Clinically evaluating genotypic interpretation systems is essential to provide optimal guidance in designing potent individualized HIV-regimens. This study aimed at investigating the ability of the latest Rega algorithm to predict virological response on a short and longer period.
Materials & Methods
9231 treatment changes episodes were extracted from an integrated patient database. The virological response after 8, 24 and 48 weeks was dichotomized to success and failure. Success was defined as a viral load below 50 copies/ml or alternatively, a 2 log decrease from the baseline viral load at 8 weeks. The predictive ability of Rega version 8 was analysed in comparison with that of previous evaluated version Rega 5 and two other algorithms (ANRS v2011.05 and Stanford HIVdb v6.0.11). A logistic model based on the genotypic susceptibility score was used to predict virological response, and additionally, confounding factors were added to the model. Performance of the models was compared using the area under the ROC curve (AUC) and a Wilcoxon signed-rank test.
Per unit increase of the GSS reported by Rega 8, the odds on having a successful therapy response on week 8 increased significantly by 81% (OR = 1.81, CI = [1.76–1.86]), on week 24 by 73% (OR = 1.73, CI = [1.69–1.78]) and on week 48 by 85% (OR = 1.85, CI = [1.80–1.91]). No significant differences in AUC were found between the performance of Rega 8 and Rega 5, ANRS v2011.05 and Stanford HIVdb v6.0.11, however Rega 8 had the highest sensitivity: 76.9%, 76.5% and 77.2% on 8, 24 and 48 weeks respectively. Inclusion of additional factors increased the performance significantly.
Citation: Vercauteren J, Beheydt G, Prosperi M, Libin P, Imbrechts S, Camacho R, et al. (2013) Clinical Evaluation of Rega 8: An Updated Genotypic Interpretation System That Significantly Predicts HIV-Therapy Response. PLoS ONE 8(4): e61436. https://doi.org/10.1371/journal.pone.0061436
Editor: Nicolas Sluis-Cremer, University of Pittsburgh, United States of America
Received: March 16, 2012; Accepted: March 14, 2013; Published: April 17, 2013
Copyright: © 2013 Vercauteren et al. This is an open-access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.
Funding: The research leading to the results has received funding from the European Community’s Seventh Framework Programme (FP7/2007–2013) under the project “Collaborative HIV and Anti-HIV Drug Resistance Network (CHAIN)” - grant agreement n° 223131 and was also supported by the Fonds Wetenschappelijk Onderzoek (FWO) grant G.0611.09 and a grant from the Interuniversity Attraction Poles Programme, Belgian State, Belgian Science Policy (IAP-VI P6/41). MP from the University of Manchester’s Health Research Center (HeRC) was funded by the Medical Research Council (MRC) grant MR/K006665/1. The data used for this study originated from several cohorts and hospitals: EuResist (IST-2004-027173; www.euresist.org), Virolab (IST-FP6-027446; www.virolab.org), University Hospitals Leuven (Leuven, Belgium), Chaim Sheba Medical Center (Tel-Hashomer, Israel) and Hospital Egas Moniz (Lisbon, Portugal). GB was funded by a Ph.D grant of the Institute for the Promotion of Innovation through Sciences and Technology in Flanders (IWT). The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.
Competing interests: Mybiodata was consulted for the data management and extraction. Mybiodata had no influence on the analysis of the data nor the results and interpretation. Mybiodata has no patents nor products under development or already marketed in the field of the paper; there are no competing interests among this private company and the PLOS ONE policy and Mybiodata presence among the authors does not alter their adherence to all the PLOS ONE policies on sharing data and materials.
