AVT Figure Template

Uy 30/8/07 10:24 Page 957

Antiviral Therapy 12:957–962 HIV genotypic resistance testing to optimize antiretroviral prescribing: is there room for improvement?

Jonathan Uy1*, John T Brooks2, Rose Baker2, Mark Hoffman2, Anne Moorman3, Richard Novak1 and the HOPS investigators†

1Section of Infectious Diseases, University of Illinois College of Medicine, Chicago, IL, USA 2Cerner Corporation, Kansas City, MO, USA 3Division of HIV/AIDS Prevention, Centers for Disease Control and Prevention, Atlanta, GA, USA †Participants are listed in the additional file online

*Corresponding author: Tel: +1 609 897 4603; Fax: +1 609 897 6068; E-mail: [email protected]

Background: Clinical utilization of genotype resistance medication for >6 months after this finding. In 33% of testing is evolving. We examined the extent to which these instances, prescribers reported these actions were HIV care providers requesting genotype resistance tests erroneous oversights. For persons taking the resistant used the information appropriately and the impact of antiretroviral at the time of the genotype test, stopping inappropriate utilization. this medication within 6 months of the test produced Methods: Data from a prospective cohort of HIV-infected greater decreases in viral load (-1.35 versus -0.43 log patients (the HIV Outpatient Study) were used in the copies/ml, P=0.025) and a greater likelihood of analysis. We analysed the frequency with which patients achieving an undetectable viral load (25.3% versus were prescribed any non-nucleoside reverse transcriptase 7.3%, P=0.012) at 9 months. Changes in CD4+ T-cell inhibitor after identification of the K103N mutation in count differed (+22.8 versus -23.0 cells/mm3), but not reverse transcriptase and the frequency of prescription of significantly (P=0.167). nelfinavir after identification of the D30N mutation in Conclusions: Following evidence of definitive resistance HIV protease; the short-term impact of this action on HIV by genotype testing, a substantial fraction of antiretroviral load and CD4+ T-cell count was assessed. viral prescriptions were continued in error leading to an Results: Among 441 patients demonstrating either attenuated therapeutic response. These data highlight mutation, 18% who were taking the resistant antiretro- the need to consider better systems to manage genotype viral at the time of the test were continued on the resistance testing data in the clinical setting.

Introduction

Testing pathogens for antimicrobial resistance is a infection and has been incorporated into clinical strategy used widely to manage many infectious treatment guidelines [4]. diseases. Early biological phenotype assays for testing Clinicians’ knowledge and understanding varies resistance of HIV to antiretrovirals (ARVs) were regarding the appropriate utilization of this novel and labour-intensive, unreliable, and impractical for evolving test [5]. Familiarity with HIV genotype resis- routine clinical practice. In the late 1990s, genotype tance assays has increased, current reports are clearer resistance testing became available, and several and easier to read, algorithms to interpret genotype randomized trials [1–3] demonstrated that its use to resistance test results are accessible on the internet [6], guide changes in ARV therapy improved virological and innovative bioinformatic techniques are further outcomes compared with ARV changes without resis- improving genotype interpretation [7]. Still, the scien- tance testing. Subsequent approval by the Food and tific literature is sizeable, technical, and rapidly Drug Administration (FDA) of several genotype resis- changing, thus challenging clinicians’ ability to reach tance assays permitted expansion of the technology, consensus regarding the interpretation of genotype which is now available through most clinical labora- resistance test results. tory services. Genotype resistance testing has become To examine the extent to which care providers part of the standard of care for treatment of HIV requesting genotype resistance tests are using the

J Uy et al.

