Real Versus Virtual Phenotype to Guide Treatment in Heavily Pretreated Patients: 48-Week Follow-Up of the Genotipo-Fenotipo di Resistenza (GenPheRex) Trial

We compared viroimmunologic response after real phenotype (r-PHT) versus virtual phenotype (v-PHT) in patients failing highly active antiretroviral therapy (HAART). A total of 201 patients with >2 years of exposure, more than six experienced drugs, >1000 HIV RNA copies/mL, and on stable HAART for >6 months were randomized to the r-PHT or v-PHT arm. The primary end point was the proportion of HIV plasma viral load (pVL) <400 copies/mL. Secondary end points were absolute pVL change, proportion of pVL reduction >0.5 log₁₀ copies/mL, and absolute CD4 cell change. In the intention-to-treat-last observation carried forward analysis, study outcomes were not significantly different between arms over 48 weeks of follow-up: 20% and 24% pVL <400 copies/mL; 58% and 61% pVL reduction >0.5 log₁₀ copies/mL; -0.92 and -0.94 log₁₀ copies/mL mean pVL decrease; and +41.6 and +94.4 cells/mm³ mean absolute CD4 increase in the r-PHT and v-PHT arms, respectively. On-treatment analyses gave similar results. In the multivariate analysis of pVL <400 copies/mL, the following covariates were independent predictors at week 48: adherence (OR = 0.25;p = .002), baseline CD4 (OR = 4.39;p = .007), intravenous drug use as risk factor for HIV acquisition (OR = 0.33;p = .024), and sensitivity score of the new regimens by biologic cut-offs (OR = 1.84;p = .029). Prescribed drugs for which patients were naive resulted in marginal prediction (OR = 1.93;p = .054). In conclusion, virologic and immunologic outcomes did not differ when r-PHT or v-PHT was used in this cohort of heavily pretreated patients. Several factors should be considered to take better advantage of resistance testing, including treatment history, clinical status, and patients' ability to adhere to treatment.

Highly active antiretroviral therapy (HAART) has markedly improved the prognosis of HIV-infected patients by controlling HIV replication. HAART fails to control HIV replication in an increasing number of patients as a result of a network of causes, however. There is now substantial evidence that the emergence of drug resistance is a leading cause (^1,2) [as well as consequence (^3,4)] of antiretroviral therapy failure. Several retrospective (^5,6) and prospective studies (^7-12) have indicated that both genotypic and phenotypic HIV-1 drug resistance testing results are associated with or are predictive of virologic outcome (^5-10). Although this association or prediction has not been consistently reported (^13,14) and can be moderated by other factors affecting treatment response (^10,12), international guidelines have recommended the use of resistance testing to guide treatment choices after virologic failure occurs (^15-19).

Genotypic and/or phenotypic tests are currently available. Genotypic tests are easier to perform, but as patients fail multiple drugs, resistance patterns become increasingly complex and are often difficult for clinicians to interpret. By contrast, phenotypic tests provide a measure of HIV-1 drug susceptibility, but they are more expensive and time-consuming and require particular expertise and special laboratories to perform them; in addition, clinical cut-offs to interpret results are still undefined for most drugs (¹⁹).

In an attempt to overcome drawbacks of the two methods, an alternative way to interpret a genotype has been developed: the virtual phenotype (v-PHT). The v-PHT is a probabilistic estimate of the actual phenotype (real phenotype [r-PHT]) derived from a patient's HIV genotype by matching this with genotypes in a large proprietary database of samples that have been both genotyped and phenotyped (>28,000 clinical isolates). The phenotypes that correspond to the matching genotypes are then summed to give an average phenotype (i.e., average fold change in susceptibility compared with a reference strain) for each drug (²⁰).

Although promising, this approach has also raised many concerns, including the following: 1) only predefined mutations are searched in the database; 2) the number of matches may be limited, especially when resistance patterns are complex or rare; 3) clinical and therapeutic data are not part of the database; and 4) as for r-PHT, clinical validation of the cut-off points to interpret results is still poor (¹⁹).

