Abstract

Background. This study was performed to investigate the concordance between commonly used human immunodeficiency virus type 1 (HIV-1) drug resistance interpretation systems for didanosine (ddI) and their ability to predict responses at weeks 8 and 24.

Methods. The study included drug-experienced HIV-infected patients who had viral loads >500 copies/mL and who underwent a genotypic resistance test when beginning a ddI-containing therapy. The interpretations of the level of resistance to ddI were compared for the 6 interpretation systems. Linear and logistic regression were used to assess their ability to predict responses for weeks 8 and 24, respectively.

Results. The 1453 patients had a median viral load of 4.3 log10 copies/mL, and 31% were preexposed to ddI. Complete concordance was found for 19% of samples, partial discordance for 49%, and complete discordance for 32%. The median viral load reduction at week 8 was 1.36 log10 copies/mL, and 56% of patients had viral loads >400 copies/mL at week 24. At week 8, all systems correctly predicted a greater viral load reduction in patients with susceptible viruses than in those with resistant viruses, but only the Stanford system was able to discriminate between patients with resistant, intermediately resistant, and susceptible viruses. No systems predicted virological response correctly at week 24.

Conclusions. Our results show the need for standardized methods to establish genotypic interpretation systems.

Nucleoside analogue reverse-transcriptase inhibitors (NRTIs) remain an important component of highly active antiretroviral therapy (HAART) and have represented the backbone of virtually all antiretroviral regimens. Despite the growing number of antiretroviral drugs available, therapeutic options remain limited for HIV-1-infected patients experiencing failure of antiretroviral therapy [1]. These patients often began antiretroviral treatment with monotherapy or dual therapy and carry NRTI-resistant viruses. To guide the choice of active antiretroviral drugs, several rules have been developed by experts to interpret HIV-1 drug resistance genotyping results [2–8]. However, most genotypic interpretation rules have been developed with limited data, with different approaches and methodologies, and without any formal cross-validation to establish whether their predictive value was retained in a different study population [2, 9, 10]. Several studies recently reported substantial discordances in the interpretation of genotypic resistance and variability in their ability to predict virological response [11, 12]. Some of these discordances have been attributed to didanosine (ddI), because resistance to it may be associated with incompletely defined patterns of mutations [13].

ddI is a 2′,3′-dideoxyinosine NRTI used in antiretroviral combination therapies for the treatment of HIV-infected patients. The in vitro and in vivo patterns associated with resistance to ddI have been described in several studies. Resistance to ddI is frequently associated with a mutation at codon 74 (L74V), which confers an −4–10-fold reduction in ddI susceptibility [14]. In addition to L74V, ddI resistance has been observed for 2 patterns selected by other nucleoside analogues that are associated with NRTI multidrug resistance, the Q151M complex [15] and the family of amino acid insertions between codons 67 and 70 of reverse transcriptase [16]. Other NRTI mutations, such as M184V or K65R, can also be selected by ddI when it is given as monotherapy or in sequential NRTI therapy [17]. However, it is unusual to find these mutations in patients in whom a ddI-containing HAART regimen failed. The Jaguar study [1] has shown that ddI retains substantial antiviral activity against isolates expressing up to 3 thymidine-associated mutations. Moreover, it has been shown that the M184V mutation may have a positive impact on virological response to ddI [2], although this is partially inconsistent with the results of in vitro studies of phenotype (higher fold change in isolates with 184IV). Hence, the impact on clinical activity of ddI can be difficult to predict.

The aim of the present study was to investigate the concordance of existing HIV-1 drug resistance interpretation systems as well as their ability to predict virological response to ddI-containing regimens in antiretroviral-experienced patients enrolled in a number of clinical trials and observational studies across the Western world. This project, together with a similar one for abacavir, is intended as a first step toward greater worldwide collaboration in standardizing interpretation systems for antiretroviral drugs.

