When to Initiate Highly Active Antiretroviral Therapy: A Cohort Approach

Transcription

1 American Journal of Epidemiology Copyright 2003 by the Johns Hopkins Bloomberg School of Public Health All rights reserved Vol. 157, No. 8 Printed in U.S.A. DOI: /aje/kwg036 When to Initiate Highly Active Antiretroviral Therapy: A Cohort Approach Linda Ahdieh-Grant 1, Traci E. Yamashita 1, John P. Phair 2, Roger Detels 3, Steven M. Wolinsky 2, Joseph B. Margolick 4, Charles R. Rinaldo 5, and Lisa P. Jacobson 1 1 Department of Epidemiology, Johns Hopkins Bloomberg School of Public Health, Baltimore, MD. 2 Northwestern University Medical School and Howard Brown Health Center, Chicago, IL. 3 School of Public Health, University of California Los Angeles, Los Angeles, CA. 4 Department of Molecular Microbiology and Immunology, Johns Hopkins Bloomberg School of Public Health, Baltimore, MD. 5 Graduate School of Public Health, University of Pittsburgh, Pittsburgh, PA. Received for publication March 26, 2002; accepted for publication November 6, The appropriate immunologic stage of human immunodeficiency virus infection at which to initiate highly active antiretroviral therapy (HAART) among asymptomatic persons is a core question. A cohort approach using longitudinal data from the US Multicenter AIDS Cohort Study was used to mimic a clinical trial to assess the risk of acquired immunodeficiency syndrome (AIDS) by timing of therapy. Three treatment groups were defined according to CD4 + count (cells/µl) at HAART initiation between July 1995 and January 2000: <200 (deferral to <200, n = 127), (deferral to , n = 130), and (immediate treatment, n = 92). Survival analysis was used to compare time to AIDS between groups from the index visit until July The index visit for the immediate group was the one prior to HAART initiation. For the deferral groups, the index visit was a randomly selected, pre-haart, AIDS-free visit after July 1990 at which CD4 + counts were cells/µl. This strategy accounted for lead time bias. Compared with immediate treatment, the relative hazards of AIDS were 2.68 (p = 0.003) and 1.05 (p = 0.897) for deferral to <200 cells/µl and cells/µl, respectively. These results support recent US public health guidelines for deferring HAART initiation until a count of <350 cells/µl. Furthermore, results suggest a potential threshold for HAART initiation in the neighborhood of 275 cells/µl. cohort studies; epidemiologic methods; HIV Abbreviations: AIDS, acquired immunodeficiency syndrome; CI, confidence interval; D <200, deferred treatment group of men whose CD4 + counts were <200 cells/µl; D <350, deferred treatment group of men whose CD4 + counts were cells/µl; HAART, highly active antiretroviral therapy; HIV, human immunodeficiency virus. Introduction of highly active antiretroviral therapy (HAART) has profoundly altered the landscape of human immunodeficiency virus (HIV) infection. Although its ability to suppress plasma HIV RNA, raise circulating CD4 + counts, and reduce the incidence of acquired immunodeficiency syndrome (AIDS) and death has been well documented (1 5), the optimal stage of HIV infection at which to initiate HAART among asymptomatic persons is still under debate. Recent guidelines (6, 7) reflect a shift in opinion toward deferring HAART initiation until CD4 + counts fall below 350 cells/µl rather than the previously recommended 500 cells/µl. There is general agreement that HAART should be administered to symptomatic persons and those whose CD4 + counts are <200 cells/µl. Benefits of delayed treatment include reserving the limited number of available drugs until later in the course of HIV disease, reducing adverse effects and the expense of treatment, and minimizing the development of viral drug resistance (6). One of the primary arguments for early treatment is that it may delay progression to AIDS (8). Early treatment has also been postulated to preserve immune function that cannot later be reconstituted and to reduce the risk of viral transmission (9, 10). Dr. Ahdieh-Grant and Dr. Jacobson contributed equally to this study. Correspondence to Dr. Linda Ahdieh Grant, Johns Hopkins Bloomberg School of Public Health, Department of Epidemiology, 615 North Wolfe Street, Room E-7014, Baltimore, MD ( lahdieh@jhsph.edu). 738

2 When to Initiate HAART 739 The research and clinical communities have discussed the design and conduct of randomized clinical trials to determine the appropriate stage of HIV infection at which to initiate HAART (11). One such trial would randomize asymptomatic persons whose CD4 + counts are cells to either immediate or deferred treatment arms and monitor the incidence of AIDS and death. The principal strength of such a trial would be to minimize the effect of subjective factors related to treatment decisions. The major limitation would be the lengthy follow-up, during which therapeutics would likely change and comparability would be compromised. In addition, participants might not adhere to the original treatment if changes in practice occurred in the community. Thus, at the end of follow-up, differences in AIDS risk might not be due to the timing of treatment but rather to treatment choices. The goal of this analysis was to simulate such a trial by using data collected in an ongoing cohort study. Advantages of this setting include the comparability of treatments, given that HAART was initiated during the same calendar period for all participants, as well as improved efficiency since questions may be answered with existing data. An important limitation is the lack of randomization in treatment assignment. Another potential problem is lead time bias. In screening, lead time is the additional survival time associated with the detection of disease early in its natural history, that is, during the asymptomatic stage. In our context, lead time is the additional survival time prior to initiating therapy among those who start