Antiretroviral Concentrations in Untimed Plasma Samples Predict Therapy Outcome in a Population with Advanced Disease

Abstract

This study was designed to examine the relationship between untimed antiretroviral concentrations measured in plasma samples collected for virus-load testing and response to highly active antiretroviral therapy. Plasma nonnucleoside reverse-transcriptase–inhibitor and protease-inhibitor concentrations were retrospectively measured in all virus-load plasma samples collected during the first year of therapy, for 122 patients in British Columbia, Canada, who initiated therapy between August 1996 and September 1999 and who had CD4 counts <50 cells/μL. Drug levels were designated a priori as “low” if the concentrations were below the published C_trough-SD. A single low drug level measured shortly after initiation of therapy (median, 6 weeks) is common (30%) and is predictive of both more-rapid immunological failure (P=.06) and failure to achieve virologic success during the first year of therapy (P=.01). These results may reflect incomplete adherence, since a strong association (P<.001) was found between low drug levels and an imperfect prescription-refill record (<95%)

The use of a potent combination of antiretroviral drugs has led to dramatic reductions in the morbidity and mortality associated with human immunodeficiency virus (HIV)–1 infection [1]. However, ∼20% of therapy-naive patients fail to achieve adequate virological response during their first year of triple therapy, with ∼20% more patients experiencing virological failure during the second year [2, 3]. Variability in response to antiretroviral agents has been attributed, in part, to virological, immunological, pharmacokinetic, and adherence differences between patients [4, 5]. The first 2 issues have been partially addressed by drug-resistance testing and monitoring of patient CD4 cell counts, as a standard of care. However, the latter 2 issues have proved to be difficult to objectively evaluate in a routine clinical setting

Interpatient variability in absorption, distribution, metabolism, elimination, and protein binding may put patients at risk for subtherapeutic or toxic exposures to antiretroviral drugs. Antiretroviral drug–drug interactions and interactions with other drugs and/or supplements can also contribute to less-than-ideal exposures [6–8]. It is therefore reasonable to postulate that at least some unexplained cases of either virological failure or adverse effects may result from pharmacokinetic problems. Indeed, the association between low plasma drug levels and inferior virological response has been demonstrated for both protease inhibitors (PIs) and nonnucleoside reverse-transcriptase inhibitors (NNRTIs) [9–15]

Incomplete adherence to therapy has been shown to be a major contributor to failure of therapy [16–18]. Recent reports suggest that <95% adherence can significantly compromise virological and immunological responses to therapy [19, 20] and can have a negative effect on mortality [21]. Methods used to gauge adherence include patient and physician assessments, electronic monitoring, pill count, and prescription-refill compliance, as used in the present study. These methods have proved to be predictive of outcomes, although the results are variable [19, 20, 22, 23]. Identifying additional accurate predictors of adherence—predictors that can routinely be applied in clinical practice—may be of clinical value

Here we investigate the predictive value of measurement of both PI concentrations and NNRTI concentrations, in untimed plasma samples collected for virus-load testing in a population of antiretroviral-naive HIV-positive patients initiating highly active antiretroviral therapy (HAART). Given the pharmacokinetic characteristics of both PIs and NNRTIs, we anticipate that effectively treated patients should maintain drug concentrations above predicted limits; therefore, we hypothesize that the detection of drug levels below predicted limits could reveal potential pharmacokinetic and/or adherence difficulties

Patients, Materials, and Methods

Patient populationAll patients in British Columbia, who initiated triple therapy including a PI or NNRTI, between August 1996 and September 1999, who had baseline CD4 counts <50 cells/μL, and who had ⩾1 follow-up visit were eligible for inclusion in the study (n=129). This subset of a previously characterized observational cohort [24] represents the patients, in British Columbia, who have the highest risk of therapeutic failure. Plasma samples were collected for virus-load testing at baseline, at 4 weeks after initiation of therapy, and approximately every 3 months thereafter. All plasma samples were stored indefinitely at −20 °C

