Despite the evidence that resistance testing can predict treatment response, many doctors are cautious about its use in the clinical setting. Most see it as a potential adjunct to existing ways of determining drug regimens and decisions about switching therapy. This caution has been shaped by early results of clinical trials of resistance testing which showed, at best, only moderate benefit in the short term. These factors, combined with the limited availability of resistance assays in the United Kingdom, have meant these tests were not widely used in the past.

Guidelines

However, the most recent guidelines issued by the British HIV Association (BHIVA) recommend the routine use of resistance testing in care of people with HIV. The EuroGuidelines HIV Resistance Group has also set out guidelines on the use of HIV resistance testing to select therapy.

While acknowledging the short-comings of the current technology, the BHIVA guidelines state that routine resistance testing should be recommended prior to therapy in chronically infected HIV patients. They also warn that doctors should be aware that treatment of naive patients may be complicated by pre-existing drug resistance. United States guidelines from the Department of Health and Human Services also recommend that resistance testing may be a useful tool in selecting a regimen for use during acute or chronic infection.

Expert opinion is clearly moving towards the routine use of resistance testing. The disparity in these expert recommendations indicates the limitations of current research, and a meta-analysis published in 2004 showed only marginal benefits for genotypic resistance testing and no benefit from phenotypic resistance testing (Panidou 2004). While treatment failure has been closely linked with the presence of resistance mutations in many studies, fewer studies have shown that people who undergo resistance testing before the selection of a drug regimen have a better response to treatment than people who do not. In addition, only short-term benefit has been established. However, studies have shown that patients who are infected with drug-resistant virus do as well as those with wild-type virus after starting treatment, if drugs are chosen on the basis of resistance testing^[1].

Genotype, phenotype or treatment history

One of the largest studies to date, the CERT study, randomised 450 individuals to change treatment on the basis of treatment history, genotypic or phenotypic testing. In the genotyping group, treatment selection based on algorithm was replaced by treatment selection based on Virtual Phenotype in 52 of 124 individuals. In 99 individuals with prior non-nucleoside reverse transcriptase inhibitor (NNRTI) experience, time to virologic failure was significantly longer in the phenotyping group when compared with the genotyping group and the treatment history group. In individuals who had taken at least four drugs prior to randomisation, both phenotyping and genotyping were associated with superior virological outcomes (Wegner 2002).

The difference between the two sub-groups may be explained by the greater accuracy of phenotyping in determining sensitivity to protease inhibitors (PIs) and nucleoside reverse transcriptase inhibitors (NRTIs) in individuals with a greater degree of treatment experience.

Phenotypic testing was not found to be useful in the NARVAL study, first reported in 2000. This study investigated the impact of changing treatment based on information from phenotypic or genotypic resistance tests or patient history. Results showed no added benefit from using the more sensitive phenotypic resistance test, and only a marginal benefit to switching treatment based on genotypic test results when compared to doctor assessment based on a patient's treatment history (Meynard 2002).

The group of 541 participants had an average viral load of about 20,000 copies/ml. This study may be regarded as an assessment of resistance testing to predict effective salvage therapy as this group was highly treatment-experienced. However, no sub-group analysis has assessed the value of phenotypic testing in NNRTI-experienced individuals (24% of participants in this study).

After three months follow-up, similar numbers of patients across the three arms had viral load below 200 copies/ml and had experienced a viral load fall of greater than 1 log₁₀. However, in the 427 patients who had completed six months of a new treatment regimen, switches based on genotypic test results had been significantly more successful than switches based on the phenotypic resistance test or on treatment history. Nevertheless, the number of those who maintained undetectable viral load after six months was disappointing in all three groups: 29% in the genotypic testing group, compared with 22% in the phenotypic testing group and 17% in the control group.

Several explanations for the comparatively poor responses have been proposed:

• Those in the control group were more likely to have received NRTIs they had not taken before when they were prescribed a new regimen. This may have been more significant than the total number of new drugs prescribed or the new protease inhibitor prescribed in determining outcome.

• Those in the control group were more likely to have their drug dosage adjusted as a result of therapeutic drug monitoring.

• Treatment switches in the control group were carried out on the basis of a full treatment history, which is often unavailable in the clinical setting.

