With a few exceptions [1, 2], a number of studies have shown that drug resistance testing improves the efficacy

Transcript

1 HIV/AIDS MAJOR ARTICLE Comparison of a Rule-Based Algorithm with a Phenotype-Based Algorithm for the Interpretation of HIV Genotypes in Guiding Salvage Regimens in HIV- Infected Patients by a Randomized Clinical Trial: The Mutations and Salvage Study Nicola Gianotti, 1 Vincenzo Mondino, 4 Maria Cristina Rossi, 5 Elisabetta Chiesa, 2 Ivano Mezzaroma, 6 Nicoletta Ladisa, 8 Giovanni Guaraldi, 9 Carlo Torti, 10 Pierluigi Tarquini, 11 Paula Castelli, 12 Aldo Di Carlo, 7 Enzo Boeri, 3 Wilco Keulen, 13 Paula Mc Kenna, 14 and Adriano Lazzarin, 1 on behalf of the Mutations and Salvage (MuSa) Study Group a 1 Clinica di Malattie Infettive, Università Vita-Salute San Raffaele, 2 Clinica di Malattie Infettive, Ospedale Luigi Sacco, and 3 Diagnostica & Ricerca San Raffaele, Milan, 4 Divisione di Malattie Infettive, Ospedali Riuniti di Pallanza, Verbania, 5 Divisione di Malattie Infettive, Ospedale Ca Foncello, Treviso, 6 Dipartimento di Medicina Clinica, Università di Roma La Sapienza, and 7 Unità Operativa AIDS Rome 71, Istituto di Ricovero e Cura a Carattere Scientifico Dermosifilopatico S. Maria e S. Gallicano, Roma, 8 Clinica di Malattie Infettive, Policlinico di Bari, 9 Clinica di Malattie Infettive e Tropicali Università di Modena e Reggio Emilia, 10 Clinica di Malattie Infettive e Tropicali, Spedali Civili di Brescia, 11 Unità Operativa Malattie Infettive, Ospedale G. Mazzini, Teramo, and 12 Reparto di Malattie Infettive, Ospedale di Macerata, Italy; 13 Virology Education, Utrecht, The Netherlands; and 14 Virco, Mechelen, Belgium Background. There is still considerable uncertainty as to the best algorithm for interpreting humanimmunodeficiency virus (HIV) genotyping results. Methods. A total of 318 subjects with HIV RNA levels of copies/ml were enrolled in 41 centers throughout Italy from 2001 through 2003, stratified on the basis of their drug history, randomized (1:1) to 2 arms to have their treatments modified on the basis of the results of HIV genotyping (as interpreted by virtual phenotype analysis or with use of a rule-based interpretation system), and followed up for 48 weeks. At least 1 nucleoside reverse-transcriptase inhibitor and 1 protease inhibitor had to be included in any new regimen; nonnucleoside reverse-transcriptase inhibitor naive patients were also prescribed a nonnucleoside reverse-transcriptase inhibitor. Only drugs licensed in Italy were allowed. The primary end point was a decrease in HIV RNA level to!400 copies/ml by week 12 according to ontreatment analysis. Results. The mean ( standard deviation) values at baseline were as follows: HIV RNA level, log 10 copies/ml; CD4 + T lymphocyte count, cells/ml; reverse-transcriptase mutations, ; and protease mutations, There were 133 patients (41.8%) who were nonnucleoside reverse-transcriptase inhibitor naive and protease inhibitor experienced, 63 patients (19.8%) who were nonnucleoside reverse-transcriptase inhibitor experienced and protease inhibitor naive, and 122 patients (38.4%) who were 3-class experienced. A total of 192 patients completed 12 weeks of the treatment regimen assigned at baseline; at 12 weeks, 66.3% of patients in the virtual phenotype arm and 71.3% of patients in the rule-based interpretation arm had HIV RNA levels of!400 copies/ml ( P p.46). No statistically significant difference between arms was observed by intention-to-treat analysis. Conclusion. Both the virtual phenotype and rule-based interpretation methods of HIV genotyping can guide the selection of effective antiretroviral drugs for a salvage regimen. With a few exceptions [1, 2], a number of studies have shown that drug resistance testing improves the efficacy Received 4 November 2005; accepted 1 February 2006; electronically published 13 April Reprints or correspondence: Dr. Nicola Gianotti, Ospedale San Raffaele, Divisione di Malattie Infettive, Via Stamira d Ancona 20, Milano MI, Italy (nicola.gianotti@hsr.it). Clinical Infectious Diseases 2006; 42: by the Infectious Diseases Society of America. All rights reserved /2006/ $15.00 of antiretroviral therapy in HIV-infected patients with treatment failure, compared with the empirical selection of drugs for a salvage regimen [3 7]. Identifying an effective algorithm for interpreting the results is crucially important for the management of Presented in part: 10th European AIDS Conference/EACS, November 2005, Dublin, Ireland (abstract PE3.4/2). a Members of the MuSa Study Group are listed at the end of the text CID 2006:42 (15 May) HIV/AIDS

