The Changing Incidence of AIDS Events in Patients Receiving Highly Active Antiretroviral Therapy

Background  Although the incidence of most AIDS events declines after initiation of highly active antiretroviral therapy (HAART), this decline is more rapid for some conditions than others. We herein describe the decline in incidence of AIDS-defining events among 12 574 antiretroviral-naive individuals who started HAART in the Antiretorviral Therapy Cohort Collaboration and determined whether the rate of decline is similar for events with different etiologies.

Methods  Rates of AIDS were calculated for the periods 0 to 3, 4 to 6, 7 to 12, 13 to 24, and 25 to 36 months after starting HAART. Changes in incidence over time were investigated using Poisson regression.

Results  During 22 958 person-years of follow-up, 928 AIDS events developed (25.3% viral, 24.6% bacterial, 20.7% fungal, 8.1% protozoal, and 21.2% other). The incidence of any AIDS event declined from 129.3 (95% confidence interval [CI], 116.7-141.8) per 1000 person-years in the first 3 months to 13.2 (95% CI, 9.4-17.0) in the third year after starting HAART (P<.001). The rate of decline in incidence was greatest for events with a viral etiology (87.0% per year) and lowest for those with a fungal etiology (54.0% per year). In the third year, fungal events represented 37.0% of AIDS events that occurred. After adjustment for the CD4 count and human immunodeficiency virus 1 RNA level, changes in the incidence of bacterial and viral events from months 0 to 6 and 7 to 12 remained significant, suggesting that changes in these markers did not fully explain the changes in incidence seen.

Conclusion  Although the incidence of all AIDS-defining events decreased substantially after starting HAART, the pattern of decline was most pronounced for events with a viral etiology.

Treatment with highly active antiretroviral therapy (HAART) results in a dramatic reduction in the incidence of AIDS and death in individuals infected with human immunodeficiency virus (HIV).^1-6 However, as a consequence of the gradual introduction of antiretroviral treatment since the late 1980s, many of the patients included in these studies had already received monotherapy or dual therapy with nucleoside reverse transcriptase inhibitors (NRTIs) before starting HAART. Because prior treatment exposure is associated with a poorer response to HAART,⁷ inclusion of antiretroviral-exposed individuals in these studies will lead to an underestimate of the full benefits of HAART that would be seen in an antiretroviral-naive population.

In 1997, Mellors and colleagues⁸ identified the CD4 count and HIV RNA level as 2 of the most important prognostic markers for clinical progression in HIV-1 infected patients. Recently, the Antiretroviral Therapy (ART) Cohort Collaboration demonstrated that among the factors measured at the time of starting HAART, the CD4 count is the strongest predictor of progression to AIDS or death during a 3-year period, whereas the HIV RNA level measured at the same time is a weaker predictor, being associated with an increased risk of progression only when the level is greater than 100 000 copies/mL.⁹ In a subsequent analysis of this cohort, CD4 counts and viral loads measured 6 months after starting HAART were found to be even stronger predictors of clinical outcome than these markers at baseline, providing evidence of the clinical benefits of short-term immunological and virological responses to HAART.¹⁰

Several studies have demonstrated that the risk of clinical progression is higher in the first months after starting HAART than at subsequent times.^11,12 This may be explained by a delay in achieving immune restoration¹³ or by the occurrence of the immune reconstitution syndrome,^14-16 or it may be a consequence of late reporting of the clinical event that triggered the initiation of therapy. Although most clinical events have exhibited a decline in incidence in the HAART era, some conditions (eg, non- Hodgkin lymphoma) appear to exhibit less rapid declines in incidence after initiation of HAART.^17,18 At present, however, few data illustrate the changes in the incidence of specific AIDS-defining events during the first few years after starting HAART in antiretroviral-naive individuals. Consequently, it is unclear whether the pattern of decline is similar for AIDS events of all etiologies.