Since the advent of Highly Active Anti Retroviral Therapy (HAART), morbidity and mortality related with HIV/AIDS have considerably decreased in the Western world . Clinicians can now compose several efficient combination regimens using 25 approved drugs . Nevertheless, in many patients, not all options can be used, due to intolerance or side effects for certain drugs and because of the presence of antiviral drug (cross-)resistance . Extended cross-resistance has been decreasing over calendar year and drug developers have been encouraged to focus their research on new potent drugs with a better tolerability, ease of use and less toxicity . Nevertheless, resistance is and will continue to be an important issue in the management of HIV. Correct interpretation of the mutational patterns is however not straightforward, and unfortunately there is no consensus on this matter yet. Several genotypic interpretation systems have proven to significantly predict virologic response in retrospective analyses – and are therefore mentioned in treatment and resistance guidelines . However, it remains a challenge to keep those interpretation systems up-to-date and improve their usefulness for clinicians treating HIV-infected patients . New knowledge on resistance related mutations is accumulating and new drugs are still being implemented in clinical practice. Thus, guidelines stress the fact that regular updating and proper clinical evaluation of interpretation algorithms is needed .
The Rega algorithm was initiated in January 2000 and the current version Rega 8 dates from June 2009. Previous versions have been retrospectively evaluated with focus on short term viral response in HIV-1 patients (3 months) , –. Here the results of the clinical evaluation of Rega 8 for prediction of virologic response on short (8 weeks), mid-long (24 weeks) and long (48 weeks) term based on a large clinical database are presented, including a comparison with the HIVdb v6.0.11 and ANRS 2011.05 algorithms and a previous version of the Rega algorithm (Rega 5).
Materials and Methods
An integrated database was set up using data originating from different countries: Belgium, Germany, Israel, Italy, Luxembourg, Portugal, Spain and Sweden, in collaboration with the EuResist consortium. This data didn’t serve as base for the construction of the rules-based interpretation systems discussed in this paper. RegaDB was used to manage and analyse the data in the integrated database , from which treatment change episodes (TCEs) were extracted. A TCE was defined as the start of a first line or follow-up therapy with the corresponding baseline variables and follow-up viral load measurements. A baseline PR-RT genotype and viral load measurement was required between 90 days before and one week after the start of the new therapy and at least one follow-up viral load measurement. Virological outcome of each TCE was assessed at 8, 24 and 48 weeks based on the latest viral load measurement. At 8 weeks, we defined virological success as the achievement of a viral load less than 50 copies/ml or a decrease from the baseline viral load by two or more Logs. At 24 and 48 weeks, the virological success was defined as the achievement of a viral load less than 50 copies/ml. No restrictions on therapies were contemplated, i.e. suboptimal treatment regimens made of less than three drugs were allowed. However TCEs containing inhibitors against other viral proteins than protease or reverse transcriptase were excluded because only PR-RT sequences were available. The following compounds, all currently approved by the Food and Drug Administration (FDA) (http://www.fda.gov/oashi/aids/virals.html) and European Medicines Agency (EMEA) (http://www.emea.europa.eu) were considered: nucleotide/side reverse transcriptase inhibitors (N(t)RTIs): zidovudine (AZT), stavudine (D4T), zalcitabine (DDC), abacavir (ABC), lamivudine (3TC), emtricitabine (FTC), didanosine (DDI), tenofovir (TDF); non-nucleoside reverse transcriptase inhibitors (NNRTIs): efavirenz (EFV), nevirapine (NVP), etravirine (ETR); protease inhibitors (PIs): amprenavir (APV), fosamprenavir (FPV), atazanavir (ATV), indinavir (IDV), lopinavir (LPV/r), nelfinavir (NFV), saquinavir (SQV), tipranavir (TPV), darunavir (DRV), along with boosting ritonavir (RTV). If multiple TCEs were available for a single patient, one was randomly selected to ensure a single TCE per patient. Based on previous reports , , potential confounding factors were included, if data on them was available: age, gender, risk group, baseline viral load and CD4 count, year of therapy start, introduction of a new drug class in the regimen, number of previous therapy switches and information on drug class experience (defined as more than 1 year on NRTIs, NNRTIs or PIs, respectively).