2 Figure 1. Flow diagram of the study, showing patient disposition through 48 weeks of follow-up HIV-infected patients who experience failure of their current antiretroviral regimen. Two such algorithms are RetroGram (Virology Education) and VirtualPhenotype (Virco). The former is a rule-based interpretation (RBI) algorithm that takes into account updated published data concerning resistance and virological response to each considered drug in the presence of defined sets of mutations; the latter is a virtual phenotype interpretation (VPI) method that predicts the HIV phenotype on the basis of the correlation between phenotype and genotype in the presence of a given set of mutations, as inferred from a database of paired genotypes and phenotypes [8]. One retrospective analysis suggested that rule-based algorithms may be more predictive of response than an early version of virutal phenotyping [9], but only 1 study has made a randomized comparison of rule-based and phenotype prediction based algorithms [10]. It has been shown that expert advice further improves the performance of genotype resistance testing [7], but, because such advice is often unavailable in many clinical settings, it is important to identify an interpretative algorithm with results that are easily understood by most clinicians caring for HIV-infected patients. The aim of the Mutations and Salvage (MuSa) study was to compare the efficacy of a rule-based algorithm with that of a phenotype-based algorithm for interpreting HIV genotyping as a means of guiding salvage regimens under conditions that reflect the real-world scenario of HIV treatment, in which expert advice is largely unavailable. METHODS Study design. From 2001 through 2003, subjects experiencing failure of their first or second HAART regimen were recruited and enrolled in 41 centers throughout Italy. The study protocol was approved by each local ethics committee, and all of the participants gave their informed consent. After being stratified into 3 groups on the basis of their treatment history (group 1, nonnucleoside reverse-transcriptase inhibitor (NNRTI) naive, protease inhibitor (PI) experienced patients; group 2, PI-naive, NNRTI-experienced patients; and group 3, PI- and NNRTI-experienced patients), the patients were centrally randomized (1:1) at the coordinating site (San Raffaele Hospital, Milan, Italy) to receive treatments modified on the basis of the results of HIV genotyping, as interpreted by the VirtualPhenotype system, version (Virco), with the first-generation biological cut-off points [8] available during the course of the study (VPI arm) or by the RetroGram rule-based system, version i (RetroGram) (RBI arm). All of the enrolled subjects were nucleoside reverse-transcriptase inhibitor (NRTI) experienced and, on the basis of VPI or RBI results, were prescribed a new antiretroviral regimen of at least 3 drugs (including at least 1 NRTI and 1 PI). The NNRTInaive patients were also prescribed an NNRTI. Only drugs licensed in Italy were allowed to be administered. In particular, enfuvirtide was not prescribed to any of the patients enrolled in the study, lopinavir-ritonavir was available from the begin- HIV/AIDS CID 2006:42 (15 May) 1471

3 Table 1. Baseline characteristics of patients randomized to either the virtual phenotype interpretation (VPI) arm or the rule-based interpretation (RBI) arm. Characteristic All patients (n p 318) VPI arm (n p 161) RBI arm (n p 157) P Risk category.26 Drug addiction 129 (40.6) 58 (36) 71 (45.2) Homosexual intercourse 65 (20.4) 32 (19.9) 33 (21) Heterosexual intercourse 102 (32.1) 55 (34.2) 47 (29.9) Homosexual and heterosexual intercourse 12 (3.8) 9 (5.6) 3 (1.9) Other 10 (3.1) 7 (4.3) 3 (1.9) CDC stage at study entry.47 A 184 (57.9) 89 (55.3) 95 (60.5) B 48 (15.1) 24 (14.9) 24 (15.3) C 81 (25.5) 44 (27.3) 37 (23.6) U 5 (1.6) 4 (2.5) 1 (0.6) Period since first antiretroviral treatment, mean years SD No. of reverse-transcriptase drug-resistance mutations, mean SD No. of protease drug-resistance mutations, mean SD Drug history.69 2-class experienced NNRTI naive and PI experienced 133 (41.8) 71 (44.1) 62 (39.5) NNRTI experienced and PI naive 63 (19.8) 30 (18.6) 33 (21) 3-class experienced 122 (38.4) 60 (37.3) 62 (39.5) CD4 + T lymphocyte count, mean cells/ml SD HIV RNA level, mean log 10 copies/ml SD NOTE. Data are no. (%) of patients, unless otherwise indicated. CDC, Centers for Disease Control and Prevention; NNRTI, nonnucleoside reverse-transcriptase inhibitor; PI, protease inhibitor; U, undetermined. ning of the study, and tenofovir was available during the last year of the study. Virological methods. All of the patients had virus genotyped by the Virology Laboratory of San Raffaele Hospital. HIV RNA was extracted using a QIAmp Viral RNA kit (Qiagen) and retrotranscribed to cdna using Expand Reverse Transcriptase (Roche Diagnostics). The cdna was amplified by means of 2 nested reactions using the Expand High Fidelity PCR System Kit (Roche Diagnostics) and oligonucleotide primers (Virco). The amplification primers were purified using the QIAquick kit (Qiagen). Megabace 1000 (Amersham Biosciences) was used for sequencing, and the sequences were assembled by a Sequencer (Genecodes). HIV RNA level was assessed in each center by means of branched DNA (Versant RNA, version 