Methods

Study population. The data were pooled from 13 sources (clinical trials and clinic-based cohorts): the Aquitaine Cohort, France; the Adult AIDS Clinical Trials Group, United States; the Antiretroviral Resistance Cohort Analysis (ARCA) database, Italy; the British Columbia Cohort, Canada; the EuroSIDA Cohort, Europe; ICoNA (Italian Cohort Naive Antiretrovirals), Italy; the Institut d'Investigacions Biomèdiques August Pi i Sunyer (IDIBAPS) Barcelona Hospital Clinic Cohort, Spain; Jaguar Trial Bristol-Myers Squibb (BMS), France; the Narval Agence Nationale de Recherches sur le SIDA (ANRS) 88 Trial, France; the Swiss HIV Cohort Study, Switzerland; the AIDS Clinical Trials Group (ACTG), United States; the Catholic University Sacro Cuore (UCSC) Cohort, Italy; and the UK National Resistance Database and UK Collaborative HIV Cohort Study (CHIC) Database, United Kingdom. Patients were included if they had (1) virological failure (according to the clinician's judgment) with antiretroviral treatment before beginning a regimen that included ddI (either used for the first time or recycled); (2) results of a genotypic resistance test undergone during the previous regimen (measured <12 weeks before the start of the new regimen); (3) a viral load measurement of >500 copies/mL during the previous failing regimen (<12 weeks before the start of the new regimen; this was considered the baseline viral load; (4) at least 1 viral load measured 4–12 weeks (the 8-week viral load) or 16–32 weeks (the 24-week viral load) after the start of the new (ddI-containing) regimen; (5) no changes in therapy between the time of the baseline viral load or resistance test and the start of the new regimen, nor between the time of the start of the new regimen and the time of the analyses (4–12 weeks for 8-week analyses and 16–32 weeks for 24-week analyses); and (6) no evidence of inadequate adherence to the ddI-containing regimen. More details about the data extraction and items collected can be found elsewhere [18].

HIV-1 RNA extraction and sequencing. HIV-1 RNA was isolated by means of commonly used assays, and full-sequence analysis of HIV-1 protease and reverse-transcriptase reading frames were performed according to study protocols, using the genotype that was closest to the date of initiation for the ddI-containing regimen. Reference HIV strains (pNL43, HXB2, and consensus B) varied from study to study, but data were standardized so that coding for resistance was not ambiguous. Genotypic results were interpreted by using 6 commonly used rule-based interpretation systems or tables of mutations: ANRS, version 14, updated in July 2006 [4]; Detroit Medical Center, version 3 (DMC-3), updated in September 2004 [8]; Stanford HIV RT and Protease Sequence Database (Stanford), version 4.3.0 [3], updated in July 2007; Rega Institute for Medical Research, version 7.1.1, updated in July 2007 [5]; Bayer Health Care, version 10 [7]; and Grupo de Aconseilhamento Virològico-Sao Paulo University, Brazil version (Sao Paulo–2) (Advanced Biological Laboratories), updated in October 2001 [7]. All interpretation systems except Stanford are rules-based algorithms that report 3 levels of resistance: susceptible, resistant, and intermediate. The Stanford algorithm assigns a drug penalty score for each drug resistance mutation. The total score for a drug is derived by adding the scores associated with each mutation. The program uses the total drug score to assign one of the following levels of inferred drug resistance: susceptible, potential low-level resistance, low-level resistance, intermediate resistance, and high-level resistance. For this study, we evaluated the virological responses predicted by the Stanford algorithm, and we grouped the closest levels to obtain 3 levels of resistance. Therefore, “susceptible” was considered susceptible; “potential low-level resistance,” “low-level resistance,” and “intermediate resistance” were considered intermediate; and “high level-level resistance” was considered resistant (tables A1 and A2 in Appendix A).

Table A1

Publicly available algorithms for didanosine.

Table A1

Publicly available algorithms for didanosine.

Table A2

Stanford, version 4.3.0: levels of resistance.

Table A2

Stanford, version 4.3.0: levels of resistance.

Statistical analyses. Percentages of isolates classified as resistant, intermediate, or susceptible according to the 6 interpretation systems were calculated and compared in a descriptive fashion. Interpretations were considered concordant between interpretation systems if all 6 algorithms assigned the same level of resistance (resistant, intermediate, or susceptible). Partial discordances occurred when the interpretations split between susceptible and intermediate or between intermediate and resistant. If isolates were classified as susceptible by at least 1 algorithm and as resistant by another, the interpretation systems were said to be discordant [11].