We used data from the ART Cohort Collaboration to describe changes in the overall incidence of AIDS events during a 3-year period in antiretroviral-naive patients initiating HAART, to evaluate whether the pattern of decline is similar for AIDS events of different etiologies (viral, bacterial, fungal, protozoal, or other) and whether the decrease in incidence of these events can be fully explained by changes in CD4 counts and HIV-1 RNA levels.

The ART Cohort Collaboration is an international collaboration between the investigators of 13 cohort studies from Europe and North America.^9,10 Studies were eligible if they had enrolled at least 100 patients with HIV-1 infection 16 years or older who had not previously received antiretroviral treatment and who had started ART with at least 3 drugs, including NRTIs, protease inhibitors, and/or non-NRTIs (NNRTIs), with a median duration of follow-up of at least 1 year. Cohorts included were EuroSIDA; the French Hospital Database on HIV; the Italian Cohort of Antiretroviral-Naive Patients (ICONA); the Swiss HIV Cohort Study (SHCS); the AIDS Therapy Evaluation Project, the Netherlands (ATHENA); Collaborations in HIV Outcomes Research (CHORUS); the Frankfurt HIV Cohort; the Antiprotease Cohort (APROCO); the Aquitaine Cohort; the British Columbia Centre for Excellence in HIV; the Royal Free Hospital Cohort; the South Alberta Clinic Cohort; and the Koln/Bonn Cohort. All cohorts provided anonymous data on a predefined set of demographic, laboratory, and clinical variables that were then pooled and analyzed centrally.^9,10

For this analysis, events meeting the 1993 Centers for Disease Control and Prevention criteria for AIDS¹⁹ and person-years of follow-up (PYFU) in the first 3 years after starting HAART were split into the following 5 periods: 0 to 3 (baseline), 4 to 6, 7 to 12, 13 to 24, and 25 to 36 months after the start of HAART. Patient follow-up was censored at the date of the last clinic visit after starting HAART or on the date of death, if this occurred before a patient had his or her first post-HAART visit. The incidence of new AIDS events was calculated as the number of events occurring in each period divided by the total PYFU in the period. The incidence of new events was calculated for AIDS events of different etiologies (viral, fungal, bacterial, protozoal, or other). Viral events included cytomegalovirus disease, herpes simplex virus disease, Kaposi sarcoma, and progressive multifocal leukoencephalopathy. Bacterial events included recurrent bacterial pneumonia, pulmonary and extrapulmonary tuberculosis, atypical mycobacteriosis, and recurrent Salmonella sepsis. Fungal events included pulmonary or esophageal candidiasis, extrapulmonary cryptococcosis, and Pneumocystis carinii pneumonia (PCP). Protozoal events included brain or disseminated toxoplasmosis, isosporiasis, and cryptosporidiosis. Other events included wasting syndrome, HIV-related encephalopathy, invasive cervical cancer, non-Hodgkin lymphoma, primary brain lymphoma, and events in which the etiology could not be determined (n = 9). Although this other-event group includes events that may have very different etiologies, the small number of each specific event (eg, wasting syndrome developed in only 37 patients) prevented us from using any finer groupings. Although only the first incidence of each specific AIDS-defining event after initiation of HAART was counted (recurrences of earlier AIDS events were not reported to the study center), individuals may have experienced multiple events with the same type of etiology (ie, a patient in whom cytomegalovirus disease and Kaposi sarcoma developed would have been included as having 2 events of viral etiology). Thus, incidence rate ratios were estimated by using generalized estimating equations to fit univariable and multivariable Poisson regression models (using the GENMOD procedure in SAS version 8²⁰), allowing for multiple events in the same individual.

Analyses of the change in incidence over time were adjusted for baseline characteristics (ie, sex, age, risk group, clinical stage, cohort, and year of starting HAART). Initially, time period was included in the model as a categorical variable, and incidence rates in each time period were compared with those in the first period (0-3 months). However, to estimate overall rates of decline during the 3-year period, time was subsequently modeled as a continuous variable (with the values 0.125, 0.375, 0.75, 1.50, and 2.50 attributed to each of the 5 periods to represent the midpoint of each period). Visual inspection of the rates in each of the time periods confirmed that the assumption of a log-linear relationship with time was reasonable. The percentage decline in incidence rate per year was derived from the rate ratio as 100 × (1 − rate ratio).