3.0 [Bayer]), Amplicor (Monitor test, version 1.5 [Roche Diagnostics]) or nucleic acid sequence based amplification (Nuclisens, version 2.1 [Bio- Merieux]). All of the tests had a quantification limit of!400 copies/ml. VPI and RBI genotype interpretation and definition of drug activity. In the VPI assay [8], the patient s set of mutated amino acids is processed against a constantly growing genotypephenotype correlated dataset (at the time of the study, the dataset contained 128,000 sets of clinically isolated HIV-1 genotypes and matching drug-susceptibility phenotypes) and used to predict drug susceptibilities by retrieving phenotypic results for isolates with the same mutational profile from the genotype-phenotype correlated database. Two possibilities can be distinguished at this point: if the database returns 10 matches, a mean fold-resistance is calculated for each drug; if it returns!10 matches, VPI falls back on a rules-based interpretation of the mutation profile of the new isolate in combination with all of the available quantitative data and offers a qualitative prediction of likely or unlikely drug resistance. RetroGram addresses the use of each antiretroviral drug by ranking them according to published data concerning their anticipated activity against pathogens with a given set of mutations. Each drug is classified into 1 of 5 ranks: A, the drug should be preferentially used; B, the drug should be used only if no rank A drug is available or can be administered; C, the drug should be used only if no drug in rank A or B is available or can be administered; D, the drug should be used only if no 1472 CID 2006:42 (15 May) HIV/AIDS

4 Figure 2. Reverse-transcriptase (RT) drug-resistance mutations at baseline in the rule-based interpretation (RBI) arm and the virtual phenotype interpretation (VPI) arm of the study. Asterisk indicate mutations for which a statistically significant difference between arms was detected ( P!.05). drug in rank A, B, or C is available or can be administered; and U (i.e., unranked), no indication concerning the drug can be provided on the basis of the available data. To compare data for antiretrovirals in the VPI and RBI arms, the VPI outputs of resistance (i.e., a higher estimated foldresistance than the biological cut-off point) and resistance likely were pooled, and the proportion of patients with virus resistant to each drug was compared with the proportion of patients with virus for which RBI placed the same drug in rank CorD. Each patient was similarly assigned a regimen activity score (i.e., the number of drugs initially received that could be considered active on the basis of the HIV genotype interpretation provided by the algorithm to which the patient was randomized); this score was 3 if all 3 drugs of the new treatment regimen were classified as sensitive or resistance unlikely by VPI or ranked A or B by RBI. The patients starting a new regimen with only 1 or 2 drugs classified as sensitive or resistance unlikely by VPI or ranked A or B by RBI were assigned a regimen activity score of 1 or 2. Statistical analyses. The primary end point was a decrease in HIV RNA level to!400 copies/ml at week 12, as assessed by on-treatment (OT) analysis (including only those patients who were still receiving the regimen prescribed on the basis of the results of their baseline resistance testing). The secondary end points were the quantitative decrease in HIV RNA level and the increase in CD4 + T lymphocyte count at week 4, 12, and 48 and the probability of maintaining an HIV RNA level of!400 copies/ml at week 48, also assessed by OT analysis. A sample size of 480 patients was required for the estimated probability of reaching the primary end point to be 60% in arm A and 45% in arm B; the expected dropout rate was 25%, HIV/AIDS CID 2006:42 (15 May) 1473

5 Figure 3. Protease drug-resistance mutations at baseline in the rule-based interpretation (RBI) arm and the virtual phenotype interpretation (VPI) arm of the study. and power was fixed at 80% ( b p 80%, with a p 5% ). The OT analyses excluded the patients whose treatment was modified for any reason, thus considering only patients still receiving treatment with the same regimen that had been prescribed according to the results of resistance testing at baseline. The intention-to-treat (ITT) analyses were made on the basis that missing data or noncompletion of treatment equaled treatment failure. Categorical variables were compared between arms using distribution frequencies, with the x 2 or Fisher s exact test being used to identify any statistically significant differences; the latter was used when the former was inappropriate, because the expected frequency of observations in at least 1 cell was!5. Continuous variables were compared between arms by calculating their mean and median values, SDs, and interquartile ranges, with Student s t test being used to determine whether the differences were statistically