Virologic Rebound on HAART in the Context of Low Treatment Adherence Is Associated With a Low Prevalence of Antiretroviral Drug Resistance

The contributions of viral drug resistance, drug pharmacokinetics, and treatment adherence to failure of protease inhibitor (PI)-based highly active antiretroviral therapy (HAART) have been explored in isolation. We examined the interactions between these factors. Patients on their first PI-based HAART combination for >6 months with plasma HIV viral load (VL) >1000 copies/mL (viremic group) were compared with patients with a stable VL <50 copies/mL (nonviremic group). Data were collected on adherence (measured electronically), PI plasma levels (sampled randomly, at trough and twice after an observed dose), viral genotype, and phenotype. Mean adherence was significantly lower in the viremic group (84.3% versus 100.1%;p = .005). In the viremic group, substitutions in HIV protease were detected most frequently in the following positions in subjects on indinavir (IDV): L10I/V (35.7%), M46I/L (35.7%), A71T/V (35.7%), V82A (42.9%); and for subjects on nelfinavir (NFV): D30N (50.0%), V77I (56.3%), N88D (37.5%). Phenotypic resistance was detected in 20% of isolates from patients on IDV and 42% on NFV. Lower adherence was associated with detection of fewer resistance substitutions (r = 0.52;p = .005) and less phenotypic resistance (ρ = 0.56, p = .005). There were no differences between the groups in pharmacokinetics. However, in viremic subjects, lower postdose NFV levels were associated with higher resistance (ρ = -0.61, p = .02).

Protease inhibitor (PI)-based highly active antiretroviral therapy (HAART) has greatly reduced AIDS-associated mortality and morbidity (^1,2). Current treatment guidelines recommend a reduction in plasma viral load (VL) to <50 copies/mL within 6 months of commencing HAART (³). However, even in clinical trials, 10% to 40% of subjects fail to reach this target, whereas in clinical cohorts this may range from 9% to 63% (^4-6). Among those whose VL does fall below the limit of quantification virologic rebound occurs in 8% at 1 year (⁷) and 20% to 40% at 2 years (⁴). The cause of treatment failure has been the focus of many studies, notably Trilége, in which subjects failing indinavir (IDV)-based regimens had a low incidence of drug resistance substitutions in the viral protease gene (Pr) (⁸). They also displayed lower adherence and had lower random plasma PI levels than their virologically successful counterparts. A number of questions arose from this study, including whether the lower drug levels were due to interindividual differences in pharmacokinetics or simply lower adherence. In addition, the relationship between adherence and the emergence of viral resistance is not well described. We sought to answer these questions by examining adherence, pharmacokinetics, and viral resistance in detail within our own clinical population.

STUDY POPULATION

The study was performed in a publicly funded specialist HIV clinic where treatment was free of charge. Subjects were HIV-1-seropositive adults (>18 years old) on their first PI-based HAART combination for at least 6 months consisting of either IDV (800 mg three times daily) or nelfinavir (NFV) (750 mg three times daily or 1250 mg twice daily) plus two nucleoside reverse transcriptase inhibitors (NRTIs). Two groups were recruited: first, subjects whose VL remained consistently below the limit of quantification (50 copies/mL) on at least three occasions over the preceding 6 months (nonviremic group); second, subjects whose VL was >1000 copies/mL on two consecutive tests (viremic group). The following were excluded from the study: patients whose medication was administered by another person, those who were not able to read and understand English or who had clinical cognitive impairment, and those taking concomitant medications that are inducers or inhibitors of cytochrome P₄₅₀.