Each study source provided a viral load measured between 4 and 12 weeks, defined as the week 8 viral load. A Kaplan-Meier approach was used to estimate the overall median week 8 viral load reduction from baseline. For each interpretation system, a linear regression model [19] of the week 8 reduction in viral load from baseline was fitted. The linear model accounted for the censoring of viral load measurements due to assay lower limits by using a program designed for parametric survival analysis models (PROC LIFEREG in SAS, DIST = NORMAL option) [20].

The following covariates were included because of their potential roles as confounders, as stated in the predefined analysis plan: (1) ddI susceptibility (as a 3-category variable scored as susceptible/intermediate/resistant, with the resistant group as the base) based on the interpretation systems; (2) baseline viral load (fitted as log10 transformed, as a continuous variable); (3) the exact number of weeks from the start of the regimen to the 8- or 24-week viral load measurement (fitted as a continuous variable, untransformed); and (4) the number of drugs in the new regimen to which virus is susceptible (at time of resistance test). Each drug in the new regimen was scored as susceptible (score of 1), intermediate (score of 0.5), or resistant (score of 0), according to the Rega interpretation system, in the main analysis. Analyses were repeated using the ANRS and Stanford interpretation systems.

The viral load measured between 16 and 32 weeks was used for analysis of the week 24 viral load response. The criterion for virological failure at week 24 was an HIV RNA level >400 copies/mL, with the exclusion of patients who had discontinued ddI or for whom the week 24 viral load was not available. For each interpretation system, a multivariable logistic regression model [19] of the week 24 binary outcome (failure: yes/no) was fitted on the same covariates described for the week 8 analysis. Analyses were performed with the SPSS software package (version 15.0 for Windows; SPSS), the SAS software package (version 9.1 for Windows), and the STATA software package (version 10 for Mac OS; Stata).

Results

Baseline characteristics of the study population. We identified 1453 patients who met the inclusion criteria and were evaluated at week 8, and 1074 of them were also evaluated at week 24. The 379 patients excluded were those who stopped taking ddI or did not have a week 24 viral load measurement. The baseline characteristics of patients are shown in table 1. Most of the patients were men (78%); the median age was 39 years (interquartile range [IQR], 35–45 years), the median HIV-1 RNA level was 4.3 log10 copies/mL (IQR, 3.7–4.9 log10 copies/mL), and the median CD4 cell count was 280 cells/µL (IQR, 167–420 cells/µL). The median number of previously used antiretroviral drugs was 4 (range, 1–12), including medians of 3 NRTIs (range, 0–6) and 1 protease inhibitor (PI; range, 0–4). Thirty-one percent of patients were preexposed to ddI before enrolment. Patients were enrolled between May 1993 and August 2004, with a median in September 2000.

Table 1

Baseline patient characteristics.

Table 1

Baseline patient characteristics.

Antiviral drugs prescribed with ddI. The median number of drugs in the new regimen coprescribed with ddI was 2 (IQR, 2–2). The most frequent NRTI drugs coprescribed with ddI were stavudine (50%), lamivudine (17%), zidovudine (16%), and abacavir (15%) (figure 1). Twenty-one percent of patients received neither non-NRTIs (NNRTIs) nor PIs in the regimen, 25% received at least 1 NNRTI and no PI, 44% received at least 1 PI and no NNRTI, and 10% received at least 1 NNRTI and at least 1 PI.

Figure 1

Drugs coprescribed with didanosine. 3TC, lamivudine; ABC, abacavir; APV, amprenavir; AZT, zidovudine; D4T, stavudine; DDC, zalcitabine; EFV, efavirenz; IDV, indinavir; LPV/r, lopinavir/ritonavir; NFV, nelfinavir; NVP, nevirapine; RTV, ritonavir; SQV, saquinavir. Asterisks indicate categories that include APV/ritonavir (3%), IDV/ritonavir (4%), and SQV/ritonavir (6%).

Figure 1

Drugs coprescribed with didanosine. 3TC, lamivudine; ABC, abacavir; APV, amprenavir; AZT, zidovudine; D4T, stavudine; DDC, zalcitabine; EFV, efavirenz; IDV, indinavir; LPV/r, lopinavir/ritonavir; NFV, nelfinavir; NVP, nevirapine; RTV, ritonavir; SQV, saquinavir. Asterisks indicate categories that include APV/ritonavir (3%), IDV/ritonavir (4%), and SQV/ritonavir (6%).