The viral genotypes were subtyped using version 2 of the Rega HIV-1 subtyping tool  and resistance interpretation was performed by Rega 5 (version 5.5; December 2001) and Rega 8 (version 8.0.2; June 2009) , version 6.0.11 (March 2011) of the Stanford HIVdb algorithm  and version 20 (May 2011) of the ANRS algorithm . These algorithms (available as supplementary material) were used to assign a resistance score to each drug in the administered regimen. The Rega 5 algorithm and the ANRS algorithm consider 3 levels of resistance (corresponding with scores 0, 0.5, 1). The Stanford HIVdb algorithm considers 5 levels (0, 0.25, 0.5, 0.75 and 1). In the Rega 8 algorithm, the resistance scores are weighted based on the estimated potency and genetic barrier of the corresponding drug or drug class. The scores of boosted PIs are changed to 0, 0.75 and 1.5 and those for NNRTIs (except etravirine) become 0, 0.25 and 1 for resistant, intermediate resistant and susceptible respectively. Subsequently, the arithmetic sum of the scores for all drugs in the regimen was calculated to achieve the genotypic susceptibility score (GSS) of the regimen. Rules are removed from the algorithm when a drug is no longer used in clinical practice and new rules are implemented as soon as drug resistance mutations are known for a newly available drug. In order to compare Rega 5 and 8 on the same dataset, rules for an absent drug were copied from the other algorithm. Rules for boosted and un-boosted PIs were interchanged after correcting the weights. Rules for FTC and 3TC and for APV and its pro-drug FPV were interchanged. Full versions of the Rega algorithm can be found on the website of the KU Leuven http://regaweb.med.kuleuven.be/software/rega_algorithm/.
As the primary objective, the performance of Rega 8 was checked by univariate and multivariate logistic regression on the full dataset stratified for the three follow-up times and the odds ratio (OR), 95% confidence interval (CI) and p-values are reported. As a secondary objective, a comparison of Rega 8 with Rega 5, ANRS and Stanford HIVdb was performed. The individual performances were compared by means of a Receiver Operation Characteristic (ROC) analysis and Area Under the ROC Curve (AUC) was adopted by using 10-fold cross-validation. AUCs were then compared by using a Wilcoxon signed-rank test . P-values were corrected for multiple testing using the Benjamini-Hochberg procedure. Additionally, the sensitivity was investigated for each individual algorithm using the GSS cut-off of 3 as suggested by recent guidelines . Sensitivity (true positive rate) was defined as the ratio of the number of TCEs with a GSS of at least 3 and with virological success over the total number of virological successes. Specificity (true negative rate) was the ratio of the number of TCEs with a GSS less than the GSS cut-off and with virological failure over the total number of virological failures.
9231 TCEs were extracted from the integrated database (Table 1). Median patient age was 40 years. 71% of patients were men, 35% were heterosexual, 26% were men who have sex with men and 20% were intravenous drug users. Most patients were infected with a subtype B virus (72%). Other prevalent subtypes were subtype A (6%), subtype C (6%), CRF 02_AG (3%), subtype G (2%) and subtype F (2%). The ethnicity was unknown for most patients (69%). At baseline, the median CD4 cell count was 232 cells/mm3 and mean HIV-RNA load (viral load) 5.23 log10 copies/ml. 90% of the TCEs had a follow-up viral load measurement at 8 weeks, 93% at 24 and 82% at 48 weeks. The success rate dropped from 60% at 8 weeks to 50% at 24 weeks and to 47% at 48 weeks. The start date of the TCEs ranged from 1994 to 2010. 62% of the TCEs started for the first time with an NRTI, 8% with an NNRTI and 38% with a PI. The most prevalent drug administrated in the TCEs was 3TC, followed by TDF, LPV/r and AZT (Figure 1). At the start of the new regimen, experience for more than 1 year with an NRTI occurred in 53%, with an NNRTI in 23% and with a PI in 39% of the patients. The mean GSS reported by the algorithms ranged from 2.40 to 2.80 with standard deviations between 0.78 and 0.93 (Figure 2).
Nucleotide/side reverse transcriptase inhibitors (N(t)RTIs): lamivudine (3TC), tenofovir (TDF), zidovudine (AZT), emtricitabine (FTC), didanosine (DDI), abacavir (ABC), stavudine (D4T), zalcitabine (DDC); non-nucleoside reverse transcriptase inhibitors (NNRTIs): efavirenz (EFV), nevirapine (NVP), etravirine (ETR); protease inhibitors (PIs): lopinavir (LPV/r), atazanavir (ATV), nelfinavir (NFV), saquinavir (SQV), indinavir (IDV), fosamprenavir (FPV), darunavir (DRV), amprenavir (APV), tipranavir (TPV), along with boosting ritonavir (RTV).