Here, we present the results of a multicenter, prospective, randomized trial aimed at comparing viroimmunologic outcome after therapy has been changed on the basis of r-PHT versus v-PHT results, both interpreted with expert advice, in a population of heavily pretreated patients. Moreover, we have investigated the influence of other factors that could have had an impact on the clinical response so as to determine the best way to use such testing in clinical practice.

METHODS

Patients

HIV-1-infected patients were screened and enrolled in the Genotipo-Fenotipo di Resistenza (GenPheRex) study from 18 Italian centers participating in the nationwide Management Standardizzato di Terapia Antiretrovirale (MaSTeR) cohort. The study protocol was approved by the local ethics committees from all participating study sites. Written informed consent was obtained from all study participants before randomization.

The criteria for inclusion were as follows: 1) to have at least 2 years of previous exposure to antiretrovirals and more than six experienced drugs in the treatment history, 2) to have a plasma HIV-1 RNA load >1000 copies/mL, and 3) to be on stable antiretroviral HAART for >6 months.

Study Design

This is a randomized, open-label, multicenter trial. Patients were consecutively assigned at a 1:1 ratio to change treatment according to the results of either r-PHT (arm A) or v-PHT (arm B) testing.

Randomization was carried out by the coordinating center. In both arms, a panel consisting of 3 expert clinicians who were in close contact both before and during the study provided independent advice on the treatment options, taking into account the results of resistance testing as well as evaluation of the treatment history (including relevant reasons for changing drugs as a result of previous toxicity or subjectively determined poor patient adherence), clinical picture, and standard immunologic and virologic parameters. Time of starting therapy was considered as baseline in this study.

The primary end point of the study was the proportion of patients with a plasma HIV-1 RNA viral load (pVL) <400 copies/mL at 48 weeks. Absolute change from baseline in pVL (log₁₀ copies/mL), proportion of patients with pVL reduction >0.5 log₁₀ copies/mL, and absolute change from baseline in CD4 lymphocyte counts (cells/mm³) were also analyzed throughout the follow-up period.

Study Monitoring and Enrollment

At baseline, r-PHT or v-PHT resistance testing, pVL, and CD4 lymphocyte counts were performed, along with demographics, risk factors for HIV acquisition, clinical stage of HIV disease (Centers for Disease Control and Prevention 1993 classification), previous and current antiretrovirals, and drug adverse event history. Subjectively determined poor adherence causing treatment switching or simplification in the treatment history was also reported by the treating clinicians.

At weeks 4, 16, 32, and 48 after baseline, the patients were asked to return for a pVL, CD4 count, routine hematologic and biochemical parameters, and clinical assessment of adverse events. Adherence was also recorded by the treating provider at weeks 16, 32, and 48 by means of patient interview and was ranked into three classes with regard to proportion of missed doses during the last week (i.e., >50% = poor adherence, 50%-20% = fair adherence, <20% = good adherence).

Laboratory Measurements

Phenotypic drug susceptibility analysis (r-PHT, Antivirogram) (²¹), HIV RNA extraction, and genotyping with subsequent determination of v-PHT (²⁰) were performed by Virco (Mechelen, Belgium). Neither virtual nor real phenotypic results were performed in 57 of 201 (28.4%) patient samples for any of the drugs because of technical reasons (20/101 [19.8%] patient samples in the r-PHT arm, mean of 5.14 missing drug results per patient, range: 1-14 of the 14 tested drugs; 37/100 [37%] patient samples in the v-PHT arm, mean of 6.35 missing drug results per patient, range: 1-14 of the 14 tested drugs). CD4 T-cell counts were performed at each center using standard flow cytometry. Plasma HIV concentrations were quantified by means of the Amplicor HIV-1 Monitor (version 1.5; Roche Diagnostic System, Meyland, France) or with the HIV b-DNA test (version 3.0; Chiron Corporation, Emeryville, CA).

