The correlation between plasma concentrations of protease inhibitors, medication adherence and virological outcome in HIV-infected patients

Transcription

1 Antiviral Therapy 9: The correlation between plasma concentrations of protease inhibitors, medication adherence and virological outcome in HIV-infected patients Julie M Yasuda 1, Chris Miller 2,3, Judith S Currier 4, Donald N Forthal 5, Carol A Kemper 6, Gildon N Beall 7, Jeremiah G Tilles 5, Edmund V Capparelli 2, J Allen McCutchan 2, Richard H Haubrich 2,3 * and the California Collaborative Treatment Group (CCTG) 1 San Diego State University, San Diego, Calif., USA 2 University of California San Diego, San Diego, Calif., USA 3 CCTG Data and Biostatistical Unit, San Diego, Calif., USA 4 University of California Los Angeles, Los Angeles, Calif., USA 5 University of California Irvine, Irvine, Calif., USA 6 Santa Clara Valley Medical Center, San Jose, Calif., USA 7 Harbor-UCLA Medical Center, Los Angeles, Calif., USA *Corresponding author: Tel: ; Fax: ; rhaubrich@ucsd.edu Background: Although adherence clearly influences response to antiretroviral therapy (ART), accurate assessment of adherence is problematic. The objective of this analysis was to assess the independent predictive value of protease inhibitor (PI) concentrations as a supplement to self-report as markers of medication adherence. Methods: This retrospective analysis was conducted from a prospective clinical trial designed to compare the outcomes of frequent versus infrequent HIV RNA measurement used to manage antiretroviral therapy. For 131 patients, self-reported medication adherence, HIV RNA levels, CD4 counts and PI concentrations (unannounced, random samples) were measured at baseline (when patients changed to a new regimen) and every 2 months thereafter. The change in HIV RNA from baseline to month 6 (area-based measure) was used to evaluate overall response. The proportion of measured PI concentrations below the detection limit was used as an alternative marker of adherence. An undetectable concentration would be expected after missing a single dose. Results: The mean baseline CD4 count was 125 cells/mm 3 and the mean HIV RNA level was 4.7 log 10 copies/ml. The mean change in log 10 HIV RNA was copies/ml. The mean percentage of self-reported adherence was 91% (range: %) and the mean proportion of undetectable PI concentrations was 27% (range: 0 100%, mean 2.5 samples/patient). The correlation between the two measures was 0.23 (P=0.009). In a multivariate model, percentage of visits with undetectable PI concentrations (P=0.02), percentage of medication adherence (P=0.02), baseline HIV RNA level (P=0.005), prior PI use (P=0.0004), prior lamivudine (3TC) use (P=0.0009) and randomization to the frequent HIV RNA measurement group (P<0.0001) were all related to change in HIV RNA. After accounting for adherence, patients who always had detectable PI concentrations had an average of 0.4 log 10 additional HIV RNA reduction compared with those who had no detectable concentrations. Conclusions: Repeated, random PI concentration values are independently predictive of virological response and may add to self-report of adherence in understanding the response to ART. Introduction The development of highly active antiretroviral therapy (HAART) has had a major impact on the course of HIV infection, resulting in a significant decrease in opportunistic infections and an improvement in survival for infected patients [1]. Although it has been well demonstrated that antiretroviral agents, especially the inhibitors of the HIV protease, have had a therapeutic benefit both in clinical trials and in practice, therapeutic failures have also occurred. Well-known risk factors for antiretroviral therapy (ART) failure include having a high baseline viral load, having a low baseline CD4 cell count, having been previously treated with another protease inhibitor (PI), having inadequate concentrations of ART drugs and being non-adherent to a 2004 International Medical Press /02/$

