AAC Accepts, published online ahead of print on 18 August 2008 Antimicrob. Agents Chemother. doi: /aac

Transcription

1 AAC Accepts, published online ahead of print on August 00 Antimicrob. Agents Chemother. doi:0./aac Copyright 00, American Society for Microbiology and/or the Listed Authors/Institutions. All Rights Reserved. Comparison of Genotypic Scores for LPV/r Title: Virologic Response to Lopinavir/Ritonavir-Based Antiretroviral Regimens in a Multi-Center International Clinical Cohort: Comparison of Genotypic Interpretation Scores Authors: Grant, Philip, MD* Wong, Eric C., MS Rode, Richard, PhD Shafer, Robert, MD Deluca, Andrea, MHS Nadler, Jeffrey, MD, FACP Hawkins, Trevor, MD Cohen, Calvin, MD, MS Harrington, Robert, MD Kempf, Dale, PhD Zolopa, Andrew, MD Stanford University, Palo Alto, CA Abbott Laboratories, Abbott Park, IL Università Cattolica del Sacro Cuore, Rome, Italy University of South Florida, Tampa, FL University of New Mexico, Santa Fe, NM Community Research Initiative of New England, Boston, MA University of Washington, Seattle, WA *Reprints or correspondence: Dr. Philip Grant, Division of Infectious Diseases and Geographic Medicine, 00 Pasteur Drive, Grant Building Room S-, Stanford, CA 0-0; fax: 0--0; telephone: (0) -0; pmgrant@stanford.edu Presented in part: "Genotypic Predictors of Response to Lopinavir/ritonavir in Clinical Practice", nd International AIDS Society Conference on HIV Pathogenesis and Treatment, Paris, France, July -, 00, abstract no.

2 0 Abstract Background: Several genotypic interpretation scores have been proposed for the evaluation of susceptibility to lopinavir/ritonavir (LPV/r) but have not been compared using an independent dataset. Methods: This study was a retrospective multi-center cohort of patients initiating LPV/r-based therapy. Virologic response (VR) was defined as a viral load of <00 copies/ml at week. The genotypic interpretation scores surveyed were the Lopinavir Mutation Score, the ViroLogic Score, the ATU Score, the Stanford Database Score, and the International AIDS Society-USA mutation list. Results: Of the 0 patients included in the analysis, % achieved VR at weeks. For scores with clinical breakpoints defined (Lopinavir Mutation Score, ViroLogic, ATU, and Stanford), over 0% of patients below the breakpoints achieved VR, while 0% or less above the breakpoints responded. Protease mutations at positions 0,, and and at positions,, and 0 were associated with lack of VR in the univariate and multivariate analyses, respectively. The area under the receiver operator characteristic curves for the genotypic interpretation scores studied ranged from

3 Conclusions: The study confirms that the currently available genotypic interpretation scores which are widely used by clinicians performed similarly well and can be effectively used to predict virologic activity of LPV/r in treatment experienced patients.

4 0 Introduction Treatment of HIV-infected individuals with combination antiretroviral therapy has significantly reduced morbidity and mortality associated with HIV (, ). However, the efficacy of antiretroviral therapy (ART) can be impaired by several factors including the development of antiretroviral-resistant HIV quasispecies (). Lopinavir/ritonavir (LPV/r), FDA-approved in 000, has been widely used in the management of treatment-naïve and treatment-experienced patients (, ). Genotypic interpretation of the impact of protease mutations on boosted protease inhibitors (PI/r) is complicated, as clinically relevant resistance generally requires multiple mutations and can develop through the interplay of major and minor mutations in a variety of patterns. Phenotypic resistance testing is generally considered easier to interpret for boosted PI s; however, it is more expensive than genotypic resistance testing and, in clinical studies, has not been shown to improve effectiveness outcomes above that achieved using genotypic resistance testing (). Several genotypic interpretation scores have been developed, but there is no consensus on their relative value. 0 The "Lopinavir Mutation Score" was developed by comparing genotypes and phenotypes of viral isolates from early clinical trials (). Mutations at amino acid positions in protease were found to be correlated with increased