Definition of Active Drug During the Study and Sensitivity Score

To interpret both r-PHT and v-PHT results, technical cut-offs (<4-fold resistance, 4-10-fold resistance, and >10-fold resistance) were used for decision-making purposes during the study. Technical cut-offs available at the time of study are derived from fold resistance variability of multiple tests on a single wild-type virus in vitro reference strain. After completion of enrollment, biologic cut-offs were created, taking into account the natural variability (×2 SD) in drug susceptibility of viruses isolated from a large population of untreated individuals (²⁰). As a post-hoc analysis, the actual fold resistance results were reinterpreted according to the new biologic cut-offs generated by Virco so as to investigate the correlation of virologic outcomes with the results of the resistance testing. With this objective in mind, the sensitivity score for each regimen (i.e., the number of prescribed drugs to which the virus remained sensitive) was calculated for each patient (⁵). Ritonavir was not considered in the score because it was always prescribed as a boosting dose. The score for lopinavir (for both r-PHT and v-PHT arms) was not available, because results were not provided at the time of the study for this drug. The score obtained was then correlated to virologic outcome and tested either in univariate or multivariate models together with other factors that could have had an impact on treatment response.

For description of the HIV resistance patterns, prevalences of amino acid changes at positions in the HIV reverse transcriptase (RT) and protease (P) genes compared with the gene sequence of the wild-type reference strain HXB2 are presented for both r-PHT and v-PHT arm patient samples; among these substitutions, those occurring at positions indicated by the International AIDS Society as being resistance related (²²) have been selected and analyzed.

Statistical Methods

Normal distribution was tested by means of the Shapiro-Wilk test. Comparisons of demographic characteristics at baseline were performed using χ² tests or unpaired t tests as appropriate.

Intention-to-treat (ITT) analysis was the primary analysis, and the last observation carried forward (LOCF) method was used to impute the information for patients still on antiretroviral therapy at each time point of the follow-up. On-treatment (OT) analysis was also performed and included only randomized patients with measurements at each follow-up time on the original treatment combination.

Power

A two-group χ² test with a 0.050 one-sided significance level will have 87% power to detect the difference between a Group 1 proportion of 0.400 and a Group 2 proportion of 0.200 (OR = 0.375) when the sample size in each group is 80.

Repeated-measure ANOVA was used to test for statistically significant changes over time for pVL and CD4 cell count. The Scheffe test for multiple comparison was used for comparing each pair of group means.

Linear logistic regression (bivariate and multivariate models) was used to model the relation between virologic outcome at each time point and (predictors) independent variables. Variables considered for these analyses were as follows: study arm (v-PHT versus r-PHT), gender, age (years), HIV risk factor (intravenous drug use versus others), baseline CD4⁺ T-cell count (>200 versus <200 cells/mm³), baseline viral load (<4.7 versus >4.7 HIV RNA log₁₀ copies/mL), adherence (poor and fair versus good), duration of previous treatment experience (years), number of experienced treatment lines, resistance sensitivity score of the new regimen (0 versus 1-2 and 3-4), and number of drugs never experienced by patients in the new regimen. The variables found to be significant (p ≤ .2) in univariate analysis or clinically meaningful were entered into multivariate models to identify predictive factors. Missing drug resistance results have been considered resistant in the sensitivity score in a first analysis. The analysis has been repeated excluding regimens with missing resistance results to provide confirmation. ORs were calculated with 95% CIs. Likelihood ratio tests were used to perform analysis between pairs of maximum likelihood models.

A p value of less than 0.05 was considered statistically significant. All tests were two-sided. Analyses were performed with STATISTICA (StatSoft, Tulsa, OK) and with STATA (Stata Statistical Software, release 7.0; StataCorp, College Station, TX).

RESULTS

Characteristics of Patients at Baseline and Resistance Testing Results

Enrollment took place between May and July 2000. A total of 101 patients were randomized into the r-PHT arm of the study, and 100 were randomized into the v-PHT arm. Of those who were randomized, 14 patients dropped out from the study in each arm before starting therapy (13 because resistance testing was not possible due to technical reasons, although pVL was >1000 copies/mL and analysis was repeated twice in samples taken on different occasions; 6 because treatment interruption was decided; 1 because of death; and 8 because they did not return after resistance testing).

The two arms were balanced at baseline by demographics, risk factors for HIV acquisition, and CD4 cell counts, whereas mean pVL was slightly higher in the r-PHT arm at a statistically significant level (p = .028). Treatment history was also comparable between arms (Table 1). Types of antiretroviral combinations at the time of resistance testing are shown in Table 1.