Performance of Rega 8
The odds ratios, 95% confidence intervals and p-values of the logistic regression models were determined for 8, 24 and 48 weeks response using the univariate approach (Table 2). The GSS as reported by Rega 8 was predictive for the virological success at all time-points. Per unit increase of the GSS, the odds on having a successful therapy response on week 8 increased by 81% (OR = 1.81, CI = [1.76–1.86], ), on week 24 by 73% (OR = 1.73, CI = [1.69–1.78], ) and on week 48 by 85% (OR = 1.85, CI = [1.80–1.91], ) (Table 2). The ROC analysis of the 10-fold cross-validation showed that the inclusion of all available covariates significantly increased the performance of Rega 8 on all time points (, and for week 8, 24 and 48 respectively). Also ANRS and HIVdb improved significantly when adding covariates to the model (Figure 3 for 24 weeks results).
Comparison with Rega 5, ANRS and Stanford HIVdb Algorithms
Rega 8 and Rega 5 performed equally well in predicting virological success on all time-points as the AUC was not statistically different: , and for week 8, 24 and 48 respectively (Table 2). However, Rega 8 had a higher sensitivity than Rega 5∶76.9% versus 67.4%, 76.5% versus 66.6% and 77.2% versus 67.4% on 8, 24 and 48 weeks respectively (Table 3). Without applying weights to the rules in Rega 8, the sensitivity decreased to 73.4%, 72.5% and 73.2% on 8, 24 and 48 weeks respectively.
The performance of Rega 8 was further compared to those of the ANRS and Stanford HIVdb algorithms. No significant differences in AUC were found after correcting for multiple testing (Table 2). Both ANRS and Stanford HIVdb had a lower sensitivity than Rega 8: respectively 74.5% and 65.8% compared to 76.9% on 8 weeks, 74.1% and 64.1% compared to 76.5% on 24 weeks and 74.8% and 64.9% compared to 77.2% on 48 weeks.
In this study the Rega HIV drug resistance interpretation algorithm version 8 was retrospectively evaluated and compared with other systems using in vivo data from 9231 patients. We applied the algorithm to the baseline genotype and evaluated the association with the follow-up virologic response. Virological suppression means undetectable viral load but this threshold may differ according to the period in which measurements were done, with the newer viral load assays having a cut-off of 50 copies/ml. Excluding all measurements with a cut-off different from 50 copies/ml was difficult as we had a large dataset from various countries and laboratories that updated their assays at different times. However, we included the start date of therapy in the multivariate analysis to correct (partly) for those differences in assays.
The studied algorithms, Rega, ANRS and HIVdb, were able to significantly predict virological response of the TCEs. On average, per unit increase of the genotypic susceptibility score of the administrated regimen, the odds on having a successful therapy response almost doubled. Inclusion of treatment history and baseline patient- and viral derived characteristics enhanced the predictive value of the model. However these factors are not always available to the clinician nor are they automatically taken into account by the online available interpretation systems.
The predictive value reported here can only be a combined evaluation of the rules for those drugs that are present in all algorithms. Rega 5 contains rules for six NRTIs (AZT, DDC, DDI, 3TC, D4T, ABC), one NtRTI (TDF), three NNRTIs (NVP, DLV, EFV) and six PIs (RTV, IDV, SQV, NFV, APV, LPV/r). Rega 8 differs from Rega 5 in respect to drugs included: one NRTI (FTC), one NNRTI (ETV), four boosted PIs (FPV(r), ATV(r), TPVr and DRVr), two entry inhibitors (T20 and MVC) and one integrase inhibitor (RAL) are in Rega 8 and not in Rega 5. DDC is no longer in Rega 8 (discontinued since 2006), whereas the rules for APV are replaced by the ones for its prodrug FPV(r). Furthermore, existing rules are updated, a score-based system for PIs is implemented instead of logical rules and drug weighting factors are incorportated. Finally, Rega 8 has specific rules for the interpretation of HIV-2 drug resistance (which were not evaluated in this study). We saw that these differences between Rega 8 and 5 didn’t influence significantly the performance of the Rega algorithm in predicting therapy outcome. Moreover, Rega 8, ANRS and HIVdb reached the same performance on all time-points and using the univariate as well as the multivariate approach. In this discussion, we explore some of the possible reasons for this.