METHODS

Ethical committee approval for the study was secured. At baseline, informed consent was obtained and subjects were asked the time of their last PI dose. Blood sampling was performed for VL (Chiron bDNA 3.0), CD4⁺ lymphocyte count, random plasma PI level, and HIV resistance genotype and phenotype. Without assistance, participants completed an adherence self-report (SR), having been assured that responses were confidential and would not be shared with their physician. The SR included items about the number of doses of PI subjects had missed 1, 2, and 3 days before and in the 2 weeks before that. Subjects were also asked to indicate on a visual analogue scale their best guess about the proportion of their PI doses that they had taken in the last month. The scale ranged from 0 to 100% in 10% intervals. Each subject was given a bottle containing a 6-week supply of their PI closed with a MEMS TrackCap (MC) (Medication Event Monitoring Systems, Aprex Corp., Menlo Park, CA, U.S.A.). These are pill bottle caps containing a microprocessor that records the time the bottle is opened as a presumptive dose (⁹).

Subjects were informed that adherence was one of the factors under scrutiny in the study. They were not made aware that the MC recorded their doses but were told that the cap would ensure the stability of their PI during the study to reduce variability in the results of the pharmacokinetic measurements. Subjects were asked to dose directly from the MC bottle, to use no other container until the end of the study, and to keep the bottle closed between doses. MC data of those who admitted to deviating from these instructions were excluded from statistical analysis. Subjects returned the MC and unused medication after 1 month. A pill count (PC) was performed, and participants completed the SR questionnaire again. Data were retrieved from the MC via a MEMS Communicator and processed using MEMS View 2.61 software (Aprex Corp.). We have expressed MC data using 2 indices. TC is defined as: (time under observation - sum of time doses were delayed beyond period of drug activity) / time under observation; possible range expressed as a percentage 0%-100%. TC indicates adherence both to the prescribed number of doses and also to the correct time intervals between doses. O/E is defined as number of presumed dosing events recorded on the MC as a proportion of that expected. The range may exceed 100% when observed dose events exceed those expected. Although O/E is the index most often cited, it gives an inflated estimate of adherence if the patient opens the bottle between doses without removing medication, or plays with the MC.

Measurement of Plasma Protease Inhibitor Levels

On an agreed date during the study period, subjects attended the clinic just before a PI dose was due. Blood was withdrawn for measurement of trough PI level (time of preceding dose confirmed from MC), and the PI was administered according to the manufacturers' instruction: 800 mg of IDV in the fasted state or 750 mg or 1250 mg of NFV with food. Further samples for PI plasma levels were drawn after the time of expected peak concentration (T_max) of each drug: for IDV (T_max = 0.8 hours, (¹⁰)) at 2 and 4 hours; for NFV (T_max = 3.4 hours) (¹¹) at 4 and 6 hours. Plasma samples were centrifuged and separated within 2 hours then refrigerated at -70°C until analyzed in batches by a validated high performance liquid chromatography assay (D. Back, Department of Pharmacology and Therapeutics, University of Liverpool, U.K.). NFV plasma concentration was measured as both the native compound and its active hydroxy-t-butylamide (or M8) metabolite. The assay limit of detection was 1 ng/mL for each substance.

Viral Drug Resistance Analysis

Drug resistance analysis was performed by Virco NV (Mechelen, Belgium) (¹²). After extraction and amplification of HIV RNA, genotypic analysis was performed for sequences in Pr and reverse transcriptase (RT). Phenotype was measured as the 50% inhibitory concentration (IC₅₀) of the PI in vitro, and also as the fold change in IC₅₀ relative to wild-type virus. Sensitive phenotype for IDV was defined as less than a threefold change, intermediate as three-to tenfold and resistant as more than tenfold change; for NFV, the cut-off values were defined as less than four-, four- to ten-, and more than tenfold change, respectively (¹³).