information appropriately, and the impact of inappro- incidence of ARV prescription inconsistent with priate utilization on outcomes, we analysed data genotype resistance testing results in two instances collected from a cohort of HIV-infected patients where there is broad agreement that a specific muta- receiving care at a selection of US medical practices tion confers definitive resistance: prescription of any participating in the HIV Outpatient Study (HOPS). We non-nucleoside reverse transcriptase inhibitor analysed the frequency with which patients were (NNRTI) after identification of the K103N mutation prescribed an ARV to which their virus demonstrated in reverse transcriptase and prescription of nelfinavir unequivocal resistance, the circumstances under which (NFV) after identification of the D30N mutation in these apparent inconsistencies occurred, and the the HIV protease. Although continuing NFV in the impact of these inconsistencies on subsequent HIV presence of the D30N mutation might reduce viral viral loads and CD4+ T-cell counts. fitness, this practice also promotes development of additional protease inhibitor resistance mutations Methods that could undermine future treatment options. We defined two categories of inconsistency The HOPS is an ongoing prospective, observational (Figure 1): Mismatch, remaining on an NNRTI or NFV cohort of HIV-infected patients seen at ten HIV- after a genotype resistance test reported a K103N or specialty clinics in seven US cities: Tampa; Washington, D30N mutation, respectively; and Misstart, starting on DC; Oakland; Denver; Chicago; Stony Brook, New an NNRTI or NFV under the same circumstances. York; and Philadelphia [8]. Since 1993, over 8,000 With regard to patients in the first category patients have been enrolled and followed at over (mismatch), to allow for lag time that might have 250,000 visits contributing more than 30,000 person- occurred between reporting and reviewing a genotype years of observation. Data are abstracted from resistance test result and changing a patient’s regimen patients’ medical records and entered directly into an accordingly, we only considered as a mismatch electronic database by trained abstractors. Abstracted prescriptions that continued >6 months after the date information includes basic demographic information, of the test. Patients in whom the ARV was stopped risk factors for HIV infection, and all symptoms, diag- within 6 months of the genotype resistance test result noses, medications and laboratory results. The data are date were considered to have been treated appropri- compiled centrally, regularly controlled for quality, and ately (no mismatch). We evaluated the impact of updated monthly. For genotype resistance testing, all mismatches in this group by comparing quantitative resistance mutations and ARV susceptibility interpreta- HIV RNA levels (viral loads) and CD4+ T-cell counts at tions on the report form are abstracted. The present baseline, defined as the date of the genotype resistance analysis is based on the HOPS database updated as of test result, and at 9 months thereafter between the two March 31, 2005. groups (mismatch versus no mismatch). Baseline values Recognizing that consensus on the interpretation were defined as the values closest in time to the date of of genotypes is evolving, we chose to examine the the genotype resistance test result, and ranged from 180 days before to 7 days after. Nine-month values were defined as the value closest to the date 9 months after the genotype resistance test result within the Figure 1. Algorithm used for defining mismatches and 6–12 month period after the test. misstarts in the HIV outpatients study, 1998–2005 With regard to patients in the second category, we considered as misstarts initiation of treatment with an Resistance mutation present NNRTI or NFV ≥4 weeks after the date that the corre- at first known genotype test sponding resistance mutation was first reported and that continued for ≥7 days. We then reviewed all available genotype resistance tests between the date on which the Off drug to which On drug to which resistant by genotype resistant by genotype resistance mutation was first detected and the date of misstart to determine whether there might have been (6 months) intervening genotype tests indicating the mutation of interest was absent (that is, the virus had reverted to wild type) on the premise that such reversions could have led Drug Drug ‘Mismatch’ stopped continued to the prescribing of an NNRTI or NFV despite the likely permanent significance of archived resistant virus. We ≥ > ( 4 weeks) (Drug later stopped for 4 weeks) could not analyse the clinical impact of misstarts in terms + Drug started or restarted ‘Misstart’ of their impact on subsequent changes in CD4 T-cell count and viral load, because of the difficulties inherent

