Introduction

The Big Heart Disease Lie

Cure Heart Disease Permanently

Get Instant Access

The first randomized trial of antiviral therapy in HIV type 1 * infection included 282 patients with AIDS or advanced AIDS-related complex (ARC) and was stopped early in 1986 after an average follow-up of 4 months because of a substantial reduction in mortality in the group who received zidovudine (ZDV) (Fischl et al., 1987). The era of anti-HIV treatment had begun. This chapter will discuss some of the issues faced by clinical trialists and governmental regulatory agencies in the evaluation of therapies for HIV disease over the subsequent years as new anti-HIV drugs have been developed requiring evaluation in clinical trials.

A number of features specific to HIV infection have influenced trial design and interpretation. Even without treatment, the disease has a long asymptomatic phase, on average about 10 years. During this time HIV-infected individuals are essentially well although some laboratory markers, principally the CD4 lymphocyte count and viral load, as measured by HIV RNA in plasma or serum, are indicative of disease progression. As the

* Throughout, HIV will be used to denote HIV-1.

disease progresses, individuals become increasingly susceptible to a number of different opportunistic infections and tumors, some of which are life threatening. They may also develop a number of nonspecific symptoms (e.g., fever and weight loss) and hematological and neurological symptoms which may be due to the direct effect of HIV.

It is difficult to find exact parallels between HIV and other diseases. The design of current treatment strategies reflects present ideas about development of drug resistance which have much in common with the chemotherapy of tuberculosis and malignant tumors. However, the need for long-term suppressive therapy, which is likely to be an essential feature of the management of the disease, has much in common with other chronic diseases such as diabetes, hypertension, rheumatoid arthritis, ulcerative colitis, and multiple sclerosis. An additional factor not present in many chronic diseases is the substantial toxicity and burden to the patient of the current effective highly active antiretroviral therapies (HAART). In particular, over the last few years it has become apparent that a substantial proportion of patients taking HAART long term are suffering from metabolic abnormalities and/or significant fat redistribution (''lipodystrophy'' ) that, in combination with other risk factors, may put them at higher risk of cardiovascular disease (Shevitz et al., 2001; Egger et al., 2001). The balance between assessing short-term effects of treatment regimens and long-term effects of treatment strategies [essentially the difference between testing the direct effect of a treatment regimen under idealized conditions (efficacy) and pragmatic trials of effectiveness which assess the impact of a treatment regimen in clinical practice] has yet to be found.

Many of the issues discussed in other chapters, such as the analysis of failure time and longitudinal data, methods for multiple endpoints and early stopping, and methods for assessing compliance, are highly relevant to clinical trials in HIV infection. HIV trials face several other practical problems which affect their design and analysis and may threaten their successful outcome (Ellenberg et al., 1992; Foulkes, 1998; Albert and Yun, 2001).

As many HIV-infected individuals take large numbers of drugs (both antiretrovirals and prophylaxis), there has been considerable interest in trial designs which maximize the information gained on drugs while minimizing the number of participants involved and the time spent on inferior drug combinations. Factorial designs are attractive to HIV research for two reasons. First, they are the only trial designs that allow investigation of synergistic or antagonistic interactions when these are thought to exist: second, when interactions are assumed to be small, multiple drug effects may be estimated more efficiently in one trial (DeGruttola et al., 1998; Ellenberg et al., 1992; Schoenfeld, in Finkelstein and Schoen-feld, 1995). However, they do have lower power to detect interactions than main treatment effects. When the same endpoint is used to compare more than one treatment regimen, the sample size required to assess one treatment effect must be inflated to incorporate the likely smaller number of events due to both effective treatments. For fixed power, the inflation factor is the ratio of the probability of an event in a nonfactorial trial where there is only one effective treatment to the probability of an event in the factorial trial where both treatments are effective. Therefore in a 2 x 2 factorial trial, where treatments 1 and 2 have hazard ratios a and h, respectively, compared with placebo, and 1 — k is the event rate on the double placebo arm on which the sample size is based, the inflation factor can be shown to be

The inflation factor is only large when event rates are low and treatments moderately efficacious. A practical (rather than statistical) method for reducing patient time spent on inferior drug combinations has been the increasing use of coenrollment in more than one protocol of a clinical trials organization (Larntz et al., 1996). Coenrollment can be more realistic than factorial designs in HIV infection, where the type of interventions to be tested depend on disease stage. However, factorial designs have great potential for trials in HIV infection, in particular in small populations such as children, and also to simultaneously assess different types of treatment strategies, such as testing two regimens of antiretroviral drugs together with two criteria for defining treatment failure and change of treatment.

