Reply to Badri et al. on ‘Limited benefit of antiretroviral resistance testing in treatment-experienced patients: a meta-analysis’

Badri and colleagues claim that genotypic resistance testing has a greater benefit in patients with limited antiretroviral exposure, and that expert advice is likely to have a particularly beneficial effect in these patients. The evidence that they provide is derived from non-randomized data. We would caution that although these data may be useful, it is common for observational non-randomized studies to suggest greater treatment benefits than is shown by randomized trials [1]. Moreover, although their suggestions seem reasonable, the available randomized data do not really support them. First, contrary to what Badri and colleagues claim, a considerable proportion of patients in the five genotyping resistance testing trials had taken only one protease inhibitor before randomization. This percentage ranged from 33% in the NARVAL study [2] to 48% in the GART study [3] across the five trials. With the exception of the NARVAL trial [2], the mean or median duration of protease inhibitor regimens did not exceed 2 years. Therefore, even though the remaining treatment options were not as extensive in these trial populations as they are currently, considerable data are available on patients with limited previous exposure to highly active antiretroviral regimens. Data from the HAVANA [4], ARGENTA [5], and GART [3] trials show that the difference between the genotyping resistance testing and control arms is not significantly different in patients with limited previous antiretroviral experience versus those with more extensive previous antiretroviral experience. Only the NARVAL trial [3] discussed the presence of an interaction effect with the previous antiretroviral experience, but this applied only to the 12-week data (not the 24-week data), and even then it was not formally statistically significant (P = 0.17). Finally, the only long-term trial to date did not show any benefit from genotypic resistance testing despite the fact that it enrolled a considerable proportion of patients with limited previous exposure to protease inhibitors (68% had two or fewer protease inhibitors) [6]. Although one might anticipate that expert advice would make a difference, this is not shown by the randomized data. Table 4 of our meta-analysis [7] shows the difference in the proportion of patients with viral loads below detection with genotypic resistance testing versus controls. We apologize that because of an uncorrected typesetting error the title of the table refers to ‘virtual phenotypic resistance testing’ rather than the correct ‘genotypic resistance testing’. As shown, the effect size is very similar in trials that employed expert advice and in those that did not. Moreover, the only available head-to-head comparison of expert advice versus no expert advice [4] did not show any overall difference. Interaction terms in complex multivariate models should be viewed cautiously when they are based on only a small number of patients.