HIV genotypic resistance testing to optimize antiretroviral prescribing

in defining valid comparison patients from this dataset Patients in the mismatch group with NNRTI resistance given the large variance in the lengths of time between were significantly more likely to have been infected first reporting of the resistance mutation and the misstart. through high-risk heterosexual activity, have AIDS, have For determining whether an inconsistency with lower baseline CD4+ T-cell counts at the time the resis- treatment occurred, we considered in our analysis all tance mutation was detected, and to have experienced genotypes from 1998 (when genotype resistance tests lower nadir CD4+ T-cell counts. Mismatch patients were became widely available) to 2004. Genotype resistance also more likely to be female, non-white, publicly tests were excluded if patients did not have a visit insured and to have taken a greater median number of recorded ≥6 months after the test or if the test ARVs, but these differences were not statistically signif- occurred after 1st October, 2004, in order to allow icant (Table 1). Patients in the NFV mismatch group adequate time for follow up to evaluate treatment were more likely to have been male, non-white, a man decisions. Patients inactive as of 31st March, 2005 who had sex with other men (MSM), privately insured with open antiretroviral treatment dates had treat- and younger; however, none of theses differences were ments censored 90 days past their date of last contact statistically significant. (that is, after the typical maximum length of prescrip- The clinicians caring for the 42 total mismatch tion) or at 31st March, 2005, whichever occurred first. patients were queried and in 33 (79%) cases provided We included in our final analysis the first genotype responses regarding reasons for continuing the ARV result for any patient who reported either a K103N or despite genotypic evidence of resistance. For 12 (36%) D30N mutation. of these patients, the ARV was prescribed in error. For Analyses were performed using SAS version 8.2 eight (24%) patients, the ARV was prescribed because (Cary, NC, USA). The Wilcoxon rank sum test (for there were limited other options. For seven (21%) continuous variables), and Fisher’s exact test or the patients, the ARV was prescribed because of adherence Pearson χ2 test (for discrete variables) were used in issues or patient preferences. Reasons for the remaining unadjusted comparisons of patients in the mismatch 6 patients included clinicians not being aware of the analysis. For patients in the mismatch analysis, we also genotype test results or the results not being available, used general linear modelling to assess the effect of age, participation in a clinical trial and slight delays in nadir CD4+ T-cell count and number of ARV drug changing the ARV regimen to a time beyond the 6 exposures prior to the genotype resistance test on the month window. Clinicians were unable to locate two continuous outcomes (mean change in CD4+ T-cell records with information for nine of the 42 patients. count and mean change in log HIV viral load) and on Mismatches occurred at all HOPS sites. one discrete outcome (achieving an undetectable viral We compared the impact of mismatches on viral load). We adjusted each model for baseline CD4+ T-cell loads and CD4+ T-cell counts among those patients with count or log HIV viral load, depending on the outcome paired measurements at baseline and at 9 months. For being evaluated. A variable was considered to be signif- this analysis, we combined patients with both the icantly associated with the outcome of interest if the K103N and D30N resistance mutations. Four mismatch P-value was <0.05. patients each had two genotypes collected at different times in which one genotype indicated resistance to Results NNRTIs and the other resistance to NFV; we included only the data from the first resistant result so that each Among 2,122 genotype test results performed on patient contributed to only one set of observations in 1,269 patients during the study period, 2,016 test the analysis. In bivariate analyses, persons in the no results involving 1,218 patients had adequate follow- mismatch group compared with persons in the up data to permit evaluation of treatment decisions. mismatch group experienced greater decreases in log Of the 374 patients who had ≥1 genotype with a viral load (-1.35 versus -0.43, P=0.025), and were more K103N result, 198 (53%) were on an NNRTI at the likely to achieve an undetectable viral load (25.3% time this mutation was first reported. Among these versus 7.3%, P=0.012). CD4+ T-cell count increases at 9 patients, 163 (82%) had their NNRTI discontinued months were greater in the no mismatch group, but the within 6 months and the remaining 35 (18%) patients difference was not statistically significant (Table 2). continued the NNRTI past 6 months. Of the 67 Changes in viral load and CD4+ T-cell count were patients who had ≥1 genotype with a D30N result, 41 evaluated in multivariable analyses, which included (61%) were on NFV at the time this mutation was variables for treatment group (no mismatch versus first reported. Among these patients, 34 (83%) had mismatch), age, nadir CD4+ T-cell count, number of their NFV discontinued within 6 months and the ARV exposures before genotype testing, and either remaining seven (17%) patients continued NFV past baseline CD4+ T-cell counts or viral load, depending 6 months. on the clinical outcome being evaluated. We also

Antiviral Therapy 12:6 959 Uy 30/8/07 10:24 Page 960

J Uy et al.