Table 1

More than 70% of patients' isolates retained sensitivity to stavudine, didanosine, zalcitabine, abacavir, and amprenavir. Prevalence of resistance in patients' isolates was similar in the two arms, except for zidovudine. Population resistance categorized by the <4-fold, 4-10-fold, and >10-fold resistance laboratory cut-offs is shown in Figure 1a.

Fig. 1

Genotypic analysis as amino acid changes at positions in the HIV RT and P genes compared with the wild-type reference strain HXB2 showed a mean number of RT substitutions of 25.33 (±7.91 SD) in the r-PHT samples and of 25.56 (±5.82 SD) in the v-PHT samples, whereas the mean number of P substitutions was 11.73 (±4.4 SD) in the r-PHT samples and 11.85 (±3.79 SD) in the v-PHT samples. Among these substitutions reported by Virco, those recognized by the International AIDS Society as being resistance related (²²) have been selected, whose prevalence and distribution in the study arms are described in Figure 2.

Fig 2

Many substitutions of unknown significance have also been detected at resistance-related positions in both HIV-1 RT and P genes. Among the isolates with codons different from wild type, the proportions of those with substitutions not related to drug resistance found in >10% of cases in both arms were as follows: 40% (RT position 188), 38.4% (RT position 106), 37.8% (RT position 219), 16.6% (RT position 151), 15.6% (RT position 74), 13.3% (RT position 67), 10.9% (RT position 190), 10.8% (RT position 215); 57.1% (P position 33), 51.7% (P position 20), 48.9% (P position 73), 21.4% (P position 53), 16% (P position 63), 14.4% (P position 71), 10.9% (P position 54), and 10.6% (P position 36).

Finally, to gather a better description of the resistance patterns, we have analyzed mutations in relation to drug selective pressure at the time of resistance testing. Interestingly, among 59 patients who were not on protease inhibitor (PI) resistance selective pressure at the time of testing, 33 (55.93%) harbored viruses with more than one primary PI mutations, whereas 56 (94.91%) had viruses with more than one secondary PI mutations; all patients with any PI resistance-associated mutations were PI experienced. Primary PI mutations were absent in the HIV isolated from the remaining 3 patients whose HIV did not harbor secondary PI mutations. Among 104 patients who were not on nonnucleoside reverse transcriptase inhibitor (NNRTI) selective pressure, 56 (53.85%) harbored viruses with more than one NNRTI mutation, with 5 of them being NNRTI naive.

Prescribed Treatments and Outcomes

Intensity of treatment combinations prescribed for salvage as well as adherence to the expert recommendations was comparable between the two arms (see Table 1). Of note, the number of drugs was low in both arms and slightly less in the v-PHT arm, although not to a statistically significant level. At least three sensitive drugs were prescribed in the new regimens for 64.4% patients in the r-PHT arm and for 61.4% patients in the v-PHT arm, however. Stavudine and didanosine were the most frequently prescribed drugs, followed by abacavir, amprenavir, saquinavir, and lamivudine, and this was consistent with the reported prevalences of resistance in the cohort (see Fig. 1b).

In the ITT-LOCF analysis, rates of pVL <400 copies/mL were not significantly different between arms over the follow-up period (Fig. 3); OT analysis gave comparable results (11/70 = 16% at week 4, 8/53 = 15% at week 16, 7/43 = 16% at week 32, and 5/25 = 20% at week 48 in the r-PHT arm versus 10/71 = 14% at week 4, 10/52 = 19% at week 16, 9/41 = 22% at week 32, and 8/24 = 33% at week 48 in the v-PHT arm). Despite such a low prevalence of HIV RNA <400 copies/mL, 49% to 68% of patients had a pVL reduction >0.5 log₁₀ copies/mL at ITT-LOCF analysis (see Fig. 3) and 49% to 71% had a pVL reduction >0.5 log₁₀ copies/mL at OT analysis, without a significant difference over the follow-up period between the two treatment arms. The mean pVL decrease ranged from -0.55 to -1.11 log₁₀ copies/mL over the follow-up period on ITT-LOCF analysis (see Fig. 3) and from -0.61 to -1.03 log₁₀ copies/mL on OT analysis, again without a significant difference between the two treatment arms. Results were not influenced by the method used for pVL quantification. On ITT-LOCF analysis, the mean CD4 cell increase was marginally higher in the v-PHT arm over the follow-up period, but this difference was not statistically significant (+48.4 cells/mm³ at week 4, +43 cells/mm³ at week 16, +22.7 cells/mm³ at week 32, and +41.6 cells/mm³ at week 48 in the r-PHT arm, and +42.7 cells/mm³ at week 4, +52.6 cells/mm³ at week 16, +77.7 cells/mm³ at week 32, and +94.4 cells/mm³ at week 48 in the v-PHT arm); OT analysis gave comparable results.