Statistical Analysis

SR items relating to number of doses actually taken or missed were converted to percentages of doses expected. Parametric data are presented as mean with standard error (SE) and nonparametric data as median with interquartile range (IQR). Where quantitative data showed gaussian normal distribution, comparison between two independent groups was made using the unpaired t test; comparison among more than two groups used one-way analysis of variance (ANOVA) and compared repeated measures using the paired t test. Where data showed a hypergeometric distribution, comparison between two independent groups was made using the Mann-Whitney U test. Differences between the groups in pharmacokinetic indices were explored for IDV, NFV, and its M8 metabolite separately and were then repeated including IDV and either NFV or its M8 metabolite simultaneously. We used χ² test statistics to assess the association between numbers of subjects in each group with a trough PI level above or below the theoretical minimum effective concentration (100 ng/mL for IDV; 400 ng/mL for NFV). Association between quantitative data with gaussian normal distribution were plotted and assessed using the Pearson correlation coefficient (r). Spearman rank correlation (ρ) was used to test association where data showed a hypergeometric distribution. Regression coefficients are presented with 95% confidence intervals (CI). All p values are presented using two-tailed tests. Statistical analysis was performed using SPSS 9.0.0 for Windows software (SPSS Inc., Chicago, IL, U.S.A.).

RESULTS

Four patients declined involvement in the study. Of the 75 patients recruited between January 18, 1999 and February 10, 2000, 7 were excluded from analysis, 3 were lost to follow-up, 2 withdrew consent, and 2 initially viremic subjects were excluded because the repeat VL test at baseline was below the limit of quantification. Of the 68 subjects completing the study, 80.9% were white, 85.3% were gay men, 8.8% were women, and 7.3% had a history of intravenous drug use. AIDS had been diagnosed in 45.6%. The PI was IDV for 32 subjects (15 viremic and 17 nonviremic) and NFV for 36 (17 viremic and 19 nonviremic). NFV was prescribed twice daily in 15 of 17 of viremic and in 18 of 19 of nonviremic subjects. Subjects had been diagnosed HIV-1 seropositive for a mean of 6.2 (SE, 0.5) years. The mean number of new drugs prescribed at the start of the current combination was 2.8 (SE, 0.1), and the mean daily pill burden was 13.5 (SE 0.5). Two subjects from each group started no new NRTIs with their PIs. The median CD4⁺ lymphocyte count on commencing the PI was 158 (IQR, 27-261) cells/mL; median VL was 5.1 (IQR, 4.3-5.6) log₁₀ copies/mL. The mean period of observation during the study was 29.4 (SE 0.7) days. There was no significant difference between the groups for any of these characteristics. However viremic and nonviremic groups differed significantly in age, treatment history and CD4⁺ lymphocyte count at recruitment (Table 1).

Table 1

In the viremic group median VL was 3.5 (IQR, 2.8-4.0) log₁₀ copies/mL; VL had never fallen below the limit of quantification for 13.3% of subjects on IDV and for 52.9% of subjects on NFV (p = .028). In patients in the viremic group who had achieved undetectable viremia initially but subsequently displayed virologic rebound, median time from first rebound to study entry was 112 (IQR, 63-182) days.

Adherence to Protease Inhibitors

Six subjects admitted to not using the MC for all their doses (1 viremic subject on IDV and 5 subjects on NFV: 1 viremic and 4 nonviremic); MC data from these individuals was excluded from statistical analysis. Adherence measured by MC was significantly lower among viremic subjects than their nonviremic counterparts. These data are summarized in Table 2. We also found significantly lower self-reported adherence at recruitment in viremic subjects. However, at follow-up this difference was no longer apparent because of a significant increase in SR adherence from recruitment to follow-up for viremic subjects: mean 3-day SR at baseline = 86.7% (SE, 3.9%), at follow-up = 96.9% (SE, 1.8) (mean difference, 10.2%; [SE 4.2%];p = .022); mean adherence over preceding month at baseline = 88.0% (SE 2.5%), at follow-up = 91.8% (SE 2.0) (mean difference = 3.8% (SE 1.7%), p = 0.033). During the same period, there was no significant change in SR adherence in nonviremic subjects.