These factors may have biased the outcome somewhat in favour of the control group, according to investigators.

The VIHRES study randomised 144 individuals who had experienced failure of at least two antiretroviral therapy regimens to undergo genotypic or phenotypic testing with a subsequent switch in therapy based on an expert panel's interpretation of this result. After 24 weeks, the proportions with viral load below 200 copies/ml did not differ between the two arms (Blanco 2002).

In summary, these studies indicate that selecting a regimen on the basis of resistance testing results in a greater viral load reduction after 16 weeks, but many people who receive a resistance test still fail to achieve undetectable viral load after switching a regimen, particularly if they have few remaining treatment options.

Genotyping or treatment history

The French VIRADAPT assessed 108 patients who were failing treatment and were allocated a new combination based on either their doctor's assessment plus clinical guidelines, or the results of genotypic resistance testing. Six months after switching treatment, the group who switched on the basis of genotypic testing had significantly lower viral loads (Clevenbergh 2000). Experts have questioned whether the superiority of the genotypic arm could be entirely attributed to genotypic testing, since more aggressive management of people in the genotypic arm of VIRADAPT could simply be appropriate to this group.

Further analysis of the VIRADAPT study demonstrated that people whose treatment was adjusted according to the results of genotypic testing did better if they also maintained optimum concentrations of their prescribed protease inhibitor during the first six months of the study (Durant 2000). These data indicate that both genotypic resistance testing and drug concentrations predict response to treatment.

CPCRA 046, also known as the GART study, recruited 153 patients who had evidence of viral load rebound after more than 16 weeks on a triple combination regimen. Patients were randomised either to receive a genotypic antiretroviral resistance testing (GART) and expert advice to guide selection of their new combination, or to switch to a new regimen selected according to the drugs previously taken. After 12 weeks the GART group had a significantly greater viral load reduction. On average, the difference between the GART and non-GART group was 0.44 log₁₀ (Baxter 2000).

The study also showed that the number of new and active drugs was a significant predictor of response, regardless of whether genotypic testing was carried out (Baxter 2002). This finding has led some experts to suggest that genotypic testing may not be necessary after the failure of first-line therapy. Instead, they argue, all the drugs should be changed and blood should be stored for genotypic testing if the second-line therapy fails. However, opponents of this view argue that a completely new regimen does not rule out the risk of cross-resistance, and point to the indifferent responses in many 'salvage' studies unless switching occurs quickly after the detection of virological rebound.

Perhaps the most telling evidence of the potential value of genotypic testing to come from the GART study was the virological response of patients whose doctors switched therapy in response to expert interpretation of the genotypic test. After 12 weeks the patients whose doctors followed this advice more than 80% of the time had viral load nearly 0.5 log lower than those whose doctors followed the advice in less than 60% of cases.

The ARGENTA study randomised 174 patients with detectable viral load despite antiretroviral therapy to switch therapy on the basis of genotyping or to receive the standard of care. Treatment experience varied, with a quarter having failed more than three drug regimens. Responses at 12 weeks showed a benefit in the group that was tested by genotype, but this was lost by 24 weeks (Cingolani 2002).

When patients were stratified by baseline viral load, genotypic testing showed sustained benefit at month six in those with viral loads below 10,000 copies/ml, but no such difference was observed at month three or six in those with baseline viral loads above 10,000 copies/ml.

Phenotyping or treatment history

Results of the VIRA 3001 study demonstrated that phenotypic resistance testing could also produce superior treatment responses in the short-term. Two hundred and seventy-two participants were randomised to phenotypic resistance testing (PRT) or standard of care (SOC) in determining a new drug combination, having failed one protease inhibitor and multiple NRTIs. After 16 weeks, significantly more of the PRT group had viral loads below 400 copies/ml compared with the SOC group. The benefit of resistance testing was greatest when individuals who commenced an NNRTI were excluded from analysis. The high drop-out rate undermines the impact of these results (Cohen 2002).

A second prospective, randomised study of phenotypic resistance testing found that short-term benefits of resistance testing were not borne out at week 16. In this study of 115 heavily pre-treated individuals, there was a significantly greater viral load reduction and CD4 increase in the group who received phenotypic testing (Melnick 2000).

