Commentary: The changing face of AIDS.

Abstract

The paper from the CASCADE collaboration 1 is a welcome addition to the growing body of evidence from observational cohort studies showing that the incidence of AIDS-defining diseases has declined dramatically since 1997 when highly active antiretroviral therapy (HAART) was widely introduced. The CASCADE collaboration follows patients with known date of seroconversion and therefore adjustments for duration of HIV infection can be made. The results from the present analyses give rise to several questions. Here I will focus on two issues: firstly, are estimates of disease incidence from seroprevalent cohorts comparable with those from seroincident cohorts? Secondly, what difference is there in the interpretation of a cause-specific and a competing risks model? The authors state that it is necessary to use a seroincident rather than a seroprevalent cohort to estimate the effect of treatment changes on the incidence of AIDS over time. If we compare the raw incidence rates with the adjusted relative hazards, we can see the importance of adjusting for time since seroconversion. Does this mean that seroprevalent cohorts cannot be used to estimate treatment effects over time because adjustment for time from infection cannot be made? In fact the requirement for an unbiased analysis is for patients at a similar point in the disease process to be compared. An alternative way of characterizing the stage of the infection is to adjust for CD4 cell count and human immunodeficiency virus type 1 (HIV-1) RNA in the analyses, as patients with similar immune suppression and viral load are at a similar risk of AIDS defining diseases. This point was very clearly illustrated in a paper by Tarwater et al. 2 using the Multicenter AIDS Cohort Study (MACS). Gay men were enrolled into the MACS cohort some of whom were seronegative and later converted to seropositive, therefore their dates of seroconversion were well documented. They contrasted two methods of assessing population effectiveness of therapies in HIV cohorts. First, they looked at all patients whose date of infection was known. The relative hazard of AIDS was 1.52 for no therapy, mono therapy was the reference group, 0.91 for dual and 0.30 for triple therapy in the seroincident cohort controlling for duration of infection. In the second method, they used patients whose date of seroconversion was not known and calculated relative hazards adjusting for CD4 cell count and HIV-1 RNA. The corresponding figures for the seroprevalent cohort were 1.52, 1.03 and 0.31. Knowledge of duration of infection did not add significant information beyond that provided by CD4 cell count and levels of HIV-1 RNA. The conclusion was that the two methods worked equally well and that the population effectiveness of drugs could be measured in cohorts that did not have dates of seroconversion.