2 JM Yasuda et al. medication regimen [2 11]. Among these factors related to treatment failure, non-adherence is a primary factor influencing ART response [12,13]. Non-adherence to study medications in clinical trials can bias study results and reduce the sensitivity of statistical tests of treatment differences. Thus, clinical trials testing the efficacy of anti-hiv medications would benefit from a standard method to measure medication adherence in order to determine which participants in the study actually took the drug regimen as prescribed. There is currently no standard method for measuring a patient s adherence [14,15], although the one most commonly used because of its low cost and convenience is the self-report questionnaire. This analysis was initiated to determine if a combination of methods to assess adherence might provide a more accurate estimate than using a single method alone. In this analysis, we assessed the independent predictive value of PI concentrations as a supplement to self-reported adherence in explaining the magnitude of virological response to ART. Methods The data used for this analysis represent a subset of HIV-infected patients enrolled in a 12-month prospective clinical trial to determine the utility of monitoring individual patients HIV RNA levels (California Collaborative Treatment Group 570). The details of the study design have been published [16], but, in brief, patients were randomized 1.5:1 for frequent or infrequent feedback of viral load results. All patients had plasma HIV RNA and CD4 cell counts measured at baseline (when a new antiretroviral regimen was started) and every 2 months thereafter. A central laboratory determined HIV RNA levels, which were provided to the treating physicians. Study visits occurred every 2 months, during which antiretroviral regimens and clinical events were recorded. All patients signed an informed consent approved by the Human Subjects Committee at each institution. Patients who were included in this analysis received a PI regimen at baseline, had at least one PI concentration measured throughout the study period, completed at least one adherence questionnaire during either month 2 or month 6 of the study period and had HIV RNA data measured up to 6 months. Protease inhibitor concentrations Plasma concentrations of PIs were measured at baseline and every 2 months thereafter for indinavir, nelfinavir, saquinavir or ritonavir by specific HPLC methods with UV detection. Briefly, the validated reverse phase (RP)-HPLC method measured all drugs simultaneously using a phosphate buffer mobile phase at ph 4.0 and acetonitrile (63:37 v/v) with UV detection at 205 nm [17]. The laboratory carries out proficiency testing four times a year with the international inter-laboratory quality control program for therapeutic drug monitoring (TDM) and the Aids Clinical Trials Group (ACTG) quality assurance/quality control (QA/QC) patient program. Patients were unaware that their medication concentrations were being measured, so the time of the last medication dose was not available. For analysis, the time of last medication dose was imputed to be 7.30 am based on surveys of medication dosing [18], and the time after dose was calculated to be the time from 7.30 am to the time of the blood draw. Monte Carlo simulations based on phase I pharmacokinetic studies were performed to generate the distribution of drug concentrations throughout the dose interval. Measured concentrations were compared with these distributions to determine the percentage of the population mean as well as the expected population percentile that the measured drug concentrations represented. From this analysis, several metrics of exposure were evaluated: (i) the percentage of visits with PI concentrations below the quantitative limits of the assay used (lower limits of quantification were as follows: indinavir, 0.1 µg/ml; saquinavir, µg/ml; ritonavir, 0.2 µg/ml; and nelfinavir, 0.2 µg/ml); (ii) the percentage of visits with the PI concentration ratio (observed concentration over the predicted concentration) less than 20%; (iii) the percentage of visits with PI concentrations less than the 5th, 20th and 25th percentile of the population predicted value; and (iv) similar to (iii), but using log-transformed values to account for the skewed nature of the concentration data. The percentiles (for example, 25th percentile) are taken from the population predicted models and yield the concentration at any time point where a given percentage (25%) of the population is below that concentration. Since the standard deviation for the concentration at any time point does vary, the percentiles may or may not be the same as the percentage of predicted concentration. When saquinavir plus another PI were in a regimen, only saquinavir concentrations were analysed and drug interactions (concentration boosting) were taken into account when determining the predicted saquinavir concentrations. Adherence questionnaire A self-administered questionnaire developed for this study was completed at baseline and at months 2 and 6. To promote patient candour, providers were blinded to patient responses. Questionnaires were administered and collected in sealed envelopes by study coordinators and delivered unopened to the data centre. The questionnaire asked about ART adherence, reasons for missing doses, recreational use of drugs or alcohol and International Medical Press