Study proceduresPlasma HIV RNA levels were measured by use of the Amplicor HIV-1 monitor (version 1.0 or 1.5; Roche Diagnostic Systems). All available stored plasma samples collected during the first year of therapy were tested for concentrations of PIs and NNRTIs. At the time of sample collection, patients were not aware that drug levels would be tested. To assure patient confidentiality, all patient data were recoded before outcome comparisons were made, according to guidelines approved by the University of British Columbia Ethics Review Committee (St. Paul’s Hospital site)

The plasma concentrations of all commercially available PIs and NNRTIs were determined by a sensitive, validated, simultaneous assay using reverse-phase (Zorbax XDB-C18; Agilent Technologies) high-pressure liquid chromatography (HPLC) (HP 1100; Agilent Technologies) coupled with tandem mass spectrometry (MS-MS) (API-2000; Applied Biosystems). In brief, acetonitrile (190 μL) was used to precipitate proteins from plasma (50 μL), and proteins were separated by centrifuge filtration, with a normalized molecular-weight limit of 30,000 (Millipore), before analysis by HPLC–MS-MS. All drugs were quantified relative to a reserpine (Sigma-Aldrich) internal standard. Duplicate plasma quality-control samples were analyzed at low, medium, and high concentrations, in each of 11 analytical batches. The lower limits of quantification and the intraday and interday coefficients of variations, for each quality-control sample and for each drug, are given in table 1

Table 1

Intraday and interday coefficients of variation for quality-control samples (low, medium, and high) tested for each drug

For most PIs and NNRTIs, plasma levels were designated a priori as “low” if the concentrations were less than the reported steady-state trough (or, in the case of efavirenz [EFV], the minimum) concentrations minus 1 SD (C_trough-SD), as published in the product monographs [25–30]. It should be noted that the calibre of these data was variable, in terms of the number and HIV status of patients used in establishing these values. The numbers of patients in each study were as follows: 10 for ritonavir (RTV), 11 for nelfinavir (NFV), 16 for indinavir (IDV), 21 for lopinavir (LPV), 35 for EFV, 67 for delavirdine (DLV), and 242 for nevirapine (NVP). The product monographs indicate that the pharmacokinetics were studied in HIV-infected patients, except in the case of IDV and NVP, for which the HIV statuses of the study populations were not specified. In spite of the variability among these reference values, we felt that the use of these data was a consistent approach. Since the original dosing recommendations were based on these data, we expected that patients experiencing concentrations at the lower end (i.e., below the 16th percentile) of these distributions were at the greatest risk of therapeutic failure. Furthermore, we anticipated that, during the dosing period, the majority (∼84%) of patients would maintain concentrations above this limit. Thus, “low” was defined as any concentration falling below the following: 2610 ng/mL, for NVP; 2289 ng/mL, for DLV; 761 ng/mL, for EFV; 1485 ng/mL, for LPV; 500 ng/mL, for NFV; and 1110 ng/mL, for RTV. For IDV and saquinavir (SQV), drug levels below the lower limits of quantification (101 ng/mL and 56 ng/mL, respectively) were designated as “low.” Any measurement above these respective limits was designated as “normal.” In the case of IDV, the limit of quantification was greater than C_trough-SD (43.5 ng/mL), and, for SQV, no C_trough was reported in the product monograph

Outcome MeasuresMortality, HIV RNA levels, CD4 cell counts, adherence, and changes in HAART regimens were compared in patients with low plasma PI or NNRTI concentrations versus patients with normal plasma PI or NNRTI concentrations. In this study, “virologic success” was defined as achievement of 2 consecutive plasma viral-RNA tests of ⩽500 copies/mL. The “time to immunologic failure” was defined as the time between the initiation of therapy and either the first CD4 cell count below baseline values or death

