We estimated that initiation of cART reduced the overall incidence of mortality of HIV-infected individuals by 50%. In absolute terms, this mortality reduction was translated into a 5% increase in 5-year survival for cART initiation compared with no initiation. However, the absolute benefit depended on the baseline CD4 cell count: the survival increase ranged from about 46% in individuals with less than 100 cells/μL to about 1% in those with more than 500 cells/μL. Similarly, the absolute benefit was greatest in injection drug users compared with other transmission groups.
Because our collaboration includes a wide range of HIV-infected individuals living in Europe and the United States—including representative clinical cohorts, seroconverters, and U.S. veterans—we could confirm that cART initiation was followed by 40–60% mortality reduction in all of these groups, with perhaps an even greater reduction among seroconverters. The survival benefit was clear even among U.S. veterans, who experienced a higher mortality rate (40 deaths per 1000 person-years) than the other cohorts (6–13 deaths per 1000 person-years). The high mortality rate among U.S. veterans—partly explained by their older age, predominantly male sex, and different death ascertainment procedures—is in line with those reported in similar populations.16, 37, 38
Previous studies reported mortality hazard ratios for cART vs. no cART in the range 0.1 to 0.5. However, most previous studies on cART and mortality have been based on ecological comparisons and/or did not appropriately adjust for time-dependent confounders,12–18
which makes the estimates hard to interpret. Lack of appropriate adjustment for time-varying CD4 cell count and HIV RNA levels may result in either underestimation or overestimation of the effect of cART. In contrast to previous studies, we used a prospective cohort design and inverse probability weighting to appropriately adjust for measured confounders. It has been previously shown that effect estimates obtained from observational HIV cohorts via inverse probability weighting, but not those obtained via conventional adjustment methods, can replicate the findings of randomized clinical trials on cART and time to AIDS.3, 20
In the current analysis, we found empirical proof of the bias of conventional methods in that further adjustment for the strong time-varying confounding by indication resulted in a weaker (from 0.79 to 0.83), rather than stronger, effect estimate.
The validity of our effect estimates, however, relies on the untestable assumption that the measured covariates were sufficient to adjust for confounding by indication. We believe this assumption may approximately hold here because the most important clinical information used by physicians as indications for cART initiation (i.e., CD4 cell count and HIV-RNA) was collected and used in the analysis. An incomplete or imperfect measurement of confounders would result in attenuation of the effect because the bias due to confounding by indication is upwards in this setting. As a further protection against unmeasured confounding, we estimated the effect of cART initiation rather than the effect of continued cART use. This strategy makes it unnecessary to adjust for joint determinants of treatment discontinuation and death, which are less well-measured in most observational studies. The price to pay for this strategy is again an attenuation of the effect estimate towards the null (as it would happen in the intention to treat analysis of a placebo-controlled randomized clinical trial) if a substantial proportion of individuals stopped taking cART during the follow-up. Thus, our effect estimates could be viewed as conservative estimates of the effect of continuous cART. Note, however, that the effect of continuous cART use would not be clinically interesting if most subjects who discontinue cART do so for toxicity-related reasons.
Other studies have used surrogate biological markers, like HIV RNA and CD4 cell count, in hopes of rapidly inferring the effect of cART on death. Several randomized studies, however, suggest that surrogate markers may not be good proxies for clinical endpoints like death.6–9
In the MRC Delta Trial, CD4 cell count and HIV RNA changes overestimated clinical benefit (defined as survival or no disease progression).9
HIV RNA may be an inadequate surrogate for death because death may be caused mainly by immune system damage or unintended harmful effects via other mechanisms;39
CD4 cell count may be inadequate due to biological and assay-based variability, differences between peripheral blood and lymphatic levels, and the existence of other important markers of immune function that are affected by treatment, like CD8 cells.9
Even if surrogate biological markers provide some indication of short-term clinical endpoints, it is debatable whether these surrogate effects translate into long-term clinical effects, which depend on the interplay among immune factors, drug resistance, and toxicities that may not be evident in the short-term. For example, a regime might be perceived to be favorable in the short-term because its potency allows patients to maintain higher CD4 count and lower viral load, but over the long-term, this same potency might cause toxicities that lead to death more quickly. It is best to directly evaluate the effect of cART on mortality in large follow-up studies as we did here. An alternative would be to evaluate the effects of cART on cause-specific morbidity and mortality, but this information is not available in all HIV cohorts.
In summary, in the absence of data from long-term randomized clinical trials, appropriate adjustment for time-varying confounding in observational cohorts provide the best available evidence on the effects of cART on the overall mortality of HIV-infected individuals. We estimated that cART initiation halved the mortality rate of HIV-infected individuals in developed countries, and that the absolute reduction in mortality was stronger in those with worse prognosis at the start of follow-up. Thus the 46% survival increase estimated for individuals who start cART at CD4 cell count less than 100 cells/μL demonstrates the benefits of being treated even at the most advanced stages of immunosupression. This finding, however, does not imply one should delay cART initiation until the CD4 cell count drops below 100 because, besides the possibility of dying during the waiting period, the 5-year mortality risk of treated individuals with less than 100 cells/μL at baseline (11%) was almost 4 times greater than that of treated individuals with more than 500 cells/μL (3%).