Our paper confirms the results of colleagues Frentz et al.  to that extent that the three commonly used HIV drug resistance interpretation systems - ANRS, Rega and HIVdb - predict virological response to the same extent. However, our paper evaluates more recent versions of the algorithms that are currently being used in clinical practice. Moreover it compares the latest Rega algorithm (version 8) with the previous evaluated version 5 to check consistency and improvement. Finally, it is for the first time that Rega, ANRS and Stanford HIVdb algorithms were evaluated on such an extensive dataset: 9231 TCEs from patients from several European countries. However, we have to acknowledge that the dataset still has its limitations. Despite the fact that the Rega algorithm is set up to be used for all HIV subtypes, our dataset contained 72% subtype B strains (Table 1), while the other subtypes were underrepresented . Additionally, there is a bias present in the administered drugs: NRTIs that are commonly used to form the backbone regimen are highly prevalent in the dataset (e.g. 3TC was part of the regimen in almost half of the the cases), while more recently approved drugs (e.g. ETR) are underrepresented (Figure 1). Moreover, the prevalence of resistance mutations in the dataset follows a similar distribution with NRTI resistance being more prevalent than NNRTI or PI resistance (Table 1). This means that studies investigating the performance of interpretation systems specifically for recently approved drugs, less prevalent mutations and non-B subtypes are needed . Moreover, only PR or RT genotypes were available so drugs targeting other regions had to be excluded from the dataset and the corresponding rules were thus not evaluated in this analysis.
In agreement with other reports ,  we noticed that the predictive value of the weighted rules in Rega 8 was higher compared to the unweighted rules in Rega 5, however not significant. This may in part be due to the fact that NNRTIs and PIs, for which weighting factors were introduced, are less represented in our clinical data compared to the N(t)RTIs (Figure 1). This result corresponds with that of another study  in which the introduction of weighting did not change the GSS to a great extent.
Rega 8 had the highest sensitivity with respect to the other studied algorithms, although its specificity was the lowest (Table 3). This shift from specificity to sensitivity was a deliberate choice to extend the application of the Rega algorithm. Initially its purpose was solely to score resistance, so the previous versions of the Rega algorithm were heavily taking into account the genetic barrier to resistance associated with a large number of minor resistance mutations. In the latest version of Rega this has less effect. As a result, the focus of the algorithm shifted from predicting resistance and consequential therapy failure to predicting virological success in order to find an active, potent regimen. This change makes the Rega algorithm more in alignment with the other algorithms. How sensitivity and specificity should be balanced depends on the number of antiviral drugs that are still active. Therefore the Rega algorithm recommends different GSS cut-offs depending on the patient characteristics: in case of therapy-naïve persons with no indications of transmitted drug resistance and in case of therapy-experienced persons, in case of therapy-naïve persons with indications of transmitted drug resistance and in case of therapy-experienced persons with limited treatment options. A higher risk for virologic failure and resistance development at mid-term can be expected when therapy changes with GSS below the target GSS of 3 are installed. Using this cut-off, Rega 8 reached a sensitivity score of 77.2% and a specificity score of 45.8% at 48 weeks follow-up, despite the fact that confounding factors like adherence, resistance- and treatment history, minor variants, baseline viral load and CD4 cell count were not taken into account. Kaplan-Meier analysis that was performed on the time to virological failure stratified by the GSS reported by Rega 8 showed that the Kaplan-Meier curves differed significantly between the group with a GSS of 2 and the group with a GSS of 3 () (results not shown). This supports the clinical relevancy of the Rega algorithm, despite the relatively small differences that were observed in comparison with the previously evaluated version. Regimens with 3 active drugs are required for long-term virologic response and are therefore currently recommended. However, virologic success with less potent regimens can be observed at the short- and mid-term time-points that were investigated in this analysis. As was in the pre-HAART era, when bi-therapies were prescribed , , some patients in our study still succeeded therapy at 48 weeks (22.8%, 116/507) although their GSS was 2.5 at the most. In contrast, regimens with 3 drugs lead to 48.5% (3430/7069) virological successes at 48 weeks. Three active drugs or a target GSS of at least 3 doubles the rate of success in the dataset. However, because genotypic interpretation systems have been used in clinical practice since 10 years, there is a bias towards a target GSS of 3 in the dataset (Figure 2). Therefore, not enough data is available in our to suggest that a target GSS of 2.5 would give similar success rates.