HAART at a later disease stage, that is, the time required for that person to progress from the early stage (i.e., a high CD4 + count) to the later stage (i.e., a low CD4 + count), at which time therapy is then initiated. Acknowledgment of lead time is essential when comparing survival times for persons at different disease stages. For example, a person initiating HAART with a CD4 + count of 150 cells/µl may appear to develop AIDS more quickly compared with a person initiating HAART with a CD4 + count of 450 cells/µl. However, the former person initiated HAART at a later disease stage, and the time it took to reach 150 cells/µl from 450 cells/µl must be considered. Accounting for lead time also permits observation of events that may occur during this progression from early to later disease stages. These concepts are illustrated in figure 1. One approach to account for lead time is to use group-level data to estimate the distribution of time spent progressing from a given CD4 + count category to a lower category (12, 13). An alternate approach, well suited for application in longitudinal cohort studies, is to capitalize on prospectively collected, standardized data on CD4 + counts prior to HAART initiation. The lead time for each person can then be determined by randomly selecting a visit at which the CD4 + count was cells/µl and then measuring the time from this point to HAART initiation. Thus, the AIDS-free time for each person is composed of two intervals: the lead time to HAART initiation (or to AIDS if it occurred before HAART initiation) and the time from HAART initiation to AIDS or censoring (if the person remained event free at HAART initiation). By using this latter strategy, we mimicked a clinical trial intended to determine whether early FIGURE 1. Illustration of lead time for comparison of treatment groups regarding time to acquired immunodeficiency syndrome (AIDS). Lead time is defined as the time from an earlier disease stage to a later disease stage among those who defer highly active antiretroviral therapy (HAART) initiation to when their CD4 + cell counts are lower. A) Naïve analysis, where AIDS-free time is defined as time from HAART initiation to AIDS (or censoring if AIDS free). For groups defined according to CD4 + cell count at HAART initiation (CD4 + cell count cells/µl, cells/µl, and <200 cells/ µl), the origin for the survival analysis would be when CD4 + cell counts were cells/µl, cells/µl, and <200 cells/µl, respectively. Persons who developed AIDS prior to HAART initiation would not be captured. B) Current analysis accounting for lead time (dotted line) in the deferred treatment arms. For the groups defined above, their AIDS-free times are determined from the same stage of disease, i.e., the index visit when CD4 + cell counts are cells/ µl. AIDS-free time is defined as the lead time in addition to the time from HAART initiation to AIDS (or censoring if AIDS free). Persons who developed AIDS prior to HAART initiation are captured in the analysis that incorporates lead time. versus later HAART initiation was associated with the same subsequent disease progression, where early and late were defined by CD4 + count. MATERIALS AND METHODS Multicenter AIDS Cohort Study The Multicenter AIDS Cohort Study, an ongoing cohort study of homosexual men in Baltimore, Maryland; Chicago, Illinois; Los Angeles, California; and Pittsburgh, Pennsylvania, enrolled 4,954 men in and 668 mainly African Americans in Written informed consent was obtained from all participants, in accordance with protocols approved by local institutional review boards at each site. Study participants returned every 6 months for visits that included an extensive interview about behavioral practices, medical history, and health care utilization; a physical examination; and a psychosocial self-assessment (14 16). In addition, participants provided blood specimens for storage in local/national repositories and for concomitant laboratory analyses including standardized flow cytometry (17, 18) and plasma HIV RNA determinations using the Roche Amplicor

3 740 Grant et al. assay (19). Men who had AIDS as well as men whose CD4 + counts were <200 cells/µl were contacted at least every 3 months to obtain information on recent clinical events, confirmed by medical record review. Reports by proxy contacts and passive surveillance were also used to determine vital status. Study questionnaires are available at the following Internet Web site: macs.html. Use of antiretroviral therapy At each visit, participants were asked for names of the antiretroviral medications used since their previous visit. This information was used to identify men who initiated HAART between July 1, 1995, and January 1, The definition of HAART (20) was guided by the US Department of Health and Human Services/Kaiser Family Foundation Panel guidelines (6) and the International AIDS Society USA Panel (7). The date of HAART initiation was the midpoint between the first visit at which a person reported HAART and the preceding visit. Outcome The outcome for this analysis was clinical AIDS, defined by the opportunistic infections and malignancies identified by the Centers for Disease Control and Prevention 1993 criteria (21). Asymptomatic men whose CD4 + counts were <200 cells/µl were not considered to have AIDS. All AIDSdefining events occurring prior to July 1, 2000, were included. Nested cohort study design FIGURE 2. Algorithm describing the assignment of treatment groups and the identification of the index visit from longitudinal retrospective data. H represents the visit at which study participants first reported highly active antiretroviral therapy (HAART). Persons were assigned to treatment groups according to the average CD4 + cell counts at the two visits (H-1 and H-2) preceding HAART initiation. Persons in the immediate treatment group initiated HAART with an average CD4 + cell count at H-1 and H-2 of