Table 2

Antiretroviral Drug Resistance

Amplification for genotyping failed in 1 viremic subject on IDV and NFV, respectively. In subjects on IDV the commonest substitutions detected in Pr were L10I, I54V, and V82A. In subjects on NFV the commonest substitution detected was V77I, followed by D30N and N88D (Table 3). The prevalence of substitutions in HIV reverse transcriptase (RT) in the viremic group was as follows: M41L: 20.0%, D67N: 20.0%, K70R: 16.7%, Q151M: 3.3%, M184V: 63.3%, L210W: 3.3%, T215F: 6.7%, T215Y: 13.3%, K219Q: 10.0%, and G333E: 13.3%. The genotype of 6 viremic subjects revealed no substitutions in Pr. In 3 of these subjects (2 on NFV and 1 on IDV), the only detectable substitution in RT was M184V; in a single subject on IDV the only detectable substitution in RT was G333E, whereas in a further subject on IDV no substitutions in either RT or Pr associated with drug resistance were detected. However, one subject on IDV had substitutions in RT associated with multiple drug resistance (D67N, F116Y, and Q151M). All except one of these 6 subjects had been treated with NRTIs before starting their PI.

Table 3

A viral phenotype was available for 10 viremic subjects on IDV: 20% were classified as resistant, 40% as intermediate and 40% as sensitive. The median fold change in IC₅₀ was 4.3 (IQR, 1.7-9.9). A viral phenotype was available for 15 viremic subjects on NFV; the respective proportions were 54%, 13%, and 33%. The median fold change in IC₅₀ was 4.0 (IQR, 1.5-18.1).

Association Between Adherence and Antiretroviral Drug Resistance in the Viremic Group

Fold change in IC₅₀ was significantly higher in subjects displaying high adherence (n = 23; TC, ρ = 0.55, p = .007; O/E, ρ = 0.56, p = .005, see Fig. 1A). Subjects whose virus had a sensitive phenotype had significantly lower adherence than those with either intermediate or resistant phenotype-where mean TC if sensitive phenotype was 63.3% (SE, 7.7%), intermediate phenotype was 85.1% (SE, 12.1%), and resistant phenotype was 90.8% (SE, 1.6%) (p = .016) and mean O/E if sensitive phenotype was 68.7% (SE, 7.7%), intermediate phenotype was 92.7% (SE, 4.8%), and resistant phenotype was 99.0% (SE, 2.1%) (p = .009). There was a similar association between adherence and Pr genotype. Lower adherence measured by MC (n = 27; TC, r = 0.51;p = .007; O/E, r = 0.52, p = .005; see Fig. 1B) or by adherence SR at baseline only (r = 0.37, p = .046) was associated with detection of fewer amino acid substitutions. There was no association between resistance to PIs and length of time subjects had taken a PI, total duration of ART or time since first VL rebound.

Fig. 1

Subjects on NFV with the D30N substitution displayed significantly higher adherence (mean TC, 90.5% [SE 2.9%]; mean O/E, 97.5% [SE 3.8%]) than those without this substitution (mean TC = 61.6% [SE 10.2%]), p = .010; mean O/E, 66.8% [SE 10.8%];p = .011). A similar difference was observed for the N88D substitution (mean TC, 92.7% [SE 0.9%] vs. 67.2% [SE 8.7%];p = .022; mean O/E, 100.9% [SE 1.4%] vs. 71.9% [SE 9.0%];p = .015). In subjects on IDV the presence of any individual substitution in Pr was not associated with any difference in adherence. However, when data for both PIs were examined together, the presence of the L90M substitution was associated with significantly higher adherence (mean TC, 93.1% [SE 3.1%] vs. 74.4% [SE 4.8%];p = .003; mean O/E was 99.5% [SE 0.5%] vs. 80.9% [SE 4.9%];p = .001). We found no association between adherence and substitutions in RT either for the viremic group as a whole or for the subset of patients who were treatment naive on starting the PI.

Pharmacokinetic Indices

The timing of samples did not differ between the groups. For example the median time period between last reported dose and time of random sampling at recruitment for viremic subjects was 245 (IQR, 183-410) minutes compared with 223 (IQR, 160-348) minutes for nonviremic subjects. There was no statistically significant difference between the groups for any of the pharmacokinetic indices measured (Table 4). A single viremic subject taking IDV had chronic renal impairment and was on hemodialysis; after excluding this individual a significant difference between the two groups did not emerge. PI was undetectable in the plasma of 8 subjects at various timepoints (see Table 4). In a single viremic subject the random NFV level was low (175 ng/mL), and the M8 metabolite could not be detected; also in a single nonviremic subject, the trough NFV level was low (540 ng/mL) and the M8 metabolite could not be detected.