Fig. 3

The numbers of patients who were not on the prescribed treatment combination at the end of the follow-up period were 25 of 173 (14.45%) due to adverse events, 52 of 173 (30.06%) due to virologic failure, and 14 of 173 (8.09%) because of different reasons. The distribution of those 91 patients between the two arms of the study was not statistically different at each time point of the follow-up period.

There were only 12 patients who had interruptions in therapy during the study, and they were not differently distributed between the study groups: 9 (5 because of patient choice and 4 because of toxicity) in the r-PHT arm and 3 (2 because of patient choice and 1 because of toxicity) in the v-PHT arm. The months at stopping in these patients ranged from 1 to 8, and follow-up ceased at the time of treatment interruption.

A total of nine hard clinical end points of the study were recorded over the follow-up period (three HIV-related deaths in the r-PHT arm and two in the v-PHT arm and four AIDS-defining events in the r-PHT arm).

Resistance Results According to the Biologic Cut-Offs and Predictors of Outcomes

Prevalence of resistant HIV isolates for each drug showed a similar pattern when biologic cut-offs (see Fig. 1c) were applied compared with that obtained with the technical cut-offs (see Fig. 1a).

The study of factors with a potential influence on treatment outcome was performed by either bivariate or multivariate logistic regression analyses. Because of the possibility of a time-dependent influence of the considered (predictors) independent variables, the analyses were repeated for each time point, both on ITT-LOCF and OT analyses. Moreover, a composite end point between 32 and 48 weeks of follow-up was considered because of the limited number of outcomes at the last time points. The analysis presented below considers missing drug resistance results as resistant; however, complementary analyses performed excluding regimens with missing data provided confirmation (data not shown).

At ITT-LOCF univariate analysis, variables associated with pVL <400 HIV RNA copies/mL were baseline CD4 cell count, pVL, adherence, and sensitivity score (Table 2). A multivariate model confirmed these covariates to be independently predictive of virologic outcome at the week 32 through 48 composite time point, with ORs of 4.39, 0.25, and 1.87 for CD4 cell count, adherence, and sensitivity score, respectively. Intravenous drug use as a risk factor for HIV acquisition emerged as an independent variable in the multivariate analysis with an OR of 0.33. Intriguingly, over the follow-up period, the predictive value of adherence was found to be stable, whereas that of the CD4 cell count increased; for the sensitivity score, the predictive value was found to decrease over the subsequent time points of the study (see Table 2).

Table 2

On OT analysis, adherence was the only factor independently associated with virologic outcome, and this was consistent across all time points of the study (OR = 0.36, 95% CI: 0.117-1.1, p = .074 at week 16; OR = 0.15, 95% CI: 0.04-0.6, p = .007 at the week 32-48 composite time point). In the same multivariate analysis, the sensitivity score appeared to be independently associated with pVL <400 HIV RNA copies/mL at borderline significant values (OR = 2.13, 95% CI: 0.89-5.07, p = .08 at week 16; OR = 2.41, 95% CI: 0.93-6.64, p = .057 at the week 32-48 composite time point).