Prevalence of mutations at baseline. In these treatment-experienced subjects, the range of International AIDS Society-USA panel (IAS-USA) resistance mutations (updated in September 2006 [6]) observed at baseline was extensive (table 2). The median number of NRTI mutations, according to IAS-USA, was 3 (IQR, 1–4). The prevalence of ddI resistance mutations at baseline, according to IAS-USA, was 2% for K65R and 4% for L74V.

Table 2

Frequency of baseline HIV-1 reverse-transcriptase (RT) and protease (PR) resistance mutations.

Table 2

Frequency of baseline HIV-1 reverse-transcriptase (RT) and protease (PR) resistance mutations.

Concordance in predicted susceptibility between interpretation systems.Figure 2 displays the proportions of patients in each susceptibility category, according to interpretation system. Complete discordance was found for 32% of samples, partial discordance for 49%, and complete concordance for 19%. The median number of active drugs in addition to ddI in the prescribed combination was 1 for Rega (IQR, 0–1; range, 0–5), 1 for ANRS (IQR, 0–1; range, 0–4.5), and 1 for Stanford (IQR, 0–1; range, 0–4).

Figure 2

Genotypic resistance interpretations for didanosine showing the percentages of patients in each susceptibility category, according to interpretation system. ANRS, Agence Nationale de Recherches sur le SIDA; DMC, Detroit Medical Center.

Figure 2

Genotypic resistance interpretations for didanosine showing the percentages of patients in each susceptibility category, according to interpretation system. ANRS, Agence Nationale de Recherches sur le SIDA; DMC, Detroit Medical Center.

Week 8 virological response. Overall, the Kaplan-Meier estimate of the median week 8 viral load decline was 1.36 log10 copies/mL (IQR, 1.25–1.47 log10 copies/mL). Figure 3A shows the absolute mean week 8 viral load reductions in patients grouped according to whether they had a predicted ddI-resistant, intermediately resistant, or susceptible virus, with a median short-term evaluation of 8 weeks (IQR, 6–11 weeks).

Figure 3

Virological response at week 8, according to level of resistance and interpretation system. A, Median change in week 8 viral load (log10 copies/mL). ANRS, Agence Nationale de Recherches sur le SIDA; DMC, Detroit Medical Center; KM, Kaplan-Meier. B, Univariate and adjusted differences between mean (95% confidence interval [CI]) week 8 viral load (log10 copies/mL) reductions in those with intermediately resistant (I) or susceptible (S) viruses, with the mean week 8 viral load reduction in those with resistant (R) viruses taken as the comparator. Analysis was adjusted for baseline viral load, Rega genotypic sensitivity score, and the exact no. of weeks to the week 8 viral load.

Figure 3

Virological response at week 8, according to level of resistance and interpretation system. A, Median change in week 8 viral load (log10 copies/mL). ANRS, Agence Nationale de Recherches sur le SIDA; DMC, Detroit Medical Center; KM, Kaplan-Meier. B, Univariate and adjusted differences between mean (95% confidence interval [CI]) week 8 viral load (log10 copies/mL) reductions in those with intermediately resistant (I) or susceptible (S) viruses, with the mean week 8 viral load reduction in those with resistant (R) viruses taken as the comparator. Analysis was adjusted for baseline viral load, Rega genotypic sensitivity score, and the exact no. of weeks to the week 8 viral load.