3 Two measures of adherence predict antiretroviral outcome activities limited by side effects from their medication in the previous 4 weeks. Patients were asked to estimate the proportion of prescribed medication they had taken in the previous 4 weeks as: <20%, 20 40%, 40 60%, 60 80%, 80 95% or >95%. Numerical scores were assigned to the categories as 0%, 20%, 40%, 60%, 80% and 95%, respectively. In a separate question, patients noted reasons for missing doses, with the first response being never missed pills. Patients who indicated both >95% medication taken and never missed pills were given an adherence score of 100%. Adherence scores were averaged for patients who completed multiple questionnaires during the study. The percentage of self-reported adherence over the study period was analysed as a continuous variable. Questions about the percentage of time during which a patient s activities were limited by medication side effects were similarly collected and analysed. The questionnaire asked whether recreational drugs or alcohol were used to feel better in the previous 4 weeks. Patients who responded in the questionnaire at either month 2 or month 6 that they used drugs or alcohol were considered to have used alcohol or drugs during the study period. Using alcohol or drugs was analysed as a dichotomous variable. Statistical methods All HIV RNA values were log 10 -transformed for analysis. The integrated area of the change from baseline of log 10 HIV RNA level was calculated using the trapezoidal rule, and was used to evaluate overall response. An HIV RNA area change from baseline of 1.0 was equivalent to having a sustained 1 log 10 reduction in HIV RNA over the study period of 6 months. Age, baseline CD4 counts, baseline HIV RNA levels, variables relating to the PI plasma concentrations, percentage of self-reported medication adherence to drug regimens and percentage of activities limited by side effects were analysed as continuous variables. Univariate analyses were performed using simple linear regression for continuous variables and one-way analysis of variance (ANOVA) for categorical variables. The correlation between the percentage of visits with undetectable PI concentrations and the percentage of medication adherence was determined by the Pearson s correlation coefficient. All variables were considered significant at the α=0.05 level. A multivariate linear regression model was constructed using the change in the HIV RNA level from baseline to month 6 (area-based measure) as the primary outcome variable. The percentage of visits with undetectable serum PI concentrations, level of selfreported adherence to a drug regimen, baseline HIV RNA levels and any other variables that were significantly related to the change in HIV RNA level from baseline to month 6 in the univariate analyses were simultaneously entered into the linear regression model. Data were analysed using the SAS System for Windows Version 8.1 (SAS Institute, Inc., Cary, NC, USA). Results Of 206 patients enrolled in the study, 131 had HIV RNA values at month 6. All 131 had plasma concentrations measured and were included in the analysis. Table 1 summarizes the baseline demographic characteristics for these 131 patients. Ninety-one percent of the population was male and most (70%) are believed to have acquired HIV through homosexual contact. Patients were predominantly Caucasian (47%) and Hispanic (38%) with African American patients comprising 11% of the sample. The mean baseline HIV RNA level was 4.7 log 10 copies/ml and mean CD4 cell count was 125 cells/mm 3. The majority of the patients in this study had been previously treated with ART (for a mean of 26 months), with only 6% of patients reporting being naive to ART at baseline. Thirty-seven Table 1. Baseline demographic characteristics Characteristic n* % Sex Female 12 9 Male Ethnicity African American Hispanic/Latino Asian/other 5 4 Caucasian Risk factors for HIV Homosexual contact Heterosexual contact Received blood products 7 5 Injection drug use Other risk factor 6 5 Mean age (range) 40 (22 63) Treatment group Infrequent HIV RNA monitoring Frequent HIV RNA monitoring Alcohol or drug use in past month Mean log 10 HIV RNA (range) 4.7 ( ) Mean CD4 cell count (range) 125 (3 515) Prior PI use Mean months prior PI use (range) 2 (0 18) Prior 3TC use Mean months prior 3TC use (range) 6 (0 23) Antiretroviral naive 8 6 Mean months prior treatment (range) 26 (0 95) *Two participants declined to state their age, gender and ethnicity. Nine declined to state alcohol or drug use at baseline. Antiviral Therapy 9:5 755