The CD4 counts and HIV-1 RNA levels were available in the pooled data set at 2 time points: baseline (before HAART) and 6 months after start of HAART. AIDS events that occurred during the first 6 months (corresponding to the baseline CD4 count) and the second 6-month period (7-12 months, corresponding to the 6-month CD4 count) were stratified by the CD4 count and HIV-1 RNA level at the start of each respective period. For the purposes of these analyses, the CD4 count and HIV-1 RNA level were stratified into 5 groups (CD4 count, <50, 50-199, 200-349, 350-499, and ≥500 cells/μL; HIV-1 RNA level, ≤500, 501-9999, 10 000-99 999, 100 000-499 999, and ≥500 000 copies/mL). To assess whether changes in incidence between these 2 time periods could be explained by changes in the CD4 count and/or HIV-1 RNA level at the start of each period, incidence rates were compared in the 2 periods, before and after adjustment for the CD4 count and HIV-1 RNA level at the start of each period. Because the pooled data set contained CD4 counts and HIV-1 RNA levels at these 2 time points only, we were unable to describe the relationship between changes in CD4 count and HIV-1 RNA level and clinical events during a longer time period. It should be noted that information on prophylaxis against PCP and atypical mycobacteriosis was not available, and therefore we were not able to assess the impact of changes in the use of this on the incidence of AIDS.

A total of 12 574 individuals were included in the analysis (Table 1). The CD4 counts at the time of starting HAART were low (median, 250 cells/μL). Most of the patients (n = 10 773 [85.7%]) started a protease inhibitor–containing HAART regimen; 1383 (11.0%) started a NNRTI-containing regimen; 164 (1.3%) started a regimen including a protease inhibitor and NNRTI; and 254 (2.0%) started a regimen including NRTIs only. The most common NRTIs received were lamivudine (76%), zidovudine (59%), and stavudine (40%). Indinavir sulfate (43%) was the most commonly used protease inhibitor, followed by nelfinavir mesylate (21%), ritonavir (14%), and saquinavir mesylate (14%). Nevirapine and efavirenz were used by smaller proportions of patients (8% and 4%, respectively).

Patients in the study contributed a total of 22 958 PYFU, during which time 928 AIDS events occurred. Seven hundred ninety-two patients (6.3%) experienced at least 1 clinical event during the study period (680 patients experienced only a single event; 92, 2 different events; 16, 3 different events; and 4, 4 different events). Three hundred twenty-five patients (2.6%) are known to have died; of these, 230 (70.8%) died without development of a new AIDS-defining event. Overall, 235 (25.3%) of the 928 events had a viral etiology; 228 (24.6%) had a bacterial etiology; 192 (20.7%) had a fungal etiology; 75 (8.1%) had a protozoal etiology; 197 (21.2%) had other etiologies; and 1 AIDS event was unspecified and was excluded from further analyses (Table 2).

The incidence of any AIDS event declined from 129.3 (95% confidence interval [CI], 116.7-141.8) per 1000 PYFU in the first 3 months after starting HAART to 13.2 (95% CI, 9.4-17.0) in the third year after starting HAART. This decline remained highly significant after adjusting for the demographic and clinical features of the individuals during follow-up in each of the 5 periods. The rate of decline was greatest for events with a viral etiology (with a decrease in incidence of 87.0% per year) and was lowest for those with a fungal etiology (54.0% per year). This had an impact on the relative frequencies with which each of the events occurred. In the first 3 months after starting HAART, a third of clinical events (33.3%) had viral etiologies, and a quarter (25.1%) had bacterial etiologies. Over time, the proportion of events with viral etiologies dropped and those with fungal or other/unknown etiologies increased. Thus, by the third year after starting HAART, fungal events (37.0%) were the most common etiology reported, whereas events with a viral etiology represented only 6.5% of the events that occurred (Table 2).