Among the genotype interpretation systems tested, we found some variability in the prediction of the week 8 virological response. In the multivariable model, after adjustment for baseline viral load, the number of active drugs, and the exact number of weeks of follow-up, all interpretation systems predicted patients with susceptible viruses to have greater reductions in viral load than patients with resistant viruses, with adjusted differences ranging between 0.28 (P=.003) (Sao Paulo-2) and 0.68 (P<.0001) (Stanford) log10 copies/mL. For some interpretation systems (e.g., Sao Paulo-2), there were greater virological reductions in patients with predicted intermediate resistance than in those with predicted susceptible viruses, but other systems (e.g., ANRS, Rega, and DMC-3) were not able to discriminate between patients with intermediate and those with resistant viruses (figure 3B). Only 2 interpretation systems (Bayer and Stanford) showed significantly greater reductions in patients with intermediate viruses than in those with resistant viruses (Bayer, 0.31 [P=.002]; Stanford, 0.36 [P=.0001]) (figure 3B). On the other hand, Stanford was the only system that showed significantly greater reductions in patients with susceptible viruses than in those with intermediate viruses (0.32; P<.0001). Results were similar when the model was adjusted for the number of active drugs calculated using the ANRS or Stanford interpretation system instead of the Rega interpretation system (data not shown)

Among the factors included in the model because they were identified as potential confounders, baseline viral load (P=.029) and the number of active drugs in the regimen (P<.0001) were both strongly associated with the week 8 virological response, but adjustment for these variables did not change the predicted response to ddI according to the level of resistance (figure 3B).

Week 24 virological response. Overall, 56% of patients had treatment failure at week 24 (599 failures in 1074 patients who were still receiving ddI by week 24 and for whom viral load information was available). Figure 4A displays the proportion of patients with predicted week 24 treatment failure according to the interpretation systems, with a median midterm evaluation of 24 weeks (IQR, 22–25 weeks).

Figure 4

Virological response at week 24, according to level of resistance and interpretation system. A, Percentages of patients with virological failure (plasma HIV RNA level >400 copies/mL) at week 24. B, Univariate and adjusted odds ratios (ORs) with 95% confidence intervals (CIs) for week 24 virological failure, obtained by fitting a logistic regression model. Analysis was adjusted for baseline viral load, Rega genotypic sensitivity score, and the exact no. of weeks to the week 24 viral load. The comparator used was patients with resistant viruses. ANRS, Agence Nationale de Recherches sur le SIDA; DMC, Detroit Medical Center; I, intermediately resistant viruses; S, susceptible viruses.

Figure 4

Virological response at week 24, according to level of resistance and interpretation system. A, Percentages of patients with virological failure (plasma HIV RNA level >400 copies/mL) at week 24. B, Univariate and adjusted odds ratios (ORs) with 95% confidence intervals (CIs) for week 24 virological failure, obtained by fitting a logistic regression model. Analysis was adjusted for baseline viral load, Rega genotypic sensitivity score, and the exact no. of weeks to the week 24 viral load. The comparator used was patients with resistant viruses. ANRS, Agence Nationale de Recherches sur le SIDA; DMC, Detroit Medical Center; I, intermediately resistant viruses; S, susceptible viruses.

Among the factors included in the logistic regression that were identified as potential confounders, the following were strongly associated with the risk of treatment failure at week 24: baseline viral load (odds ratio [OR], 2.21 per log10 copies/mL higher [95% confidence interval {CI}, 1.85–2.62]; P=.0001), the number of active drugs in the regimen (Rega OR, 0.49 per additional sensitive drug [95% CI, 0.41–0.58; P=.0001]; ANRS OR, 0.47 [95% CI, 0.39–0.56; P=.0001]; Stanford OR, 0.45 [95% CI, 0.37–0.53; P=.0001]), and the exact number of weeks to the week 24 viral load (OR, 1.05 per additional week [95% CI, 1.01–1.09]; P=.017). Contrary to the week 8 virological response analysis, adjustment for these variables modified the association between virological response and the level of predicted resistance, showing that these variables confounded the association between response and interpretation systems (figure 4B). For example, when the Stanford interpretation system was considered, the unadjusted and adjusted ORs were 0.55 (P=.001) and 0.88 (P=.511), respectively, for patients with predicted intermediate viruses compared with patients with resistant viruses and were 0.30 (P<.001) and 0.56 (P=.009) for patients with predicted susceptible viruses compared with patients with predicted resistant viruses.