Because mutations in the RT gene can confer wide cross-resistance within the class of nucleoside reverse transcriptase inhibitors (NRTIs) and because NRTIs were invariably prescribed for our patients, we assessed the impact of the number of these mutations on virologic outcome. Of note, a significantly higher number of mutations were present in the group of patients who did not respond (pVL >400 HIV RNA copies/mL) at the week 32 through 48 composite virologic primary end point with respect to responders (mean mutations: 4.6 ± 2.2 versus 2.8 ± 2.2;p < .0001). The number of mutations has not been considered as a variable in the logistic regression analyses presented here, because an intrinsic correlation exists between them and the sensitivity score.

DISCUSSION

In the current study, we found similar virologic and immunologic outcomes over 48 weeks of follow-up in heavily pretreated patients randomized to guide the change of therapy according to r-PHT resistance results and in those who received v-PHT testing, together with expert advice in both cases. The number of sensitive drugs in the salvage regimens showed an independent correlation with virologic outcome. Moreover, in a multivariate analysis, CD4⁺ cell count, patient adherence, and, at borderline significance, the number of drugs used for salvage that were never used by patients in the treatment history were independent predictors of virologic response.

The VIRADAPT study showed that the application of genotype resistance testing provides better virologic suppression over the standard of care in patients with limited experience for antiretroviral drugs (^7,8). The Havana study aimed at clarifying the role of expert advice independently of genotyping (¹⁰). Results at 24 weeks indicated a better response with either genotype or expert advice; however, the best response was in the group that received both genotype and expert advice. The CPCRA 046 study confirmed the benefit of genotypic testing offered with expert advice, but this advantage was reduced over time (⁹). In more heavily pretreated patients, the ARGENTA study demonstrated the short-term utility of genotyping at the third month, but at the sixth month, this benefit vanished to nonsignificant levels (¹²); however the Narval study showed no benefit of resistance testing over the standard of care unless patients were experiencing a first PI failure (¹⁴).

The same considerations on the utility of resistance testing apply to the phenotypic tests. In fact, the VIRA3001 study (¹¹) showed significant HIV RNA reduction in the pVL in the phenotype arm compared with the standard of care in patients who had experience for a single HAART regimen, including those on PIs and those who were NNRTI naive. By contrast, in more heavily pretreated patients, the Narval trial failed to demonstrate significant pVL reductions (¹⁴). Finally, there was no overall difference between patients randomized to phenotyping and those assigned to the standard of care in the CCTG 575 study. There was some advantage to phenotyping in the most treatment-experienced group, however (¹³).

The cohort enrolled in the GenPheRex study reflects the heavily pretreated situation of patients in clinical practice today as a result of the sequential introduction of drugs over a long history of antiretroviral therapy. This is also demonstrated by the higher prevalence of resistance mutations detected in our study cohort. For instance, the L90M primary resistance mutation, which confers wide cross-resistance across PIs, was reported at a high prevalence in the ARGENTA (35.8%) (¹²) and Narval (45%) (¹⁴) studies but was found in 52% of cases in this study, and substitutions at position 10 (which are considered markers of intense previous exposure to PIs) were found in 63.6% of the cases in this study. Resistance mutations were correlated with previous or current drug experience, except for NNRTI resistance-associated mutations in patients naive for this drug class. A recent publication on primary HIV infection has demonstrated an increased prevalence of NNRTI resistance but not of PI resistance (²³). Moreover, not all substitutions at positions known for conferring resistance were resistance related, probably as a result of previous specific selective pressure, a concept indicated as intermediate or revertant mutations. It has been recently demonstrated that revertant substitutions can rapidly reselect for the original mutations under the resumption of specific drug-selective pressure, even though they may not increase phenotypic resistance (^24-26). Therefore, these substitutions could potentially be missed in a phenotypic resistance assay when scored as drug susceptible as well as in current v-PHT and rule-based algorithms in which they are not taken into account. It is difficult to assess from this study of heavily pretreated patients whose resistance patterns were complex whether these substitutions could potentially affect viral load and CD4⁺ T-cell count. It is plausible that they may be disadvantageous by improving viral fitness in the presence of drugs with respect to wild-type HIV at those positions, implying that further studies are necessary.