However, another study CCTG 575, failed to show benefit from phenotypic testing. CCTG 575 was designed to evaluate the usefulness of phenotypic testing (using the Virologic test, PhenoSense) when changing treatment. This randomised, prospective study compared treatment response in two groups of participants. The first group (PHENO) changed their failing anti-HIV therapy with the results from a phenotypic resistance test performed while still on the failing regimen; whilst the second (SOC; standard of care) changed without a test. CCTG 575 found no difference overall between those who received a phenotypic test and those who did not in their response to their new regimen at either six or twelve months (Haubrich 2005).

However, there was a subgroup of patients in the PHENO arm who benefited from phenotyping. Of those who began with resistance to at least three PIs, PHENO participants were more likely to have viral load below 400 copies/ml at six months than SOC participants. Phenotypic testing did provide a benefit to people who had resistance to at least four PIs at baseline.

The CCTG 575 study team has proposed a series of possible explanations for the lack of benefit associated with phenotypic testing observed in their study, given that this result does not appear to fit with that of other trials. In comparison with other studies, participants in CCTG 575 were not particularly drug experienced and many had yet to use drugs from the NNRTI class. It is suggested therefore that participants in both trial groups should have had a range of viable treatment options available to them.

A further important consideration is that phenotypic tests are currently basing predictions of drug sensitivity on break-points which have not been clinically validated. While this affects all current antiretrovirals to a greater or lesser extent, it is suggested that it may be a particular problem for the NRTIs ddI (didanosine, Videx / VidexEC) and d4T (stavudine, Zerit). In CCTG 575, there was a tendency for these two drugs to be used more often in the PHENO arm than the SOC arm. If the break-points used for these drugs were inaccurate, then one would expect the PHENO arm to have under-performed. In the final published findings of this study, the authors reported that alternative NRTI sensitivity cut-offs produced more accurate recommendations about which NRTIs would be effective. This confirms the role inaccurate cut-offs for the NRTIs may have played in reducing the efficacy of phenotypic testing.

Genotype versus genotype plus phenotype

The ERA study evaluated whether any extra benefit is derived from performing a phenotypic resistance test alongside a genotypic resistance test. This study found no additional benefit to using the more expensive phenotypic test (Loveday 2003).

Expert interpretation of resistance data needed

One of the problems with implementing resistance testing in clinical practice will be ensuring correct interpretation of genotypic test results. Several methods of analysis have been used with varying degrees of success: expert analysis; computer analysis of genotypic data to produced a 'virtual phenotype'; and a set of rules known as an algorithm which doctors can use to assess genotypic results.

The HAVANA study looked at the impact of expert interpretation of genotypic test results on subsequent viral load response when a new regimen was started. The researchers randomised 326 patients failing treatment to receive one of the following:

• Genotypic resistance testing with expert interpretation of the results.

• Genotypic resistance testing, no interpretation.

• Standard of care based on medical history with advice from an expert.

• Standard of care based on medical history without expert advice.

'Expert advice' was formulated by a panel of experts, while 'standard of care' was determined by the individual judgement of the treating physician.

People who received treatment based on genotypic testing with expert interpretation had the best viral load response (Tural 2002).

Genotypic interpretation may be improved by the use of a virtual phenotype compiled from a database of known genotype / phenotype relationships, such as the one developed by Virco, manufacturer of two resistance tests. This database will calculate the probability of reduced susceptibility based on all the available information about the genotypic resistance pattern presented to it.

The reliability of virtual phenotyping has been investigated in several studies. The first was an analysis of VIRA 3001, where the actual phenotype established during the study was compared with a virtual phenotype derived from genotypic data and the Virco database. There was an 86% agreement between actual and virtual phenotypes, with major discrepancies in only 3% of cases (Cohen 2002).

Interpretation systems are proliferating as it becomes evident that clinicians will not be able to remember all the patterns of mutations which may compromise treatment response.

However, interpretation systems may provide different results depending on the data used to establish the rules, or algorithm.

A US-French collaboration considered the differences between online algorithms and their rates and causes for comparative discordance. This study aims to establish a site where algorithms can evolve and converge through ongoing inter-algorithm assessments and be validated using clinical data. The group analysed genotypic sequences from over 2230 individuals between 1997 and 2000 (Shafer 2001).