cells/µl. Persons in the deferred <350 cells/µl group initiated HAART at H with an average CD4 + cell count at H-1 and H-2 of cells/µl. Persons in the deferred <200 cells/µl group initiated HAART at H with an average CD4 + cell count at H-1 and H-2 of <200 cells/µl. For the immediate group, the index visit was H-1. For the deferred groups, the index visit was randomly selected from all eligible visits, defined as visits following July 1, 1990, prior to HAART initiation at which CD4 + cell counts were cells/µl and persons were free of acquired immunodeficiency syndrome (AIDS). Our objective was to mimic a clinical trial in which AIDSfree persons with CD4 + counts of cells/µl would be randomly assigned to either immediate (defined as starting HAART when CD4 + counts were cells/µl) or one of two deferred treatment groups: those starting HAART when counts were <200 cells/µl and cells/µl. Our study population consisted of HIV-infected men who had a CD4 + count of cells/µl while AIDS free and who later reported use of HAART. Only visits after July 1, 1990, were included to account for prognostic changes due to historical events, for example, use of combination medications, prophylaxis for opportunistic infections, and diagnostic practices. Based on his CD4 + count preceding HAART initiation (defined as the average of the CD4 + counts measured at the two semiannual visits immediately prior to HAART initiation), each participant was assigned to one of three categories ( cells/µl, cells/µl, and <200 cells/µl), as illustrated in figure 2. The first group corresponds to the immediate treatment arm in a clinical trial, whereas the second and third groups correspond to deferred treatment arms. Hereafter in this paper, the groups are referred to as the immediate group, the deferred group <350 (D <350), and the deferred group <200 (D <200). In the analogous clinical trial, follow-up would begin at the date of randomization. In our study, the origin for followup was the index visit. For the immediate group, the index visit was the semiannual visit approximately 3 months prior to HAART initiation. For each person in the deferred groups, the index visit was randomly selected from all study visits at which a person was eligible for the study, that is, AIDS free and having CD4 + counts of cells/µl. The random selection was performed by using the sample function in S- Plus software, version 5.1, release 1 (MathSoft, Inc., Seattle, Washington). Although participants were required to be AIDS free at the index visit, it was possible for AIDS to occur prior to HAART initiation. Alternatively, one could select the first or last visit when a person s CD4 + count was cells/µl. This would overestimate or underestimate lead time, respectively, because in a clinical trial, participants would have had CD4 + counts of cells/µl for varying (and unknown) lengths of time when the trial was initiated. Statistical methods We compared groups at the index visit for prior use of antiretroviral therapy, demographic characteristics, and markers of disease stage. Wilcoxon tests were used to compare groups for differences in continuous variables, and chisquare tests were used to assess the significance of differences in discrete variables.

4 When to Initiate HAART 741 TABLE 1. Descriptive characteristics of men in the Multicenter AIDS* Cohort Study at the index visit (July 1990 September 1999) according to treatment group, United States * AIDS, acquired immunodeficiency syndrome; IQR, interquartile range; HIV, human immunodeficiency virus. We conducted a time-to-event analysis to determine whether treatment deferrals were associated with a decreased probability of remaining AIDS free (22). Men who developed AIDS before July 1, 2000, contributed time from the index visit to the date of diagnosis; those who developed AIDS after July 1, 2000, were censored at this date. Men who withdrew from the study before July 1, 1999, were censored at the time of withdrawal; men followed after this date were administratively censored at July 1, Time from the index visit to AIDS by treatment group was depicted graphically by using Kaplan-Meier survival analysis. Cox proportional hazards models were used to assess the effect of treatment timing on progression to AIDS. RESULTS Deferred group, <200 cells/µl (D <200) A total of 689 participants initiated HAART between July 1, 1995, and January 1, 2000; 546 had their CD4 + counts measured in the year prior to initiation, 437 of whom had CD4 + counts of <500 cells/µl. Of these men, 88 either were not AIDS free after July 1, 1990, or did not contribute any visits for which their CD4 + counts were cells/µl during this period. Thus, 349 participants remained: 92 in the immediate group and 127 and 130 in the D <200 and D <350 groups, respectively. The median numbers of eligible visits for D <200 and D <350 were two (range, one to seven) and three (range, one to nine), respectively. The overall median (25th, 75th percentiles) follow-up time from the index visit was 4.6 (3.2, 7.1) years. All AIDS-free men had contact with the study within 9 months of the date of analysis; of these, four died who did not have AIDS (two in the D <350 group Treatment group Deferred group, cells/µl (D <350) No Median (IQR*) date of index visit Prior antiretroviral therapy (%) Median (IQR) CD4 + cell count Median (IQR) log 10 HIV* RNA May 1992 (May 1991, April 1993) October 1993 (July 1992, February 1995) Immediate group, cells/µl October 1996 (April 1996, May 1997) p value p value (D <200 (D <200 and compared D <350 compared with D <350) with immediate group) <0.001 <0.001, < <0.001, (389, 460) 415 (378, 443) 410 (382, 451) , (4.16, 4.95) (n = 101) 4.44 (3.89, 4.79) (n = 116) 4.35 (3.90, 4.94) (n = 92) , Median (IQR) age (years) 39.3 (35.4, 44.3) 40.4 (34.5, 45.5) 42.6 (38.3, 47.7) <0.001, Race (%) White , African American Other Median (IQR) no. of years 6.7 (4.7, 8.0) 7.5 (5.1, 9.4) 11.3 (7.2, 