Table 1. Demographic and clinical characteristics of patients in the HIV Outpatient Study with a K103N or D30N mutation, 1998–2005 First instance of K103N mutation First instance of D30N mutation while receiving an NNRTI while receiving NFV NNRTI discontinued NNRTI continued NFV discontinued NFV continued Characteristic (no mismatch) n=163 (mismatch) n=35 P-value (no mismatch) n=34 (mismatch) n=7 P-value

Gender male, n (%) 135 (82.8) 25 (71.4) 0.120 27 (79.4) 6 (85.7) 1.000 Race, n (%) 0.440 0.410 White 97 (59.5) 17 (48.6) 18 (52.9) 2 (28.6) African-American 44 (27.0) 13 (37.1) 0 (–) 0 (–) Other/unknown 22 (13.5) 5 (14.3) 16 (26.5) 5 (71.4) HIV risk behavior, n (%) 0.020 0.685 MSM 100 (61.4) 19 (54.3) 15 (44.1) 4 (57.1) High-risk heterosexual 33 (20.2) 14 (40.0) 0 (–) 0 (–) Other/unknown 30 (18.4) 2 (5.7) 19 (26.5) 3 (42.9) Insurance, n (%) 0.149 0.419 Private 86 (52.8) 14 (40.0) 17 (50.0) 5 (71.4) Public 60 (36.8) 19 (54.3) Other/unknown 17 (10.4) 2 (5.7) 17 (50.0) 2 (28.6) AIDS, n (%) 121 (74.2) 32 (91.4) 0.028 19 (55.9) 4 (57.1) 1.000 Median age, 41.7 (37.0–47.6) 40.2 (36.6–47.7) 0.577 42.6 (34.2–49.1) 39.4 (35.9–43.6) 0.931 years (IQR) Median baseline CD4+ T-cells, 280 (176–485) 174 (128–373) 0.046 416 (279–540) 403 (262–913) 0.765 cells/mm3 (IQR) Median nadir CD4, 163 (56.1–275) 70 (21–163) 0.011 228.5 (36.5–324) 223 (98.1–913) 0.599 cells/mm3 (IQR) Median number of ARV 7 (5–10) 8 (7–10) 0.172 4 (3–6) 3 (3–5) 0.519 drugs experienced (IQR)

ARV, antiretroviral; IQR, interquartile range; MSM, men having sex with other men; NFV, nelfinavir; NNRTI, non-nucleoside reverse transcriptase inhibitor.

Table 2. Clinical impact of continuing use of an NNRTI or NFV (to which a resistance mutation was reported) for 9 months after the report among patients in the HIV Outpatient Study, 1998–2005 Discontinued NNRTI/NFV <6 months Continued NNRTI/NFV >6 months Unadjusted Clinical factor after finding mutation (no mismatch) after finding mutation (mismatch) P-value

Mean CD4+ T-cell count change*, 22.8 -23.0 0.167 cells/mm3 Mean viral load change†, -1.35 -0.43 0.025 log copies/ml Patients achieving an 25.3 7.3 0.012 undetectable viral load‡, %

*Among 152 no mismatch and 30 mismatch patients with available data. †Among 154 no mismatch and 30 mismatch patients with available data. ‡Among 194 no mismatch and 41 mismatch patients with available data. NFV, nelfinavir; NNRTI, non-nucleoside reverse transcriptase inhibitor.