Regardless of general trial design, the endpoints used to evaluate anti-HIV therapy have changed markedly over the last 15 years. Endpoints used in HIV trials to date include mortality, various measures of morbidity, biological markers of disease progression (surrogate markers), and adverse events. Mortality is the natural choice of endpoint in the definitive evaluation of therapy in a fatal disease such as HIV. It is clearly relevant to patients, it is a unique endpoint, and all trial participants are at risk. However, trials using death as an endpoint need to be much larger and last longer than trials which use earlier endpoints. With the advent of HAART in clinical practice from 1997 [usually consisting of at least two nucleoside analog reverse transcriptase inhibitors (NRTI), with an additional drug which is usually either a nonnucleoside analog reverse tran-scriptase inhibitor or a protease inhibitor], the use of survival as the primary outcome is likely to be impractical, except in patients with late HIV disease.

As a result, the most commonly used clinical outcome measure is time to the first new AIDS-defining event or death. Delaying the onset of the first AIDS event and preventing subsequent new AIDS events is certainly clinically relevant. However, two issues remain: (1) the implications for the size and duration of the trial and (2) the composite nature of this endpoint. Although progression to a new AIDS-defining event or death is more common than mortality, with the advent of HAART this endpoint is likely to be rare enough to still require large numbers of patients to be recruited and followed over a long period, particularly in early HIV disease. However, several international trials with AIDS as a primary clinical endpoint are currently recruiting patients (see, for example, Emery et al., 2002). Some trials in the early 1990s used other (nonAIDS) clinical events (such as AIDS-related complex, ARC) in addition to AIDS or death. These events may provide early evidence of a treatment effect, but their usefulness is questionable because they are largely subjective relatively minor symptoms and are clinically much less important than AIDS. AIDS-defining illnesses, in contrast, are more clinically relevant and can more easily be assessed objectively. However, the current definition of AIDS includes a variety of over 20 different conditions including opportunistic infections and malignancies (Centers for Disease Control, 1992) in addition to a CD4 cell count of less than 200 cells/ml (ignored for the purpose of this discussion). A major issue in using progression to a new AIDS defining event or death is the composite nature of this endpoint, which treats all events equally regardless of their clinical significance. Furthermore, information on second or subsequent AIDS events are ignored (or sometimes not even collected) as is the total number of events experienced by a participant. In the Delta trial (Delta Coordinating Committee, 1996) a total of 1451 AIDS-defining events and 498 deaths were observed in 2765 participants who were AIDS-free at entry. When the relative risks of death associated with different AIDS events were simultaneously estimated from a Cox proportional hazards model using the occurrence of these events as time-dependent covariates, the impact of the different types of AIDS events on mortality ranged between no effect to an increase of about 20-fold. The composite endpoint of progression to AIDS or death utilized 936 (48%) of the total observed events. Thus one clear disadvantage of the composite endpoint of AIDS or death is that more than 50% of all events and the great majority of severe events have not been utilized. Methods for the analysis of multi-variate failure time data can be used to include all events in the analysis, and to investigate differences in treatment effects across different AIDS events (see Section 2).

