The best way to compare the impact of 2 or more regimens in a given population is to conduct a blinded, randomized study. However, these kinds of trials have important limitations. In a randomized controlled trial, there may be better adherence than in the "real world." Narrow entry criteria may make the outcomes difficult to apply broadly. In prospectively followed populations -- even with long follow-up times -- deaths do not occur often enough to do meaningful analysis of this all-important endpoint.

Although there are now many antiretroviral combinations, only a few at a time can be studied in a randomized controlled trial. The largest number of regimens ("arms") that can be studied at a time, in even large randomized controlled trials, is 4. Regimens that have been discredited can't be ethically included in randomized controlled trials, even though some clinicians still may use these regimens and they are relevant to resource-poor areas.

Case-control studies can surmount these limitations. They are, however, subject to their own limitation, which is known as "confounding." Say, for instance, you found that patients who took regimen A had better survival than patients who took regimen B. We could then say regimen A is associated with better survival. But the association might not be causal if it turned out that clinicians were more likely to use regimen B in patients who had more advanced disease. The association would have been confounded by this tendency.

Epidemiologists try to counter this problem by "matching" cases and controls. In the foregoing example, they would try to find a control with the same CD4+ cell count at the start of treatment for each case. They can also "adjust" for these differences between the case and control populations using a statistical technique called multivariate analysis.

The San Francisco AIDS Surveillance dataset is an unusually fertile source for the study of treatment-related survival. It is large, it has been around long enough to have included enough deaths to perform meaningful analysis, it has collected enough data to "deconfound" many of the potential confounders, and it has been extremely well analyzed.

Details of This Study

Chen et al presented a case-control study of the impact of antiretroviral regimens on death in persons receiving highly active antiretroviral therapy (HAART) as their initial regimen. The study encompassed San Francisco AIDS Surveillance data through 2002. Of the patients in that dataset, 310 had died and 1,161 controls were still alive.

The authors analyzed 39 regimens that were initiated by more than 9 persons. Controls were matched with cases by year of AIDS diagnosis, CD4+ cell count in the 6 months prior to starting HAART, and year of HAART initiation. There was adjustment for intravenous drug use, age, race and homelessness.

The most commonly initiated HAART regimen in San Francisco, until the end of 2002, was nelfinavir (NFV, Viracept) plus stavudine (d4T, Zerit) plus lamivudine (3TC, Epivir).

The 11 most common regimens all consisted of 2 nucleoside reverse transcriptase inhibitors (NRTIs) and either an unboosted protease inhibitor (PI) or a non-nucleoside reverse transcriptase inhibitor (NNRTI). The 4 NNRTI-containing regimens were lamivudine plus stavudine or zidovudine (AZT, Retrovir), with the addition of either nevirapine (NVP, Viramune) or efavirenz (EFV, Sustiva, Stocrin). These 4 regimens are approved by the World Health Organization (WHO) for use in resource-poor areas, and were significantly associated with the best survival.

The study found that none of the 4 regimens were significantly better than any of the others.

Significance for Patients and Clinicians

This complicated analysis reinforces data on HIV treatment obtained by many randomized controlled trials. This is more than a bit comforting. Most randomized control studies only compare the effects of various treatments on laboratory outcomes such as viral load, which is not what we really care about (that is, illness-free survival).

The study also supports the WHO's regimen choices for resource-poor nations. Little consideration was paid to boosted PI regimens, but such regimens are less important in this context as they are less practical in the Third World. We will have to confront stavudine-related lipoatrophy soon, but not immediately.

The study had important limitations, of course. It only reported on the outcome as a function of initial therapy; subsequent regimens were not considered. Data from HIV-infected persons who never developed AIDS were not included. Information on adherence, except to the extent that adherence affected mortality, was not compiled. The mortality considered was "all cause" mortality, not AIDS-related mortality. Toxicity was not considered.

Nevertheless, this presentation stimulated as lively a discussion as occurred at the conference. It was clear that the authors had many ideas for refining their analysis, several of which arose during the discussion. We can look forward to the final publication as an important contribution to our understanding of the relative efficacy of various treatment regimens.