12.2) <0.001, <0.001 from first HIV-positive visit to index visit CD4 + slope per year prior to index visit (IQR) 30 ( 69, 7) 27 ( 51, 6) 29 ( 49, 14) , and two in the immediate group) and 96.3 percent of the others had visits after July The groups did not differ significantly (p < 0.05) by race, CD4 + count, or HIV RNA levels, although the D <200 group had slightly higher HIV RNA values at the index visit (table 1). In addition, the rate of decline in CD4 + count prior to index (approximately 30 cells/year overall) was similar across groups (p = 0.77). The groups differed in the proportion reporting prior antiretroviral therapy, however. Specifically, 46.5 percent of the D <200 group and 59.2 percent of the D <350 group reported prior antiretroviral therapy compared with 73.9 percent of the immediate group. The observed differences are explained by the study s design; the index visits for persons in the immediate group occurred in the late 1990s compared with the early 1990s for the deferred groups. The design also explains why men in the deferred groups were younger (age 39.3 and 40.4 years, respectively) than those in the immediate group (age 42.6 years) and why the deferred groups had less seropositive time on study prior to index, since the date of the first HIV seropositive visit was similar across groups. From the direction of these differences (i.e., more experience with antiretroviral therapy and older age), and because these factors were prognostic in the pre-haart era, the immediate group may be expected to subsequently progress more rapidly. By design, time from index to HAART initiation was approximately 0.25 years for the immediate group. The median times from index to HAART initiation were 3.1 and 4.3 years for the D <350 and D <200 groups, respectively, corresponding to the lags that would exist in the deferred groups in a clinical trial from randomization to treatment

5 742 Grant et al. FIGURE 3. Probability of remaining free of acquired immunodeficiency syndrome (AIDS) following the index visit, according to treatment group, among men in the Multicenter AIDS Cohort Study, United States, July 1, 1990 July 1, Unadjusted relative hazards (RH) are presented for the deferred group with <350 cells/µl (D <350) and the deferred group with <200 cells/µl (D <200) compared with the immediate group. Treatment groups were defined according to the average CD4 + cell count (cells/µl) at the two visits preceding HAART initiation. initiation. Although there was sufficient overlap in HAART initiation time, men with the lowest CD4 + counts initiated HAART the earliest, as expected from previous studies (4, 23); 68.5 percent of those in the immediate group initiated HAART before July 1997 compared with 77.7 percent and 85.0 percent in the D <350 and D <200 groups, respectively (p = 0.004). All of the men initiated treatment within a 4-year period. The components of the initial HAART regimens reported did not differ significantly across treatment groups and represented those used by the overall Multicenter AIDS Cohort Study population (20). In addition, the treatment groups did not differ in the proportion of men reporting more than three drugs in their initial HAART regimen (data not shown). Slightly more of the immediate group initiated HAART with a protease inhibitor-sparing regimen, not surprising given that more of them started HAART after July Figure 3 depicts AIDS-free time according to treatment group. A total of 103 cases of AIDS occurred in the three groups. As shown, 28 men (21.5 percent) in the D <350 group and 64 men (50.4 percent) in the D <200 group developed AIDS following the index visit. Of these, 60.7 percent (17 of 28) and 75.0 percent (48 of 64), respectively, developed AIDS prior to HAART initiation. Since 71 percent of the AIDS cases in the deferred groups occurred prior to HAART initiation, these men would not be affected by the use of HAART (i.e., drug classes, number of antiretroviral therapies, or treatment adherence). Compared with the immediate group, the relative hazards of AIDS were 1.05 (95 percent confidence interval (CI): 0.52, 2.13) for men in the D <350 group and 2.68 (95 percent CI: 1.40, 5.14) for those in the D <200 group. Characteristics that differed between the deferral groups and the immediate group were not significantly prognostic after accounting for treatment. This finding occurred because the survival likelihoods associated with values obtained under HAART are not equivalent to those of similar values long before HAART was initiated (20). However, since the two deferral groups contributed sufficient pre-haart time, we restricted the analysis comparing these two groups to the 217 men for whom all determinants were available at the index visit (unadjusted relative hazard = 2.19, 95 percent CI: 1.35, 3.57). The D <200 group still had a significantly worse prognosis (relative hazard = 2.00, 95 percent CI: 1.23, 3.26) than the D <350 group after adjustment for HIV RNA, prior antiretroviral therapy use, and seropositive follow-up time at the index visit. To investigate a possible threshold between 200 and 349 cells/µl at which the risk of AIDS increased, we examined those men in the D <350 group whose CD4 + counts were above and below 275 cells/µl, given that the median CD4 + count at HAART initiation was 270 cells/µl. Compared with the immediate group, the 67 men who deferred HAART to a CD4 + count of <275 cells/µl had an increased, but nonsignificant risk of AIDS (relative hazard = 1.40, p = 0.385), whereas the relative hazard of AIDS for those 63 men who deferred HAART to a CD4 + count of 275 cells/µl was 0.69 (p = 0.411), suggesting a potential threshold below 300 cells/ µl. The D <350 group was composed of men whose index visit spanned the period Since all persons had