Real and virtual phenotypic resistance is frequent either with technical or biologic cut-off interpretations in our patients. Low-resistance prevalences were found for drugs extensively used in the treatment history of our patients, such as stavudine, didanosine, and zalcitabine. A definitive explanation for this finding has yet to be proved. It has been demonstrated that the difficulty in detecting didanosine resistance may be an artifact caused by inefficient conversion of this drug to the active compound ddATP in stimulated lymphocytes in vitro (²⁷), and similar technical limitations in detecting resistance to stavudine have been suggested (^28,29). It has been demonstrated that modest increases in IC₅₀ or IC₉₀ for stavudine had an important impact on the virologic response either to stavudine monotherapy (³⁰) or to combination therapy (^31,32). Resistance to stavudine or zidovudine (and to NRTIs in general) relies on a common mechanism of hydrolytic removal of the monophosphate drugs from the terminated cDNA chain of HIV-1 through a common set of mutations at positions in the RT enzyme (nucleoside excision mutations [NEMs]) (³³). For this reason, it has been postulated that stavudine genotypic resistance may be more indicative of the clinical outcome than the phenotypic ones; however, genotypic correlates of resistance to stavudine merit further investigation, because a constellation of mutations have been suggested to play a role in resistance to this drug (^34,35). Moreover, it has been suggested that the presence of genotypic resistance does not invariably predict immunologic response to stavudine monotherapy (³⁶). For more recent drugs such as abacavir and amprenavir, prevalence of cross-resistance was also low, and this may rely on either a high genetic barrier against resistance or poor characterization of phenotypic clinical cut-offs. Complete loss of response to abacavir appears to require the combination of three or more NEMs together with the mutation M184V (³⁷), and 4.5-fold clinical resistance have been proposed as the phenotypic cut-off for abacavir (³⁸). A variable virologic response to abacavir-containing regimens was found in patients whose viruses became abacavir sensitive based on the 3-fold biologic cut-off in an analysis from the current study (³⁹), possibly indicating that more studies are necessary. In vitro drug susceptibility studies suggest that patients failing other PIs often have isolates that retain susceptibility to amprenavir and saquinavir (^40,41). Neither drug has demonstrated usefulness when administered as salvage therapy without ritonavir boosting (^42,43), however. Altogether, these observations suggest that further studies are urgently needed to assess clinically relevant cut-offs for most drugs.

Even though many drugs have been recycled and mega-HAART has been avoided in this study because of concerns regarding safety and adherence, average viral load reductions compare favorably with those of other studies in similar cohort of patients. For instance, HIV RNA reduction was -0.66 log₁₀ in the third failure group in the Havana trial at week 24 (¹⁰) versus -0.2 to -0.6 log₁₀ in stratum 3 of the Narval study at week 12 (¹⁴) versus -0.55 to -1.11 log₁₀ over 48 weeks of follow-up in our study. Moreover, an increase in the CD4 cell counts was achieved consistently, with few clinical events recorded over 1 year of follow-up. These results provide indirect confirmation that therapy should be continued in such patients and support the notion that hard-to-reach undetectability of the viral load is not a requirement to obtain temporary clinical stabilization in heavily pretreated patients.

No statistical differences were observed between therapy guided by the r-PHT or v-PHT arm with regard to virologic outcome. Other studies have found a high level of concordance between r-PHT and v-PHT results both in a large unselected cohort (⁴⁴) and in patients with limited drug experience (^45,46). A retrospective long-term follow-up showed comparable viroimmunologic outcomes correlated with either r-PHT or v-PHT results (⁴⁷). In patients who failed at least one antiretroviral regimen in a recent prospective comparative trial (⁴⁸), v-PHT performed as well as or even better than r-PHT in the short term. Our study provides the first prospective randomized evidence that both tests perform equally in the long term. Nevertheless, more studies are necessary to assess possible discordances between genotype- and phenotype-based interpretations. In fact, several potential differences have been recently pointed out and discussed by Parkin et al. (⁴⁹), such as the capacity to detect the presence of virus mixtures and the diverse appreciation of the complex relation between the amino acid sequence of RT and P proteins and their enzymatic activity in the presence of drug.