The use of biological markers of disease progression to assess different treatments is attractive because it may provide direct evidence of treatment activity and lead to smaller and shorter trials. Disadvantages in using markers to measure treatment effects include their large within-patient biological variability and problems with quality control. A more relevant criticism is that a biological marker generally measures activity in only one mechanism of action of a drug regimen, be it efficacy or toxicity. Furthermore, the choice of timing of marker measurements and final clinical outcome will clearly affect the degree to which a marker measures the treatment effect on the clinical endpoint. There are a number of examples from other diseases where the inappropriate use of such surrogate markers has led to misleading conclusions and consequently to the inappropriate treatment of many patients, the most notable being the use of anti-arrhythmic drugs (see, for example, Fleming and DeMets, 1996). Candidate markers in HIV include CD4 lymphocyte count and viral load measured by plasma or serum HIV RNA. CD4 lymphocytes are the main target of the HIV virus and a key part of the defence against infection provided by the immune system. Declining numbers of CD4 cells are therefore associated with an increase in susceptibility to infections to which a person would not usually succumb (opportunistic infections). HIV RNA levels in plasma or serum directly measure the number of circulating copies of the virus, and quantitative plasma HIV RNA measurements are now the most commonly used primary outcome measures in phase III trials. Although the prognostic significance of both viral load and CD4 cell count is beyond dispute (Mellors et al., 1997), neither viral load nor CD4 cell counts are particularly strong surrogates for clinical outcome in the evaluation of therapy (HIV Surrogate Marker Collaborative Group, 2000). The assessment of surrogacy of biological markers is discussed in Section 3.1. However, even assuming surrogacy of a marker, a number of issues of analysis remain. If a marker is analyzed as a continuous variable, then repeated measures methods must be employed and informative dropout accounted for (Section 3.2). HIV RNA levels are also often reported as below a limit of assay detectability, so this censoring of continuous data must also be considered. Alternatively, biological markers can be synthesized into failure time data; in this case, methods for interval censored failure time data should be used (Section 3.3).

When drug regimens are taken over the long term, the relative contribution of toxicity and efficacy becomes more important, particularly when endpoints are markers of efficacy only, such as levels of HIV RNA. For example, since 1998 a sizeable proportion of individuals infected with HIV have begun to present with severe disturbances of metabolic parameters and body fat redistribution or ''lipodystrophy'' (both at levels associated with increased cardiovascular disease risk) (Carr, 1999; Shevitz et al., 2001). The precise relationship between these changes and antiretro-viral therapy is currently unclear, although epidemiological studies are ongoing (Friis-Moller, 2002). Toxicity data are a second example of multivariate failure time data with the additional problems of sparse data. The methods described in Section 2 can be used to investigate the effect of treatment on the time to multiple adverse events, but may have little power when there are small numbers of events.

Independently of the choice of endpoint, however, the long asymptomatic phase of HIV infection means that trials which are designed to evaluate the effectiveness of therapy in early disease need to have long-term follow-up. To date more than half of the published phase III trials with clinical endpoints have a median follow-up of 1.5 years or less. Long-term trials are considered undesirable by many clinicians and patient groups because of the urgency of making new treatments available, and the rapid changes in perceptions about therapy. In addition, HIV drugs to date have demonstrated only transient benefit, and so the emphasis has shifted to determining efficacy over short periods of time (particularly in studies sponsored by the pharmaceutical industry for regulatory purposes). This is achieved through trials of relatively short duration using endpoints such as biological markers which occur earlier in the disease and are thus more proximal to randomization. However, there is a clinical need for extra information beyond that required for regulatory purposes: It is precisely because the effect of treatment might be of only very short duration that long-term follow-up is needed to assess durability of the effect as well as the effectiveness of treatment strategies. The transient nature of the benefit of ZDV was only established after the completion of trials with relatively longer duration [the Concorde trial (Concorde Coordinating Committee 1994), and the extended follow-up of the ACTG 019 (Volberding et al., 1994)]. This has been further confirmed by the extended follow-up of the Concorde and Opal trials (Joint Concorde and Opal Coordinating Committee 1998) and the overview of trials of ZDV in asymptomatic and early symptomatic infection (HIV Trialists' Collaborative Group, 1999). The difference between assessing biological efficacy and clinical effectiveness embodied in the difference between explanatory and pragmatic trials (Schwarz and Lellouch, 1967) is central to the arguments for and against longer term follow-up (Schoenfeld, 1996) and the level of agreement expected between short- and long-term results.