6 When to Initiate HAART 743 TABLE 2. Challenges to the study of when to initiate highly active antiretroviral therapy and strategies to address them in clinical trials and nested cohort studies Challenge Clinical trial Strategy to address challenge Nested cohort study Selection of treatment Randomization Restrict window of HAART* initiation 1) by total length and 2) to the period prior to publication of established guidelines Imbalance between treatment groups of factors related to progression Check and adjust Check and adjust Maintenance of the protocol for treatment deferral Waiting for the threshold to start treatment Attending minimally spaced study visits Representation of the population for underlying disease progression * HAART, highly active antiretroviral therapy. similar CD4 + counts when HAART was initiated ( cells/µl) and similar CD4 + counts at the index visit ( cells/µl), those whose index visits occurred earlier progressed more slowly prior to HAART. When we stratified the D <350 group according to index visit date (median, approximately January 1993), the relative hazards for those whose index visits occurred before and after January 1993 were 1.18 (95 percent CI: 0.49, 2.83) and 1.06 (95 percent CI: 0.48, 2.32), respectively, compared with the immediate group. These similar relative hazards suggest that the difference in AIDS progression prior to HAART did not account for the lack of difference between the D <350 group overall and the immediate group. DISCUSSION 1) Emphasize importance of protocol adherence 2) Conduct intent-to-treat analysis and report lack of protocol adherence 3) Limit analysis to those who adhere to the protocol Compare with historical controls, if available Given increasing concern regarding the challenges of maintaining HAART regimens, early initiation for asymptomatic persons may not be justified unless it substantially improves prognosis. Consistent with our results, previous studies showed that a low CD4 + count at the time of HAART initiation significantly predicted time to both AIDS and death (20, 24 30). However, these studies did not account for the lead time prior to HAART initiation. Using data collected by the Multicenter AIDS Cohort Study, we found that the risk of AIDS did not differ between those delaying treatment until their CD4 + counts were cells/µl and those who initiated treatment with CD4 + counts of >350 cells/µl. Men who delayed treatment until they had <200 CD4 + cells/µl had a significantly worse prognosis. These results do not suggest compelling risks associated with delaying HAART initiation until CD4 + counts are cells/µl. Furthermore, the lack of an increased risk of AIDS among those who initiated treatment with CD4 + counts of cells/µl suggests a potential threshold for HAART initiation in the neighborhood of 275 cells/µl. Other 1) Analyze all participants regardless of the window between the threshold and treatment 2) Limit analysis to those who initiate HAART within a window of crossing the threshold (e.g., 1.5 years) Compare progression with peers in the cohort benefits and risks of early versus late treatment will require additional studies. Limitations of this analysis are the lack of randomization to group, a relatively extended period before treatment onset after reaching initiation thresholds, and potentially different underlying progression in the three groups (table 2). We investigated each of these limitations and discuss their implications below. First, external factors may influence a participant s choice of treatment; randomization in clinical trials removes these individual selection factors. In our study, the upsurge of treatment initiation from 1995 to 1999 not only resulted in the use of HAART by 85 percent of the cohort in a relatively short period but also minimized differences in the types of treatments selected (20). Also of note is that this upsurge in HAART use occurred prior to, and thus was not affected by, dissemination of established guidelines regarding when to initiate HAART. Second, one must evaluate and adjust for imbalances among treatment groups with regard to factors related to disease progression. Longitudinal cohort studies with ongoing follow-up allow for such investigation of confounders and adjustment as needed. On the issue of confounding, although biomarkers, that is, CD4 + count, have been shown to predict outcome in the HAART era, the progression associated with a given value in the HAART era differs from that observed long before HAART was available. For example, we previously showed that although CD4 + count at HAART initiation predicts progression, a person starting HAART when the CD4 + count is <200 cells/ µl has a progression likelihood similar to that for someone with approximately 300 more CD4 + cells/µl in the natural history era (20). Thus, the effects of these characteristics measured in different eras do not have the same confounding relation. Regarding differences in prior use of therapy, although previous studies, including the Multicenter AIDS

7 744 Grant et al. Cohort Study, showed that prior drug exposure was associated with a short-term effect of HAART on biomarkers (4, 31), in our study it was not a confounder for progression to AIDS, a finding consistent with other reports (12, 20, 27). In addition, prior antiretroviral therapy in this study refers to use of medications prior to the index visit and not prior to HAART; at HAART initiation (the time when prior antiretroviral therapy may modify HAART s effect), the groups did not differ significantly, with 83.5 percent, 85.4 percent, and 73.9 percent of the D <200, D <350, and immediate groups, respectively, reporting a history of antiretroviral therapy. A third important challenge involves maintaining, for the deferred groups, the protocol for initiating treatment. In a clinical trial, the protocol dictates that persons assigned to the deferred arms initiate HAART the first time they cross the designated treatment threshold. In a cohort study, however, those in the deferred groups do not necessarily