Timing of HAART Initiation and Clinical Outcomes in Human Immunodeficiency Virus Type 1 Seroconverters FREE

The Quiz Ref IDintroduction of highly active antiretroviral therapy (HAART) in 1996 reduced morbidity and mortality rates in human immunodeficiency virus type 1 (HIV-1)–infected individuals.¹ Randomized controlled trials^2- 3 conducted in immunocompromised patients (eg, those with a CD4 cell count ≤200/μL) demonstrated that rates of AIDS or death were halved in patients starting HAART compared with rates in patients treated with drugs from only 1 class during approximately 1 year.

A central unresolved issue in the care of HIV-1–infected patients is when HAART should be initiated. Randomized evidence is unlikely to be available before 2015.⁴ Observational studies of 3 large multicenter seroprevalent cohorts^5- 7 have suggested clinical benefit to initiating therapy at CD4 cell counts greater than 350/μL, but the magnitude and thresholds for benefit were quite different.

The objective of the present study was to provide clinically relevant information about the relative and absolute benefits of HAART initiation at different CD4 cell counts to support treatment decisions for AIDS-free, HAART-naive individuals living with HIV. We applied a novel approach to a cohort of 9455 HIV-1 seroconverters to estimate the benefit of initiating vs deferring HAART on long-term disease progression and death.

Quiz Ref IDPatients included in this analysis were enrolled in 1 of 23 clinical cohorts in Europe, Australia, and Canada participating in the CASCADE (Concerted Action on SeroConversion to AIDS and Death in Europe) Collaboration, which pools data on individuals with a well-estimated date of seroconversion (<2 years between the last negative and first positive HIV test results).⁸ Individuals 13 years and older at seroconversion were included in this analysis.

All the clinical cohorts participating in the CASCADE Collaboration received approval from their individual ethics review boards except the Danish cohort, which received approval from the National Data Registry Surveillance Agency because Danish law allowed collection and pooling of anonymous clinical data with approval from this agency alone. Two ethics review boards deemed their cohort participants exempt from providing signed informed consent. Signed informed consent was obtained from all others. Approval was also given by all ethics review boards to pool anonymous data for analyses and dissemination. This analysis was reviewed by the institutional review board at the University of North Carolina and was determined to be exempt from further review.

We created a set of sequential nested subcohorts (a special case of a nested structural model^9- 10) rather than a marginal structural model, as used in a recent analysis.⁵ We first considered all individuals who were eligible as of January 1, 1996, and imagined a cohort study in which the subsequent disease progression of those who initiated HAART during this month was compared with that of patients who did not initiate HAART during this month (Figure 1). In patients who remained HAART naive and otherwise eligible at the end of January 1996, we defined a new cohort for February 1996 to compare individuals who first initiated HAART in this month with those who did not initiate HAART during this month. We created a new subcohort with all eligible individuals for each month between January 1996 and May 2009, classified each treatment-naive individual in the subcohort according to whether they initiated HAART in the index month, pooled data across all 161 subcohorts, stratified data into separate analyses based on CD4 cell count at baseline, and, finally, estimated the absolute and relative measures of association with HAART initiation. We used a robust variance¹¹ to account for the fact that the same individual could contribute to more than 1 subcohort. To emulate the clinical scenario in which treatment decisions are made, we did not select a single alternative treatment strategy. Rather, we allowed the comparison group to encompass the range of treatment strategies present in this population. Thus, the survival times of patients who deferred HAART in the index month were used to represent the average population prognosis of individuals who were AIDS free and HAART naive with a CD4 cell count in the specified stratum but did not start HAART immediately, weighted by the number of trials each individual contributed in the CD4 stratum.

HAART was defined as any regimen containing 3 or more antiretroviral agents. Patients were eligible if they (1) were HAART naive as of the first of the month, (2) had not experienced the end point of interest (ie, AIDS or death) as of the end of the month, (3) had no more than 21 days (cumulative) of monotherapy or dual therapy, and (4) had a qualifying CD4 cell count (<800/μL ≥180 days after seroconversion and in the previous 365 days). Eligibility criteria were time varying. A patient who did not have a qualifying CD4 cell count available at the time of the first subcohort for which he or she was otherwise eligible could still be included in a subsequent subcohort as soon as a qualifying CD4 cell count was recorded.

The primary outcome of interest was the combined end point of time to first AIDS diagnosis or death from any cause. Analyses were repeated using death from all causes as the sole outcome. For each subcohort, follow-up began on the first day of the next month. Patients who did not experience an outcome of interest during follow-up were censored when they were last known to be alive.