4 JM Yasuda et al. percent of patients had taken a PI prior to study entry and 82% reported having taken lamivudine (3TC). All patients started a PI-based regimen at baseline: 68% started indinavir (800 mg every 8 h), 5% nelfinavir (750 mg three times daily), 12% saquinavir (600 mg three times daily), 13% ritonavir (600 mg twice daily) and 3% ritonavir plus saquinavir (400 mg of each twice daily). The mean self-reported adherence rate was 91% (range: %). Of 328 PI concentrations measured (mean 2.5 per patient), the mean proportion of those with undetectable PI concentrations was 27% (range: 0 100%). Most of the PIs were given as single agents (not boosted); only 7% (22/328) of the PI levels were from a dual PI regimen (saquinavir plus ritonavir) and only one of these was undetectable. Sixty-five percent (215/328) of the levels were from patients on indinavir and 20 39% of the levels were from patients to whom nelfinavir, ritonavir or saquinavir was given as a single PI regimen. For the study population, the mean change in the HIV RNA level from baseline to month 6 was 0.73 log 10 copies/ml. In the univariate analyses, there were no statistically significant differences between the mean change in HIV RNA by categories of sex, ethnicity, level of education, risk factors for HIV transmission or level of monthly income. However, prior therapy with PIs (P=0.0004), prior therapy with 3TC (P<0.0001), recreational use of alcohol or drugs during the study period (P=0.005) and assignment to the infrequent versus frequent viral load measurement group (P=0.0004) were associated with lower HIV RNA reductions (Table 2). Age, baseline CD4 counts and the reported percentage of activities limited by side effects were not associated with the change in HIV RNA from baseline to month 6 (Table 3). However, there was a statistically significant relationship between the percentage of selfreported adherence to medication regimens (P=0.002, Figure 1) and the baseline viral load (P=0.007) and change in HIV RNA level. Increases in both factors resulted in greater HIV RNA reductions at month 6. The relationships between viral response and various metrics of PI concentration were explored (Table 3). There were no statistically significant associations found between the change in HIV RNA to month 6 and the percentage of visits with the PI drug concentration less than the 5th, 20th or 25th percentile of the population predicted values (using normal and log-transformed values). For indinavir, the 5th, 20th and 25th percentiles 8 h after a dose were 0.040, and µg/ml, respectively. However, the percentage of visits with undetectable PI concentrations significantly contributed to the linear prediction of the change in the HIV RNA level (P=0.0007). Patients with 100% of visits with detectable PI concentrations Table 2. Univariate analysis: mean change in HIV RNA from baseline to month 6 versus categorical variables Mean change in log 10 copies/ml Variable HIV RNA to month 6 P Demographic characteristics Gender 0.93 Male 0.74 Female Ethnicity 0.41 Caucasian African American Hispanic Asian/other Risk factors Homosexual contact 0.92 Yes No Heterosexual contact 0.86 Yes No Injection drug use 0.55 Yes No Prior PI use Yes No Prior 3TC use < Yes No Treatment group Infrequent HIV RNA monitoring Frequent HIV RNA monitoring Alcohol or drug use Yes No PI, protease inhibitor; 3TC, lamivudine. had an additional 0.63 log 10 HIV RNA reduction compared with those with no visits with detectable concentrations (Figure 2). In addition, the percentage of visits with the PI concentration ratio less than 20% [method (iii) listed above in the protease inhibitor concentrations section in Methods] also significantly contributed to the linear prediction of the change in the HIV RNA level (P=0.025). To determine if PI concentration data and adherence questionnaires provide unique insight into overall adherence, we examined the correlation between them. The correlation between the percentage of visits with undetectable PI concentrations and the percentage of self-reported adherence was 0.23 (P=0.009). Thus, the relationship between the two variables was significant, but weak, suggesting that these variables may provide complementary information International Medical Press