Table 4

Adherence SR at baseline was lower in the 4 subjects with undetectable random plasma PI levels than in those with detectable PI levels (median adherence over preceding 3 days: 86.1% (IQR, 20.8%-97.2%) and 100.0% (IQR, 88.9%-100.0%), respectively, p = .053; median adherence over preceding 2 weeks: 95.8% (IQR, 88.1%-97.3%) and 100.0% (IQR, 96.4%-100.0%), respectively, p = .03; median adherence over preceding month: 80.0% (IQR, 53.0%-87.5%) and 97.7% (IQR, 91.1%-100.0%), respectively, p = .007). However, during the month on study there was no significant difference in adherence between subjects with detectable and undetectable trough PI levels.

Association Between Pharmacokinetic Indices and Antiretroviral Drug Resistance in the Viremic Group

Fold change in IC₅₀ was significantly lower in viremic subjects who had high plasma levels of NFV measured at either 4 hours (n = 15; ρ = -0.54, p = .037) or 6 hours (n = 14; ρ = -0.61, p = .020) postdose. There was a similar relationship for Pr genotype: detection of fewer amino acid substitutions was associated with high levels of NFV measured at 4 hours (n = 16; ρ = -0.51, p = .042) and 6 hours (n = 15; ρ = -0.52, p = .047) postdose. Viral drug resistance was not associated with NFV levels measured at random or at trough, nor was it associated with levels of M8 metabolite. Plasma IDV levels were also not associated with viral drug resistance. Subjects on NFV with the N88D substitution had significantly lower plasma NFV levels 4 hours postdose (n = 6; median, 1855 ng/mL [IQR, 1102-2528 ng/mL]) and 6 hours postdose (n = 5; median, 1226 ng/mL [IQR, 622-1666 ng/mL]) than those without this substitution (n = 10; median, 3524 ng/mL (IQR, 2083-4798 ng/mL);p = .031 and n = 10, median = 2375 ng/mL (IQR 1327-3303 ng/mL) p = .040, respectively). A similar difference was not found for other PK indices or for any other individual substitution in Pr.

Association Between Adherence, Pharmacokinetic Indices, and Antiretroviral Drug Resistance in Viremic Subjects on Nelfinavir

Viremic subjects on NFV with high-level resistance on viral genotype displayed significantly higher adherence by MC and lower postdose plasma NFV levels than those with low-level resistance (Table 5). When the group of viremic subjects on NFV was divided at median fold change in IC₅₀, those with high resistance (fold change ≥4.3) also displayed significantly higher adherence by O/E (n = 7; mean, 95.8% [SE 4.4%]) than those with low resistance (n = 6; mean, 68.9% [SE 11.8%];p = .044), but there was no significant difference in other adherence indices or pharmacokinetics.

Table 5

DISCUSSION

This study demonstrated that when compared with subjects taking PI-based HAART who maintained long-term viral suppression, subjects who failed virologically displayed significantly lower adherence when measured at baseline by self-report and by MEMS Cap. Isolates from the majority of these subjects had the commonly described substitutions associated with PI resistance. However, a minority of subjects did not display such substitutions. Low adherence was associated with the absence of NFV-associated substitutions and the presence of a smaller total number of substitutions in Pr. A similar relationship with adherence was found for Pr phenotype.

Although failure of HAART may be defined as clinical progression, failure of CD4⁺ lymphocyte count to rise or detectable viremia (³), the presence of detectable viremia precedes CD4⁺ decline and disease progression in the majority of patients. Virologic failure may occur because of inadequate drug exposure (whether due to alterations in conventional pharmacokinetics or to changes at the cellular level in for example p-glycoprotein), inadequate adherence (whether due to missing doses, taking reduced doses, or failing to follow dietary instructions) or inadequate drug potency. Virologic failure also risks the emergence of antiviral drug resistance, which limits future treatment options. Numerous previous studies (detailed below) have examined elements of this process. However, we report the first attempt to bring these factors together and study them prospectively in a clinical cohort.