significant. Statistical significance was defined as a 2-tailed P value of!.05. Because of the skewed nature of the plasma HIV RNA dataset, a logarithmic transformation was considered. Multiple-logistic regression analyses were used to estimate the adjusted between-arm OR (arm A vs. arm B) of having HIV RNA levels of!400 copies/ml 12 weeks after randomization. The variables included in the multivariable analysis (those considered to be the most clinically relevant in the studied circumstances) were baseline HIV RNA level, CD4 + T lymphocyte count, Centers for Disease Control and Prevention disease stage (A or B vs. C), drug experience (3-class vs. 2-class experienced), number of reverse-transcriptase (RT) mutations, and number of protease mutations. All analyses were performed by SAS software, version 8.2 (SAS). RESULTS Enrollment began in June 2001 and was terminated 2.5 years after the first patient was enrolled, because no further patients 1474 CID 2006:42 (15 May) HIV/AIDS

6 Table 2. No. (%) of isolates determined to be resistant according to virtual phenotype interpretation (VPI) or the classified C or D according to the rule-based interpretation (RBI) method for the specified drug at baseline. Drug VPI (n p 159) RBI (n p 151) P Zidovudine 69 (43.4) 72 (47.7).42 Lamivudine 116 (73) 102 (67.5).34 Abacavir 42 (26.4) 69 (45.7).0003 Stavudine 17 (10.7) 92 (61)!.0001 Didanosine 20 (12.6) 38 (25.2).005 Zalcitabine 19 (6.3) 42 (27.9)!.0001 Tenofovir 3 (2.6) 7 (18).0007 Nevirapine 71 (44.7) 72 (47.7).55 Delavirdine 63 (39.6) 71 (47.1).15 Efavirenz 74 (46.5) 60 (39.8).25 Saquinavir 28 (17.6) 43 (28.5).02 Indinavir 36 (22.6) 38 (25.2).58 Ritonavir 42 (26.4) 38 (25.2).88 Nelfinavir 59 (37.1) 76 (50.3).01 Amprenavir 25 (15.7) 20 (13.3).57 Lopinavir-ritonavir 20 (12.6) 4 (2.7).001 could be recruited by any center, probably because of the widespread availability of new drugs through expanded access programs and the wider availability of resistance testing. The data refer to the 318 patients who were randomized up to 20 December Figure 1 shows patient disposition throughout the 48 weeks of the study, and table 1 summarizes the patients characteristics at baseline; the only statistically significant between-arm difference at baseline was with respect to the mean number of RT mutations per subject, which was higher in the VPI arm (5.2 vs. 4.3; P p.005). There was no statistically significant difference between the 2 arms in terms of ongoing antiretroviral treatment at enrollment. Figures 2 and 3 show the prevalence of RT and protease drug-resistance mutations at baseline. The only statistically significant differences between the VPI and RBI arms were with respect to the proportion of patients with the RT 41L (45.9% vs. 35.1%; P p.03) and RT 184V mutation (76.4% vs. 67.6%; P p.04), which was higher in the VPI arm, and the proportion of patients with the RT 75T/I mutation (5.4% vs. 0.6%; P p.02), which was higher among those randomized to the RBI arm. The VPI and RBI systems differently assessed the use of 8 (50%) of 16 drugs (table 2). Only 4 (25%) of the 16 drugs were actually prescribed in significantly different proportions in the 2 arms; the drugs that were more frequently prescribed in the VPI arm were stavudine (51.6% vs. 35.5%; P p.006) and nelfinavir (19.3% vs. 5.8%; P p.0006), and the drugs that were more frequently prescribed in the RBI arm were lamivudine (34.8% vs. 23.6%; P p.03) and lopinavir-ritonavir (46.5% vs. 31.7%; P p.01). The mean ( SD) regimen activity scores in the VPI and RBI arms were and , respectively ( P p.15). Primary end point. The OT analysis showed that 132 (68.8%) of 192 patients had plasma HIV RNA levels of!400 copies/ml at week 12, including 65 (66.3%) of 98 patients in the VPI arm and 67 (71.3%) of 94 patients in the RBI arm ( P p.46) (figure 4). The risk difference was 4.95 (95% CI, 8.13 to 18.03). Secondary end points. The ITT analysis showed that the proportion of patients with plasma HIV RNA levels of!400 copies/ml at week 12 was 40.4% in the VPI arm and 42.7% in the RBI arm ( P p.68). There was no between-arm difference in the proportion of patients with HIV RNA levels of!400 copies/ml at week 4, 24, 36, or 48 according to either OT (figure 4) or ITT analysis. In particular, by ITT analysis, the proportion of patients with plasma HIV RNA levels of!400 copies/ml at week 48 was 23.6% in the VPI arm and 29.9% in the RBI arm ( P p.2). The risk difference at this time point by ITT analysis was 6.33 (95% CI, 3.38 to 16.05). Kaplan- Meier analysis showed that there was also no between-arm difference in the cumulative probability of maintaining a virological response at week 48 ( P p.45, by log-rank test). There were also no between-arm differences in the mean HIV RNA concentrations or in CD4 + T lymphocyte counts at any time point during the 48 weeks of observation, by either OT (figure 5) or ITT analysis. At week 12, the proportion of 2-class experienced patients with HIV RNA levels