They found:

• Application of three rules-based algorithms resulted in 84% concordance in deciding whether an isolate was susceptible or resistant to a particular anti-HIV drug.

• Four drugs - amprenavir (Agenerase), abacavir (Ziagen), ddI, and d4T - were responsible for two thirds of the discordances.

• Several common mutational patterns were responsible for most of the discordances.

• Discordances between algorithms were less marked when SIR (sensitive, intermediate, resistant) scoring was used compared with SR scoring.

A comparison of four algorithms - Stanford, geno2pheno, Retrogram 1.4 and TruGene - used a scoring system to assess the match between predictions of sensitivity and subsequent virological response in 131 consecutive patients. The study found that between 66 and 77% of responses to drugs were predicted correctly, and between 67 and 75% of therapeutic failures were also predicted correctly (Ehret 2002).

See Resistance in non-B HIV sub-types in Anti-HIV therapy: Resistance for further information on the performance of assays in different HIV subtypes.

The GUESS study compared the ability of 12 experts to predict phenotypic resistance from genotypes by taking 50 genotypes from the Virco database. Large variations between drugs were seen. While 74% predicted 3TC sensitivity correctly, only 25 and 26% predicted abacavir and nelfinavir resistance correctly. The consensus on treatment recommendations was lowest in the presence of ddI resistance, and generally poorer for NRTIs (Zolopa 2002).

Yet another study sought to establish the correlation between protease and reverse transcriptase mutations and antiviral drug resistance. The study also aims to generate models that predict phenotypic drug resistance from sequence information. Their geno2pheno approach (available free online at http://cartan.gmd.de/geno2pheno.html) models permutations of identified mutants and directs them along decision-tree classifiers that identify them as resistant or susceptible. These pathways provide a sophisticated analysis based on the juxtaposition of different mutations. For example, they cite a mutational pair, 41M and 77L as resistant, but when these are combined with 215T and 75T/V, they report the same virus as being susceptible.

The use of an algorithm to interpret genotypic resistance results has been associated with a superior virological response. After three months of salvage therapy, patients whose resistance results were assessed using a rules-based algorithm were more likely to have a viral load below 500 copies/ml, even after controlling for other factors such as number of new drugs and baseline viral load (van Laethem 2002).

Virtual phenotype vs phenotype

A randomised study in Spain showed that a similar number of drugs defined as active were prescribed to 260 patients who received phenotypic or virtual phenotype testing, but by as treated analysis (which excludes patients who may have dropped out of the study due to virologic failure or adverse events), significantly more of those in the virtual phenotype group had viral load below 400 copies at week 24 (Perez-Elias 2004).

The GenPherex study, conducted in Italy, demonstrated equivalence between phenotypic and virtual phenotype testing in 106 patients randomised to switch therapy on the basis of resistance testing. Participants had been exposed to an average of 9 drugs. Approximately 60% of patients in each arm were sensitive to at least three drugs in the new regimen, and after 12 months similar proportions in each arm had viral load below 400 copies/ml (Mazzotta 2002).

The open-label PhenGen study involving 303 people has shown that virtual phenotype and phenotype were equivalent. After six months of treatment, 55% of the virtual phenotype group had undetectable viral load compared to 53% of the phenotype group and mean reductions in viral load were nearly identical (Aracino 2004).

Whilst the Virtual Phenotype lags behind other resistance testing systems in terms of evaluation of its use in patient management, it is already in use in limited settings, such as the Chelsea and Westminster Hospital, London.

Virtual phenotyping was also compared with rules-based algorithms in a retrospective analysis of ACTG 372B/D. When the data was controlled for baseline viral load and number of new drugs received, the virtual phenotype at baseline was significantly more predictive of virological failure than rules based interpretation of genotypic data (Hammer 1999).

The need for more sophisticated analyses

As the divergent results of interpretations studies show, resistance testing is still a relatively crude method of selecting treatment. In order to select new regimens more accurately, clinical studies will need to define the following for each drug that is available:

• A genotype and phenotype correlation from clinical isolates, in order to establish the degree of susceptibility to new agents.