A previous report  that had as primary endpoint to evaluate the performance of the Rega algorithm (version 5.5) described the predictive value at 3 months. Viral load changes at that time point reflect more the activity of the treatment on the predominant virus population detected by baseline genotyping. However, clinicians are more interested in the prediction of longer-term virologic response (in our study up to 48 weeks), but then the outcome can be more influenced by the genetic barrier to resistance. We corrected (partly) for this bias by including information on treatment history as a surrogate for expected HIV drug resistance when calculating the performance of the algorithm.
In conclusion, we convincingly demonstrated that Rega 8 is significantly predictive for virological response and that additional variables should be taken into account to ensure an effective regimen. There was no distinguisable trend in the performance of the consecutive versions of the Rega algorithm, nor were there significant differences in AUC between Rega, ANRS and HIVdb. The clinical dataset used in this study may not have the power to show the gain of the sometimes minor improvements to excisting rules, of the specific rules for divergent subtypes, or of the rules for new drugs. Nevertheless, the reported analyses need to be done and published to give confidence to the users of the algorithms.
Conceived and designed the experiments: JV GB MP PL AMV KVL. Performed the experiments: JV GB PL SI. Analyzed the data: JV GB. Contributed reagents/materials/analysis tools: MP RC BC ADL ZG RK AS CT EVW JCS MZ AMG. Wrote the paper: JV GB ZG AMV KVL.
- 1. Mocroft A, Vella S, Benfield TL, Chiesi A, Miller V, et al. (1998) Changing patterns of mortality across Europe in patients infected with HIV-1. EuroSIDA Study Group. Lancet 352: 1725–30.
- 2. Thompson MA, Aberg JA, Cahn P, Montaner JSG, Rizzardini G, et al. (2010) Antiretroviral treatment of adult HIV infection: 2010 recommendations of the International AIDS Society-USA panel. JAMA : the journal of the American Medical Association 304: 321–33.
- 3. Wilson LE, Gallant JE (2009) HIV/AIDS: the management of treatment-experienced HIV-infected patients: new drugs and drug combinations. Clinical infectious diseases : an official publication of the Infectious Diseases Society of America 48: 214–21.
- 4. Vercauteren J, Deforche K, Theys K, Debruyne M, Duque LM, et al. (2008) The incidence of multidrug and full class resistance in HIV-1 infected patients is decreasing over time (2001–2006) in Portugal. Retrovirology 5: 12.
- 5. Vercauteren J, Vandamme AM (2006) Algorithms for the interpretation of HIV-1 genotypic drug resistance information. Antiviral research 71: 335–42.
- 6. Fox ZV, Geretti AM, Kjaer J, Dragsted UB, Phillips AN, et al. (2007) The ability of four genotypic interpretation systems to predict virological response to ritonavir-boosted protease inhibitors. AIDS (London, England) 21: 2033–42.
- 7. Helm M, Walter H, Ehret R, Schmit JC, Kurowski M, et al. (2007) Differences of nine drug resistance interpretation systems in predicting short-term therapy outcomes of treatment-experienced HIV-1 infected patients: a retrospective observational cohort study. European journal of medical research 12: 231–42.
- 8. Vandamme AM, Camacho RJ, Ceccherini-Silberstein F, De Luca A, Palmisano L, et al. (2011) European Recommendations for the Clinical Use of HIV Drug Resistance Testing: 2011 Update. AIDS reviews 13: 77–108.