of!400 copes/ml was significantly greater than that of 3-class experienced patients (75.7% vs. 58.4% by OT analysis; P p.01). However, the randomization arm did not influence this result, because OT analysis showed that the end point was reached by 75.4% of the 2-class experienced subjects in the VPI arm and 75.9% of those in the RBI arm ( P p.95), and the corresponding proportions of 3-class experienced subjects in the 2 arms were 51.4% and 65%, respectively ( P p.22). Multivariable analysis. The results of the multiple-logistic regression analysis are given in table 3. The OT analysis showed that the only independent correlates of virological response at week 12 were HIV RNA level at baseline (OR per log 10 copies/ ml, 0.55; 95% CI, ; P p.032) and CD4 + T lymphocyte count at baseline (OR per 50-cell increment, 1.08; 95% CI, ; P p.048). The ITT analysis confirmed only baseline CD4 + T lymphocyte count (OR per 50-cell increment, 1.07; 95% CI, ; P p.008) as independently predicting the 12-week virological response. HIV/AIDS CID 2006:42 (15 May) 1475

7 Figure 4. On-treatment analysis of virological response (defined as an HIV RNA level of!400 copies/ml) through 48 weeks of follow-up for patients randomized to either the virtual phenotype interpretation (VPI) arm or the rule-based interpretation (RBI) arm. The primary end point was at 12 weeks. BL, baseline. DISCUSSION Despite the proven usefulness of HIV drug-resistance testing, there is still considerable uncertainty as to the best algorithm for interpreting genotyping results [11]. Various algorithms have been proposed, but their clinical efficacy has been tested in only 1 other randomized clinical trial, in which Torti et al. [10] found no statistically significant between-arm difference in the efficacy of salvage regimens of antiretroviral drugs selected (with the help of an expert) on the basis of the HIV genotype interpretation offered by an RBI algorithm (Visible Genetics) or VPI algorithm. Our results show that both algorithms were effective in guiding salvage therapies. No statistically significant differences between them were observed for our study population, although the smaller number of RT mutations at baseline in the RBI arm may have slightly favored the outcomes of the patients randomized to receive a new regimen on the basis of the results of this test. However, it seems unlikely that this small, betweenarm difference at baseline had any meaningful effect on the overall results. Because the biological cutoff points of the VPI algorithm used in the course of this study are no longer applied and because some of the rules of the RBI algorithm used in the course of this study have changed, we cannot exclude the possibility that our results might have been different if we had used the updated or future versions of the 2 algorithms. The algorithms proved to be comparably efficacious, despite significant differences in the way in which each algorithm addressed the use of individual drugs and the consequent significant difference in their actual use in the 2 arms. A possible reason for this is that errors in the use of 1 or more drugs in 1 arm were compensated for by errors made in the other arm. However, a significant between-arm difference in actual drug prescription was found for only one-quarter of the patients. It is probable that differences in genotype interpretation do not always lead to differences in drug prescription, because the latter involves the global clinical evaluation of a patient and not just the evaluation of the patient s viral genotype. We used 12-week virological outcome as our primary end point to allow comparison with other studies [1, 4 6] and because long-term changes in antiretroviral regimens are frequent and usually caused by toxicity and decreasing levels of compliance, rather than by virological failure [12]. However, our short-term result was confirmed by our extended, 48-week observations. Previous studies have shown that expert advice significantly increases the chance of identifying an effective salvage regimen [7], but we decided to test the efficacy of the 2 tests in a real world setting, in which expert advice is not easily available. Our findings indicate that, if the results of drugresistance testing (i.e., the interpretation provided by an algorithm) are presented in a clear fashion, even clinicians who are not expert in the field of HIV drug resistance can select effective drugs for a salvage regimen. However, we cannot exclude the possibility that the accompaniment of expert advice would lead to even better results. The study was initially planned to enroll 480 patients and to have 80% power in detecting a 15% difference between the 1476 CID 2006:42 (15 May) HIV/AIDS