For the comparison between patients with susceptible viruses and those with resistant viruses, the Rega-adjusted ORs of having a viral load >400 copies/mL were 0.73 for ANRS (95% CI, 0.53–1.01; P=.06), 0.56 for Stanford (95% CI, 0.36–0.87; P=.009), 0.61 for Rega (95% CI, 0.44–0.86; P=.004), 0.58 for DMC-3 (95% CI, 0.39–0.86; P=.006), 1.02 for Sao Paulo–2 (95% CI, 0.70–1.47; P=.92), and 0.56 for Bayer (95% CI, 0.42–0.75; P=.0001). For the comparison between patients with intermediate viruses and those with resistant viruses, the Rega-adjusted ORs of having a viral load >400 copies/mL were 0.65 for ANRS (95% CI, 0.20–2.08; P=.47), 0.88 for Stanford (95% CI, 0.59–1.30; P=.51), 1.27 for Rega (95% CI, 0.88–1.83; P=.20), 0.89 for DMC-3 (95% CI, 0.56–1.41; P=.62), 0.53 for Sao Paulo–2 (95% CI, 0.40–0.72; P=.0001), and 0.72 for Bayer (95% CI, 0.49–1.07; P=.10).

Overall in this analysis, after adjustment for all factors identified as potential confounders, 4 interpretation systems (Stanford, Rega, DMC-3, and Bayer) predicted greater odds of treatment failure in patients with resistant viruses than in those with susceptible viruses. The 2 interpretation systems that were not able to discriminate between patients with resistant and susceptible viruses (ANRS and Sao Paulo-2) predicted greater (but not significantly greater) odds of treatment failure in patients with susceptible viruses than in those with intermediate viruses, whereas the Rega system predicted greater odds of week 24 treatment failure in patient with intermediate viruses than in those with resistant viruses. No interpretation system except Sao Paulo-2 predicted significantly greater odds of week 24 treatment failure in patients with resistant viruses than in those with intermediate viruses (figure 4B).

Discussion

Standardized methods for interpretation of genotypic resistance tests are needed, especially in view of drug labels recommending the use of other active drugs (i.e., drugs for which the virus is predicted as being susceptible) to be used with new drugs. Moreover, clinical trials for drug approval need a definition of active drugs in the regimen to which the new drug is added. No mechanism currently exists for the systematic review and validation of genotypic algorithms and interpretation systems. This is the objective of the initiatives of the Forum for Collaborative HIV Research for developing and comparing genotype and phenotype interpretation systems [18]. In this analysis, which represents a first step toward greater worldwide collaboration and standardization in developing interpretation systems, the purpose was to examine the ability of 6 HIV-1 drug resistance interpretation systems to predict virological response to ddI-containing regimens in a large population of antiretroviral-experienced patients.

Our results show a high level of discordance in the interpretation of genotyping results for ddI among the algorithms used and revealed that, among the HIV-1 drug resistance interpretation systems compared, only the Stanford system correctly predicted the short-term (week 8) virological responses, and none of the systems correctly predicted the week 24 response. By correct prediction, we mean a better response in patients with susceptible viruses than in those with intermediate viruses, and a better response in patients with intermediate viruses than in those with resistant viruses.

Regarding the discordance issue, our results are in agreement with previous reports [11–13]. Given the relatively large sample size in the present study, we believe that the lack of ability of current interpretation systems to predict virological response at either 8 or 24 weeks is not due simply to a lack of power but rather reflects limitations in the interpretation systems for ddI and in the current genotype technologies.

We believed that viral load change from baseline to week 8 (between 4 and 12 weeks) is the best end point for analyses relating the chance of virological failure to the presence of resistance mutations present before the start of therapy. This is because a genotypic test can detect only the mutations present in the majority of the population of viruses at baseline, and virological failure occurring after week 12 could be partially determined by mutations that are present at low levels at baseline and subsequently grow with the pressure of therapy. Longer-term responses, at week 24 or 48, are also much more likely to be affected by confounding factors [10]. Moreover, our study showed that factors identified as potential confounders—baseline viral load, the number of active drugs coprescribed with ddI, and the exact number of weeks of follow-up—changed the predicted response to ddI at week 24 according to the level of resistance in the multivariate analysis compared with the univariate analysis, showing that these variables confounded the association between response and interpretation systems at week 24. This was not observed at week 8, with similar results in adjusted and unadjusted analyses. Our data were pooled from 13 sources. This pooling can produce some variability in the data collection and decrease the precision of estimates, but we made a great effort to standardize the data and ensure that genotypic information was comparable. Of note, when the analysis was restricted to patients not previously exposed to ddI, no interpretation system predicted the virological response correctly at either week 8 or 24 [21].