5 Two measures of adherence predict antiretroviral outcome Table 3. Univariate analysis: change in HIV RNA level versus continuous variables Variable n β P Age Percentage of self-reported adherence Baseline log HIV RNA Baseline CD4 cell count Percentage of activities limited by side effects Concentration variables % visits with undetectable PIC % visits with PIC ratio <20% % visits with PIC <25th percentile % visits with PIC <20th percentile % visits with PIC <5th percentile % visits with PIC <25th percentile-log transformed % visits with PIC <20th percentile-log transformed % visits with PIC <5th percentile-log transformed PIC, protease inhibitor concentrations. β, regression coefficient, describes the magnitude of the association between the variables, that is, for a 10% increase in the % undetectable PIC, the reduction in HIV RNA would be 0.07 log 10 copies/ml fewer. Figure 1. Change in HIV RNA level from baseline to month 6 versus percentage of self-reported adherence 1 0 Chnage in HIV RNA (month 6) Percentage of self-reported adherence Antiviral Therapy 9:5 757

6 JM Yasuda et al. Figure 2. Change in HIV RNA level from baseline to month 6 versus percentage of visits with undetectable PI concentrations 1 0 Chnage in HIV RNA (month 6) Percentage of visits with undetectable PI concentrations Multivariate analyses Two linear multivariate models were created by simultaneously entering variables that were statistically related to the change in the HIV RNA level from baseline to month 6 in the univariate analyses. In the first model, the variable for the percentage of visits with undetectable PI concentrations was not included (Table 4). In this model, percentage of self-reported adherence (P=0.006), baseline HIV RNA level (P=0.007), prior PI use (P=0.0007), prior 3TC use (P=0.0004), recreational drug or alcohol use (P=0.03) and frequent HIV RNA measurement group (P=<0.0001) were all significantly related to the change in the HIV RNA level. In the second model, all variables that were statistically related to the change in the HIV RNA level from baseline to month 6 were simultaneously entered into the linear regression model. In this model, percentage of visits with undetectable PI concentrations (P=0.02), percentage of self-reported adherence (P=0.02), baseline HIV RNA level (P=0.005), prior PI use (P=0.0004), prior 3TC use (P=0.0009) and frequent HIV RNA measurement group (P=<0.0001) were all related to change in the HIV RNA level. Recreational alcohol or drug use was not statistically related to the change in the HIV RNA level after controlling for the other variables (P=0.09). After accounting for self-reported adherence, having 100% of PI concentrations detectable resulted in an additional 0.4 log 10 HIV RNA reduction compared with having no detectable concentrations. Discussion Adherence is one critical factor influencing the response to ART. In clinical trials, it is important to account for adherence when assessing new regimens or treatment strategies. There is currently no standard method for measuring a patient s adherence to a medication regimen. In this analysis, we determined International Medical Press

7 Two measures of adherence predict antiretroviral outcome Table 4. Multiple linear regression models evaluating factors associated with change in HIV RNA from baseline to month 6 Factors in the model β P Model 1 Percentage of self-reported adherence Baseline HIV RNA Prior PI use Prior 3TC use Alcohol/drug use Frequent HIV RNA monitoring group < Model 2 Percentage of visits with undetectable PI concentrations Percentage of self-reported adherence Baseline HIV RNA Prior PI use Prior 3TC use Alcohol/drug use Frequent HIV RNA monitoring group < PI, protease inhibitor; 3TC, lamivudine; β, regression coefficient. that repeated measurements of PI concentrations add to and were independent of self-reported adherence measurements in understanding the response to ART. In this study, self-reported adherence to ART was good; the mean proportion of adherence was 91% with 64% of patients reporting being adherent at least 95% of the time. The adherence rate is similar to other studies; in one, the mean adherence rate was 89% and 77% of patients had a rate of adherence greater than 95% [19]. As was found in other studies [12,20,21], patients with higher levels of self-reported adherence had greater reductions in