence of early versus delayed treatment modifications on subsequent CD4 T cell counts was studied using history-adjusted marginal structure models. These models consider time-updated covariates to address potential confounding by prior and current HIV RNA levels, CD4 T cell counts, treatment history, opportunistic diseases, adherence, and drug use.

Results:  A total of 100 patients was evaluated, most of whom were heavily pre-treated. The median CD4 T cell count and viral load at time zero (when failure occurred) was 268 cells and 3947 copies/mL, respectively. The median time to switch after onset of failure was 6 months. Cross-validated data-adaptive regression demonstrated that treatment was more likely to be modified in patients who had lower CD4 T cell counts and lower levels of adherence, and in patients with an opportunistic infection. After adjusting for confounders, the effect of waiting each additional month on expected CD4 T cell count 8 months in the future was 0.05(CI 0.02 to 0.09) of baseline CD4 –13 (CI –22 to –4). Thus, waiting to switch resulted in an overall loss of CD4 T cells for individuals with <260 CD4 T cells at the time of failure, and an overall gain in those with higher CD4 T cell counts. Among individuals who had already spent 5 or more months on a partially suppressive therapy, the effect of additional time until switching therapy had a negligible effect on future CD4 T cell count, irrespective of current CD4 T cell count.

Conclusions:  Delaying a switch in therapy for patients with low CD4 T cell counts at the time of virologic failure was associated with a long-term loss of CD4 T cells, while waiting to switch therapy for patients with higher CD4 T cells counts was not associated with immunologic harm. The effect of delaying a switch in therapy for all groups waned over time; after 5 months, delaying treatment modification further was neither advantageous nor detrimental.