included sex, race and insurance status in the that treatment group and baseline log viral load were adjusted analysis; however, because inclusion of these significant predictors of change in viral load variables did not significantly alter the initial results, (P=0.020 and P<0.001, respectively). Similarly, treat- these variables were excluded from the final analysis ment group and baseline viral load were significant to keep the model as parsimonious as possible. We predictors of whether an undetectable viral load was found that treatment group and nadir CD4+ T-cell achieved (P=0.030 and P=0.012, respectively; count were significant predictors of changes in CD4+ Table 3). The mismatch group was one fourth as T-cell count (P=0.049 and P<0.001, respectively) and likely to achieve an undetectable viral load as

HIV genotypic resistance testing to optimize antiretroviral prescribing

Table 3. Multivariable analysis of factors associated with changes in CD4+ T-cell count and viral load after finding a K103N or D30N mutation while on an NNRTI or NFV, respectively, in the HIV Outpatient Study, 1998-2004 Mean change in Mean change Achieved undetectable Clinical factor CD4+ T-cell count P-value in log viral load P-value viral load P-value

‘No mismatch’ treatment group 0.049 0.020 0.030 Age 0.906 0.194 0.129 Nadir CD4+ T-cell count <0.001 0.469 0.253 Number of prior ARV exposures 0.529 0.868 0.329 Baseline CD4+ T-cell count 0.516 – – Log baseline viral load – <0.001 0.012

ARV, antiretroviral; NFV, nelfinavir; NNRTI, non-nucleoside reverse transcriptase inhibitor.

patients in the no mismatch group (odds ratio 0.254, and heterogeneous nature of genotype resistance 95% confidence interval 0.073–0.89). testing utilization, which continues to evolve with new Of 374 patients with a K103N mutation, 56 (15%) knowledge and ARV drugs. We also studied only two started or restarted an NNRTI ≥4 weeks after this resistance mutations and not the full spectrum of HIV mutation was first reported (misstart), of whom 14 genotype data. As with most studies examining clinical (25%) patients were started on an NNRTI multiple endpoints and HIV resistance testing, we were able times after this result. The median time from the resis- only to examine short-term correlates and were not tance finding to NNRTI initiation was 480 days able to study impact on long-term changes in CD4+ (interquartile range [IQR]=168–818 days). Among T-cell counts or morbidity and mortality. Studying these 56 patients, 19 (34%) had a subsequent genotype only patients in HIV-specialty clinics could also under- that reported no K103N mutation (that is, reversion to represent the incidence and impact of these errors. wild type) after the initial K103N result and prior to As the use of HIV resistance testing expands, as new starting an NNRTI. Three (4%) of 67 patients with a ARVs become available and as algorithms for inter- D30N mutation started or restarted NFV ≥4 weeks preting genotype resistance tests become more after this mutation was first reported. The median time complex, the risk that ARVs will be prescribed after from the resistance finding to NFV initiation was 467 genotypic evidence of resistance might increase. days (IQR=30–855). All of these patients’ subsequent Reporting results in clear formats and educating clini- genotype tests demonstrated the D30N mutation (that cians on the interpretation and use of genotype resis- is, no reversions to wild type). Although technical limi- tance tests could ameliorate some of these risks. Better tations of our observational data did not permit analysis access to all prior genotype resistance test results, such of the clinical impact of misstarts by comparison to an as through centralized electronic medical record equivalent group of patients with no misstart, we did systems [9], could decrease prescription of an ARV in observe that among 32 misstart patients with paired situations where a historical genotype resistance test baseline and 6-month values, the change in log viral showed resistance but more recent subsequent tests load was only -0.06. indicate reversion to wild type. Ultimately, automated decision support systems integrated with an electronic Discussion medical record [9] could prove beneficial, especially where expertise in genotype resistance testing is Among patients in the care of a group of experienced lacking, such as in resource-limited settings. Until such HIV clinicians, 18% continued to be treated with an systems are widely implemented, manual systems to ARV despite genotypic resistance tests that demon- follow up the results of and resulting actions from HIV strated definitive resistance to the medication. Our resistance tests, possibly as part of practice quality analysis showed that prescription of ARVs inconsistent improvement programs, might be able to increase the with genotype resistance testing was associated with a likelihood that HIV clinicians utilize these valuable but significantly attenuated reduction in viral load. Unlike complex tests optimally. previous studies that examined the impact of an HIV resistance test on virological outcomes, this study is the Acknowledgements first to look at outcomes based on clinician responses to resistance test results. The authors acknowledge financial support from the Limitations of this observational cohort study Centers for Disease Control and Prevention (contract include selection bias, particularly given the complex 200-2001-00133) and the Agency for Healthcare

Antiviral Therapy 12:6 961 Uy 30/8/07 10:24 Page 962

J Uy et al.