5 0 phenotypic resistance to LPV. Later studies linked a reduction in LPV/r activity to the presence of or more LPV mutations (). The "ViroLogic Score" was developed using clinical samples available from the Monogram Biosciences (then ViroLogic) database for which both a genotype and phenotype for LPV/r had been performed (n=) (). Parkin, et al. evaluated discordant samples for which the genotype showed susceptibility according to the Lopinavir Mutation Score (i.e., less than LPV/r associated mutations) but the phenotype showed clinical meaningful reduced susceptibility (i.e., fold change >0) to identify additional mutations that contributed to decreased LPV susceptibility. The weighted ViroLogic Score consists of mutations at protease positions. Viruses with a score of or more are considered to have reduced susceptibility to LPV. Although this score was based on many more isolates than the Lopinavir Mutation Score, the investigators were limited to making genotype-phenotype correlations as they did not have access to the corresponding virologic response (VR) data for the isolates. 0 Recently, a large database from France (the "ATU" database) was used to evaluate VR to LPV/r in treatment-experienced individuals and generate the "ATU Score" (). Within this patient set, mutations at 0 amino acid positions in protease were found to better predict VR than the original Lopinavir Mutation

6 0 Score. Breakpoints of 0-, -, and > mutations best distinguished between responders and non-responders. In addition to these studies, many web-based interpretation systems (,,, 0) and lists of mutations from various organizations can be used to predict susceptibility to LPV/r. We chose to evaluate the scores generated by the Stanford HIV Resistance Database and International AIDS Society-USA (IAS-USA) given their widespread availability and usage. The "Stanford Score" generates a weighted score using mutations at different positions in protease and produces different outputs (susceptible, potential low-level resistance, low level resistance, intermediate level resistance, and high-level resistance) depending on the score (0). The "IAS-USA Score" for LPV/r includes mutations at positions with mutations at positions considered major (0). In contrast to other prediction systems, the IAS-USA score does not define clinical breakpoints. 0 LPV/r has been the standard boosted PI for treatment experienced patients with PIrelated resistance mutations. However, with the recent approval of tipranavir (TPV/r) and darunavir (DRV/r) for PI-experienced patients, it has become increasingly important for clinicians to be able to choose between these newer PI/r s and LPV/r for these patients. Newer PI's have activity against strains resistant to LPV/r (, ), but in LPV/r susceptible patients, DRV/r has not shown a statistically

7 0 significant improvement in response (). Given the many different interpretation tools available, it can be difficult for the clinician to decide when LPV/r will be the best choice or when one of the newer PI/r's should be used. In this study, we used an independent multi-center cohort to evaluate VR to LPV/r and to better define the factors that influence VR. We performed univariate and multivariate analyses to determine which mutations were associated with virologic failure. In addition, we used our cohort to determine how well each genotypic interpretation score predicted VR to LPV/r and which genotypic interpretation score performed best. Materials and Methods Recruitment and Data Collection The Clinic-Based Investigator Group is comprised of 0 academic, private, and public clinics. Participating clinics (n=) retrospectively identified previously LPV/r-naïve patients who had received a LPV/r-based regimen and had a baseline viral load and genotype, and had follow-up viral loads. Participating clinics were sent a standardized seven-page data acquisition form.