- 9. De Luca A, Di Giambenedetto S, Maserati R, Gianotti N, Narciso P, et al. (2011) Interpretation of genotypic HIV-1 resistance to darunavir and virological response: validation of available systems and of a new score. Antiviral therapy 16: 489–97.
- 10. Prosperi MCF, Altmann A, Rosen-Zvi M, Aharoni E, Borgulya G, et al. (2009) Investigation of expert rule bases, logistic regression, and non-linear machine learning techniques for predicting response to antiretroviral treatment. Antiviral therapy 14: 433–42.
- 11. De Luca A, Perno CF (2003) Impact of different HIV resistance interpretation by distinct systems on clinical utility of resistance testing. Current opinion in infectious diseases 16: 573–80.
- 12. De Luca A, Cozzi-Lepri A, Perno CF, Balotta C, Di Giambenedetto S, et al. (2004) Variability in the interpretation of transmitted genotypic HIV-1 drug resistance and prediction of virological outcomes of the initial HAART by distinct systems. Antiviral therapy 9: 743–52.
- 13. Rhee SY, Fessel WJ, Liu TF, Marlowe NM, Rowland CM, et al. (2009) Predictive value of HIV-1 genotypic resistance test interpretation algorithms. The Journal of infectious diseases 200: 453–63.
- 14. Van Laethem K, De Luca A, Antinori A, Cingolani A, Perno CF, et al. (2002) A genotypic drug resistance interpretation algorithm that significantly predicts therapy response in HIV-1-infected patients. Antiviral therapy 7: 123–9.
- 15. Libin P, Deforche K, Van Laethem K, Camacho RJ, Vandamme Am (2007) RegaDB: An Open Source, Community-Driven HIV Data and Analysis Management Environment. In: Fifth European HIV Drug Resistance Workshop. Cascais, Portugal.
- 16. de Oliveira T, Deforche K, Cassol S, Salminen M, Paraskevis D, et al. (2005) An automated genotyping system for analysis of HIV-1 and other microbial sequences. Bioinformatics (Oxford, England) 21: 3797–800.
- 17. Liu TF, Shafer RW (2006) Web resources for HIV type 1 genotypic-resistance test interpretation. Clinical infectious diseases : an official publication of the Infectious Diseases Society of America 42: 1608–18.
- 18. Meynard JL, Vray M, Morand-Joubert L, Race E, Descamps D, et al. (2002) Phenotypic or genotypic resistance testing for choosing antiretroviral therapy after treatment failure: a randomized trial. AIDS (London, England) 16: 727–36.
- 19. Altmann A, Däumer M, Beerenwinkel N, Peres Y, Schülter E, et al. (2009) Predicting the response to combination antiretroviral therapy: retrospective validation of geno2pheno-THEO on a large clinical database. The Journal of infectious diseases 199: 999–1006.
- 20. Frentz D, Boucher CAB, Assel M, De Luca A, Fabbiani M, et al. (2010) Comparison of HIV-1 genotypic resistance test interpretation systems in predicting virological outcomes over time. PloS one 5: e11505.
- 21. Hemelaar J, Gouws E, Ghys PD, Osmanov S (2006) Global and regional distribution of HIV-1 genetic subtypes and recombinants in 2004. AIDS (London, England) 20: W13–23.
- 22. Zazzi M, Prosperi M, Vicenti I, Di Giambenedetto S, Callegaro A, et al. (2009) Rules-based HIV-1 genotypic resistance interpretation systems predict 8 week and 24 week virological antiretroviral treatment outcome and benefit from drug potency weighting. The Journal of antimicrobial chemotherapy 64: 616–24.
- 23. Katlama C, Ingrand D, Loveday C, Clumeck N, Mallolas J, et al. (1996) Safety and efficacy of lamivudine-zidovudine combination therapy in antiretroviral-naive patients. A randomized controlled comparison with zidovudine monotherapy. Lamivudine European HIV Working Group. JAMA : the journal of the American Medical Association 276: 118–25.
- 24. Staszewski S, Hill AM, Bartlett J, Eron JJ, Katlama C, et al. (1997) Reductions in HIV-1 disease progression for zidovudine/lamivudine relative to control treatments: a meta-analysis of controlled trials. AIDS (London, England) 11: 477–83.