| |
Limited benefit of antiretroviral resistance testing in treatment-experienced patients: a meta-analysis
|
| |
| |
AIDS: Volume 18(16) 5 November 2004
Panidou, Ermioni Ta; Trikalinos, Thomas Aa,b; Ioannidis, John PAa,b
From the aClinical Trials and Evidence-based Medicine Unit, Department of Hygiene and Epidemiology, University of Ioannina School of Medicine, Ioannina, Greece and the bInstitute for Clinical Research and Health Policy Studies, Tufts-New England Medical Center, Boston, Massachusetts, USA.
Abstract
Objective: To estimate the effectiveness of resistance assessments based on viral sequencing (genotypic antiretroviral resistance testing, GART), phenotypic antiretroviral resistance testing (PART) or virtual PART (vPART) in the management of treatment-experienced HIV-1-infected patients.
Design: Meta-analysis of randomized controlled trials comparing treatments aided by GART, PART and vPART, and controls.
Methods: The meta-analysis synthesized data on the proportion of patients with undetectable plasma viral load, the decrease in viral load, and the increase in CD4 cell count at 3 and 6 months after randomization.
Results: Ten trials were analyzed (total 2258 participants). Compared with controls, at 3 and 6 months GART increased the proportion of patients with viral load below detection by 11% [95% confidence interval (CI), 6-16], and 10% (95% CI, 5-16), respectively. The difference in viral load change was 0.27 log10 copies/ml (95% CI, 0.11-0.43) and 0.21 log10 copies/ml (95% CI, 0.09-0.34), respectively. However, no improvement was observed in the CD4 cell count at either time point: the difference in CD4 cell count -5.7 × 106 cells/l (95% CI, -18.8 to 7.3) and 1.2 × 106 cells/l (95% CI, -15.0 to 17.4), respectively, at 3 and 6 months. For PART, there was no clear evidence for any benefit versus no testing (three trials). vPART conferred a small benefit in indirect comparisons versus no testing.
Conclusion: Evidence for benefit of antiretroviral resistance testing is sparse and limited to small short-term improvements of virologic response, mostly with GART and less with vPART. Current guidelines widely recommending the use of antiretroviral resistance testing in clinical practice are not commensurate with the available evidence.
Introduction
The management of the increasing number of HIV-infected patients failing antiretroviral therapy may be theoretically aided by knowledge of the resistance profile of HIV-1 strains. The most frequently used resistance assessments are based on sequencing analysis of the viral genome (genotypic antiretroviral resistance testing, GART), on phenotypic in vitro antiretroviral resistance testing (PART) or on virtual PART (vPART), a computer-predicted virtual phenotype generated by cross-linking the tested sample's genotypic results with detailed databases of phenotypes. Resistance testing has been strongly recommended in clinical guidelines developed by experts and is used increasingly in clinical practice. However, knowledge about antiretroviral testing use is limited, even among HIV specialists. Recent data have been used to question whether access to antiretroviral testing may have any meaningful impact on long-term outcomes for HIV-infected patients. Is it possible that the benefits of antiretroviral resistance testing have been overstated? Several randomized controlled trials (RCT) have looked into the role of GART, PART and/or vPART as aids in therapeutic decision making regarding HIV-positive patients who are treatment experienced. The conducted RCT have been of modest sample size and so they leave considerable uncertainty in their estimates of efficacy. The meta-analysis described here took data from all the relevant RCT in order to assess the exact magnitude of the impact on virological and immunological outcomes of using GART, PART or vPART to assist treatment decisions versus treatment without such aids.
Methods
Eligibility criteria and search strategy
All RCT were considered where GART-, PART- or vPART-guided therapy was compared either against each other or against therapy without such testing. The study focused on trials targeting antiretroviral treatment-experienced HIV-1-seropositive people. MEDLINE and EMBASE were searched (1998-2004; last search February 2004) using combinations of the terms HIV, genotypic resistance testing, phenotypic resistance testing, drug resistance; terms such as randomized controlled trial, controlled trial, and clinical trial were used in conjunction with the primary terms. Search was complemented by perusing meeting abstracts (Conferences on Retroviruses and Opportunistic Infections, World AIDS Conferences, Interscience Conferences on Antimicrobial Agents and Chemotherapy, Infectious Disease Society of America) in the same period, as well as the references of the retrieved articles. Experts in the field were contacted in order to identify still unpublished relevant studies and to obtain additional information and clarifications. There were no language restrictions.