Research and Quality (Grant 1R18HS011800-01. The be accessed via the Volume 12 Issue 6 contents page findings and conclusions from this review are those of for Antiviral Therapy, which can be found at the authors and do not necessarily represent the views www.intmedpress.com (by clicking on ‘Antiviral of the Centers for Disease Control and Prevention. Therapy’ then ‘Journal PDFs’.

Conflicts of interest References Dr J Uy has served as a consultant for and/or is on the 1. Durant J, Clevenbergh P, Halfon P, et al. Drug-resistance speakers’ bureau of Abbott Laboratories, Bristol- genotyping in HIV-1 therapy: the VIRADAPT randomised Myers Squibb, Gilead and GlaxoSmithKline and has controlled trial. Lancet 1999; 353:2195–2199. received research grants from Abbott Laboratories, 2. Baxter JD, Mayers DL, Wentworth DN, et al. A randomized study of antiretroviral management based on plasma geno- Bristol-Myers Squibb and GlaxoSmithKline. Dr typic antiretroviral resistance testing in patients failing R Novak has served as a consultant for Abbott therapy. CPCRA 046 Study Team for the Terry Beirn Laboratories, Boehringer-Ingelheim, Pfizer, Tibotec and Community Programs for Clinical Research on AIDS. AIDS 2000; 14:F83–F93. Gilead; and has received research funding from Wyeth, 3. Cohen C, Hunt S, Sension M, et al.A randomized trial GlaxoSmithKline and Merck. R Baker and M Hoffman assessing the impact of phenotypic resistance testing on are employed by Cerner Corporation which has had antiretroviral therapy. AIDS 2002; 16:579–588. contracts with Gilead, Amgen, Bristol-Myers Squibb, 4. Hirsch MS, Brun-Vezinet F, Bonaventura C, et al. Antiretroviral drug resistance testing in adults infected with Novartis, Berlex, Allergan, Sanofi, Roche, Wyeth, human immunodeficiency virus type 1: 2003 recommenda- GlaxoSmithKine, Cephalon, Schering-Plough and tions of an international AIDS Society-USA panel. Clin Serono. All other authors have no conflicts of interest. Infect Dis 2003; 37:113–128. 5. Salama C, Policar M, Cervera C. Knowledge of genotypic resistance mutations among providers of care to patients Ethical considerations with human immunodeficiency virus. Clin Infect Dis 2003; The investigation followed the guidelines of the US 36:101-104. Department of Health and Human Services regarding 6. Shafer RW. HIV Drug Resistance Database. http://hivdb.stanford.edu/, accessed February 2007. protection of human subjects. The study protocol was 7. Larder BA, Kemp SD, Hertogs K. Quantitative prediction approved and renewed annually by each participating of HIV-1 phenotypic drug resistance from genotypes: the institutions’ ethical review board. All study participants virtual phenotype (VirtualPhenotype). Antivir Ther 2000; 5 Suppl 3:49. provided written, informed consent. 8. Palella FJ Jr, Delaney KM, Moorman AC, et al.Declining morbidity and mortality among patients with advanced Additional files human immunodeficiency virus infection HIV Outpatient Study Investigators. N Engl J Med 1998; 338:853-860. 9. Bates DW, Ebell, M, Gotlieb E, et al. A proposal for The additional file ‘The HIV Outpatient Study electronic medical records in U.S. primary care. J Am Med (HOPS) Investigators, December 2006–present’ can Inform Assoc 2003; 10:1–10. Accepted for publication 31 May 2007