8 0 0 Inclusion Criteria Patients were included for analysis if they had at least one viral load determination within weeks prior to initiation of LPV/r, a baseline genotype within weeks prior to and up to weeks after LPV/r initiation, at least one viral load measurement during the first weeks on LPV/r, and at least one viral load measurement between and 0 weeks after LPV/r initiation (used as the " week" viral load). Definitions Given the use of different viral load assays having different lower limits of quantification, VR was defined by a viral load <00 copies/ml at weeks of therapy (i.e., the lower limit of quantification of the least sensitive assay). A genotypic susceptibility score for the ART was calculated using the Stanford HIV Resistance Database. For each drug used in the new LPV/r-based regimen, a score was given based on the interpretation of the genotype (susceptible=, potential lowlevel resistance=0., low-level resistance=0., intermediate resistance=0., and high-level resistance=0). The genotypic susceptibility score was calculated by adding the activity of all the antiretroviral medications used in the regimen. In addition, a partial genotypic susceptibility score was defined as the genotypic susceptibility score minus the PI's contribution to the regimen.

9 0 Scores: Not all mutations used to calculate the scores were captured in the data acquisition form. Not all the labs performing genotypes reported all the mutations of interest in their genotype reports, while some of the mutations of interest had not been described at the time of the study. Lopinavir Mutation Score () unweighted sum of the number of mutations appearing at amino acid positions 0, 0,,,,,,,,, and 0. ViroLogic () weighted sum of the following of mutations: 0I/F, E, 0I/M, I, F, Q, T, L/I, V, M/V, 0V, A/M/S/T/V, E, T, T, S, A/F/S, and I/M. All mutations were given a weight of one except for those at positions 0,, and which were given a weight of three. The following mutations were not included in the data acquisition form and were excluded from the calculations: E, 0I, Q, T, M, S, T, T, S, I, and M. ATU () unweighted sum of mutations appearing at positions 0, 0,,,,,,,, and. 0 Stanford (0) weighted sum of the following mutations (points for each mutation in parentheses): 0F(), 0I(), 0R(), 0V(), 0Y(), I(), I(), F(),

10 0 I(0), A(), F(), I(), L(), V(0), V(0), A(0), V(0), M(0), A(), S(), T(), 0V(0), L(), Y(), L(), M(), S(0), T(0), V(), A(0), T(), V(), I(), C(), S(), T(), A(), V(), A(0), F(0), S(0), T(0), M(0), L(0), C(0), A(0), V(), C(0), V(), I(), T(), and 0M(0). The following mutations were not included in the data acquisition form and were excluded from the calculations: 0R, 0Y, I, F, A, V, A, A, M, S, T, Y, S, T, I, V, S, C, L, M, A, C, I, T, and V. IAS-USA (0) unweighted sum of the following mutations: 0F/I/R/V, 0 M/R, I, I, F, I/L, A/V, 0V, L, V/L/A/M/T/S, P, V/T, S, V, A/F/T/S, V, and 0M. The following mutations were not included in the data acquisition form and were excluded from the calculations: 0R, A, S, T, V and S. 0 Laboratory Methods A variety of genotype methods were used across the participating sites. Thirty-six percent of the genotypes were run at hospital-based labs, and % were run at commercial reference labs. 0

11 0 0 Statistical Analysis Virologic response was evaluated in pre-defined demographic and clinical subgroups. Chi-squared tests were used to assess statistical significance within each subgroup at the α=0.0 level (two-tailed). Fisher s exact test was used to evaluate the association between VR and the presence or absence of mutations in protease at each position of interest. The Benjamini-Hochberg procedure was used to correct for multiple comparisons. Statistical significance was assessed at the α=0.0 level (two-tailed) (). Multivariate stepwise logistic regression models were used to assess the association between VR and demographic characteristics, clinical predictors, and mutations in protease. Demographic characteristics consisted of age, clinic, and gender. Clinical predictors included baseline log 0 viral load, baseline CD+ T-cell count, and partial genotypic susceptibility score. Mutations in protease found to be significant from the univariate analysis, and the twenty mutations thought a priori to be important to LPV/r resistance (based on the Lopinavir Mutation, ViroLogic, ATU, Stanford, and IAS-USA Scores) were included in the starting model. The criterion for entry into the stepwise models was at the α=0.0 level. Predictors were retained in the stepwise models if they remained significant at the α=0.0 level.