viral load than those with less adherence. Since adherence can vary with time, we used a composite of reports from three measurements. Multiple measures of adherence may be a better predictor of true adherence and better linked to outcomes [13]. As expected, other factors were important determinants of reduced HIV RNA response, including prior treatment (with 3TC or a PI), lower baseline HIV RNA, use of recreational drugs or alcohol and randomization to the infrequently monitored group [16,20,22]. The percentage of visits with undetectable PI concentrations and the percentage of visits with the PI concentration ratio less than 20% were significantly related to the change in viral load at month 6. Other modelled pharmacokinetic variables were not predictive of outcome. Lack of association could be due to the necessity of imputing the time of the last dose of PI taken, as these data were not collected in the study. The accuracy and reliability of these variables were dependent on correctly estimating the drug concentration expected in an individual at a given time after the last ingestion of the drug. It is possible that collection of data on the time of the previous three doses might have improved the value of PI concentrations in predicting response and would help separate the degree to which PI concentration might reflect adherence versus pharmacokinetics. However, patient knowledge of measurement of concentration might influence adherence behaviour just prior to the clinic visit (see below). The two pharmacological variables that were significantly related to the change in viral load were not as dependent on the assumption of the last dosage time. This suggests that sophisticated population pharmacokinetic modelling may not be necessary to gain information about those who are non-adherent to their drug regimens, and about those who are unable to obtain adequate plasma drug concentrations because of difficulties with drug absorption or metabolism. Taking multiple plasma PI concentration measurements over a study period, and determining whether they are detectable, may be a simple method to gain this information. In addition, keeping the patient masked to measurement of concentration may facilitate accuracy of assessment. Since patients were unaware that we were measuring their plasma drug concentrations, this may have reduced the propensity for patients to increase their adherence behaviour shortly before clinic visits. It is likely that multiple concentration measurements are needed to assess adherence. In this study, we found that PI levels were undetectable 27% of the time. Since adherence behaviour varies over time, the probability of finding an unmeasurable concentration increased with repeated sampling. This study did not determine the exact number of samples that would be optimal to estimate adherence. There was a significant, inverse correlation found between the percentage of visits with undetectable PI concentrations and level of self-reported adherence. However, the correlation was weak (r= 0.23), suggesting that these two variables may be measuring complementary information. Although self-report is the most common method employed to measure adherence, these estimates are overly optimistic [23]. Studies have indicated that although patients who report nonadherence are probably candid, those reporting full adherence may not always provide an accurate estimate [24]. Thus, additional objective measures, such as drug concentrations, may improve the overall assessment of adherence. The multivariate analysis used in this study demonstrated that the percentage of visits with an undetectable PI concentration was an independent predictor of viral load response to a regimen. Since self-report and drug concentration were in the model, they both added to the prediction of response. Patients that had measurable PI concentrations at all Antiviral Therapy 9:5 759