Our study also shows that the intermediate level of resistance was not correctly defined by the majority of the current interpretation systems. For example, for the ANRS interpretation system, this intermediate level was added by experts to the genotypic score defined in the Jaguar study [2]. This intermediate resistance level is represented by the presence of the K65R mutation, which has a very low prevalence in treatment-experienced patients (2% in our study). This is an example of an interpretation system that has, without additional validation, mixed a statistically defined score with external data on the acquisition of resistance mutations by viruses in patients in whom treatment was failing.

In conclusion, the present study shows a high degree of discordance in the interpretation of genotyping results among currently used algorithms for ddI. Only 1 interpretation system predicted the short-term virological response, and none predicted the 6-month response. This illustrates the need for standardized approaches as well as global collaboration in putting together data sets of the necessary size, because no mechanism currently exists for systematic review and validation of genotypic interpretation systems. The Forum for Collaborative HIV Research, with its mission to provide a neutral and independent mechanism to address HIV research questions, brings together researchers and databases from private and publicly funded sources; it is therefore well suited to work on this project and will continue to address these questions.

Standardization and Clinical Relevance of HIV Drug Resistance Testing Project of the Forum for Collaborative HIV Research

The subgroup planning committee included Daniel Kuritzkes and Veronica Miller (chairs), Françoise Brun-Vezinet and Andrew Zolopa (genotypic interpretation system), Richard Haubrich and Joe Eron (phenotypic cutoffs), Lisa Demeter and Rob Schuurman (technology and standardization), and Victor DeGruttola, Dominique Costagliola, and Andrew Phillips (statistical analysis).

Acknowledgments

The data providers included H. Ribaudo, P. Bohlin, M. Wantman, ACTG Trials, Statistical and Data Analysis Center, Boston, Massachusetts; B. Masquelier, Aquitaine Cohort, France; M. Zazzi, ARCA, University of Siena, Italy; B. Hogg, R. Harrigan, Homer Cohort, British Columbia, Canada; J. Lundgren, J. Kjaer, EuroSIDA Cohort, Europe; A. d'Arminio Monforte, ICoNA, Italy; J. Gatell, IDIBAPS Barcelona Hospital Clinic Cohort, Spain; D. Seekins, Jaguar Trial BMS, France; D. Costagliola, F. Brun-Vézinet, NARVAL ANRS 88 Trial, France; B. Ledergerber, Swiss HIV Cohort, Switzerland; R. Shafer, Stanford University, Palo Alto, California; A. De Luca, University Sacro Cuore, Rome, Italy; D. Pillay, D. Dunn, C. Sabin, UK Resistance Database and UK CHIC, United Kingdom. The interpretation system providers included D. Costagliola, F. Brun-Vézinet (ANRS, version 14); R. McArthur (DMC-3); A. M. Van Damme (Rega, version 7.1.1); R. Diaz (Sao Paulo–2); R. Ziermann and Q Baldwin Bayer V10; and R. Shafer (Stanford, version 4.3.0.).

Appendix A

References

1
Molina
JM
Marcelin
AG
Pavie
J
, et al.  . 
Didanosine in HIV-1 Infected patients experiencing failure of antiretroviral therapy: a randomized placebo-controlled trial
J Infect Dis
 , 
2005
, vol. 
191
 (pg. 
840
-
7
)
2
Marcelin
AG
Flandre
P
Pavie
J
, et al.  . 
Clinically relevant genotype interpretation of resistance to didanosine
Antimicrob Agents Chemother
 , 
2005
, vol. 
49
 (pg. 
1739
-
44
)
3
Stanford University HIV Drug Resistance Database
  