8 Figure 5. On-treatment analysis of mean HIV RNA level and mean CD4 + T lymphocyte count through 48 weeks of follow-up for patients randomized to either the virtual phenotype interpretation (VPI) arm or the rule-based interpretation (RBI) arm. Bars, 95% CI. BL, baseline. arms in terms of the primary end point. The enrollment target was not reached (meaning that the power ratings of the OT and ITT analyses were 50% and 70%, respectively) but, given the small difference found in virologic outcome between arms, even if the target sample size had been reached, it is unlikely that we would have found a difference 115%. Nevertheless, we cannot exclude the possibility that minor (but potentially clinically relevant) differences exist. We concentrated on the OT analysis, because we believe that this is the best means of evaluating the efficacy of any diagnostic tool. We believe that the performance of drug-resistance testing is best assessed in patients receiving all of the drugs selected on the basis of the results of the test (the best possible regimen in the given setting) and that any therapeutic failures due to treatment withdrawal or missed visits should not be attributed to the test itself. However, we also made an ITT analysis, and the results did not lead us to change our conclusions. Our findings are consistent with those of the Resistance and HIV/AIDS CID 2006:42 (15 May) 1477

9 Table 3. Multiple-logistic regression on-treatment analysis for independent predictors of virological response, defined as an HIV RNA level of!400 copies/ml at week 12. Variable OR (95% CI) P Randomization arm (VPI vs. RBI) 0.67 ( ).242 Baseline HIV RNA level, per log 10 copies/ml increase 0.55 ( ).032 Baseline CD4 + T lymphocyte count, per 50-cell increment 1.08 (1 1.16).048 Baseline CDC stage, A and B vs. C 0.71 ( ).411 Drug history, 3-class experienced vs. 2-class experienced 0.55 ( ).096 No. of reverse-transcriptase drug-resistance mutations, per mutation 0.92 ( ).187 No. of protease drug-resistance mutations, per mutation 0.89 ( ).112 NOTE. CDC, Centers for Disease Control and Prevention; RBI, rule-based interpretation; VPI, virtual phenotype interpretation. Dosage Adapted Regimens Study, in which the genotype interpretations obtained using the VPI or RBI algorithms did not lead to any major differences in the virological outcome, probably as a consequence of the availability of expert advice [10]. However, because our study was free of this possible bias, our findings can be generalized to the many clinical settings in which the interpretation of HIV genotyping by experts is not a usual component of care or part of the normal clinical decision-making process when a patient no longer responds to a given antiretroviral therapy. The consistency between our results and those of the Resistance and Dosage Adapted Regimens (RADAR) Study also suggests that the efficacy of VPI and RBI algorithms is comparable. Finally, although the long-term efficacy of drug-resistance testing for managing the cases of patients who are experiencing failure of antiretroviral drugs has been questioned [13], the sustained reduction in viral load and the duration of immune recovery observed throughout the 48 weeks of our study suggest that this diagnostic procedure provides more than a short-term benefit. In conclusion, the versions of both the RBI and VPI algorithms used during the course of our study were able to guide the selection of effective antiretroviral drugs for a salvage regimen, with no large differences between treatment arms with regard to outcome. MEMBERS OF THE MUSA STUDY GROUP Marco Dini and Romana Del Gobbo (Divisione Malattie Infettive Azienda Ospedaliera Umberto I, Ancona, Italy); Maria Montroni and Luca Butini (Servizio di Immunologia Clinica e Tipizzazione Tissutale, Ancona, Italy); Giorgio Scalise, Fausto Ancarani, Simona Di Cesare, and Andrea Giacometti (Clinica di Malattie Infettive, Azienda Ospedaliera Umberto I, Ancona, Italy); Alberto Biglino and Maria Degioanni (Reparto di Malattie Infettive Ospedale Civile di Asti, Asti, Italy); Maurizio Paoloni, Rinalda Mariani, and Giulio Calella (Reparto di Malattie Infettive, Ospedale Civile di Avezzano, Avezzano, Italy); Giuseppe Pastore, Nicoletta Ladisa, and Giuseppe Mennea (Cattedra di Malattie Infettive Dipartimento di Clinica Medica Immunologia e Malattie infettive, Policlinico di Bari, Bari, Italy); Francesco M. Gritti and Giovanni Fasullo (U. O. Malattie Infettive, Ospedale Maggiore di Bologna, Bologna, Italy); Francesco Chiodo and Marco Borderi (Istituto di Malattie Infettive, Università degli Studi di Bologna, Bologna, Italy); Gianpiero Carosi, Francesco Castelli, Carlo Torti, and Cristina Uccelli (Clinica di Malattie Infettive e Tropicali, Spedali Civili di Brescia, Brescia, Italy); Giuliano Rizzardini and Raffaella Visonà (Reparto di Malattie Infettive, Ospedale di Busto Arsizio, Busto Arsizio, Italy); Antonino Salvo and Alfonso Averna (Divisione di Malattie Infettive, Ospedale S. Elia, Caltanissetta, Italy); Rosario Russo, Benedetto Maurizio Celesia, and Stefano Cosentino (Istituto di Malattie Infettive, Università di Catania, Azienda Ospedaliera Garibaldi, S. Luigi, S. Currò, Ascoli-Tomaselli, Catania, Italy); Enrico