It is important to discuss several potential drawbacks of the current study. First, similar outcomes in the two arms may be a result of the fact that neither r-PHT nor v-PHT gave any clinical benefit, because a standard of care arm was not present. A standard of care arm was not included in the design, because avoiding resistance testing to choose treatment was not considered to be ethical at that time, especially in this cohort of heavily pretreated patients, whose potential risk of clinical progression was high. In fact, results that questioned clinical benefit from resistance testing in heavily pretreated patients were presented after the GenPheRex study was designed (^12,14). Moreover, ITT analysis conducted through biologic cut-offs showed a significant correlation between sensitivity scores and virologic outcome over the long-term, suggesting that resistance testing was of benefit in this cohort of patients. We are investigating whether sensitivity scores obtained through rule-based genotype interpretation are more predictive for virologic outcome on both OT and ITT analyses. A large prospective study showed no significant differences in virologic or immunologic outcomes between patients randomized to receive a genotype plus rule-based interpretation or the same interpretation plus a v-PHT report in a preliminary analysis after 6 months of follow-up, however (⁵⁰).

Second, technical cut-offs were used for decision making during the study, but biologic cut-offs were used for studying the association between resistance and virologic outcome. Therefore, the latter analysis does not provide insight regarding the randomized trial, even though it maximized the knowledge obtained from the study. In addition, technical cut-offs are no longer used in phenotype-driven interpretations of the Virco assays, although biologic cut-offs may be more close to the clinical ones.

Third, the advice for treatment was provided by experts with access to all the results, and this may greatly reduce the applicability of the study results for the general HIV treating clinician. Moreover, it is possible that the availability of expert advice may have had a diluting effect; thus, the study may have been underpowered to show a difference between arms. Although expert advice was shown to be of clear benefit to patients with intermediate experience with antiretroviral drugs in the Havana study, in patients with more heavy pretreatment experience, this did not seem to have any effect (¹⁰).

Fourth, the differences between turnaround times for receiving a t-PHT or a v-PHT from a patient sample in clinical practice are not addressed in this study, because all samples were sent to Virco for genotyping and phenotyping and time to results was similar in both arms. Thus, the advantage of obtaining a v-PHT in a shorter time with respect to r-PHT results was not appreciated in this study but should be taken into account.

Post-hoc analysis showed a significant correlation of the number of sensitive drugs with virologic outcome. Because this association was found in an ITT analysis that included patients who switched therapy during the follow-up, it can also be deduced that a higher number of sensitive drugs (higher sensitivity scores) reflected a lower risk of cross-resistance and thus a higher probability of response even to subsequent regimens. This is consistent with the inverse correlation found between the number of NRTI mutations at baseline and virologic outcome. Other factors may have had an impact, however. In particular, the independent association found between lower CD4 cell counts at baseline and a worse virologic outcome suggests that resistance testing should not be delayed in patients failing therapy. Antiretroviral treatment history was also important; as more new drugs were included in the new regimens, the better was the virologic response. This finding points to the fact that virus species that emerge on therapy but subsequently become undetectable in plasma virus may still contribute to therapy failure; thus, in highly antiretroviral-experienced patients, consideration of multiple resistance assays and/or previous treatment failures may enhance the clinical utility of these tests (⁵¹). Moreover, another important factor was the level of patient adherence, which has been correlated to virologic outcome at each point of follow-up and with any statistical analyses performed in this study, confirming previous observations (¹²). The definition posed for optimal adherence (>80%) is anachronistic, however, because it has been further recognized that >90% to 95% adherence to the prescribed doses is necessary to provide an optimal sustained virologic response (⁵²). The positive impact of >80% adherence has been demonstrated in this study, suggesting that >90% to 95% adherence is even more important.

In conclusion, virologic outcomes did not differ when an r-PHT or v-PHT was used in this cohort of heavily pretreated patients. Given that v-PHT testing is less expensive, easier to perform, and less time-consuming than r-PHT testing, this result suggests that the former is a valuable surrogate for r-PHT testing and might be preferred from a cost-benefit perspective in clinical practice. Last but not least, this study demonstrates that additional factors have to be integrated in salvage strategies to take better advantage of resistance testing in clinical practice, such as treatment history, clinical status, and a patient's ability to adhere to treatment.