4
Agence Nationale de Recherches sur le SIDA (ANRS)
HIV-1 genotypic drug resistance interpretation's algorithms
  
5
Laboratory for Clinical and Evolutionary Virology
The Rega algorithms
  
6
International AIDS Society-USA Web site
 
7
Advanced Biological Laboratories
Algorithm and Knowledgebase updates
  
8
MacArthur
RD
An updated guide to genotype interpretation
AIDS Reader
 , 
2004
, vol. 
14
 (pg. 
256
-
66
)
9
Flandre
P
Marcelin
AG
Pavie
J
, et al.  . 
Comparison of tests and procedures to build clinically relevant genotypic scores: application to the Jaguar study
Antivir Ther
 , 
2005
, vol. 
10
 (pg. 
479
-
87
)
10
Brun-Vézinet
F
Costagliola
D
Ait Khaled
M
, et al.  . 
Clinically validated genotype analysis: guiding principles and statistical concerns
Antivir Ther
 , 
2004
, vol. 
9
 (pg. 
465
-
78
)
11
Ravela
J
Betts
BJ
Brun-Vézinet
F
, et al.  . 
HIV-1 protease and reverse transcriptase mutation patterns responsible for discordances between genotypic drug resistance interpretation algorithms
J Acquir Immune Defic Syndr
 , 
2003
, vol. 
33
 (pg. 
8
-
14
)
12
Sturmer
M
Doerr
HW
Staszewski
S
Preiser
W
Comparison of nine resistance interpretation systems for HIV-1 genotyping
Antivir Ther
 , 
2003
, vol. 
8
 (pg. 
239
-
44
)
13
Boulmé
R
Halfon
P
Diaz
R
Dugas
O
Schmit
JC
Comparative evaluation between five automated resistance interpretation algorithms
Antivir Ther
 , 
2001
, vol. 
6
 
Suppl 1
pg. 
121
 
14
St Clair
MH
Martin
JL
Tudor-Williams
G
, et al.  . 
Resistance to ddI and sensitivity to AZT induced by a mutation in HIV-1 reverse transcriptase
Science
 , 
1991
, vol. 
253
 (pg. 
1557
-
9
)
15
Van Laethem
K
Witvrouw
M
Balzarini
J
, et al.  . 
Patient HIV-1 strains carrying the multiple nucleoside resistance mutations are cross-resistant to abacavir
AIDS
 , 
2000
, vol. 
14
 (pg. 
469
-
71
)
16
Larder
BA
Bloor
S
Kemp
SD
, et al.  . 
A family of insertion mutations between codons 67 and 70 of human immunodeficiency virus type 1 reverse transcriptase confer multinucleoside analog resistance
Antimicrob Agents Chemother
 , 
1999
, vol. 
43
 (pg. 
1961
-
7
)
17
Winters
MA
Shafer
RW
Jellinger
RA
Mamtora
G
Gingeras
T
Merigan
TC
Human immunodeficiency virus type 1 reverse transcriptase genotype and drug susceptibility changes in infected individuals receiving dideoxyinosine monotherapy for 1 to 2 years
Antimicrob Agents Chemother
 , 
1997
, vol. 
41
 (pg. 
757
-
62
)
18
Forum for Collaborative HIV Research
Standardization and clinical relevance of HIV drug resistance testing
  
19
Kleinbaum
D
Kupper
L
Muller
K
Applied regression analysis and other multivariable methods
 , 
1988
2nd ed.
Belmont, CA
Duxbury Press
20
Hughes
MD
Analysis and design issues for studies using censored biomarker measurements with an example of viral load measurements in HIV clinical trials
Stat Med
 , 
2000
, vol. 
19
 (pg. 
3171
-
91
)
21
Costagliola
D
Cozzi-Lepri
A
Dalban
C
Cheng
B
Initiatives for developing and comparing genotype interpretation systems step 1: external validation of existing rules-based algorithm for abacavir and ddI evaluated on virological response. Standardization and Clinical Relevance of HIV Drug Resistance Testing Project from the Forum for Collaborative HIV Research
Antivir Ther
 , 
2005
, vol. 
10
 
Suppl 1
pg. 
S11
 
Potential conflicts of interest: none reported.
Presented in part: XIV International HIV Drug Resistance Workshop, Quebec City, 7–11 June 2005 (abstract 9); XV International HIV Drug Resistance Workshop, Sitges, Spain, 13–17 June 2006 (abstract 84).
Financial support: Institut National de la Santé et de la Recherche Médicale (INSERM).

Author notes

a
Study group members are listed after the text.