Rinaldi and Luigi Pusterla (Divisione Malattie Infettive, Ospedale S. Anna, Como, Italy); Giuseppe Carnevale, Placido Mondello, and Corrado Petrini (Divisione di Malattie Infettive, Azienda Ospedaliera di Cremona, Cremona, Italy); Florio Ghinelli and Laura Sighinolfi (Divisione Malattie Infettive, Arcispedale S. Anna, Ferrara, Italy); Francesco Mazzotta, Sergio Lo Caputo, and Piera Pierotti (Divisione di Malattie Infettive, Ospedale S. Maria Annunziata, Firenze, Italy); Francesco Leoncini, Gaetana (Katia) Sterrantino, and Paola Corsi (Divisione Malattie Infettive, Policlinico Careggi, Firenze, Italy); Gioacchino Angarano and Annalisa Saracino (Cattedra di Malattie Infettive, Dipartimento di Clinica Medica Immunologia e Malattie Infettive, Policlinico di Bari, Bari, Italy); Antonio G. Pompei and Maria Teresa Di Toro (Unità Operativa AIDS, Ospedale S. Giovanni di Dio, Fondi, Italy); Antonio Giuseppe Pompei and Francesco Purificato (Reparto di Malattie Infettive, Ospedale Dono Svizzero di Formia, Formia, Italy); Francesco Indiveri, Maurizio Setti, and Giuseppe Murdaca (Clinica di Medicina Interna ad Orientamento Immunologico, Dipartimento di Medicina Interna, Università di Genova, Genova, Italy); Augusto Iannessi, Antonio Cellini, and Adriano Mariani (Divisione di Malattie Infettive, Ospedale Regionale S. Salvatore, L Aquila, Italy); Antonio Scasso and Michele De Gennaro (Unità Operativa Malattie Infettive, Os CID 2006:42 (15 May) HIV/AIDS

10 pedale Campo di Marte, Lucca, Italy); Alessandro Chiodera and Paula Castelli (Reparto di Malattie Infettive, Ospedale di Macerata, Macerata, Italy); Katia Maltempo and G. Monolo (Ambulatorio HIV, Servizio per le Tossicodipendenze di Magenta, Magenta, Italy); Adriano Lazzarin and Nicola Gianotti (Clinica di Malattie Infettive, Università Vita-Salute San Raffaele, Ospedale San Raffaele, Milano, Italy); Liliana Caggese and Irene Schlacht (Divisione di Malattie Infettive, Ospedale Niguarda Ca Granda, Milano, Italy); Antonietta Cargnel, Chiara Atzori, and Valeria Micheli (II Divisione di Malattie Infettive, Ospedale Luigi Sacco, Milano, Italy); Mauro Moroni, Teresa Bini, Ivana Marcatto, and Elisabetta Chiesa (Clinica di Malattie Infettive, Ospedale Luigi Sacco, Milano, Italy); Gianmarco Vigevani, Barbara Argenteri, and Amedeo Capetti (I Divisione di Malattie Infettive, Ospedale Luigi Sacco, Milano, Italy); Roberto Esposito, Giovanni Guaraldi, and Barbara Beghetto (Clinica di Malattie Infettive e Tropicali, Policlinico di Modena, Modena, Italy); Antonio Chirianni and Miriam Gargiulo (III Divisione di Malattie Infettive, Ospedale D. Cotugno, Napoli, Italy); Nicola Abrescia, Maurizio D Abbraccio, and Tina Busto (IV Divisione di Malattie Infettive, Ospedale D. Cotugno, Napoli, Italy); Carla Marchitelli and Maurizio Santirocchi (Servizio per le Tossicodipendenze Narni [Azienda Sanitaria Locale 4- Terni], Narni Scalo, Italy); Vincenzo Abbadessa and Salvatrice Mancuso (Servizio di Riferimento Regionale per la Diagnosi di AIDS, Azienda Universitaria Policlinico di Palermo, Palermo, Italy); Francesco Meneghetti and Marzia Franzetti (Divisione di Malattie Infettive e Tropicali, Azienda Ospedaliera di Padova, Padova, Italy); Carlo Ferrari and Paolo Pizzaferri (Divisione di Malattie Infettive e Immunopatologia Virale, Azienda Ospedaliera di Parma, Parma, Italy); Giuliano Stagni and Francesco Di Candilo (Unità Operativa Malattie Infettive, Policlinico Monteluce di Perugia, Perugia, Italy); Enzo Petrelli and Maria Balducci (Divisione di Malattie Infettive, Ospedale San Salvatore di Pesaro, Pesaro, Italy); Francesco Menichetti, Marina Polidori, and Carlo Tascini (Unità Operativa Malattie Infettive, Azienda Ospedaliera Pisana, Ospedale Cisanello, Pisa, Italy); Aldo Di Carlo, and Guido Palamara (Unità Operativa AIDS Rome 71, Istituto Dermosifilopatico S. Maria e S. Gallicano, Roma, Italy); Andrea Antinori and Giuseppina Liuzzi (III Day Hospital, Istituto Nazionale di Malattie Infettive, Istituto di Ricovero e Cura a Carattere Scientifico Lazzaro Spallanzani, Roma, Italy); Fernando Aiuti, Ivano Mezzaroma, Elena Pinter, and Barbara Barbone (Dipartimento di Medicina Clinica, Università di Roma La Sapienza, Roma, Italy); Vincenzo Vullo, Paola Massetti, and Serena Dell Isola (I Cattedra Malattie Infettive, Policlinico Umberto I, Roma, Italy); Maria Stella Mura, Marco Mannazzu, and Rossana Delias (Istituto di Malattie Infettive, Ospedale Civile Vecchio, Sassari, Italy); Dante Di Giammartino and Pierluigi Tarquini (Unità Operativa di Malattie Infettive, Ospedale G. Mazzini, Teramo, Italy); Franco Marranconi and Roberto Ferretto (UOA Malattie Infettive, Ospedale Civile S. Camillo De Lollis, Schio, Italy); Pietro Caramello and Gian Carlo Orofino (Unità Operativa Divisione A di Malattie Infettive, Ospedale Amedeo di Savoia, Torino, Italy); Tommaso Cimino and Caterina Bramato (Ambulatorio Malattie Infettive, Unità Operativa Reparto Detenuti, Ospedale Amedeo di