The most commonly occurring events in the first 3 months were atypical mycobacteriosis, Kaposi sarcoma, cytomegalovirus disease, and PCP (Figure 1). By the third year, the rates of all events had declined. However, of those events that did occur, PCP, tuberculosis, esophageal candidiasis, and non-Hodgkin lymphoma were the most frequent.

By 6 months after the start of HAART, CD4 counts (measured from 3-9 months; median, 185 days after the start of HAART) had increased to a median of 343 cells/μL (range, 0-2063 cells/μL); 2795 (27%) of 10 540 patients had a CD4 count of less than 200 cells/μL. The HIV-1 RNA levels measured at the same time were 500 copies/mL or less in 7368 (72%) of 10 208 patients (maximum value, 6.70 log₁₀ copies/mL). Event rates were generally lower in the second time period, even after stratifying by CD4 count (Figure 2) and/or HIV-1 RNA level (Figure 3). After adjusting for the CD4 count and HIV-1 RNA level, the changes in the incidence of bacterial and viral events were attenuated, but remained significant. In contrast, changes in the incidence of fungal, protozoal, and other events were no longer significant after adjustment (Figure 4).

Using data from a large multicohort collaboration, we have demonstrated that the incidence of all AIDS-defining events decreased substantially during the first 3 years after HAART initiation. Although events of a viral etiology experienced the most rapid decline in incidence, this trend was seen for all events, irrespective of etiology. Patients were not excluded from the analysis if they stopped or changed drugs. Because a substantial number of patients would be expected to switch therapy within the first 3 years as a result of toxicities or treatment failure,^21,22 it is reassuring to note that there was no sign of any upturn in the incidence of clinical events toward the end of the 3 years.

As reported by others,^11,12 the absolute drop in incidence of each type of clinical event was particularly rapid in the early months of therapy. This drop in incidence mirrors the biphasic pattern to CD4 changes that has beenreported.^23,24 However, a number of other possible explanations may exist for the rapid reduction in disease incidence seen in the initial months after starting HAART. In particular, some of the events may be a consequence of immune restoration syndrome in the first few months as the CD4 count increases rapidly.^14-16 Alternatively, the rapid decline in incidence may reflect a bias of recording, whereby there is some delay in the recording of the clinical event that triggered the initiation of treatment in the first place. The correlation with the CD4 count would suggest, however, that recording bias is unlikely to fully explain this rapid decline.

Our data confirm results from the Swiss HIV Cohort study¹² in which the incidence of opportunistic infections decreased in the early HAART era. Because of the large number of patients included in our study, we have been able to demonstrate differences in the rates of decline according to etiology. In particular, events of viral etiology declined more rapidly during the first 3 years of HAART than did events of other etiologies, whereas fungal events declined less rapidly. The reason for these differences is not clear. Before the availability of HAART, the use of disease-specific prophylaxis had already reduced the incidence of PCP, esophageal candidiasis, and atypical mycobacteriosis in those with low CD4 counts.^25,26 Although the incidence of bacterial and fungal infections may have been reduced as a result, there is no standardized prophylaxis for viral infections²⁷ and, thus, patients would not already have benefited from a reduction in incidence of these events before HAART. A more rapid decline in the incidence of these events may therefore be expected. In the pre-HAART era, some viral infections, including cytomegalovirus, often occurred at much lower CD4 counts than other types of AIDS events.²⁸ Thus, our results may simply reflect the fact that smaller CD4 increases are required to protect against viral infections than against events of other etiologies.