initiate HAART when their CD4 + counts first cross the respective thresholds. In our cohort, these median delays in HAART initiation for those in the D <200 and D <350 groups were 1.8 years (25th, 75th percentiles = 0.8, 2.9) and 3.7 years (25th, 75th percentiles = 1.7, 6.1), respectively. When we limited the analysis to those who initiated treatment within 1.5 years of crossing their respective thresholds, the primary inferences held; that is, the excess risk associated with deferral to <200 cells/µl was 2.78 (p = 0.006), and the risk associated with deferral to <350 cells/µl was 0.70 (p = 0.546). Limiting the analysis to protocol adherers may improve validity in a clinical trial but would limit power and the inferences that can be drawn regarding treatment effectiveness. Finally, a fourth issue to consider when interpreting the results from both a clinical trial and a cohort study is the generalizability of study findings to the underlying population. In most clinical trials, it is largely unknown, although comparisons can be made with external published data if available. In an observational cohort, we can compare disease progression in study participants and their peers not included in the analysis. Since all participants in our study were treated with HAART, two groups may not have been represented: 1) persons whose CD4 + counts did not decline to <350 cells/µl and therefore did not receive treatment and 2) those who progressed so rapidly that they developed AIDS and died before receiving treatment. When we examined all Multicenter AIDS Cohort Study participants whose CD4 + counts were cells/µl during the same calendar period as the index visit of HAART users, approximately 23 percent of the cohort did not progress to <350 CD4 + cells/µl after 3 years, up to July Including these slower progressors in our deferral groups would have minimized the observed differences with the immediate group. As mentioned, very rapid progressors also may have been excluded. Of note, we did not exclude men who developed AIDS prior to the use of HAART. Because we allowed men who developed AIDS prior to initiating HAART to be included, our groups contained fast progressors, that is, those who rapidly transitioned from having CD4 + counts of >350 cells/µl to AIDS. This is the primary reason that we did not include death as an outcome; data for men who died prior to the initiation of HAART could not be captured. If death had been considered an outcome, it would have occurred only after treatment and not during the deferral period, thus introducing a bias across groups. However, the fastest progressors, that is, those whose disease progressed so rapidly that they died prior to July 1995, may have been excluded. When we included all AIDS-free men with counts of >350 cells/µl during the same calendar period as the index visit for HAART users, approximately 20 percent developed AIDS and died prior to July Given the rapidity of progression in these persons, it is unlikely that, even in a clinical trial, there would be ample opportunity to offer treatment. It is also not clear whether such treatment would be effective enough to prevent disease. Inclusion of these men in our deferral groups would have heightened the differences observed between the immediate and deferred groups. In summary, the biases introduced by excluding the slowest and fastest progressors from our deferral groups most likely counteracted each other. As noted in the Results section of this paper, when we dichotomized the D <350 group to examine the slower progressors, no appreciable changes in the risk of AIDS were found. Thus, these slower progressors in the D <350 group did not explain the observed lack of difference in the risk of AIDS progression between the D <350 group and the immediate group. If a proportion of the immediate group was comprised of slower HIV progressors, it may partially explain the prognostic difference between the D <200 and immediate groups. In fact, the data suggest that slower progressors were included in the immediate group since their first seropositive visits occurred at the same time as those in the deferred groups, yet they were able to remain AIDS free and start HAART when their CD4 + counts were higher. Thus, the lack of difference between immediate treatment and deferral to <350 cells/µl is conservative, given this group of slower progressors. Interestingly, recruitment began in early 2002 for a large, randomized trial to address the timing of HAART initiation. The study, known as SMART (Strategies for Management of Anti-Retroviral Therapies), will randomly assign approximately 6,000 persons to immediate (>350 cells/µl) or deferred (<250 cells/µl) treatment and follow them for 9 years (32). Given the time until definitive results are available from this trial, nested longitudinal cohort studies such as ours play an important role in providing timely, complementary information regarding patterns of therapeutic use as well as the population effectiveness of antiretroviral therapies (33). Because HAART was unavailable until late 1995 and was then initiated by persons at all stages of HIV infection, we capitalized on the forced deferral of initiation. Furthermore, collection of standardized longitudinal data enabled us to incorporate lead time. One outstanding question is whether there is a threshold between 200 and 350 cells/µl at which HAART initiation should ideally occur. While the data presented did not indicate a significant excess risk associated with deferral to <275 cells/ µl, our ability to investigate this question was limited by sample size considerations. By coincidence, this threshold was similar to that of 250 cells/µl currently being used in the SMART study (32). To address this question further, it will be of interest to pool similar data from multiple cohorts as well as to compare these results with those of clinical trials addressing similar issues.