Results
Studies included in the meta-analysis
Eleven potentially eligible RCT reports were identified, and two additional RCT were available only in meeting abstracts. Two published reports were excluded because they focused on HIV patient subgroups from previously described RCT, and one because antiretroviral testing was optional in the experimental arms (not used by all participants randomized in the experimental arms).
A total of 10 RCT with direct comparisons were included in the final analysis. Five (1113 randomized participants) provided data on the GART versus control comparison and four (959 randomized participants) provided data on PART versus control. Two RCT (501 randomized participants) compared PART with vPART, and one (382 randomized participants) compared GART with PART. There were no direct randomized comparisons of vPART and control.
In three trials, the median baseline viral load was reported to be at least 25 000 copies/ml. Expert advice was used in all randomized arms in one trial, as a factorial second randomization in another, and in the GART arm only in a third. Active monitoring of the drug concentrations was employed by only one trial. The randomization mode was adequately described in all but three of the published trials. All but two published reports clearly described the method of allocation concealment.
Genotypic antiretroviral resistance testing versus control
The proportion of patients having viral load below detection was statistically significantly higher by 11% in patients receiving GART-guided treatment versus controls at 3 months and by 10% at 6 months. There was no substantial between-study heterogeneity. The absolute fall in viral load was significantly greater in the GART group, by approximately 0.2-0.3log10 copies/ml, both at 3 and 6 months post-randomization. A moderate amount of between-study heterogeneity was evident at 3 months but not at 6 months. There was no difference between GART-aided treatment and empiric therapy regarding the increase in the CD4 cell counts at either 3 or 6 months and the 95% confidence intervals excluded large differences.
Phenotypic resistance testing versus control
Limited data were available for the PART versus control contrast; information on specific outcomes could not be retrieved from some unpublished studies and one of the two published studies had only 3 months of follow-up. There was no statistically significantly difference in the proportion of virologic or immunological responses at 3 months after randomization (Table 2). Virologic results were similar at 6 months after randomization. A small difference in CD4 cell counts at 6 months was based on only one trial.
Phenotypic resistance testing versus virtual phenotypic resistance testing
Similarly, there were limited data on the comparison between PART and vPART, and the accumulated randomized sample was small to detect any differences. Although the 95% confidence intervals excluded large differences in the virologic outcomes, there was a tendency in favor of vPART-guided treatment when the immunological outcomes were considered.
Indirect comparisons
Inferences on the vPART-assisted treatment versus control (empiric therapy) could be based only on indirect evidence (vPART versus PART and PART versus control), since a direct randomized comparison had not been made in any RCT. vPART was not formally statistically significantly better than empiric therapy in virological outcomes, although a modest effect was possible and a significant increase in the CD4 cell count was also seen at both 3 and 6 months.
Subgroup and bias analyses for genotypic antiretroviral resistance testing versus control
The summary estimates regarding the proportion of patients with viral load decreasing below detection were the same across the two subgroups of studies defined by differing HIV RNA assay detection limits (400-500 and 200 copies/ml) and there was no statistically significant variability in studies with high or low median baseline viral load, although the benefit tended to be larger in studies with higher baseline viral load. Moreover, the benefit of GART was remarkably similar in studies where expert advice was available to both arms, to one arm only or not at all.
At least five trials in a comparison were available only for the GART versus control contrast at 3 months. The funnel plots for the proportion of patients with viral load below detection and for the decrease in viral load and increase in CD4 cell count appeared asymmetric. In the same syntheses, the rank correlation test was statistically significant (P = 0.086 for all three). Therefore, smaller studies showed more prominent efficacy than larger studies.
Author Discussion
This meta-analysis of RCT on therapy-experienced HIV-1-infected patients quantified the magnitude of the efficacy of antiretroviral resistance testing strategies. Virological efficacy was demonstrated only for GART in direct comparison with no resistance testing, but the benefit was relatively small, amounting to a 10% difference in the proportion of patients with viral load below detection at 3 and 6 months. This benefit was dissociated from immunological response in this short-term follow-up, although one might argue that CD4 cell count responses may take longer to become evident. Some indirect evidence was also available for a modest efficacy of vPART versus control; the indirect nature of the data adds more uncertainty to this observation. To-date, data on PART do not suggest any demonstrable benefit. Even for GART, the meta-analysis suggested that the benefit was larger in smaller trials, and this is often considered to be suggestive of bias. Finally, contrary to current opinion, the available data suggest that the efficacy of GART is not modified by the use or not of additional expert advice.