12 0 0 To compare the performance of the different scores, receiver-operator characteristic (ROC) curves were constructed and their corresponding areas under the curve (AUC) were calculated. The AUC of the ROC curve is used to evaluate the predictive value of a given test. An AUC of denotes a perfect test, while an AUC of 0. reflects a test without discriminatory power. The precision of the AUC estimates was evaluated by constructing % BCa bootstrap confidence intervals (B=000 replicates) (). All statistical analyses were done using SAS.. (Cary, NC). Institutional review board approval was obtained from participating centers. Results Patient Characteristics One hundred three (0) patients were included in the study. The median age was. years. (Table ) The patients were predominantly male (%) and Caucasian (%). Patients were enrolled from the US (%) and from Italy (%), with the majority being treated in academic centers (%). The majority of patients were antiretroviral- (%) and PI-experienced (%). The most commonly prescribed protease inhibitor prior to enrollment was nelfinavir (%) followed by indinavir (%). The median baseline viral load and CD+ T-cell count were,000 copies/ml and cells/µl, respectively. The median number of PI, non-

13 0 nucleoside reverse transcriptase inhibitor (NNRTI), and nucleoside reverse transcriptor inhibitor (NRTI) mutations as defined by the IAS-USA was,, and, respectively. The frequencies of PI mutations are summarized in Figure. Virologic/Immunologic Response The median CD+ T-cell count increase from baseline was cells/µl. Overall, % of patients achieved VR at weeks (see Table ). As expected, more ART- and PI-naïve patients experienced VR at Week compared to their ART- and PI- experienced counterparts (ART-naïve vs. ART-experienced: 00% vs. %; PI- naive vs. PI-experienced: % vs. %); however, these differences were not statistically significant (p=0. and p=0., respectively). There were no significant differences between responses in patients across the different clinic settings or by gender or race. There was a trend towards an improved response in those with baseline viral load less than 00,000 copies/ml (0% vs. 0%, p=0.0) but there appeared to be no difference in VR in patients with CD T-cell counts <00 cells/µl compared to those with >00 cells/µl (% vs. %, respectively, p=0.). 0 Virologic response was improved in patients with higher baseline genotypic susceptibility score. In particular, patients with genotypic susceptibility scores of <, -<.,.-<, and + had virologic responses of %, %, 0% and 0%, respectively (p=0.00).

14 0 Importantly, all of the LPV/r scores with defined breakpoints (Lopinavir Mutation Score, ViroLogic Score, ATU Score, and Stanford Score) were able to distinguish virologic responses with similar accuracy. Using the Lopinavir Mutation Score, % of patients with a score less than experienced VR at week compared to only 0% of patients with a score of or greater (p=0.000). For the ViroLogic Score, % of patients with a score less than experienced VR at week compared to % of patients with a score of or greater (p<0.000). Similarly, % of patients with an ATU Score less than experienced VR at week compared to 0% of patients with an ATU Score of or greater (p=0.00). Using the Stanford Score, % vs. % vs. % of patients responded with increasing levels of resistance to LPV/r (susceptible+potential low-level resistance vs low-level resistance vs intermediate+high-level resistance) (p=0.000). Univariate Analysis In the univariate analyses, mutations at positions 0, and were significantly associated with lack of VR (p=0.00, p=0.00, and p=0.0, respectively). Additionally, there was a trend towards a significant association with lack of VR and mutations at positions 0 and 0 (p=0.0 and p=0.0, respectively) (Table ) 0 Multivariate analysis

