|Home | About | Journals | Submit | Contact Us | Français|
Given the recent availability of antiretroviral therapy (ART) in resource-limited settings and the significant burden exacted by Kaposi’s sarcoma (KS) in these areas, we reviewed data regarding the impact of ART on KS incidence. We summarized the sizeable literature in resource-rich settings as well emerging data from resource-limited regions. Importantly, we delineated ways impact can be defined, including a) individual patient-level effectiveness; b) population-level effectiveness; c) change in population-level incidence; and d) residual risk of KS.
In resource-rich settings, there are now ample data demonstrating beneficial individual patient-level and population-level effects of ART on KS incidence. There is, however, considerable variability between studies and important methodologic shortcomings. Data from resource-limited settings are much more limited; while they preliminarily indicate individual patient-level effectiveness, they do not yet provide insight on population-level effects.
ART has had a substantial impact on KS incidence in resource-rich settings, but more attention is needed on validly quantifying this effect in order to determine whether additional interventions are needed. Emerging data from resource-limited regions also suggests beneficial impact of ART on KS incidence, but — given the scope of KS in these settings — more data are needed to understand the breadth and magnitude of the effect.
The advent of HIV/AIDS in 1981 transformed Kaposi’s sarcoma (KS) from a medical oddity to an epidemic . In resource-rich settings, the cumulative lifetime incidence of KS among HIV-infected homosexual men reached nearly 40% . In many resource-limited settings, particularly sub-Saharan Africa, the extent of the HIV epidemic resulted in KS becoming the most common malignancy not just among HIV-infected individuals but among all adults . Yet, just as abruptly as AIDS impacted KS, the advent of potent antiretroviral therapy (ART) has transformed HIV disease. In resource-rich areas where ART is routinely available, the lifespan of HIV-infected individuals has nearly normalized . ART has come more slowly to resource-limited settings, but by the end of 2010 over 5 million patients in sub-Saharan Africa alone had initiated ART  with overall reduction in mortality comparable to resource-rich settings .
While the general impact of ART on morbidity and mortality has been well documented, this review will summarize the specific influence of ART on KS incidence. We shall summarize the abundant data from resource-rich settings and examine emerging data from resource-limited settings. Importantly, we will clarify what is referred to as the “impact”, “influence” or “effect” of ART on KS occurrence. We believe that patients and practitioners may be interested in as many as five questions:
In reviewing the impact of ART on KS incidence, we focused on the last four questions, namely individual patient-level effectiveness, population-level effectiveness, change in KS incidence in the ART era, and residual risk.
We reviewed published literature from 1996 to 2012, searching Medline and the Web of Science, which described the impact of ART on KS incidence in either resource-rich or resource-limited settings. Search terms included: Kaposi’s sarcoma, HIV malignancy, neoplasm, incidence, epidemiology, antiretroviral therapy, and highly active antiretroviral therapy.
We discuss the impact of ART on KS incidence in terms of individual patient-level and population-level effectiveness.
With variable findings (Table 1), nine studies [7–12••,13–15••] across three continents have evaluated individual patient-level effectiveness of ART on KS incidence (Question 2). The lowest KS incidence among ART users was 109 per 100,000 person-years  while the highest incidence was 700 per 100,000 person-years . Of special interest is the absolute KS incidence among ART users who have achieved a CD4+ T-cell count of ≥350 cells/μl. This is important because the majority of ART users will spend the rest of their lives in this immunological state, and some notable case series have suggested that KS may continue to be common in this range [19, 20]. Unfortunately, few reports provide formal estimates of KS incidence among such patients. Among individuals with a CD4 count of ≥350 cells/μl, Francheschi et al.  and Lodi et al. [12••] reported incidences of 118 and 368 per 100,000 person-years, respectively. A lower incidence (89 per 100,000 person-years) was reported among ART users with CD4 count ≥500 cells/μl [12••].
After adjustment (in some studies) for various confounding factors, the KS incidences among ART users represented declines between 19% and 93% compared to non-ART users (Table 1). Studies from Europe had a range between 61 % and 93% [7–12••,13], a 57% decline was observed in a multi-country study in Europe, Australia and Canada [12••], while a 19 to 39% reduction was seen in the U.S. [14, 15••]. The substantial variability between studies could have several explanations. First, the populations differed. For instance, Mocroft et al.  performed their work between 1994 and 1998, a period when ART users were heavily treatment-experienced and likely enriched for those unable to achieve full virologic suppression on combination ART regimens compared to patients who started ART in later years. This could in part explain the smaller 61% reduction in KS incidence that was found compared to other European estimates. Second, although rarely reported, ART impact also depends on compliance. For example, there was a 7-fold increase in KS risk in the CD4 count-guided episodic ART arm of the SMART trial . Third, the magnitude of losses to follow-up was also rarely reported, leaving open the possibility of selection bias. While some studies [8–11] were less susceptible, the potential for differential losses to follow-up between ART users and non-users is uninterpretable in others [7, 12••, 13]. Fourth, with few exceptions , the mode of KS diagnosis was not reported. While biopsy was surely performed in many instances, we also know that clinical diagnosis alone was common in some areas. The non-specificity of clinical diagnosis will typically attenuate the apparent effect of ART.