• A correlation of genotype and virological outcome, in order to establish the degree of response that can be expected.

Clinicians might also want to know about the correlation between genotypic resistance and treatment history.

The first type of analysis will need to be carried out with large databases, and has already been addressed to come extent by Virco's Virtual Phenotype database. This has also been validated by using the virtual phenotype to predict response in clinical trials. See Testing for resistance in Anti-HIV therapy: Resistance above.

The second type of analysis, which is critically dependent on establishing a genotype / phenotype relationship, also requires a large database. Some companies such as Abbott have carried out analyses to correlate genotypic and phenotypic resistance with virologic response, but this is not an evolving database and cannot accommodate such factors as non-B subtypes or the impact of new mutational patterns contributed by new drugs.

As interpretation becomes more complex, it raises the prospect that information regarding resistance will become a proprietary product, and that clinics which can pay for access to interpretation systems will have better outcomes than clinics where budgets are restricted.

Phenotypic or genotypic susceptibility score

Several studies have now reported baseline susceptibility and responses using a phenotypic or genotypic susceptibility score (PSS or GSS). This measure assigns a numerical value to a drug based on the degree of susceptibility shown to it. It allows susceptibility to be expressed in terms of partial or full susceptibility, thus placing value on the contribution of drugs to which an individual might be partially susceptible.

An analysis of the ACTG 364 study showed that using a continuous susceptibility score rather than a binary score more accurately predicted treatment outcome for up to three years after changing treatment on the basis of a phenotypic resistance test (Katzenstein 2003).

Detecting rare but significant mutations

The case for resistance testing prior to starting anti-HIV therapy in primary infection was strengthened by data presented at the Fifth Resistance Workshop held in June 2001.

Genotypic analysis of 603 recently infected people, diagnosed between 1997 and 1999 found a 3% prevalence of HIV variants with unusual mutations at position 215 in the reverse transcriptase gene. Whilst the T215Y mutation is well established as conferring a high level of resistance to AZT, other mutations, at this position, including 215C, S and D do not cause resistance to AZT, but have been shown previously to increase HIVs replicative capacity. In the presence of AZT, these mutants revert to T215Y in preference to wild-type. Infection with these unusual 215 variants therefore has important implications for choice of therapy.

In seven individuals infected with either 215C or D, who subsequently began anti-HIV treatment with an AZT-containing regimen, reversion to the AZT mutant T215Y occurred in an average of 25 or 31 days respectively. For comparison, three people infected with wild-type who later developed T215Y did so in an average of 63 days (Garcia-Lerma 2001).

Selecting drugs: using therapeutic drug monitoring with resistance testing

The next step in the refinement of resistance testing will be to marry the results of genotypic resistance tests which can establish a virtual phenotype with the results of therapeutic drug monitoring (TDM), which can establish whether adequate drug levels are being achieved when supposedly 'active' drugs are added to a new regimen.

An analysis of plasma drug levels in the GART or CPCRA 046 study was carried out. This study did not sample plasma drug levels at standardised times after dosing. The study sought to correlate plasma drug levels at week 12 with viral load responses, and classified individuals as having drug levels above or below the median. Each drug in the regimen which had a plasma level above the median was associated with a viral load reduction of 0.40 log₁₀ (the same level as seen for each genotypically active agent), and the viral load reduction increased with each genotypically active agent used which had a plasma level above the median (Baxter 2002). When four or five drugs were used which achieved above-average concentrations, the median viral load reduction was 1.44 log₁₀.

The way in which this study was designed makes it difficult to determine whether the very small additional benefit of having above average drug levels is worth pursuing. Because this was a short study, we do not know whether people with above average drug levels would be more likely to discontinue or switch therapy due to side effects related to high drug levels. Also, the method used to sample drug levels make it difficult to interpret.

The GENOPHAR study, in which 137 people with treatment failure were randomised to change treatment on the basis of either genotypic resistance testing or genotypic resistance testing coupled with the results of therapeutic drug monitoring for protease inhibitors and NNRTIs at weeks 4 and 8. After 24 weeks, 58% of the genotypic group and 66% of the TDM group had viral load below 200 copies/ml, and the authors concluded that TDM did not add anything to the selection of a new regimen on the basis of genotypic resistance testing (Bossi 2002).