Adherence to therapy was determined on the basis of the rate of refill compliance during the first year of therapy, as described elsewhere [18]. In British Columbia, antiretroviral medications are provided by a single agency, the British Columbia Centre for Excellence in HIV/AIDS Drug Treatment Program, and prescription-refill records are available through this program. In brief, “adherence” (expressed as a percentage) was defined as the ratio of the total doses dispensed to the total doses that would be required for the entire follow-up period. In the present analysis, patients were classified as “incompletely adherent” if they received <95% of their antiretroviral therapy during the first year of therapy. The “time to change,” for HAART regimen, was defined as the time between the initiation of therapy and either the change of any antiretroviral in the original regimen or the addition of an antiretroviral drug

Statistical AnalysisPatients were classified as either “low” or “normal,” according to the results of their first antiretroviral drug concentration measured after initiation of therapy. In a second longitudinal analysis, results from the serial plasma samples collected during the first year of therapy were dichotomized on the basis of whether >50% of samples tested showed normal drug levels. Statistical comparisons were conducted by use of distribution-free methods. Pairs of categorical variables were compared by Pearson’s χ² test, and difference in the location (median) of a continuous variable, across 2 groups, was tested by Wilcoxon&amp;rank-sum test. Both cumulative mortality and cumulative failure rates were obtained by product-limit estimates. Cox proportional-hazard models were used to calculate hazard ratios (HRs) [31]. Event-free patients were right-censored, as of the last contact date

Results

Proportion of low-drug-level measurementsFour hundred thirty plasma samples collected from 129 eligible patients were available for analysis, a figure reflecting a median of 4 plasma viral-RNA tests/patient/year. No plasma samples were available for 7 patients, and those patients were therefore excluded from all further analysis. For the remaining 122 patients, the number of available plasma samples per patient varied from 1 to 9 (23 patients with 1, 14 with 2, 29 with 3, 33 with 4, and 23 with ⩾5 samples). Thirty percent (129/430) of all plasma samples tested had low levels of either PIs or NNRTIs, including 17% (75/430) in which no drugs were detected; 63% of patients (77/122) presented with low drug levels at least once during the first year of therapy

The association of baseline characteristics and plasma drug levels at first follow-upAn examination of either the PI concentrations or the NNRTI concentrations, from the first plasma HIV-RNA–level test after initiation of therapy (median, 6 weeks; interquartile range [IQR], 4–12 weeks), revealed that 42% of patients (51/122) had low drug levels at this time. Demographic and clinical baseline characteristics were compared in patients with low PI or NNRTI concentrations versus patients with normal PI or NNRTI concentrations (table 2). In univariate analysis, only female sex (P=.04) and injection drug use (P<.001) were associated with low drug levels

Table 2

Comparison of baseline demographics and clinical characteristics, for patients with “low” drug levels versus patients with “normal” drug levels, in plasma samples collected for the first human immunodeficiency virus RNA–level test after initiation of therapy

The association of drug levels at first follow-up and outcomesIn univariate analysis, a finding of low drug levels in the first untimed plasma sample collected after initiation of therapy was significantly associated with (1) failure to achieve virologic success during the first year of therapy (odds ratio [OR], 4.4; P<.001) (table 3), (2) more-rapid changes during therapy (HR, 2.1; P=.001), and (3) an increased rate of immunological failure (HR, 2.5; P=.007) (figure 1A). For these patients, low PI and NNRTI concentrations in these plasma samples were also significantly associated (P<.001) with imperfect prescription-refill records (i.e., <95%). Interestingly, low drug levels at this time also appear to be associated with more-rapid progression to HIV-related death, although this association was not significant (P=.116) (figure 1B)

Table 3

Univariate and multivariate analyses of factors associated with virologic success (confirmed VL <500) during the first year of therapy, in patients who initiated triple therapy and who had baseline CD4 cell counts <50 cells/μL