8 JM Yasuda et al. visits had a 0.4 log 10 additional viral load reduction compared with those that had no detectable PI levels. This difference in HIV RNA reduction is clinically meaningful, based on prior studies in which a log 10 copies/ml difference in HIV RNA resulted in a reduction in the relative risk for progression to AIDS [25]. Although randomized clinical trials are designed to balance, as best as possible, many confounding factors that contribute to treatment response, understanding the causes of failure contributes to our understanding of therapeutic response. Drug potency, tolerability, intrinsic pharmacodynamics and drug resistance all contribute to the success or failure of therapy. Although studies are designed to determine efficacy, the outcome of a clinical trial represents a mixture of effects. By separating out the impact of adherence, the clinical trial can better define the role of other factors, particularly drug potency. Since patients were allowed to change therapy during the study, this analysis was limited by an inability to account for the effect of different regimens. This is partially mitigated by use of an area-based outcome, which captures dynamic changes in viral load due to regimen changes. More frequent sampling of PI concentration could also have improved the accuracy of prediction. Additionally, the majority of regimens used in this study were based on a single PI (mostly non-boosted indinavir), which are more difficult to adhere to than current regimens and are more likely to result in undetectable levels. Contemporary regimens often include a ritonavir-boosted PI, which would probably result in higher drug concentrations and fewer unmeasurable concentrations. However, boosting only prolongs exposure while the ritonavir is present and elimination of the boosted PI would be more rapid after ritonavir disappears, which typically occurs a little more than 12 h post-dose. For example, with Kaletra, a fixed combination of ritonavir and lopinavir, a single missed dose would result in a lopinavir concentration less than 0.2 µg/ml (the assay limit) at 24 h after the last dose in approximately 25% of patients, based on population models. Thus, a single missed lopinavir dose could result in an undetectable level in a significant percentage of patients and setting another threshold (that is, below the 5th percentile) could identify additional patients who missed doses. In this analysis, repeated, randomly timed measurements of PI concentrations were independently predictive of virological response, and may add to self-reported adherence in understanding the response to ART. These findings do not, however, provide a rationale for the use of untimed plasma concentrations in routine clinical practice, nor has such use been recommended in clinical guidelines [26]. Future clinical trials could use this marker as an additional surrogate for adherence. Summarizing self-reported adherence with multiple self-administered questionnaires and multiple drug concentrations may provide a better control for a patient s true adherence than measuring adherence at a single point in time. Acknowledgement This work was supported by funding from the University of California Universitywide AIDS Research Program (UARP) CC99-SD-003 and CC02-SD-003. References 1. Palella FJ Jr, 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. New England Journal of Medicine 1998; 338: Deeks SG. Determinants of virological response to antiretroviral therapy: implications for long-term strategies. Clinical Infectious Diseases 2000; 30(Suppl. 2):S Ledergerber B, Egger M, Opravil M, Telenti A, Hirschel B, Battegay M, Vernazza P, Sudre P, Flepp M, Furrer H, Francioli P & Weber R. Clinical progression and virological failure on highly active antiretroviral therapy in HIV-1 patients: a prospective cohort study. Lancet 1999; 353: Deeks SG, Barbour JD, Martin JN, Swanson MS & Grant RM. Sustained CD4+ T cell response after virologic failure of protease inhibitor-based regimens in patients with human immunodeficiency virus infection. Journal of Infectious Diseases 2000; 181: Lucas GM, Chaisson RE & Moore RD. Highly active antiretroviral therapy in a large urban clinic: risk factors for virologic failure and adverse drug reactions. Annals of Internal Medicine 1999; 131: Mathews C, Haubrich R, Barber E & Havlir D. Patterns and prognostic implications of prior antiretroviral therapy (ART) at UCSD Owen Clinic: th Conference on Retroviruses & Opportunistic Infections. 30 January 2 February 2000, San Francisco, Calif., USA. Abstract Acosta E, Havlir D, Richman D, Zhou X, Hirsch M, Collier A, Tebas P & Sommadossi J-P. Pharmacodynamics (PD) of indinavir (IDV) in protease-naive HIV-infected patients receiving ZDV and 3TC. 7th Conference on Retroviruses & Opportunistic Infections. 30 January 2 February 2000, San Francisco, Calif., USA. Abstract Descamps D, Flandre P, Calvez V, Peytavin G, Meiffredy V, Collin G, Delaugerre C, Robert-Delmas S, Bazin B, Aboulker JP, Pialoux G, Raffi F & Brun-Vézinet F. Mechanisms of virologic failure in previously untreated HIV-infected patients from a trial of induction-maintenance therapy. Journal of the American Medical Association 2000; 283: Garraffo R. Independent benefit of sufficient drug levels and genotypic analysis in salvage therapy: pharmacological data of the viradapt study. 