15 0 0 In the multivariate analysis, improved VR was associated with lower baseline viral load. In addition, lack of VR was associated with mutations at positions, and 0. (Table ) Evaluation of Scoring Systems ROC curves for each of the scores are presented in Figure. Each of the genotypic interpretation scores appears to have performed reasonably well in our cohort. AUCs for the ROC curves were similar, with values ranging from and their % confidence intervals having substantive overlap. (Table ) Prediction error and misclassification rates were calculated which also showed no clear advantage for any genotypic interpretation score (data not shown). Discussion With the availability of newer protease inhibitors, it becomes increasingly important to know when LPV/r will have full activity. Prior to the approval of tipranavir and darunavir, oftentimes, patients with viral isolates having high level resistance to LPV/r received LPV/r due to limited alternative options. The TITAN study suggests that patients with limited LPV resistance do better on darunavir (). However, the superiority of darunavir (%) vs. LPV (%) in patients achieving viral load<00 copies/ml at weeks was primarily driven by the % failure rate in the LPV/r arm in patients with viral isolates with LPV fold change (FC) >0 (vs.

16 0 0 % failure rate for the darunavir group in patients with viral isolates with LPV FC>0). There was no significant difference in response for patients with viral isolates with LPV FC<0. This phenotypic breakpoint is roughly approximated by the genotypic breakpoints used in the various scoring systems (). Various resistance scores have been derived that posit to be superior to other available scoring systems in predicting susceptibility to LPV/r. However, no independent data set has been used to evaluate the relative merits of the different scores. The score created based on a given dataset will always outperform a score created using a different dataset as there is always some degree of model over- fitting. Even if a portion of the dataset is used as a training dataset and the remainder is used as a validation dataset, the similarities of the treatment histories between the training and validation datasets will likely bias the evaluation in favor of one's own scoring system. This occurs because the treatment history of the patient population has a major impact on which mutations will be found to be associated with LPV/r resistance. For instance, in the ViroLogic Score, the protease mutation 0V was found to be associated with decreased LPV susceptibility (). Our study did not find this correlation as only % of the patients from our cohort had received amprenavir which selects for this mutation. By using an independent dataset to evaluate the relative merits of multiple different resistance scores, we

17 0 have removed the inherent bias that is created when a dataset is used to both create and test the performance of a scoring system. In our study, % of the patients achieved a viral load of less than 00 copies/ml at weeks. Currently, the gold standard for combination antiretroviral therapy is to achieve a viral load of less than 0 copies/ml, because suppression to this level has been found to be most durable (). However, at the time the data from our cohort were ascertained (00-00), viral load assays sensitive to 0 copies/ml were not widely used. All the scores with defined breakpoints (Lopinavir Mutation Score, ViroLogic, ATU, and Stanford) effectively predicted patients who achieved VR. Using any of these scores, over 0% of patients who had viral isolates which fell below the breakpoints achieved VR, while 0% or less of patients with viral isolates at or above the breakpoints achieved VR. 0 With our relatively small dataset, we did not identify uncommon mutations that may be associated with reduced VR to LPV/r. The results from our multivariate analysis showed an association between mutations at positions,, and 0 in protease and lack of VR. Additionally, in the univariate analysis, mutations at positions 0, and were associated with a lack of VR. Mutations at position are often

18 0 considered more important than mutations at positions,, and 0 (,, 0). In our dataset, mutations at frequently occurred in conjunction with V, thus likely obscuring the effect of a mutation at position in the multivariate analysis. The five mutations found to be significant in our univariate or multivariate analyses constitute a subset of the mutations included in the Lopinavir Mutation Score and are found relatively commonly in PI-experienced individuals (). Comparing the shape of the ROC curves and the corresponding AUCs of the different models, all the models performed reasonably well without any one being clearly superior. This is similar to the finding in a recent paper by Fox, et al. which compared different genotypic interpretation scores for protease inhibitors and found that all of the scores effectively predicted viral load response to a diverse set of protease inhibitors and that none of the scores performed significantly better than the others (). 0 One limitation of this study is not all the mutations included in the scores were captured by our data acquisition form. While all of the mutations used in the Lopinavir Mutation Score and ATU Score were collected, we did not acquire information on some of the mutations used in the ViroLogic, Stanford, and IAS- USA Scores. This could lead these latter three scores to under-perform compared to the Lopinavir Mutation Score and ATU. However, the mutations that were not