Figure 1

A Product-limit (Kaplan-Meier) estimates of cumulative progression to immunological failure, in 122 human immunodeficiency virus (HIV)–infected patients (“low” concentrations [solid line], n=51; “normal” concentrations [dotted line], n=71) who initiated highly active antiretroviral therapy (HAART) between August 1996 and September 1999, stratified by untimed drug levels at the first HIV RNA–level test ( P=.007, log-rank test). B Product-limit (Kaplan-Meier) estimates of cumulative progression to HIV-related death, in 116 HIV-infected patients (low concentrations [solid line], n=48; normal concentrations [dotted line], n=68) who initiated HAART between August 1996 and September 1999, stratified by untimed drug levels at the first HIV RNA–level test (P=.116, log-rank test)

In multivariate analysis, after adjustment for sex, age, injection drug use, baseline viral RNA, baseline AIDS-defining illness, and first-line PI use, a single low drug level in the first on-therapy plasma sample emerged as a significant predictor of both failure to achieve virological success during the first year of therapy (table 3) and failure to refill prescriptions >95% of the time (OR, 2.63; confidence interval, 1.06–6.40; P=.04). Low drug levels at this time were also associated with more-rapid immunological failure in multivariate analysis, although this association was not significant (P=.06) (table 4)

Table 4

Univariate and multivariate analyses of factors associated with immunological failure, in 122 patients who initiated triple therapy and who had baseline CD4 cell counts <50 cells/μL

The association of drug class and low drug levels at first follow-upThe majority of patients (101/122) initiated PI-based therapies (IDV, 64/101; NFV, 23/101; SQV-SGC, 11/101; or RTV, 3/101). Of the 21 patients who were prescribed an NNRTI-based therapy, 18 initiated NVP and 3 initiated DLV. Neither drug class (PI based or NNRTI based) was associated with low drug levels at the time of the first follow-up (P=.71), although the number of patients initiating NNRTI-based therapy was relatively low and therefore may not be representative. To test whether the long half-life of NVP relative to the other drugs was biasing the results, patients receiving NVP were removed from the analysis. The associations between low drug levels and virologic success during the first year of therapy (P<.001), time to therapy switch (P=.001), and time to CD4 failure (P=.005) remained statistically significant when these patients were removed

Of importance, normal concentrations of NFV (i.e., >500 ng/mL) were detected in 80% (62/78) of samples that had been expected to contain NFV. This finding is notable because a recent study [32] has reported that NFV is unstable when stored at −20°C (as our samples were): the authors of that study have reported an 83% degradation of NFV after 19 months at this temperature. Our results do not seem to reflect their finding; our samples were frozen for a median of 23 months (IQR, 19–28 months), yet significantly more NVF samples were designated “normal” than were all other drugs combined (P=.002)

The association of multiple drug-level measurements and outcomesMultiple drug-level measurements were found to increase the predictive value of the untimed plasma drug level, compared with that of the first measurement alone. The time to immunological failure (HR, 2.5; P=.001) (figure 2A), time to HIV-related death (HR, 2.9; P=.02) (figure 2B), and failure to achieve virological success during the first year of therapy (OR, 6.5; P<.001) were all significantly associated with half or fewer of all drug-level tests falling within the normal range during the first year of therapy. In multivariate analysis, low drug levels in half or more of all samples tested remained significantly associated with both immunological failure (table 4) and failure to achieve virologic success (table 3), as above. Of interest, post hoc sensitivity analysis revealed that the association between poor outcomes and multiple untimed drug-level tests became stronger with an increasing percentage of low tests. For example, in univariate analysis, patients for whom >65% of all plasma samples tested were below the normal range were at a greater risk of both more-rapid CD4 failure (HR, 4.5; P<.001) and more-rapid HIV-related death (HR, 4.79; P<.001). Other outcomes that were found to be associated with findings of low drug levels in half or fewer of plasma samples tested during the first year of therapy were incomplete adherence (P<.001), as estimated by prescription refill (< 95%), and earlier loss to follow-up (P=.006)