39th Interscience Conference on Antimicrobial Agents & Chemotherapy September 1999, San Francisco, Calif., USA. Abstract Powderly WG, Saag MS, Chapman S, Yu G, Quart B & Clendeninn NJ. Predictors of optimal virological response to potent antiretroviral therapy. AIDS 1999; 13: Phillips AN, Staszewski S, Weber R, Kirk O, Francioli P, Miller V, Vernazza P, Lundgren JD & Ledergerber B. HIV viral load response to antiretroviral therapy according to the baseline CD4 cell count and viral load. Journal of the American Medical Association 2001; 286: International Medical Press

9 Two measures of adherence predict antiretroviral outcome 12. Paterson DL, Swindells S, Mohr J, Brester M, Vergis EN, Squier C, Wagener MM & Singh N. Adherence to protease inhibitor therapy and outcomes in patients with HIV infection. Annals of Internal Medicine 2000; 133: Mannheimer S, Friedland G, Matts J, Child C & Chesney M. The consistency of adherence to antiretroviral therapy predicts biologic outcomes for human immunodeficiency virus-infected persons in clinical trials. Clinical Infectious Diseases 2002; 34: Arnsten JH, Demas PA, Farzadegan H, Grant RW, Gourevitch MN, Chang CJ, Buono D, Eckholdt H, Howard AA & Schoenbaum EE. Antiretroviral therapy adherence and viral suppression in HIV-infected drug users: comparison of self-report and electronic monitoring. Clinical Infectious Diseases 2001; 33: Bangsberg DR, Bronstone A, Chesney MA & Hecht FM. Computer-assisted self-interviewing (CASI) to improve provider assessment of adherence in routine clinical practice. Journal of Acquired Immune Deficiency Syndromes 2002; 31(Suppl. 3):S Haubrich RH, Currier JS, Forthal DN, Beall G, Kemper CA, Johnson D, Dube MP, Hwang J, Leedom JM, Tilles J & McCutchan JA. A randomized study of the utility of human immunodeficiency virus RNA measurement for the management of antiretroviral therapy. Clinical Infectious Diseases 2001; 33: Wu EY, Wilkinson 2nd JM, Naret DG, Daniels VL, Williams LJ, Khalil DA & Shetty BV. High-performance liquid chromatographic method for the determination of nelfinavir, a novel HIV-1 protease inhibitor, in human plasma. Journal of Chromatography. B, Biomedical Sciences & Applications 1997; 695: Capparelli E. Time of first morning dose of protease inhibitor drugs in HIV-infected patients. 2001: unpublished data. 19. Tuldra A, Ferrer MJ, Fumaz CR, Bayes R, Paredes R, Burger DM & Clotet B. Monitoring adherence to HIV therapy. Archives of Internal Medicine 1999; 159: Haubrich RH, Little SJ, Currier JS, Forthal DN, Kemper CA, Beall GN, Johnson D, Dube MP, Hwang JY & McCutchan JA. The value of patient-reported adherence to antiretroviral therapy in predicting virologic and immunologic response. AIDS 1999; 13: Tebas P, Royal M, Fichtenbaum C, Blutman J, Arens M, Horgan M & Powderly W. Relationship between adherence to HAART and disease state. 5th Conference on Retroviruses & Opportunistic Infections. 1 5 February 1998, Chicago, Ill., USA. Abstract Hammer SM, Squires KE, Hughes MD, Grimes JM, Demeter LM, Currier JS, Eron JJ Jr, Feinberg JE, Balfour HH Jr, Deyton LR, Chodakewitz JA & Fischl MA. A controlled trial of two nucleoside analogues plus indinavir in persons with human immunodeficiency virus infection and CD4 cell counts of 200 per cubic millimeter or less. New England Journal of Medicine 1997; 337: Golin C, Liu H, Hays R, Ickovics J, Beck K, Miller L, Kaplan A & Wenger N. Self-reported adherence to protease inhibitors substantially overestimates an objective measure. 6th Conference on Retroviruses & Opportunistic Infections. 31 January 4 February 1999, Chicago, Ill., USA. Abstract Murri R, Ammassari A, Gallicano K, De Luca A, Cingolani A, Jacobson D, Wu AW & Antinori A. Patient-reported nonadherence to HAART is related to protease inhibitor levels. Journal of Acquired Immune Deficiency Syndromes 2000; 24: Haubrich R, Lalezari J, Follansbee SE, Gill MJ, Hirsch M, Richman D, Mildvan D, Burger HU, Beattie D, Donatacci L & Salgo MP. Improved survival and reduced clinical progression in HIV-infected patients with advanced disease treated with saquinavir plus zalcitabine. Antiviral Therapy 1998; 3: Acosta EP & Gerber JG. Position paper on therapeutic drug monitoring of antiretroviral agents. AIDS Research & Human Retroviruses 2002; 18: Received 13 January 2004, accepted 12 May 2004 Antiviral Therapy 9:5 761