19 0 0 collected are relatively uncommon. In the ViroLogic Score, we collected data on mutations which are on average present in % of PI-experienced patients. The mutations for which we do not have data are present in only % of PI-experienced patients. Similarly, for the Stanford and IAS-USA mutations, the mutations for which we do not have data are present in % and % of PI-experienced patients, respectively (). In conclusion, deciding which genotypic interpretation score to use for predicting response for LPV/r can be intimidating to the average clinician. However, we found that the Lopinavir Mutation, ATU, ViroLogic, Stanford, and IAS-USA Scores are all able to distinguish between patients who are more or less likely to respond to LPV/r. There was no substantial difference in the performance of the scores when compared using data from our independent clinical cohort. From this study we can, therefore, recommend to clinicians any of the scores evaluated as effective tools in deciding when LPV/r will be active. The fact that all scoring systems tested were roughly equivalent in their performance, should provide some comfort to prescribing clinicians who may be daunted by the task of choosing between the various genotypic scoring systems that are available. From this analysis, using an independent clinical cohort, we conclude that all of the scores evaluated are equally predictive of virologic response.

20 0 Acknowledgments Financial support for this project was provided by Abbott Laboratories, Abbott Park, IL. References. Agence nationale de rechurches sur le SIDA (ANRS). Accessed January, 00. ANRS genotypic resistance guidelines. Available at: Benjamini Y, Hochberg Y.. Controlling the false discovery rate: a practical and powerful approach to multiple testing. J Roy Stat Soc B. :-00.. Clotet B, Bellos N, Molina JM, Cooper D, Goffard JC, Lazzarin A, Wöhrmann A, Katlama C, Wilkin T, Haubrich R, Cohen C, Farthing C, Jayaweera D, Markowitz M, Ruane P, Spinosa-Guzman S, Lefebvre E; POWER and study groups. 00. Efficacy and safety of darunavir-ritonavir at week in treatment-experienced patients with HIV- infection in POWER and : a pooled subgroup analysis of data from two randomised trials. Lancet :-. 0 0

21 0. Dunn DT, Green H, Loveday C, Rinehart A, Pillay A, Fisher M, McCormack S, Babiker AG, Darbyshire JH; Evaluation of Resistance Assays (ERA) Trial Investigators. 00. A randomized controlled trial of the value of phenotypic testing in addition to genotypic testing for HIV drug resistance: evaluation of resistance assays (ERA) trial investigators. J Acquir Immune Defic Syndr. :-.. Efron B, Tibshirani R.. Introduction to the Bootstrap. Boca Raton: Chapman & Hall/CRC Press.. Fox ZV, Geretti AM, Kjaer J, Dragsted UB, Phillips AN, Gerstoft J, Staszewski S, Clotet B, von Wyl V, Lundgren JD. 00. The ability of four genotypic interpretation systems to predict virological response to ritonavir-based protease inhibitors. AIDS :0-0.. Genotype to phenotype correlations. Accessed January, 00. Available at: 0. Hicks CB, Cahn P, Cooper DA, Walmsley SL, Katlama C, Clotet B, Lazzarin A, Johnson MA, Neubacher D, Mayers D, Valdez H; RESIST investigator group. 00. Durable efficacy of tipranavir-ritonavir in combination with an

