Investigating temporal changes in the rate of HIV progression

Although the randomized clinical trial is the gold standard for evaluating treatment efficacy, temporal changes in the rate of HIV progression may indicate a treatment's effectiveness in the population as a whole. The introduction of zidovudine and prophylaxis against Pneumocystis carinii pneumonia has resulted in some slowing in the progression rate at the population level [1], and effects of the use of more potent drug combinations are already noticeable [2,3]. In addition, the rapid evolution of HIV has led to the speculation that viral strains currently infecting individuals may be more pathogenic than those infecting patients earlier in the epidemic, resulting in increased rates of HIV disease progression in individuals infected more recently. Any changes to the incubation period are of interest for those studying the natural history of HIV, especially for those involved in predicting the future of the epidemic. The study of temporal changes in the incubation period may be approached in a number of ways. Progression rates can be compared between those infected early in the epidemic and those infected in later calendar periods. If viral strains have evolved over time, then any effect on the incubation period should be clear in such an analysis. The impact of improvements in clinical care may, however, show only a limited effect using this approach, as any of the patients who remained alive long enough to experience these improvements may have benefited from slower progression rates as a result, irrespective of when they became infected. If the aim of the study is to assess the effect of these improvements on the incubation period, then an alternative approach of comparing progression rates in individuals followed up in different calendar periods, after controlling for their initial disease stage, may be preferable. Thus, we can study whether individuals currently at a certain stage of disease experience faster or slower disease progression than individuals at the same stage of disease some years earlier. This approach should reveal any changes to the incubation period that have occurred as a result of improvements in clinical care. The studies carried out to date to assess temporal changes in disease progression have shown conflicting results. In this issue, Carré et al. [4] found no evidence of a change in the incubation period according to year of seroconversion. However, among individuals infected through sexual exposure, those remaining alive in recent years progressed to AIDS more slowly than individuals under follow-up in earlier calendar years, leading the authors to conclude that early treatment with antiretrovirals and prophylaxis had been beneficial in this group. Enger et al. [5] reported improved progression rates in those individuals followed over later calendar periods, especially in those with low CD4 counts [6]. A number of groups have also reported a slowing in the incubation period in recent seroconverters [7–10]. In contrast, two studies [11,12] reported shorter incubation periods over recent years in cohorts of homosexual men. Sinicco et al. [13] also reported that individuals in Italy who were infected with HIV after 1990 experienced faster CD4 count loss and more rapid progression to AIDS than individuals infected during the 1980s, although a previous analysis, including this data, had found no evidence of such a trend [14]. Although cohort studies of HIV-infected individuals are the best source of data for evaluating temporal changes to the incubation period at a population level, certain factors may limit their interpretation. The use of clinical endpoints, such as AIDS or death, is often impractical when patients have only been followed for short periods of time, due to the low rate of clinical progression in the first few years after seroconversion. Thus, these endpoints are only useful when comparing individuals who seroconverted some time ago. The inclusion of new AIDS-defining diseases [15–17] and improvements in diagnostic ability, may lead to earlier diagnosis in those followed in more recent years, resulting in an apparent increase in progression to AIDS. In contrast, increases in the rate of pre-AIDS mortality [18,19], may lead to apparent decreases in the rate of progression to endpoints other than death, if competing causes of mortality are not considered. A relaxation in AIDS case reporting cannot be ruled out as a further source of bias when obtaining information on clinical endpoints from surveillance data. One possible alternative is to consider the time taken to reach a certain CD4 count as an endpoint [20,21]. Alternatively, the rate of CD4 decline over the first few years following seroconversion, or a study of the CD4 count at a given time after seroconversion [14,22], could also be appropriate analyses to study changes in the natural history of HIV. These endpoints are unaffected by changes in diagnostic methods or changes in definition. However, there may be other methodological drawbacks when using these approaches. In particular, the use of CD4 counts as endpoints is affected by changes in laboratory methods, the frequency of CD4 measurements and the lag-time between seroconversion and the first CD4 measurement. A more fundamental problem, however, arises in the analysis of any endpoint when censoring follow-up in individuals who have not reached that endpoint. Often, individual follow-up may be censored at either the closing date of the study or on the date of the last visit, whichever happens first. Under this strategy, a patient's follow-up may be censored differently if he/she seroconverted early in the epidemic or shortly before the end of the study. For example, an individual who had not reached the endpoint at their first visit 2 years after first testing HIV-positive would be censored on this date if he/she seroconverted early in the epidemic, but would be censored at enrolment, and thus effectively excluded from the analysis, if he/she had seroconverted shortly before the end of the study. Frequent patient monitoring may reduce this bias, but analyses are often biased by this exclusion of slow progressors from the recent seroconverters. Finally, the method chosen to estimate seroconversion dates may introduce bias. Estimating seroconversion dates as midway between the last HIV-negative and the first HIV-positive tests may produce artificially late seroconversion dates when HIV incidence is declining [23], resulting in recent seroconverters who progress more rapidly than earlier seroconverters. The exclusion of those with wide ‘window periods’ between negative and positive test results, or the estimation of seroconversion dates using alternative approaches may minimize this bias. Many of these issues have been addressed in the studies that have considered changes in the incubation period. However, it is often difficult to assess the real effect of these biases on the results of the study. Studies where these biases are minimal may be more likely to show no evidence for a change of the rate of HIV progression than studies where strong biases are present. Those analyses that show strong evidence for a shortening of the incubation period over time should be repeated, controlling for the biases and the methodological problems discussed above, before drawing firm conclusions.