A study of the genotypic inhibitory quotient in 49 people commencing ritonavir (Norvir)-boosted amprenavir treatment found that the ratio of amprenavir C_min at week eight to the number of protease mutations was a better predictor of virological response at week 12 than either genotype or plasma drug level (Marcelin 2003).

Finally, a Spanish study of 139 PI-experienced patients who commenced saquinavir/ritonavir found that genotypic inhibitory quotient was most strongly predictive of virological response after 48 weeks, when compared with either genotype or drug levels alone (Soriano 2003).

Identifying resistance to a single drug

Data from a number of studies such as ACTG 343 and Trilege indicated that viral rebound can occur when there are resistance mutations to only one drug in a combination. These findings have led to renewed interest in modifying failing combinations rather than switching to a completely new regimen. Resistance testing may have a role in determining which drug has become ineffective due to resistance during the early stages of viral rebound.

For more details of ACTG 343 and Trilege, see Induction and maintenance therapy in Anti-HIV therapy: Choosing your treatment strategy.

References

Baxter JD 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 14(9):F83-93, 2000.

Baxter JD et al. Both baseline HIV-1 drug resistance and antiretroviral drug levels are associated with short-term virologic responses to salvage therapy. AIDS 16(8):1131-1138, 2002.

Blanco JL et al. A prospective randomized study on the usefulness of genotypic resistance tests versus real phenotypic resistance tests in heavily pretreated patients with virological failure (VIHRES Study). XIV International AIDS Conference, Barcelona, abstract TuPeB4624, 2002.

Bossi P et al. GENOPHAR: a randomised study of plasmatic drug measurements associated with genotypic resistance testing in patients failing antiretroviral therapy. Ninth Conference on Retroviruses and Opportunistic Infections, Seattle, abstract 585, 2002.

Cingolani A et al. Usefulness of monitoring HIV drug resistance and adherence in individuals failing highly active antiretroviral therapy: a randomised study (ARGENTA). AIDS 16(3):369-379, 2002.

Clevenburgh P et al. Persisting long-term benefit of genotype-guided treatment for HIV-infected patients failing HAART. The Viradapt Study: week 48 follow-up. Antiviral Therapy 5(1):65-70, 2000.

Cohen C et al. A randomized trial assessing the impact of phenotypic resistance testing on antiretroviral therapy. AIDS 16(4):579-588, 2002.

De Luca A et al. A prospective, randomised study on the usefulness of genotypic resistance testing and the assessment of patient-reported adherence in unselected patients failing potent anti-HIV therapy (ARGENTA). Eighth Annual Conference on Retroviruses and Opportunistic Infections, abstract 433, 2001.

Durant J et al. Drug-resistance genotyping in HIV-1 therapy: the VIRADAPT randomised controlled trial. Lancet 353(9171):2195-2199, 1999.

Durant J et al. Importance of protease inhibitor plasma levels in HIV-infected patients treated with genotypic-guided therapy: pharmacological data from the Viradapt Study. AIDS 14(10):1333-1339, 2000.

EuroGuidelines Group for HIV Resistance. AIDS 15:309-320, 2001.

Ehret R et al. Predictive value of different drug resistance interpretation systems in therapy management of HIV-infected patients in daily routine. Antiviral Therapy 7: S77, 2002.

Emery S et al. CREST - A randomised comparison of two resistance test platforms: genotype and genotype plus Virtual Phenotype: interim results. Antivir Ther 7: S116, 2002.

Fournier S et al. Factors associated with virological response in HIV-infected patients failing antiretroviral therapy: a prospective cohort study. HIV Med 6: 129-134, 2005.

Garcia-Lerma JG et al. Increased ability for selection of zidovudine resistance in a distinct class of wild-type HIV-1 from drug-naive persons. Proc Natl Acad Sci U S A 98: 13907-13912, 2001.

Graham N et al. The virtual phenotype is an independent predictor of clinical response. Eighth Conference on Retroviruses and Opportunistic Infections, abstract 524, 2001.

Hammer S et al. Relationship of phenotypic and genotypic resistance profiles to virological outcome in a trial of abacavir, nelfinavir, efavirenz and adefovir dipivoxil in patients with virological failure receiving indinavir (ACTG 372). Antivir Ther 4: S45, 1999.