We considered the following potential confounders: female sex, injecting drug use (IDU) as likely mode of transmission, documented seroconversion illness, and hepatitis B and hepatitis C virus co-infection. Time-varying covariates included age, duration of infection, calendar year, CD4 measures (most recent, nadir, number of tests, and days since last test), and viral load measures (availability of ≥1 tests, most recent [log₁₀ copies per milliliter], peak [log₁₀ copies per milliliter], number of tests, and days since last test). All time-varying characteristics were measured before the first day of follow-up.

Kaplan-Meier survival curves were used to visualize the crude (unadjusted) effect of initiating HAART compared with not initiating HAART in the index month, pooling across subcohorts. We estimated the hazard ratios (HRs) for initiating HAART compared with deferring HAART during the index month separately for 5 CD4 strata (0-49, 50-199, 200-349, 350-499, and 500-799/μL) using Cox proportional hazards regression models. All analyses followed an intent-to-treat approach and did not consider treatment changes (ie, interruptions, discontinuations, and later initiations).

To account for potential differences in the baseline prognosis of participants who initiated HAART compared with those who deferred HAART during the index month, we estimated inverse probability-of-treatment weights as a function of baseline covariate values. We used these weights to create adjusted Kaplan-Meier survival curves,¹² to estimate adjusted HRs using weighted Cox proportional hazards regression models, and to estimate the adjusted absolute effect of HAART initiation on the cumulative risk of AIDS and death.¹³ Weights were truncated at the 0.05th and 99.95th percentiles to reduce their variability and improve the stability of the final effect estimates.¹⁴ Confidence intervals (CIs) on risk differences were obtained by bootstrap with 1000 complete resamples with replacement from these data.^15- 16 We assumed a normal approximation of the parameter distribution and used the empirical standard error.

We assessed the sensitivity of the results to alternative ways of conducting the analysis. These alternatives included (1) shortening the period during which CD4 cell counts were considered eligible from 365 days to 45 days, which decreased the number of subcohorts in which an individual participated when his or her CD4 cell count had not been obtained immediately before or during the subcohort month; (2) beginning follow-up in January 1998 rather than in January 1996 to assess the effect of early suboptimal HAART regimens; and (3) requiring a baseline viral load for subcohort eligibility. We examined the impact of nonstandard treatment in the comparison group by censoring follow-up at the earliest of the 22nd day of cumulative monotherapy or dual therapy or 6 months after the patient's first CD4 cell count less than 200/μL if he or she remained HAART naive at this point. We also conducted a second version of this sensitivity analysis censoring individuals 6 months after their first CD4 cell count less than 350/μL if they remained HAART naive. Because the effect of HAART may differ in patients with a history of IDU, we conducted subgroup analyses of patients with (IDU+) or without (IDU−) a history of IDU.

Of 18 347 patients in the CASCADE Collaboration as of May 2009, nine thousand four hundred fifty-five were included in this analysis. Most patients excluded from this analysis were no longer alive, AIDS free, antiretroviral therapy naive, or in active follow-up at the beginning of the study (January 1, 1996) or 6 months after seroconversion. Many patients were no longer AIDS free and antiretroviral therapy naive at enrollment or at the time of their first eligible CD4 cell count. Thus, we analyzed data from 9455 HIV-1 seroconverters who were eligible for 1 or more subcohorts after January 1, 1996, with 52 268 person-years of follow-up (median = 4.7 years, interquartile range [IQR] = 2.0-9.1 years). Most participants were male (n = 7367 [77.9%]) and were infected through sex between men (n = 5341 [56.5%]) or sex between men and women (n = 2363 [25.0%]). The median age at seroconversion was 30.3 years (IQR = 25.4-36.8 years), and the median duration of infection was 1.3 years (IQR = 0.8-3.4 years) at the time of entry into the first subcohort. During follow-up, 812 patients (8.6%) developed AIDS and 544 patients (5.8%) died. On average, each individual contributed to 12 (IQR = 4-26) subcohorts (eTable 1).