Figure 2

A Product-Limit (Kaplan-Meier) estimates of cumulative progression to immunological failure, in 122 human immunodeficiency virus (HIV)–infected patients (⩽50% “normal” [solid line], n=55; >50% normal [dotted line], n=67) who initiated highly active antiretroviral therapy (HAART) between August 1996 and September 1999, stratified by the results of multiple untimed drug-level tests during the first year of therapy (P=.001, log-rank test). B Product-Limit (Kaplan-Meier) estimates of cumulative progression to HIV-related deaths, in 116 HIV-infected patients (⩽50% normal [solid line], n=52; >50% normal [dotted line], n=64) who initiated HAART between August 1996 and September 1999, stratified by the results of multiple untimed drug-level tests during the first year of therapy (P=.023, log-rank test)

Discussion

Because the times of administration of the drugs and of collection of the blood were not known, interpretation of the antiretroviral concentrations in these untimed samples is limited but is not completely implausible. By employing a working definition of C_trough-SD as “low,” it is conceivable that ⩽16% of patients might present with low drug levels, but only in the unlikely event that all patients were sampled at the time of trough. Our data indicate that low drug levels, as defined in the present study, were remarkably common in this cohort. In this study, 42% of patients presented with low concentrations at the first visit. In contrast, only 1.6% of patients were found to have drug concentrations exceeding the reported maximum concentration + SD. Therefore, it is likely that the relatively high prevalence of low drug levels that has been observed in the present study reflects factors other than pharmacokinetic problems, although pharmacokinetic problems presumably also play a role

Incomplete adherence may be the most important factor contributing to the prevalence of low drug levels in this cohort, especially since more than half of all plasma samples designated as “low” exhibited no trace of drugs. Female sex and injection drug use were significantly associated with low concentrations, and both risk factors have elsewhere been linked to intermittent use of antiretroviral therapy [18, 33]. Furthermore, a strong association between imperfect prescription-refill compliance (<95%) and low drug levels has been revealed in the present study. Indeed, untimely refilling of antiretroviral prescriptions has elsewhere been associated with poor therapeutic outcomes in this cohort [34]

Of importance, the results of this study demonstrate that, regardless of the contributing factors, low PI and NNRTI concentrations detected in random, untimed plasma samples are associated with poor clinical response to therapy. After adjustment for sex, age, baseline HIV RNA level, injection drug use, baseline AIDS-defining illness, and drug class (PI or NNRTI) in the initial regimen, a single low drug level observed shortly after initiation of therapy was predictive of both failure to achieve virological success during the first year of therapy and longer-term immunological failure. Not surprisingly, the predictive power of the measurement was increased when multiple, serial plasma samples were taken into account. In a similar analysis, patients with low drug levels in >50% of all plasma samples tested during the first year of therapy were at increased risk not only of virological and immunological failure but also of death. It is interesting to note that, in the majority (>85%) of cases, patients who experienced poor outcomes had low drug levels in >65% of samples tested. This finding highlights the fact that patients failing therapy (both virological and immunological failure) consistently experience low drug levels during the first year of therapy

The relationships between plasma concentration and therapeutic response have been documented for most PIs and/or NNRTIs, with the majority of studies focussing on virological outcomes and trough concentrations [9, 10, 14, 15]. Because of the changes in plasma concentrations during the dosing period, few studies have examined the relationships between outcomes and random, untimed drug levels. In one study in which this relationship has been examined, higher untimed NVP concentrations, defined as anything above the study distribution median, were found to be associated with a more-rapid viral clearance, a shorter time to undetectable viral RNA (<20 copies/mL), and an improved chance of achievement of undetectable viral RNA, for 51 patients in the triple-therapy arm of the Italy, the Netherlands, Canada, and Australia Study [12]: Veldkamp et al. suggested that the observed associations may be due to the long plasma half-life (∼25 h) relative to the dosing schedule (twice daily) of NVP. This results in relatively uniform NVP concentrations during the dosing period. Whereas this effect likely influences our observations as well, a long plasma half-life is not the sole factor producing these results. When patients who were prescribed NVP were removed from analysis, all associations between the first drug-level measurements and outcomes remained statistically significant