As all available studies are observational, confounding could be another reason for variability. While some studies reported adjusted estimates [7, 11, 14, 15••], others were either unadjusted [8, 9, 12••, 13], or adjustment was not reported [10, 14]. Notably, only two studies controlled for CD4 count [11, 14], which would be expected to be a strong confounder. Yet, CD4 count in this context is also a time-dependent mediator of the effect of ART, which if adjusted for conventionally could attenuate the effect of ART. To manage confounding of this type, advanced approaches known as marginal structural models have emerged in the past decade . Because none of the studies used these advanced techniques, we may qualitatively know the individual patient-level effectiveness of ART on KS incidence but are yet to have an unbiased quantitative estimate.
In the one study that estimated residual KS risk (Question 5) at the individual patient-level, ART users had a 25.3-fold higher rate of KS compared to the general population  (Table 2). This report, however, used contemporaneous rates of KS from the general population, which, because virtually all KS in the general population today is derived from HIV-infected patients, tends to underestimate the parameter of interest (a comparison of ART-treated/HIV-infected persons to HIV-uninfected persons) . On the contrary, any comparison (in most Resource-rich settings) of HIV-infected persons to the general population (irrespective of era) that does not consider differences in the underlying prevalence of infection with Kaposi’s sarcoma-associated herpesvirus (KSHV) will overestimate the residual risk of KS in the HIV-infected group.
Population-level impact of ART on KS incidence has been evaluated by nine studies, all of which used regional/national cancer registries to ascertain incidence (Tables 3 and and4).4). In these studies, calendar time was the predictor variable, with pre-ART era (before 1996) compared to ART era (after 1996). Whether or not authors adjusted for one or more factors dictated the question they could address. Estimating population-level effectiveness of ART on KS incidence (Question 3) requires adjustment for whatever factors differ across calendar time. Population-level effectiveness was estimated by 5 studies [15••, 25••, 26, 28, 29•], which found reductions between 33% and 95% in KS incidence in the ART era (Table 3). Given that estimating the change in KS incidence since ART became available (Question 4) is simply a crude comparison of incidences over time, it requires no adjustment. For this question, 5 studies [23, 24, 29•, 31, 32•] provided estimates, ranging from 27% to 88% reductions in KS incidence (Table 4).
Similar to individual patient-level effectiveness, the variability in the population-level estimates may also have several explanations. First, the nature of the HIV-infected populations differed across studies. Specifically, some studies included all HIV-infected individuals [25, 31], while others examined only those with AIDS [15••, 23, 24, 26••, 28, 29•, 32•]. Furthermore, there were differences in the reference calendar periods for the pre-ART era. While some studies used a period beginning in the early 1980s [23, 25, 29], others used the period 1990–1995 [26••, 28]. Second, there again may be differences in KS ascertainment. Even though cancer registries were used to ascertain KS, most studies insufficiently described the registries’ processes. Differences across studies in the fraction of clinical KS diagnoses, which are prone to non-specificity, could contribute to the differences in the derived inferences. Finally, in the studies addressing population-level effectiveness of ART, in which differences across eras have to be addressed, there was disparity in how this was handled. For example, one study adjusted for ART use over and above calendar period, which because it is on the causal pathway would be expected to attenuate effects [15••]. Notably, no study accounted for CD4 count. Therefore, we again conclude that while we know the qualitative population-level effectiveness of ART on KS incidence, we do not have an accurate quantitative estimate.
Residual KS risk in the ART era was assessed in three populations [23–25••, 26••] (Table 2), with standardized incidence ratios ranging from 22.9 [25••] to 3640 . Differences in composition of the general population comparator groups across studies likely explain this wide range. Specifically, the highest estimates used a general population group from the late 1970’s that predated the HIV epidemic [24, 26••], and the lowest used the contemporary general population [25••]. As mentioned earlier, because almost all KS that occurs in contemporary resource-rich general populations (outside of the Mediterranean) is derived from HIV-infected persons, underestimation of the true residual risk among HIV-infected persons will result if this contemporaneous comparison group is used .
A number of non-population-based cohorts [14, 33–35•, 36–42•], typically clinic-derived, also compared incidence of KS in both the pre-ART and ART eras (Table 5). However, when viewed from the perspective of the questions we initially outlined, these studies are not directly estimating any parameter of clinical or epidemiologic relevance (which is unfortunate because many of these studies have the best measurements). In a clinic-based cohort, the population-level effectiveness of ART will be overestimated. This is because the patients analyzed in the ART era are systematically enriched for those in care and are devoid of patients not on ART. Likewise, these studies are not validly estimating individual patient-level effectiveness because ART use per se was not the predictor variable. Indeed, many patients in these studies in the ART era were not actually using ART, thus leading to likely underestimation of the individual patient-level effectiveness of ART. It is therefore in the realm of individual patient-level effectiveness of ART that these non-population-based cohort studies have their greatest contribution. Given that most are estimating >70% reduction in KS incidence, we can infer that the true value is likely in this range.