At baseline, participants who initiated HAART had a poorer prognosis in some respects (higher viral loads, shorter duration of infection, and slightly lower CD4 cell counts) compared with those who deferred HAART in a given month (Table 1). In other respects, they had a better prognosis (less likely to have a history of IDU and less likely to be co-infected with hepatitis). Across all CD4 strata, CD4 cell counts were more recent in those initiating therapy, and these patients were more likely to have available viral load measures. Most deferring patients eventually went on to HAART, generally in the same CD4 stratum or the next lower stratum (eTable 2). The only exception was the 500 to 799/μL stratum in which nearly half of patients remained HAART naive at last follow-up. Of these HAART-naive patients, most had CD4 cell counts greater than 350/μL at the last follow-up. The use of all–nucleoside reverse transcriptase inhibitor regimens containing abacavir in the first HAART regimen was similar between those who initiated and those who deferred (eTable 3).

Unadjusted incidence rates and adjusted HRs (aHRs) stratified by CD4 cell count are presented in Table 2. Considering first the combined end point of AIDS or death, the effect of initiating rather than deferring HAART in a given month was protective at CD4 cell counts lower than 350/μL. At CD4 cell counts of 350 to 499/μL, there was a 25% reduction in the hazard of AIDS or death (aHR, 0.75; 95% CI, 0.49-1.14). At CD4 cell counts of 500 to 799/μL, AIDS-free survival was not different in the 2 groups after adjusting for covariates (aHR, 1.10; 95% CI, 0.67-1.79). In the analysis of all-cause mortality, HAART initiation seemed to have a stronger effect on death than on the combined end point at CD4 cell counts of 350 to 499/μL (aHR, 0.51; 95% CI, 0.33-0.80). We observed no benefit at CD4 cell counts of 500 to 799/μL (aHR, 1.02; 95% CI, 0.49-2.12).

Weighted survival curves, stratified by CD4 cell count, are presented in Figure 2, with estimates of the absolute risk of AIDS or death or at 3 years for those initiating and deferring therapy in Table 3. At CD4 cell counts of 200 to 349/μL, the absolute difference in the proportion of patients who died or progressed to AIDS increased from −1.3% at 1 year to −6.4% at 5 years. The estimated number needed to treat (NNT) to prevent 1 event decreased from 79 to 16 at 5 years. Risk reduction was one-third as large for patients with CD4 cell counts of 350 to 499/μL with NNTs of 229 and 45 at 1 and 5 years, respectively. We found no reduction in the absolute risk of AIDS or death at CD4 cell counts of 500 to 799/μL.

When death from all causes was evaluated as the sole outcome, the absolute difference in the proportion of patients who died increased from essentially no difference at 1 year to −2.1% at 5 years for those with CD4 cell counts of 200 to 349/μL. The NNT decreased from approximately 8000 to 49 during this period. Similarly, the cumulative risk of death for patients with CD4 cell counts of 350 to 499/μL differed by −0.3% at 1 year and by −2.8% at 5 years, with corresponding NNTs of 328 and 35, respectively. In patients with CD4 cell counts of 500 to 799/μL, there was no reduction in the risk of death at 1 and 5 years, although there was a small difference at 3 years that favored HAART initiation.

Results of sensitivity analyses suggest that these findings are robust to alternative ways of defining the eligible population and censoring outcomes of those who received nonstandard treatment (Figure 3). We also found that excluding individuals with previous IDU did not have a meaningful effect on the magnitude of the association between HAART initiation and time to AIDS or death (eTable 4).

Quiz Ref IDThis analysis of 9455 HIV-1 seroconverters confirms the clinical benefit of initiating HAART with CD4 cell counts of 200 to 349/μL. We estimated a 25% reduction in the relative hazard of AIDS or death and a 49% reduction in the relative hazard of death from all causes at CD4 cell counts of 350 to 499/μL. The relatively low incidence of AIDS and death in individuals with CD4 cell counts of 350 to 499/μL indicates that patients and health care providers need to weigh the risks and benefits for each individual over an extended period of treatment.

Although many studies have compared disease progression in patients starting HAART at different stages of disease with follow-up beginning at the time of treatment initiation, it is now appreciated that this study design is not ideally suited to inform the “when to start” question due to unobserved lead time and clinical events that occur during the time when patients are deferring therapy.^17- 18 Kitahata et al,⁵ Sterne et al,⁶ and Cain et al⁷ report findings from observational analyses tailored to estimate the effect of early HAART initiation on clinical outcomes using data primarily from seroprevalent cohorts. Although the comparison groups differ and, thus, the effect estimates from these studies estimate different parameters, one can compare the conclusions of the studies in broad terms. The present findings agree with those of Kitahata, Sterne, and Cain and their colleagues, who found that deferring HAART to a CD4 cell count less than 350/μL is detrimental. Kitahata et al,⁵ but not Sterne et al,⁶ further conclude that deferring HAART to a CD4 cell count less than 500/μL is detrimental. (Cain et al⁷ began observing patients at first CD4 cell count less than 500/μL and, thus, do not report effect estimates for treatment at CD4 cell counts greater than 500/μL.) Unlike Kitahata et al, we did not observe a benefit at the population level for initiation at 500 to 799/μL after adjusting for confounding.