We found significant differences in adherence by MEMS Cap between the groups in our study. Although similar trends were observed during the study for both pill count and self-report, neither was sufficiently sensitive to detect a difference between the two groups. In part, this represents the problems of patient recall associated with self-report (¹⁴). However, it is also likely that participation in the study improved adherence in the viremic group during the study: self-report detected significantly lower adherence in this group in the period immediately preceding recruitment. Given the short observation period it is not possible to reach conclusions about causality from our data. Low adherence in certain individuals may have been the result of the individual's disappointment with their treatment response; in addition, patients with high observed adherence may have had prolonged spells of lower adherence or complete breaks with treatment in the past. An increasing number of studies report an association between virologic response and PI adherence measured by self-report (^15,16), pill count (¹⁶), and MEMS Cap (^16,17)-although the levels of adherence reported were somewhat lower than that in the current study.

Genotyping of Pr for subjects on IDV showed that the primary substitutions associated with IDV resistance (M46, V82) (¹⁸) were relatively common. However, the effects of such substitutions are known to be cumulative (¹⁹): only 2 patients exhibited a combination of these two substitutions, and in no individual was the combination of three or more substitutions detected that is thought to be necessary for significant resistance to IDV (¹⁹). In keeping with this, the prevalence of phenotypic resistance to IDV was low (20%). However, the D30N substitution alone is sufficient for resistance to NFV (^18,20). Consequently, we observed a higher prevalence of phenotypic resistance (42%) to NFV and D30N was the most commonly detected substitution.

We examined the interaction between adherence and resistance. Faced with small amounts of research data, the literature on this area lacks a clear consensus. Some authors have stated that low or intermittent adherence inevitably leads to rebound with resistant virus (^21,22). However, others have suggested that viral rebound occurring in patients who take treatment intermittently may lead to rebound with wild type virus only (¹⁸). Friedland and Williams (²³) proposed a normal distribution for probability of resistance emerging with varying degrees of adherence: they suggest that both extremely low and high adherence would be associated with a very low risk of resistance. Although it remains clear that viral resistance substitutions are selected most effectively under conditions of selective pressure (²⁴), whether adherence to the present drugs can be so high that viral resistance cannot emerge is open to question. We found a significant positive linear relationship between adherence and viral resistance, which was consistent in both genotypic and phenotypic assays. Our data therefore suggest that if a bell-shaped curve of adherence versus resistance does exist it is heavily skewed so that the greatest risk is associated with much higher adherence and that maximal selective pressure may be imposed by only marginally suboptimal adherence. It is of interest that we also found higher adherence in subjects whose virus possessed NFV-associated resistance substitutions. Although this may have arisen by chance, such an association was not observed for IDV-associated substitutions. Again, this may reflect the fact that any single such substitution is insufficient for resistance to IDV and hence does not confer a significant selective advantage (¹⁸). However, as noted above, the extent to which adherence observed during this cross sectional study reflected past behavior is unknown as is the effect of this earlier behavior on viral resistance. These observations also cannot be extrapolated to patients with viremia detectable at <1000 copies/mL. At this level a quite different relationship between adherence and resistance might exist, which is not possible to examine within the limitations of commercially available resistance assays. Three previous reports have examined the interaction of adherence and resistance in detail: the study by Vanhove et al. (²⁵) included only 7 patients who were on saquinavir monotherapy. A clear relationship between adherence and resistance did not emerge. In the study by Bangsberg et al. (¹⁶) the number of subjects with detectable substitutions was small, so correlation with adherence was not reported; but adherence measured by pill count was significantly higher in those with Pr-resistance substitutions than in those without. However, the authors point out that their findings may have occurred partly because patient management did not follow modern treatment guidelines (a large proportion of subjects simply added a PI to their previous combination without at least 1 other new drug). More recently, Gallego et al. reported differential detection of PI resistance-associated mutations in subjects failing IDV-based regimens. Adherence was measured by self-report and treated as a dichotomous variable. Resistance-associated substitutions were isolated from a minority of subjects with adherence >90% but not from subjects reporting lower adherence. However, statistical association between these variables was not reported (²⁶). Our data confirm that lower adherence is associated with a lower risk of detection of drug resistance.