Current guidelines have made relatively strong recommendations favoring the use of resistance testing among treatment-experienced patients who are failing therapy. The findings of this meta-analysis cast some doubt about whether these recommendations can be as strong. The benefits seem limited to GART-based approaches and even these are of small magnitude. Our conservative results are also in line with the results of a recent randomized trial, where the availability of antiretroviral resistance testing (either GART or PART/vPART) did not improve patient outcomes over a long-term follow-up of up to 3 years. In that study, antiretroviral resistance testing was not used by all patients in the respective arms, but it was available for use, a situation analogous to a pragmatic setting. One might argue that the utility of antiretroviral resistance testing is dependent on the ability to use the information rationally, and expert reviews have advocated that expert advice may be key to its success. However, we found no evidence of differential effects on the basis on the availability or not of expert advice.
The results of this meta-analysis should not be surprising, since all the individual trials are consistent with the overall emerging picture. However, the meta-analysis increases the power and thus excludes the possibility that major benefits can be achieved with antiretroviral resistance testing, while the confidence intervals of single trials left considerable uncertainty in this regard. The meta-analysis still leaves some uncertainty about the possible existence of subgroup differences, for example regarding the effect of expert advice. However the available summary estimates again suggest that differences would be small, if present at all.
An early meta-analysis had concluded more favorably on the merits of GART. This meta-analysis had been based on more limited data and effects were measured on an odds ratio scale; odds ratios are known to inflate the true treatment benefits. The claim that expert opinion made a difference could not be validated in our more comprehensive meta-analysis. Moreover, an early cost-effectiveness analysis had shown that the increments in costs resulting from genotyping expenses were counterbalanced by the decreased use of protease inhibitors and their attendant costs. However, this cost-effectiveness analysis used data on the efficacy of GART derived from the most favorable of all studies, and many of the assumptions are really difficult to test for their validity for long-term patient care. The cost of antiretroviral testing in resource-limited setting would be prohibitive. For resource-sufficient settings, the cost-effectiveness would have to be re-evaluated, and certainly the conclusions should not be generalized to all types of resistance testing.
The theoretical advantages of antiretroviral resistance testing are quite solid. Treatment-experienced patients have limited options. A poor choice is likely to lead to resistance to several agents in the same class; for some classes (e.g. non-nucleoside reverse transcriptase inhibitors), a single mutation can render all drugs in this class quite useless. Therefore, detailed knowledge of the resistance profile may be of paramount importance. Also theoretically, PART approaches should be optimal, since they provide the direct evidence on whether an HIV-1 isolate is resistant or not. If these statements are true, why does antiretroviral resistance testing show only modest benefits, if at all, in randomized trials and why does PART actually seem completely ineffective?
There are several possibilities. We should acknowledge that, while the apparent virologic benefit was small and its duration seemed to be limited, this was also true of old trials comparing monotherapy versus placebo or dual therapy versus monotherapy. Yet for these interventions small and transient reductions in viral load could translate into some modest survival benefit. Antiretroviral resistance testing should certainly not be completely abandoned. Morever, there are likely to be multiple confounding factors in evaluating resistance testing. One of the most important may be patient adherence. If adherence is low, any benefit of resistance testing may be lost. Not only antiretroviral therapy but also resistance testing may have to be individualized.
An alternative argument is that what matters most in antiretroviral therapy is making a wise initial choice. Subsequent choices in failing patients are likely to represent only less-important corrective actions once resistance has already been established. Alternatively, given the regular follow-up of these patients in current practice, suboptimal choices may be picked up quite rapidly on the basis of poor virologic response, before resistance affects more drugs. Another possibility is that these resistance tests are in evolution and they may continue to be improved. Therefore, their early performance in these clinical trials might not be optimal. However, this would still mean that new 'improved' tests should be evaluated with clinical trials, and findings cannot be generalized from one test to another. Standardization of interpretation algorithms and expert advice may also play an important role, but it is unclear how much of this expertise is conferred from the ideal setting to clinical practice, even within the framework of randomized trials. Nevertheless, an intervention that is not useful in real-life settings cannot be advocated simply on the basis of its theoretical benefits. The current meta-analysis suggests that, while small benefits cannot be excluded, the strength of current recommendations regarding the use of GART and possibly vPART should be tempered.
|
|
| |
| |
|
|
|