The third, and perhaps the most important, reason for short follow-up is the high rate of change from allocated trial therapy (noncompliance or treatment change). In addition to toxicity and tolerability, the reasons for change from allocated treatment include perceived or actual treatment failure, desire to try new but perhaps unproven drugs, or simply the feeling of clinicians and participants that it is'' time for a change.'' Figure 1 shows that the rate of withdrawal from allocated therapy for reasons other than disease progression increases with follow-up but does not appear to depend on the therapy or the size of the trial. The effect of a large number of treatment changes is to reduce the interest of some participants and investigators in the trial because of concerns that the initial treatment

Clinical Hypothesis Symbol

Figure 1 Rate of withdrawal from allocated therapy for reasons other than disease progression by median follow-up in antiretroviral trials with clinical endpoints in HIV infection. O, monotherapy with NRTIs; A, combination therapy with NRTIs; □, combination therapy including a PI. The size of the symbol is proportional to the size of the trial.

Figure 1 Rate of withdrawal from allocated therapy for reasons other than disease progression by median follow-up in antiretroviral trials with clinical endpoints in HIV infection. O, monotherapy with NRTIs; A, combination therapy with NRTIs; □, combination therapy including a PI. The size of the symbol is proportional to the size of the trial.

effect might be diluted, the treatment comparison might be confounded and therefore the trial will not provide useful results. That this view is not necessarily correct and should not be used to justify early stopping of a trial of effectiveness is demonstrated by experience in the Delta trial, although there is a clear difference between trials primarily concerned with estimating efficacy rather than effectiveness in the way treatment changes are considered. Delta was a multinational double-blind randomized trial which compared the policy of dual therapy with ZDV plus didanosine (ddl) or zalcitabine (ddC) compared with the policy of monotherapy with ZDV alone (median follow-up 30 months) (Delta Coordinating Committee, 1996). More than 2000 patients had not had anti-HIV therapy prior to entry (Delta 1) and over 1000 had taken ZDV for at least 3 months (Delta 2). There was a high rate of change from allocated treatment in both Delta 1 and Delta 2, with median time from randomization to stopping blinded allocated treatment only about 15 months in Delta 1 (Figure 2).

Years of follow-up

Number at risk

Delta 1 2124 1214 641 158

Figure 2 Time from randomization to stopping blinded allocated treatment in Delta 1.

Largely because of this, but also because of ''trial fatigue'' and of the availability of new drugs [notably lamivudine (3TC)], there was strong pressure to terminate Delta early. The Data and Safety Monitoring Committee (DSMC) finally recommended termination of Delta 3 months before its scheduled completion, not because of the negative impact of the high rate of treatment changes, but because of the clear evidence of the superiority of combination therapy. Table 1 shows that the effect of combination treatment on mortality in Delta 1 became increasingly apparent, and the magnitude of the treatment effect larger, with longer follow-up. Had the trial been stopped on the basis of the high rate of withdrawal from allocated treatment, an important effect would have been missed. This unexpected result (an increasing effect of treatment at least during the first 3 years in spite of a high rate of change from allocated therapy) may be due to the fact that, like Concorde, Delta 1 addressed a generic question on the effectiveness of a treatment policy (namely, initiation of treatment with combination therapy or with monotherapy). Such questions may be more robust to treatment changes of the magnitude observed in Delta, because the treatment changes are part of the policy.

Clearly statistical methods cannot be used to assess rapid changes in therapeutic options. However, transient benefit of anti-HIV therapy can be investigated, at least in the setting of survival data, by either fitting flexible time-dependent treatment effects using natural cubic splines (Hess, 1994), or by considering weighted Schoenfeld residuals (Grambsch and Therneau, 1994). Exploring the impact of treatment changes on the estimated treatment effect in a manner which avoids the introduction of selection bias is considerably complex, and will be considered in detail in Section 4.

Table 1 Relative Risk of Death (RR) in ZDV + ddl and ZDV + ddC Compared to ZDV Alone by Follow-up Time in Delta 1

Years from randomization

Total deaths

ZDV + ddl vs. ZDV

ZDV + ddC vs. ZDV

Was this article helpful?

0 0
Your Heart and Nutrition

Your Heart and Nutrition

Prevention is better than a cure. Learn how to cherish your heart by taking the necessary means to keep it pumping healthily and steadily through your life.

Get My Free Ebook


Responses

  • Tewelde
    What is the null hypothesis for treatment compliance and viral load of hiv patients on haart?
    10 months ago

Post a comment