In contrast, we found no difference in plasma PI level between the two groups at any of the measured time points. Although PI levels at the first time point after dosing were lower than the reported maximum expected plasma concentration (C_max) for either drug (3000-4000 ng/mL for NFV (¹¹), 8970 ng/mL for IDV (¹⁰)) these results are consistent with our decision to sample after T_max, because this time point will vary between individuals. The trough we observed for IDV was also lower than that published (178 ng/mL) (¹⁰), whereas that of NFV was consistent with published data (700-2200 ng/mL) (¹¹). Several published studies have sought an association between plasma PI level and treatment effect. Drusano et al. (²⁷) found no association between IDV C_min (minimum expected plasma concentration at end of dosing interval), C_max or area under the curve (AUC) and virologic rebound. However, in a study of ritonavir monotherapy, Molla et al. (²⁸) found that higher ritonavir C_min and AUC were associated with delayed emergence of ritonavir resistance. Hoetelmans et al. (²⁹) noted that plasma exposure to NFV related to rate of VL decay on a double PI-based regimen, and Burger et al. (³⁰) reported that treatment failure was more likely in subjects with lower random IDV levels; treatment adherence was not measured in these studies. However Murri et al. (³¹) reported that undetectable random plasma PI levels were strongly associated with self-reported non-adherence, a finding echoed in our own data. Acosta et al. (³²) compared IDV pharmacokinetics in 9 patients with detectable viremia and 14 with undetectable viremia-all were on their first PI. They found large interindividual variations in AUC, 5-hour postdose levels, and projected C_min; these parameters were significantly higher among patients with undetectable viremia. This study differed from ours in that C_min was estimated rather than measured and was markedly lower (median, 28 ng/mL). Mean duration of PI therapy (10 months) was also shorter than that in our study (29 months). Two alternative explanations may be proposed for the differences in observations between the latter study and our own. First, because both studies were relatively small, they were liable to either type I or II statistical errors. Second, large differences in the pharmacokinetics of PIs may occur in clinical populations and account for some cases of early treatment failure (for example, failure to clear viremia after 6 months on a PI); these patients may then have their treatment changed or intensified. This would have the effect of selecting out patients with inadequate drug handling at an early stage, so that in a more treatment-experienced cohort such as ours, differences in PI levels would be less extreme. Therefore, it is not possible from our data to suggest that interindividual differences in pharmacokinetics have no role in predicting treatment response, merely that in patients well established on therapy, other factors take precedence. In a sub-group of patients in the Viradapt study, the virologic response to therapy in subjects whose trough PI level was less than twice the published IC₉₅ for their PI was significantly less that the virologic response in subjects whose trough level was greater than twice the IC₉₅ (³³). The dose preceding this trough was not given under observation. A significant proportion of subjects had undetectable plasma PI levels; although adherence was not measured in this study, the authors conclude that this is a major cause of low PI levels. However, early data from a prospective study performed by the same group (³⁴) failed to show any benefit in virologic outcome in adjusting the PI dose according to measured plasma PI level.