22 0 optimised background regimen of antiretroviral drugs for treatment-experienced HIV--infected patients at weeks in the randomized evaluation of strategic intervention in multi-drug resistant patients with tipranavir (RESIST) studies: an analysis of combined data from two randomised open-label trials. Lancet :-.. Johnson M, Beatriz G, Rodriguez C, Coco J, DeJesus E, Lazzarin A, Lichtenstein K, Rightmire A, Sankoh S, Wilber R. 00. Atazanvir plus ritonavir or saquinavir, and lopinavir/ritonavir in patients experiencing multiple virologic failures. AIDS :-. 0. Johnson VA, Brun-Vezinet F, Clotet B, Günthard HF, Kuritzkes DR, Pillay D, Schapiro JM, Richman DD. 00. Update of the drug resistance mutations in HIV-: 00. Top HIV Med. :-.. Kempf DJ, Isaacson JD, King MS, Brun SC, Sylte J, Richards B, Bernstein B, Rode R, Sun E. 00. Analysis of the virological response with respect to baseline viral phenotype and genotype in protease inhibitor-experienced HIV-- infected patients receiving lopinavir/ritonavir therapy. Antivir Ther. :-. 0

23 0. Kempf DJ, Isaacson JD, King MS, Brun SC, Xu Y, Real K, Bernstein BM, Japour AJ, Sun E, Rode R. 00. Identification of genotypic changes in Human Immunodeficiency Virus protease that correlate with reduced susceptibility to the protease inhibitor lopinavir among viral isolates from protease inhibitor-experienced patients. J Virol. :-.. King, MS, Rode R, Cohen-Codar I, Calvez V, Marcelin AG, Hanna G, Kempf D. 00. Predictive genotypic algorithm for virologic response to lopinavir- ritonavir in protease-inhibitor experienced patients. Antimicrob Agents Chemother. :0-0.. Madruga JV, Berger D, McMurchie M, Suter F, Banhegyi D, Ruxrungtham K, Norris D, Lefebvre E, de Béthune MP, Tomaka F, De Pauw M, Vangeneugden T, Spinosa-Guzman S; TITAN study group. 00. Efficacy and safety of darunavir-ritonavir compared with that of lopinavir-ritonavir at weeks in treatment-experienced, HIV-infected patients in TITAN: a randomised controlled phase III trial. Lancet 0:-. 0. Mocroft A, Vella S, Benfield TL, Chiesi A, Miller V, Gargalianos P, d'arminio Monforte A, Yust I, Bruun JN, Phillips AN, Lundgren JD..

24 0 Changing patterns of mortality across Europe in patients infected with HIV-. EuroSIDA Study Group. Lancet :-0.. Palella FJJ, Delaney KM, Moorman AC, Loveless MO, Fuhrer J, Satten GA, Aschman DJ, Holmberg SD.. Declining morbidity and mortality among patients with advanced human immunodeficiency virus infection. HIV Outpatient Study Investigators. N Engl J Med. :-0.. Panel on Antiretroviral Guidelines for Adults and Adolescents. Accessed August, 00. Guidelines for the use of antiretroviral agents in HIV--infected adults and adolescents. January, 00. Department of Health and Human Services. Available at: Parkin N, Chappey C, Petropoulos C. 00. Improving lopinavir genotype algorithm through phenotype correlations: novel mutation patterns and amprenavir cross-resistance. AIDS :-.. REGA genotypic resistance interpretation system. Accessed January, 00. Available at: 0

25 0. Stanford University HIV drug resistance database. Accessed January, 00. Avaliable at: 0. Stanford University HIV sequence database. Accessed January, 00. Available at Walmsley S, Bernstein B, King M, Arribas J, Beall G, Ruane P, Johnson M, Johnson D, Lalonde R, Japour A, Brun S, Sun E; M- Study Team. 00. Lopinavir-ritonavir versus nelfinavir for the initial treatment of HIV infection. N Engl J Med. :0-0.. Zolopa AR, Shafer RW, Warford A, Montoya JG, Hsu P, Katzenstein D, Merigan TC, Efron B.. HIV- genotypic resistance patterns predict response to saquinavir-ritonavir therapy in patients in whom previous protease inhibitor therapy had failed. Ann Intern Med. :-.