Quiz Ref IDThe absolute risk of AIDS-related morbidity and mortality in the population can drive the degree to which HAART initiation is beneficial at a particular stage of disease. In the present study, the weighted survival curves, absolute risks of disease progression, and NNT provide additional insight regarding the benefit that patients in resource-rich settings can expect from HAART at different CD4 strata. At CD4 cell counts of 350 to 499/μL, the benefits of treatment initiation become evident only beyond 2 years, suggesting that patients need to consider the long-term course of treatment, including the risk of adverse effects of HAART during an extended period.¹⁹

The decision to initiate therapy is a dynamic process, influenced by changes in the patient's condition and readiness to adhere to the lifelong regimens that are available to treat HIV. We reflected this dynamic process in the analysis by considering each month while a patient was AIDS free and HAART naive as a point in time when therapy could have been initiated rather than representing patients at a single point in time, such as the first measured CD4 cell count in a particular range. We then observed these individuals during an average of 4.7 years as they experienced the clinical consequences of initiating HAART (or not) at that point in time. By allowing individuals to contribute to multiple subcohorts as long as they remained eligible, we effectively estimated a weighted average of the benefit of initiating therapy at any time while an individual had a CD4 cell count in a given CD4 stratum compared with the prognosis that they would have experienced if they had not initiated HAART at that time. The resulting relative and absolute effect estimates can be used to help inform patient decisions about whether the benefit of therapy at this particular stage of disease is sufficient to outweigh the challenge of adhering to treatment, the risk of adverse effects, and the financial cost of medications over a longer period of treatment.

We acknowledge that if the ultimate treatment patterns of the deferrers had been different, the results of the study would have been different. In eTable 2, we describe the type and timing (relative to CD4 cell count) of antiretroviral drug therapy received by patients who composed the deferred group for each CD4 strata. To evaluate the potential effect of individuals who were not treated consistent with the current standard of care, we censored the outcomes of those who waited too long or used suboptimal regimens, but the magnitude of the effect estimates was unaffected (S1 and S2 in Figure 3). We also considered the possibility that the null effect in the 500 to 799/μL stratum was due to individuals who deferred HAART only briefly, but these patients composed only approximately 5% of the deferred group. Although the comparison groups did not follow standardized treatment algorithms, they do represent the real-world experience of thousands of HIV-infected patients in care during the study period. We believe that these findings complement those from other recent studies^5- 7 that explicitly compared 2 specific, narrowly defined treatment alternatives.

Patient well-being is adversely affected by many serious non-AIDS–defining conditions. For example, immunodeficiency and uncontrolled viremia have been implicated in the development of cardiovascular disease^20- 21 and non-AIDS–defining malignancies.^22- 23 Although CASCADE does not pool data on non-AIDS morbidity, this analysis reflects the most serious outcome (death) due to non-AIDS conditions.

We considered several alternative approaches to conducting this analysis in an effort to assess the robustness of these findings. We examined the effect of more restrictive inclusion criteria. To address confounding, we adjusted for a set of 20 covariates that we had a priori reason to suspect were associated with different rates of disease progression. We examined a wide range of possibilities for truncating the weights before deciding on a method that controlled for confounding without introducing instability in the estimates (eTable 5). Quiz Ref IDDespite this, we cannot rule out the possibility that patients who initiated therapy had an inherently better or worse prognosis than did those who deferred therapy related to unmeasured factors. We were reassured that in the 50 to 199/μL CD4 strata, where we can compare with results from a randomized trial conducted in a resource-rich setting, the present estimate is similar to that from the trial.³

In the absence of results from well-conducted, long-term, randomized trials in patients with CD4 cell counts greater than 350/μL, treatment decisions will need to be made based on the available evidence from observational cohorts. We used a novel approach applied to a unique cohort of seroconverters to reduce the potential for lead time bias. We found that treatment initiation at CD4 cell counts of 350 to 499/μL was associated with slower disease progression. We did not observe any benefit to treatment initiated at 500 to 799/μL.