Although the reductions in incidence of AIDS events could largely be explained by improvements in the immunological and virological status of patients receiving HAART, the incidences of bacterial and viral events in the second 6-month period of follow-up were significantly lower than expected on the basis of changes in the CD4 count and HIV-1 RNA level alone. Rates of protozoal events also remained lower than expected, although this finding was nonsignificant, possibly because of the small number of events. These reductions in incidence may reflect other changes in care that often occur at the time of HAART initiation, such as closer patient monitoring in the first few months. Alternatively, HAART may exert an influence on other facets of the immune system that may lead to a reduction in incidenceof disease or may have a direct effect on bacterial or viral pathogens that is not mediated through improvements to the immune system.^3,29-31

Some aspects of our study design should also be considered as possible explanations for our findings. There is the possibility that in the second 6-month period, the study may have identified a group of healthy survivors at reduced risk of disease if those who were destined to die did so in the first 6-month period. In addition, the timing of the 6-month CD4 count and HIV-1 RNA level was subject to greater variation than that of the baseline count. Thus it may be expected that the relationship between clinical events and these markers may be weaker in the second period than in the first. Finally, the year of HAART initiation may affect the clinical outcome because of the introduction of more potent HAART regimens in recent years³⁰ or because of improvements to clinical care. All analyses were adjusted for calendar year of HAART initiation and, thus, this is unlikely to have affected our results greatly.

Three limitations of our study design should also be noted. First, only the initial occurrence of each type of AIDS event was reported. Thus, for patients who had AIDS at baseline (approximately 20% of the cohort), only new events that had not previously occurred were reported, and this group may be at reduced risk of a new event. Our analyses adjusted for previous AIDS at baseline and were also repeated after excluding those with a previous AIDS event with essentially similar results. Second, any increases in other non-AIDS clinical events that may become apparent with long-term use of HAART (eg, toxic effects or other comorbidities) would not be detected by this analysis. Consideration of all events should be made when discussing treatment options with patients. Third, each of the cohorts in this collaboration use their own methods to achieve standardization of AIDS events and HIV RNA levels. Although no further attempt was made to achieve standardization across the cohorts, it is unlikely that any minor differences in standardization across the cohorts will have had a substantial impact on our results.

In this large multicohort study we demonstrated a significant reduction of the incidence of opportunistic events, regardless of etiology. The pattern of decline was more pronounced for events of viral etiology and was maintained during the 3-year follow-up. For events of viral, bacterial, and fungal etiology, this decline was greater than expected on the basis of changes in CD4 count and HIV-1 RNA level, suggesting a benefit of HAART beyond the improvement of these surrogate markers.

Correspondence: Caroline A. Sabin, PhD, Department of Primary Care and Population Sciences, Royal Free and University College Medical School, Royal Free Campus, Rowland Hill Street, London NW3 2PF, England (c.sabin@pcps.ucl.ac.uk).

Accepted for Publication: September 22, 2004.

Financial Disclosure: None.

Funding/Support: The ART Cohort Collaboration is supported by the UK Medical Research Council (MRC), London, England, and GlaxoSmithKline. Sources of funding of individual cohorts include the Agence Nationale de Recherches sur le SIDA, Paris, France; the Institut National de la Santé et de la Recherche Médicale, Paris; the Foundation pour la Recherche Medicale, Paris; the Italian Ministry of Health, Rome; the Swiss National Science Foundation, Berne; the Dutch Ministry of Health, Welfare, and Sport, The Hague; the European Commission, Brussels, Belgium; the Province of British Columbia, Victoria; the Province of Alberta, Edmonton; the Michael Smith Foundation for Health Research, Burnaby, British Columbia; the Canadian Institutes of Health Research, Ottawa, Ontario; and unrestricted grants from GlaxoSmithKline, London, England, Roche, Basel, Switzerland, and Boehringer Ingelheim GmbH, Ingelheim, Germany. The Department of Social Medicine of the University of Bristol, Bristol, England, is the lead center of the MRC Health Services Research Collaboration.

Disclaimer: The sponsors had no input into the study design; the collection, analysis, or interpretation of data; or the decision to publish the findings. Furthermore, no drug-specific analyses were performed.

Acknowlegment: We thank all patients, physicians, data managers, and study nurses who were involved in the participating cohort studies.

Group Information: A full listing of all study centers and collaborators of the Antiretroviral Therapy Cohort Collaboration was published previously (http://www.art-cohort-collaboration.org. and Lancet. 1998;351:543-549 [http://www.lancet.com.).

Box Section Ref ID