Savoia, Torino, Italy); Giovanni Di Perri, Alessandro Sinicco, Daniela Zeme, Stefano Bonora, and Laura Trentini (Clinica di Malattie Infettive, Ospedale Amedeo di Savoia, Torino, Italy); Maria Luisa Soranzo, A. Macor, and Bernardino Salassa (Unità Operativa Divisione B di Malattie Infettive, Ospedale Amedeo di Savoia, Torino, Italy); Fabio Branz and Palma Delle Foglie (Divisione di Malattie Infettive, Ospedale Villa Igea, Trento, Italy); Alberto Vaglia and M. Cristina Rossi (Divisione di Malattie Infettive, Ospedale Ca Foncello, Treviso, Italy); Luciano Ciccone and Longino Panzolli (Centro di Riferimento Provinciale AIDS, Azienda Sanitaria Locale No.4 Medio Friuli, Udine, Italy); Paolo Grossi and Massimo Giola (Divisione di Malattie Infettive e Tropicali, Ospedale di Circolo e Fondazione Macchi, Varese, Italy); Enzo Raise and Elena Narne (Ospedale Civile Santi Giovanni e Paolo di Venezia, Italy); Antonio Poggio, Flavio Demin, and Vincenzo Mondino (Divisione di Malattie Infettive, Ospedali Riuniti di Pallanza Verbania, Italy); Giovanni Serpelloni, Marina Malena, and Oliviero Bosco (Centro di Medicina Preventiva, Sezione Screening HIV, Azienda Sanitaria Locale 20, Verona, Italy). Acknowledgments We thank Patrizio Pezzotti for performing the statistical analysis and critically reviewing the manuscript and Informa Contract Reaserach Organization for collecting and inputting data. We are also particularly grateful to Francesco Mazzotta, Gianni Di Perri, Lee Bacheler, and Giuliana Fusetti for their generous contributions to this study and their continuous support. Financial support. This study was partially supported by an unrestricted educational grant provided by Bristol-Myers Squibb, Italy. Potential conflicts of interest. N.G., G.G., and A.L. are recipients of support for research and educational programs by Glaxo-Smith Kline, Bristol-Myers Squibb, Abbott, Gilead Sciences, Roche, Boehringer-Ingelheim, Pfizer, and Virco (Johnson & Johnson). W.K. is an employee of Virology Education (Utrecht, The Netherlands). P.M. is an employee of Virco (Mechelen, Belgium). All other authors: no conflicts. References 1. Meynard JL, Vray M, Morand-Joubert L, et al. Phenotypic or genotypic resistance testing for choosing antiretroviral therapy after treatment failure: a randomized trial. AIDS 2002; 16: Haubrich RH, Kemper CA, Hellmann NS, et al. A randomized, prospective study of phenotype susceptibility testing versus standard of care to manage antiretroviral therapy: CCTG 575. AIDS 2005; 19: Durant J, Clevenbergh P, Halfon P, et al. Drug-resistance genotyping in HIV-1 therapy: the VIRADAPT randomised controlled trial. Lancet 1999; 353: Baxter JD, Mayers DL, Wentworth DN, et al. A randomized study of antiretroviral management based on plasma genotypic antiretroviral resistance testing in patients failing therapy. CPCRA 046 Study Team for the Terry Beirn Community Programs for Clinical Research on AIDS. AIDS 2000; 14:F HIV/AIDS CID 2006:42 (15 May) 1479

11 5. Cohen CJ, Hunt S, Sension M, et al. A randomized trial assessing the impact of phenotypic resistance testing on antiretroviral therapy. AIDS 2002; 16: Cingolani A, Antinori A, Rizzo MG, et al. Usefulness of monitoring HIV drug resistance and adherence in individuals failing highly active antiretroviral therapy: a randomized study (ARGENTA). AIDS 2002; 16: Tural C, Ruiz L, Holtzer C, et al. Clinical utility of HIV-1 genotyping and expert advice: the Havana trial. AIDS 2002; 16: Harrigan PR, Montaner JS, Wegner SA, et al. World-wide variation in HIV-1 phenotypic susceptibility in untreated individuals: biologically relevant values for resistance testing. AIDS 2001; 15: Torti C, Quiros-Roldan E, Keulen W, et al. Comparison between rulesbased human immunodeficiency virus type 1 genotype interpretations and real or virtual phenotype: concordance analysis and correlation with clinical outcome in heavily treated patients. J Infect Dis 2003; 188: Torti C, Quiros-Roldan E, Regazzi M, et al. A randomized controlled trial to evaluate antiretroviral salvage therapy guided by rules-based or phenotype-driven HIV-1 genotypic drug-resistance interpretation with or without concentration-controlled intervention: the Resistance and Dosage Adapted Regimens (RADAR) study. Clin Infect Dis 2005; 40: De Luca A, Cingolani A, Di Giambenedetto S, et al. Variable prediction of antiretroviral treatment outcome by different systems for interpreting genotypic human immunodeficiency virus type 1 drug resistance. J Infect Dis 2003; 187: d Arminio Monforte A, Lepri AC, Rezza G, et al. Insights into the reasons for discontinuation of the first highly active antiretroviral therapy (HAART) regimen in a cohort of antiretroviral naive patients. Italian Cohort of Antiretroviral-Naive Patients Study Group. AIDS 2000;14: Panidou ET, Trikalinos TA, Ioannidis JP. Limited benefit of antiretroviral resistance testing in treatment-experienced patients: a metaanalysis. AIDS 2004; 18: CID 2006:42 (15 May) HIV/AIDS