Despite having the vast majority of the worldwide HIV/AIDS and KS burden, resource-limited settings have yielded substantially less data regarding the impact of ART on KS. It is nonetheless useful to consider the data in terms of individual patient-level and population-level effectiveness.
In the one published study on this question, KS incidence among ART users in Uganda was 340 cases per 100,000 person-years [16••] (Table 1). This study did not have a comparator group of non-ART users, but it can, in theory, be compared to historical work in the country. In 1988–2002 (a period with minimal ART availability), Mbulaiteye et al. reported a KS incidence of 380 per 100,000 person-years among HIV-infected patients in Kampala . This would seem to indicate no effectiveness of ART, but differences in KS ascertainment between studies likely preclude any valid comparison. The recent work was in the context of a clinical trial where patients were being prospectively examined. In contrast, the older work originated in a clinic population where KS diagnoses may have been easily missed, and, furthermore, to be counted by the study, a KS diagnosis had to be captured in a local cancer registry and matched back to the patient’s clinic record. This likely resulted in substantial underestimation of KS incidence in the pre-ART report, a limitation noted by the authors .
More recent data, presented to date in abstract form only, promises to better estimate individual patient-level effectiveness. In the East Africa Consortium of the International Epidemiologic Databases to Evaluate AIDS (IeDEA) Project , 98,024 HIV-infected adults at 26 HIV clinics in Uganda and Kenya were followed for incident KS . A unique feature of the work was that histological diagnosis was made available at the participating sites and accounted for approximately 50% of KS diagnoses. In Uganda, KS incidence was 1876 per 100,000 person-years in non-ART users and 201 in ART users, translating to a 78% reduction in KS incidence . In Kenya, KS incidence was 596 per 100,000 person-years in non-ART users and 270 users in ART users, translating to a 50% reduction . Another abstract, from Southern Africa IeDEA, covering 10 clinic-based cohorts with 184,592 patients, reported an incidence of 624 per 100,000 person-years among non-ART users and 174 per 100,000 person years in ART users, translating to a 72% reduction . What accounts for the clinically important differences in ART effectiveness in Kenya versus Uganda (or South Africa) remains unclear given the unpublished nature of the work. Of note, in neither of these analyses was the role of CD4 count as a time-dependent confounder/mediator appropriately managed, again leaving us without a valid quantitative estimate of ART effect.
In the one published report of the population-level effectiveness of ART on KS incidence in resource-limited settings, Msyamboza et al. estimated an increase in KS in 2007–2010 in Malawi compared to earlier periods (Table 3) [30•]. However, with the majority of KS diagnoses being clinical (only 18% histopathologic), this study highlights the challenges of using registries to estimate population-level effects in resource-limited settings. These challenges also include issues with completeness  and unknown HIV status among the cases . As such, we view the findings from these nominally population-based studies as generally uninterpretable, and we therefore have no evaluable population-level data as it relates to KS.
There is now ample evidence to indicate substantial effectiveness of ART on KS incidence at the individual patient-level and population-level in resource-rich settings. Although in preliminary form, there is now emerging evidence to also suggest that the individual patient-level effectiveness of ART on KS incidence in resource-limited settings is comparable to that of resource-rich settings. Yet, considering the public health magnitude of KS in resource-limited settings, particularly in sub-Saharan Africa, much more data from diverse settings are needed to confirm the individual patient-level effectiveness observed in the initial studies, determine population-level effectiveness, and establish residual risk of KS given ART use. From a methodological perspective, while we can safely qualitatively conclude that ART works in preventing KS, the actual magnitude of the effect has never been properly estimated due to the improper handling of time-dependent confounding/mediation. Not knowing the actual magnitude of the ART effect is more than just academic in that it will inform whether additional interventions (e.g., anti-KSHV agents) are needed in combination with ART in order to more fully reduce KS incidence. This is critical since evaluating additional interventions will be very expensive. Finally, to understand the face of HIV-associated KS in the future, we need additional specific estimates of the incidence of KS in ART-treated patients who have achieved CD4 counts >350 cells/μl as well as an understanding why patients continue to develop KS at this stage of HIV disease.
Funding: National Institutes of Health (R01 CA119903, D43 CA153717, U01 AI069911, and P30 AI027763).
We thank Megan Wenger and Melissa Assenzio for assistance with literature review.
Conflict of interest: Authors declare no conflict of interest.
Papers of particular interest, published within the annual period of review, have been highlighted as
• of special intrest
•• of outstanding interest