Our results are consistent with those of a recent report by the Community Programs for Clinical Research on AIDS 046 Study Team [35], which found, in 137 patients from the genotypic antiretroviral-resistance testing (GART) study, an association between untimed antiretroviral concentrations (including all available drugs) and short-term virological response. In contrast to our work, the GART study involved antiretroviral-experienced patients with a history of virological failure while receiving PI therapy. Thus, baseline viral resistance was a confounding factor, although approximately half of the therapies were guided by resistance testing. These differences notwithstanding, it is notable that, for 2 distinct cohorts, untimed antiretroviral concentrations proved to be powerful predictors of clinical outcomes

One limitation of our study was the relatively poor immunological health of our study patients (median baseline CD4 cell count, 20 cells/μL). This cohort was chosen, in part, because of the relatively high frequency of clinical events for analysis. It is not clear that the same results would be observed in a healthier, drug-naive population. Also, it is possible that our study patients were failing therapy not because of low drug levels but because they were either extremely ill or close to death (i.e., that there was reverse causation). However, 73% of the deaths occurring in the group of patients with low drug levels at the time of the first follow-up occurred after 1 year of therapy, a finding suggesting that early death is likely not biasing the results. Furthermore, in the analyses, baseline AIDS-defining illness was not significantly associated with either virological or immunological failure

Another limitation of the study was the choice of lower-limit cutoff points. We chose limits a priori on the basis of data provided in the product monographs, in spite of the limitations of these references. Post hoc analyses revealed that the association between therapeutic outcomes and untimed drug levels is relatively sensitive to the choice of cutoff point. For example, to test sensitivity, the data were reanalyzed by use of alternative values, to dichotomize the drug levels; these alternatives included (1) concentrations above versus concentrations below the median measured values and (2) concentrations detected versus concentrations completely undetected. Whereas similar relationships were observed between low drug levels and outcomes, the associations were weaker at the higher cutoff points (i.e., median values). For each drug, the optimum low-level cutoff point for maximization of sensitivity and specificity will be established as more data become available

Another issue raised during the review of this study was the potential degradation of NFV in our plasma samples stored at −20°C [32]. As stated above, our observations are not consistent with this possibility, although we are unable to explain the difference. Nevertheless, if NFV were decaying, this possibility would lead to a conservative bias in the analysis, a bias actually underestimating the association between untimed drug levels and outcomes of therapy

It should also be noted that, in this study, the prevalence of low drug levels may reflect factors other than pharmacokinetic problems and incomplete adherence. For example, transient problems, such as severe diarrhea or emesis, might also contribute; however, such issues are not likely to play a prominent role and would not necessarily have a significant effect on long-term outcomes

On the basis of the results of this and other studies, measurement of random drug concentrations in untimed plasma samples collected for routine HIV RNA–level testing appears to have potential to be an early and powerful predictor of clinical outcomes, on a population basis. This approach may complement other measures of adherence, in analysis of clinical trials, particularly since it can be performed retrospectively. Whereas it is important to emphasize that untimed measurements such as these will never be used as a form of therapeutic-drug monitoring, such measurement may have use for identification of patients at greatest risk of therapeutic failure. A prospective randomized trial to determine whether untimed drug-level monitoring in clinical practice may be a useful early warning sign to trigger interventions, such as adherence counseling or therapeutic-drug monitoring, is planned

Presented in part: XIV International AIDS Conference, Barcelona, Spain, 7-12 July 2002 (abstract WePeB5844)

Financial support: British Columbia Centre for Excellence in HIV/AIDS; Canadian Institute for Health Research (postdoctorate fellowship to C.S.A.)