8 When to Initiate HAART 745 ACKNOWLEDGMENTS The authors are grateful to Dr. Alvaro Muñoz for his much-valued contribution to this intiative. The Multicenter AIDS Cohort Study includes the following: Baltimore, Maryland: Johns Hopkins Bloomberg School of Public Health Dr. Joseph B. Margolick (Principal Investigator), Dr. Haroutune Armenian, Dr. Barbara Crain, Dr. Adrian Dobs, Dr. Homayoon Farzadegan, Dr. Nancy Kass, Dr. Shenghan Lai, Dr. Justin McArthur, and Dr. Steffanie Strathdee; Chicago, Illinois: Howard Brown Health Center, The Feinberg School of Medicine, Northwestern University, and Cook County Bureau of Health Services Dr. John P. Phair (Principal Investigator), Dr. Joan S. Chmiel (Co-Principal Investigator), Dr. Sheila Badri, Dr. Bruce Cohen, Dr. Craig Conover, Dr. Maurice O Gorman, Dr. Frank Pallela, Dr. Daina Variakojis, and Dr. Steven M. Wolinsky; Los Angeles, California: University of California, UCLA Schools of Public Health and Medicine Dr. Roger Detels and Dr. Beth Jamieson (Principal Investigators), Dr. Barbara R. Visscher (Co-Principal Investigator), Dr. Anthony Butch, Dr. John Fahey, Dr. Otoniel Martínez- Maza, Dr. Eric N. Miller, John Oishi, Dr. Paul Satz, Dr. Elyse Singer, Dr. Harry Vinters, Dr. Otto Yang, and Dr. Stephen Young; Pittsburgh, Pennsylvania: University of Pittsburgh, Graduate School of Public Health Dr. Charles R. Rinaldo (Principal Investigator), Dr. Lawrence Kingsley (Co-Principal Investigator), Dr. James T. Becker, Dr. Phalguni Gupta, Dr. John Mellors, Dr. Sharon Riddler, and Dr. Anthony Silvestre; Data Coordinating Center: Johns Hopkins Bloomberg School of Public Health Dr. Alvaro Muñoz (Principal Investigator), Dr. Lisa P. Jacobson (Co- Principal Investigator), Dr. Stephen Cole, Dr. Stephen J. Gange, Dr. Linda Ahdieh-Grant, Janet Schollenberger, Dr. Eric Seaberg, Dr. Michael Silverberg, Dr. Sol Su, and Traci E. Yamashita; Bethesda, Maryland: National Institutes of Health: National Institute of Allergy and Infectious Diseases Dr. Carolyn Williams; Bethesda, Maryland: National Cancer Institute Dr. Sandra Melnick. Internet Web site: REFERENCES 1. Autran B, Carcelain G, Li TS, et al. Positive effects of combined antiretroviral therapy on CD4+ T cell homeostasis and function in advanced HIV disease. Science 1997;277: Hammer SM, Squires KE, Hughes MD, et al. A controlled trial of two nucleoside analogues plus indinavir in persons with human immunodeficiency virus infection and CD4 cell counts of 200 per cubic millimeter or less. N Engl J Med 1997;337: Schapiro JM, Winters MA, Efron SF, et al. The effect of highdose saquinavir on viral load and CD4+ T-cell counts in HIV infected patients. Ann Intern Med 1996;124: Yamashita TE, Phair JP, Muñoz A, et al. Immunologic and virologic response to highly active antiretroviral therapy in the Multicenter AIDS Cohort Study. AIDS 2001;15: Detels R, Muñoz A, McFarlane G, et al. Effectiveness of potent antiretroviral therapy on time to AIDS and death in men with known HIV infection duration. JAMA 1998;280: US Department of Health and Human Services/Henry J. Kaiser Family Foundation Panel on Clinical Practices for the Treatment of HIV Infection. Guidelines for the use of antiretroviral agents in HIV-infected adults and adolescents. February ( 7. Carpenter CCJ, Cooper DA, Fischl MA, et al. Antiretroviral therapy in adults. Updated recommendations of the International AIDS Society USA Panel. JAMA 2000;283: Opravil M, Ledergerber B, Furrer H, et al. Clinical efficacy of early initiation of HAART in patients with asymptomatic HIV infection and CD4 cell count > /l. AIDS 2002;16: Ho DD. Time to hit HIV, early and hard. N Engl J Med 1995; 333: Quinn TC, Wawer MJ, Sewankambo NK, et al. Viral load and heterosexual transmission of human immunodeficiency virus type 1. N Engl J Med 2000;342: Huff B. When to begin treatment, when to switch? A brief summary of discussions between clinicians, researchers, and statisticians at a government-sponsored meeting. DAIDS Workshop on Long-Term Studies of HIV Infections. January 11 13, 2000, Washington, DC. HIV Clin Trials 2000;1: Anastos K, Barron Y, Weiser B, et al. Risk of progression to AIDS and death in HIV infected women initiating highly active antiretroviral therapy at different stages of disease. Arch Intern Med 2002;162: Cole SR, Li R, Anastos K, et al. Lead-time adjustment in cohort studies evaluating when to initiate therapy. Presented at the XIV International AIDS Conference, Barcelona, Spain, July (Oral presentation TuOrB1143). 