Haubrich R et al. A randomized, prospective study of phenotype susceptibility testing versus standard of care to manage antiretroviral therapy: CCTG 575. AIDS 19: 295-302, 2005.

Katzenstein DA et al. Phenotypic susceptibility and virological outcome in nucleoside-experienced patients receiving three or four antiretroviral drugs. AIDS 17: 821-830, 2003.

Kemper CA et al. Sequencing of protease inhibitor therapy: insights from an analysis of HIV phenotypic resistance in patients failing protease inhibitors. AIDS 15: 609-615, 2001.

Lanier ER et al. Predicting abacavir antiviral activity using HIV-1 reverse transcriptase genotype: a comparison of 12 algorithms. Antiviral Therapy 6(1): 103-4, 2001.

Loveday C et al. A randomized controlled trial of phenotypic resistance testing in addition to genotypic resistance testing: the ERA trial. Antiviral Therapy 8 (suppl 1): S188, 2003.

Marcelin AG et al. Genotypic inhibitory quotient as predictor of virological response to ritonavir-amprenavir in human immunodeficiency virus type 1 protease inhibitor-experienced patients. Antimicrobial Agents and Chemotherapy 47(2):594-600, 2003.

Mazzotta F et al. Real versus virtual phenotype to guide treatment in heavily pre-treated patients: 48 week follow-up of Genotipo-Fenotipo di Resistenzia (GenPheRex) trial. JAIDS 32: 268-280, 2003.

Melnick D et al. Impact of phenotypic antiretroviral drug resistance testing on the response to salvage antiretroviral therapy (ART) in heavily experienced patients. Seventh Conference on Retroviruses and Opportunistic Infections, San Francisco, abstract 786, 2000.

Meynard JL et al. Phenotypic or genotypic resistance testing for choosing antiretroviral therapy after treatment failure. AIDS 16: 727-736, 2002.

Panidou ET et al. Limited benefit of antiretroviral resistance testing in treatment-experienced patients: a meta-analysis. AIDS 18: 2153-2161, 2004.

Perez-Elias MJ et al. Realvirfen Study Group. Phenotype versus virtual phenotype for choosing antiretroviral therapy after failure: a prospective, randomized study. Antivir Ther 8: 577-584, 2004.

Pillay D et al. Evidence of secondary transmission of HIV-1 with position 215 variants within reverse transcriptase. Antivir Ther 6: S114, 2001.

Saracino A et al. Selection of antiretroviral therapy guided by genotypic or phenotypic resistance testing: an open-label, randomized, multicenter study (PhenGen). J Acquir Immune Defic Syndr 37: 1587-1598, 2004.

Shafer RW. Online comparison of HIV-1 drug resistance algorithms identifies rates and causes of discordant interpretations. Antivir Ther 6: 101-102, 2001.

Soriano V et al. Predictive value of drug levels, HIV genotyping and genotypic inhibitory quotient (GIQ) at different time points along 48 weeks using a saquinavir/ritonavir salvage therapy. Forty Third Interscience Conference on Antimicrobial Agents and Chemotherapy, Chicago, abstract H-1999, 2003.

Tural C et al. Clinical utility of HIV-1 genotyping and expert advice: the Havana trial. AIDS 16: 209-218, 2002.

Van Laethem K et al. A genotypic drug resistance interpretation algorithm that significantly predicts therapy response in HIV-1-infected patients. Antivir Ther 7: 123-129, 2002.

Vray M et al. Multivariate logistic regression analysis of factors predictive of the virological response in the NARVAL trial. Antiviral Therapy 7: S145, 2002.

Wegner S et al. Long-term clinical efficacy of resistance testing: results of the CERT trial. Antiviral Therapy 7: S170, 2002.

Wensing AM et al. Analysis of the worldwide evaluation study on HIV-1 genotype interpretation: ENVA-3. Antiviral Therapy 6 (1): 101, 2001.

Zolopa AR et al. Accuracy, precision and consistency of expert HIV-1 genotype interpretation: an international comparison (the GUESS study). Antiviral Therapy 7: S97, 2002.