Although there were no differences in pharmacokinetics between viremic and nonviremic subjects, in the subgroup of viremic subjects on NFV, we found that high degrees of viral resistance were associated with lower plasma PI levels postdose and higher levels of adherence. No similar association was found for IDV. Numbers in this subgroup were small and these associations may have arisen by chance. However, they are biologically plausible and suggest two possible routes to viremia in patients taking NFV. PIs act during a relatively short part of the HIV replicative cycle. Subjects who are highly adherent but have low peak NFV levels may persistently expose their viral pool to marginally inhibitory levels of PI, thus providing prolonged selective pressure for resistant strains. In contrast, subjects with low adherence (and normal pharmacokinetics) may take doses of NFV too sporadically to provide inhibitory plasma levels. Although episodically they may achieve higher adherence, the total time their viral pool is exposed to marginally inhibitory PI levels may be less than in the former group, providing fewer opportunities for viral selection and a lower risk of viral resistance. However, this interpretation of the data is speculative. Two previous studies have compared adherence, resistance, and pharmacokinetics in subjects experiencing virologic failure or not. Both were induction-maintenance studies of IDV-based regimens. In the first, Descamps et al. (⁸) found no substitutions in Pr at the time of first virologic rebound. However, 6 weeks after first rebound a small number of such substitutions did emerge, although none was primary. Plasma IDV levels and adherence to IDV (measured by pill count) were significantly lower in the maintenance phase among subjects with virologic rebound. However, this study differed from our own in the following respects: IDV levels were taken randomly and were not timed objectively. Differences in IDV levels may therefore have been confounded by differences in adherence between the two groups and pill count has been shown to give higher estimates of adherence than MEMS Caps (^9,35-37). Importantly, because of the low prevalence of detectable resistance the authors did not examine interactions with adherence. In the second study, Havlir et al. (³⁸) found no primary substitutions or changes in phenotype in patients failing IDV and no difference in random IDV levels in comparison with those not displaying virologic rebound. The low incidence of resistance in both these studies compared with our own may reflect differences in study design: in our cohort a longer time passed between viral rebound and sampling during which further substitutions could have occurred. Given that it is not routine practice to take samples of resistance testing with each VL, our data may more closely reflect clinical reality.

This study has a number of limitations. First, participation in the study may have increased subjects' adherence during the period of observation, particularly in the days leading up to the pharmacokinetic assessment. This effect may have been further exacerbated if subjects had been aware that adherence was being directly monitored (the effect of electronic monitoring on dosing behavior is unknown) (¹⁴). Although individuals may have guessed the purpose of the MEMS Cap, we took steps to minimize this. Second, the MEMS Cap system has limitations in its ability to detect nonadherence. Despite our instructions subjects may have removed more than one dose for later consumption or have opened the bottle without removing a dose. The former has been shown to be commoner than the latter (^35,37,39,40). Third, subjects came from a clinical cohort with diverse treatment histories. There were therefore significant differences between the groups in the proportions that were treatment naive on commencing the PI. Previous NRTI exposure may have primed subjects in the viremic group for subsequent treatment failure. However, this would not account for the observed relationship between lower adherence and PI resistance. Fourth, current genotyping techniques are unable to detect minority populations of resistant virus. It is possible that resistance substitutions may have arisen in all subjects failing a PI but that in those with low adherence selective pressure was so low that wild type predominated, and resistant mutants were not detectable (⁴¹). Finally, we did not perform full pharmacokinetic profiles, and AUC may have provided a more accurate estimate of drug exposure particularly if there were large differences in C_max. However, this technique is labor intensive and does not reflect the likely role of therapeutic drug monitoring for PIs in clinical practice. Despite these limitations, we believe our findings are relevant to future research and practice.

CONCLUSIONS

Virologic failure of PI-based HAART is the result of interactions between a wide variety of viral, host, and treatment-related variables. Patients experiencing virologic failure displayed significantly lower medication adherence than those stable on such regimens, but antiviral drug resistance was most likely to emerge in subjects with higher adherence. In our population, interindividual differences in plasma PI exposure were not associated with treatment failure per se. The first HAART combination offers the best chance of long-term success, and patients face unprecedented demands on their ability to adhere. Although development of new drugs and more sophisticated diagnostic tests are important, the real challenge in the treatment of HIV is the struggle patients face each day to take each dose on time.

Acknowledgments:

Funding for measurement of plasma nelfinavir concentrations and for the MEMS TrackCaps was supported by Roche Products Ltd., U.K.