14. Kaslow RA, Ostrow DG, Detels R, et al. The Multicenter AIDS Cohort Study: rationale, organization, and selected characteristics of the participants. Am J Epidemiol 1987;126: Detels R, Phair JP, Saah AJ, et al. Recent scientific contributions to understanding HIV/AIDS from the Multicenter AIDS Cohort Study. J Epidemiol 1992;2(suppl):S11 S Dudley J, Jin S, Hoover D, et al. The Multicenter AIDS Cohort Study: retention after 9½ years. Am J Epidemiol 1995;142: Giorgi JV, Cheng HL, Margolick JB, et al. Quality control in the flow cytometric measurement of T-lymphocyte subsets: the Multicenter AIDS Cohort Study experience. Clin Immunol Immunopathol 1990;55: Schenker EL, Hultin LE, Bauer KD, et al. Evaluation of a dualcolor flow cytometry immunophenotyping panel in a multicenter quality assurance program. Cytometry 1993;14: Kirstein LM, Mellors JW, Rinaldo CR, et al. Effects of anticoagulant, processing delay, and assay method (branched DNA versus reverse transcriptase PCR) on measurement of human immunodeficiency virus type 1 RNA levels in plasma. J Clin Microbiol 1999;37: Jacobson LP, Li R, Phair J, et al. Evaluation of the effectiveness of highly active antiretroviral therapy in persons with human immunodeficiency virus using biomarker-based equivalence of disease progression. Am J Epidemiol 2002;155: revised classification system for HIV infection and expanded surveillance case definition for AIDS among adolescents and adults. MMWR Recomm Rep 1992;41(RR-17): Miller RG. Survival analysis. New York, NY: John Wiley & Sons, Ahdieh L, Gange SJ, Greenblatt R, et al. Selection by indication of potent antiretroviral therapy use in a large cohort of women infected with human immunodeficiency virus. Am J Epidemiol 2000;152: Montaner J, Hogg R, Wood B, et al. Further characterizing determinants of disease progression among HIV-1 infected

9 746 Grant et al. patients initiating triple drug therapy. Presented at the 1st IAS Conference on HIV Pathogenesis and Treatment, Buenos Aires, Argentina, July (Oral presentation LB 010). 25. Chen R, Westfall A, Cloud G, et al. Long-term survival after initiation of antiretroviral therapy. Presented at the 8th Conference on Retroviruses and Opportunistic Infections, Chicago, Illinois, February (Poster presentation 341). 26. Sterling TR, Chaisson RE, Bartlett JG, et al. CD4+ lymphocyte level is better than HIV-1 plasma viral load in determining when to initiate HAART. Presented at the 8th Conference on Retroviruses and Opportunistic Infections, Chicago, Illinois, February (Poster presentation 519). 27. Hogg RS, Yip B, Chan KJ, et al. Rates of disease progression by baseline CD4 cell count and viral load after initiating tripledrug therapy. JAMA 2001;286: Palella F, Knoll M, Moorman A, et al. Lower mortality in ambulatory HIV-infected patients who initiate antiretroviral therapy at higher CD4+ cell counts. Presented at the 9th Conference on Retroviruses and Opportunistic Infections, Chicago, Illinois, February Egger M, May M, Chene G, et al. Prognosis of HIV-1 infected patients starting highly active antiretroviral therapy: a collaborative analysis of prospective studies. Lancet 2002;360: Ledergerber B, Egger M, Erard V, et al. AIDS-related opportunistic illnesses occurring after initiation of potent antiretroviral therapy. JAMA1999;282: Palella FJ Jr, Chmiel JS, Moorman AC, et al. Durability and predictors of success of highly active antiretroviral therapy for ambulatory HIV-infected patients. AIDS 2002;16: National Institute of Allergy and Infectious Diseases. A large, simple trial comparing two strategies for management of antiretroviral therapy (The SMART Study): a multicenter study of the Terry Beirn Community Programs for Clinical Research on AIDS. Version 1.0. September ( 33. Muñoz A, Gange SJ, Jacobson LP. Distinguishing efficacy, individual effectiveness and population effectiveness of therapies. AIDS 2000;14:754 6.