26 Table : Baseline Patient Demographics Gender (n=0) Male % Female % Median Age (n=0) (Range) (-) HIV Risk Factor (n=) Men who have sex with men CDC Stage (n=0) Race (n=) Heterosexual transmission Injection drug use Blood products Unknown A (Asymptomatic) B (Symptomatic) C (AIDS-defining illness) Caucasian Black % % % % % % % 0% % 0% Clinic Setting (n=0) Latino Asian Other US Italy % % % % %

27 Academic Public 0% % Private % Median Baseline log 0 VL (n=0) (SD). (0.) Baseline Median CD Count (n=0) (SD) () ART-experienced (n=0) % PI-experienced % NNRTI-experienced % NRTI-experienced Median # Protease Inhibitors (range) Amprenavir Indinavir Nelfinavir % % % % (0-) % % % Ritonavir Saquinavir Median #NNRTIs (range) Median #NRTIs (range) Median #PI mutations (range) Median #NNRTI mutations (range) % % (0-) (0-) (0-) (0-)

28 Median #NRTI muations (range) (0-) Table : Virologic Response by Subgroups a Overall /0 (%) ART exposure Clinic ART-naïve ART-experienced PI-naïve PI-experienced / (00%) p=0. /00 (%) / (%) p=0. 0/ (%) US Italy / (%) p=0. / (%) Academic Public Private / (%) p=0. / (%) / (%) Gender Male Female / (%) p=0. / (%)

29 Race Caucasian Black Latino Asian Other By genotypic susceptibility score < -<..-< + / (%) p=0. /0 (%) / (%) 0/ (0%) / (00%) / (%) p=0.00 0/ (%) / (0%) By baseline viral load <00,000 / (0%) / (0%) p=0.0 >00,000 /0 (0%) By CD count <00 >00 / (%) p=0. / (%) By Lopinavir Mutation Score < / (%) p=0.000

30 By ViroLogic Score Comparison of Genotypic Scores for LPV/r > /0 (0%) By ATU Score < > < > By Stanford Score Susc.+Pot. Low-level Res. Low-level Resistance Int.+High-level Res. / (%) p<0.000 / (%) / (%) p=0.00 0/0 (0%) / (%) p=0.000 / (%) / (%) a P value from chi-squared test. Table : Univariate Predictors of Virologic Response b Mutation Adjusted p-value 0 pi * pi pi 0. pi0 0.0 pi

31 0 pi 0. pi 0. pi pi 0. pi pi pi 0.00 * pi pi 0. pi 0.0 * pi 0. pi0 0.0 b P value from Fisher s Exact Test, adjusted by Benjamini-Hochberg procedure. * indicates P<0.0. Table : Multivariate Predictors of Virologic Response Variable Odds Ratio (% Wald Confidence Limits) 0 Log 0 VL 0.0 (0.0-0.) pi 0.0 ( ) pi 0.00 (< )

32 pi0 0.0 (0.0-0.) 0 Table : Area under Receiver-Operator Characteristic (ROC) Curve (% confidence interval) for Genotypic Interpretation Scores c Genotypic Interpretation Score Area under ROC Curve (% Confidence Interval) Lopnavir Mutation Score 0. (0.-0.) ViroLogic Score 0. (0.-0.) ATU Score 0. (0.0-0.) Stanford Score 0. (0.-0.) IAS-USA Score 0. (0.-0.) c An area under of the ROC curve of denotes a perfect test, while an area under the curve of 0. reflects a test without discriminatory power. The precision of the AUC estimates was evaluated by constructing % BCa bootstrap confidence intervals (B=000 replicates) ().

33 Figure : Frequency of Protease Inhibitor Mutations by Position Frequency of Mutation (%) Position of Protease Inhibitor Mutation

34 Figure : Receiver-Operator Curves for Genotypic Interpretation Systems d Sensitivity Specificity ATU IAS LMS Virologic Stanford d The ROC curves depict the tradeoff between the true positive rate (sensitivity) and false positive rate (-specificity) of each genotypic interpretation system as the clinical cutoff is varied.