点击显示 收起
Global Pharmaceutical Research and Development, Abbott Laboratories, Abbott Park, Illinois
Background.
Relationships between adherence to protease inhibitor (PI)based therapy and resistance development have not been fully characterized.
Methods.
We conducted a double-blind, randomized, controlled study of lopinavir/ritonavir versus nelfinavir, each administered with stavudine and lamivudine, in 653 antiretroviral-naive, human immunodeficiency virus (HIV)1infected patients. Relationships between adherence and probability of resistance development were evaluated by local linear regression or logistic regression.
Results.
A higher risk of detectable HIV-1 RNA loads after week 24 was associated with lower adherence (odds ratio [OR], 1.08 per 1% decrease in adherence [95% confidence interval {CI}, 1.051.10]; P < .001) and nelfinavir use (OR, 2.4 vs. lopinavir/ritonavir [95% CI, 1.63.6]; P < .001). Among all nelfinavir-treated patients, a bell-shaped relationship between adherence and the risk of nelfinavir resistance was observed, with a maximum probability of 20% at 85%90% adherence. No lopinavir resistance was observed. A bell-shaped relationship was also observed for the probability of lamivudine resistance, with a maximum probability of 50% at 75%80% adherence to nelfinavir and of 15% at 80%85% adherence to lopinavir/ritonavir.
Conclusions.
Bell-shaped relationships between adherence and resistance were observed. Irrespective of adherence level, the risk of detectable HIV-1 RNA loads or of PI or lamivudine resistance was significantly higher in nelfinavir-treated patients than in lopinavir/ritonavir-treated patients.
Nonadherence to antiretroviral therapy (ART) by HIV-1infected patients has been associated with incomplete viral suppression [1], the development of drug resistance [2, 3], and disease progression and mortality [46]. The nature of the relationship between adherence and the risk of developing drug resistance has not been fully characterized. A concave or bell-shaped relationship between adherence and the risk of developing drug resistance has been proposed [7]. This pattern has been noted to occur in a small cohort of patients who received protease inhibitors (PIs) but has not been observed in patients who received nonnucleoside reverse-transcriptase inhibitors (NNRTIs) or lamivudine [8]. The relationship between adherence and resistance has not been evaluated for ritonavir-boosted PI regimens, nor has it been assessed for specific ART regimens in large, randomized clinical trials. Furthermore, prior analyses have used linear models [9], which can lead to a large modeling bias if the relationship is nonlinear, or have applied arbitrary categorical cutoffs [10], which may potentially miss important patterns or suggest relationships that are only artifacts.
Therefore, in a double-blind, randomized, phase 3 clinical trial of ritonavir-boosted PI (lopinavir/ritonavir) compared with an unboosted PI (nelfinavir), we sought to model the relationship between adherence and resistance, using novel statistical techniques that do not constrain the functional form of the relationship.
PATIENTS, MATERIALS, AND METHODS
Patients and study design.
ART-naive, HIV-1infected patients were enrolled in a double-blind, randomized, phase 3 clinical trial comparing lopinavir/ritonavir with nelfinavir. Details of the study design, clinical results, and the emergence of genotypic or phenotypic resistance over the course of 108 weeks have been reported elsewhere [1113].
A total of 653 patients were enrolled at 93 clinical sites in 13 countries on 5 continents. Patients were randomized 1 : 1, in a blinded fashion, to receive lopinavir/ritonavir (400/100 mg twice daily) plus nelfinavir placebo or to receive nelfinavir (750 mg 3 times daily, converted to 1250 mg twice daily after US Food and Drug Administration licensure of twice-daily dosing) plus lopinavir/ritonavir placebo, and all patients received standard doses of lamivudine and stavudine twice daily. Patients were evaluated at baseline, every 4 weeks through week 24, every 8 weeks through week 48, and every 12 weeks thereafter.
Resistance.
For any patient with an HIV-1 RNA load >400 copies/mL anytime at or after week 24, a sample was submitted for genotypic resistance testing. Genotype was determined at ViroLogic by population sequencing. Primary PI resistance for nelfinavir-treated patients was defined as the emergence of a D30N or L90M mutation in protease or an M46I/L mutation plus confirmed phenotypic resistance (>2.5-fold vs. wild-type [wt] HIV-1; PhenoSense, ViroLogic); resistance for lopinavir/ritonavir-treated patients was defined as any primary or active site mutation in protease (amino acid positions 8, 30, 32, 4648, 50, 54, 82, 84, and 90) [14]. Secondary PI mutations were defined as those appearing at amino acid positions 10, 20, 33, 53, 71, 73, 77, and 88, plus amino acid positions 46 and 54 for nelfinavir. Lamivudine resistance was defined as the presence of the M184V/I/T mutation in reverse transcriptase. In patients who never demonstrated HIV-1 RNA loads >400 copies/mL at or after week 24 and in patients whose samples could not be amplified for resistance testing, resistance was assumed to be absent for the purposes of the analysis.
Adherence.
At each study visit, pill counts of returned study drug were conducted. Adherence was calculated as the number of pills consumed relative to the number expected to have been consumed. Between any 2 study visits, adherence was censored at a maximum of 100%. Overall PI adherence and overall lamivudine adherence were computed. For PI adherence, only the active PI (not placebo) was used in the assessment of adherence.
Statistical analysis.
Using logistic regression, we assessed the probability of demonstrating an HIV-1 RNA load >400 copies/mL anytime at or after week 24 by overall PI adherence and treatment group. We initially explored the relationship between adherence and the probability of resistance using local linear regression [15]; this nonparametric regression technique can be useful if the shape of the underlying relationship is unknown or is difficult to recognize on a scatter plot of the data. In local linear regression, for a given point along the X-axis, a "local" linear model is created with use of only the fraction of the data closest to the point of interest. This process is repeated across the range of the independent variable. Data points included in each local analysis are weighted by their proximity to the point of interest. The size of the local neighborhood, referred to as the "bandwidth," determines how many data points are included in each local analysis. For a bandwidth of 10, all data points within 10 units on the X-axis are included, whereas, for a bandwidth of 1, only data points within 1 unit on the X-axis are included. Thus, larger bandwidth values result in a "smoother" fit, because a larger fraction of the data are included in each local model and each local model has more points in common with the next nearest local model. An infinite bandwidth generalizes the local linear model to the traditional linear model, because each local model includes all data points and weights them equally. At the other end of the spectrum, a bandwidth of 0 essentially interpolates the data, akin to connecting the dots of the data points. The advantage of such nonparametric regression methods is that the data are allowed to search for a suitable functionthe model is not restricted to a particular functional form. In a traditional linear or logistic regression model, if the data do not fit the assumed functional form, a large modeling bias may result.
On the basis of the local linear regression exploration, a second-order logistic regression model was used to assess the probability of resistance by adherence and to compare treatment groups. The probabilities of primary PI resistance, of secondary PI mutations, and of lamivudine resistance were assessed among all enrolled patients who remained in the study for at least 24 weeks and among patients with viremia for whom genotypic resistance data were available. In the former group, patients who did not have viral rebound or for whom genotypic data were unavailable were considered not to have resistance.
RESULTS
Predictors of detectable HIV-1 RNA loads.
Of 653 patients enrolled in the study, 590 (292 in the lopinavir/ritonavir group and 298 in the nelfinavir group) remained in the study for at least 24 weeks and were included in the analysis. Anytime at or after week 24, a detectable HIV-1 RNA load (>400 copies/mL) was observed in 74 (25%) lopinavir/ritonavir-treated patients and in 123 (41%) nelfinavir-treated patients. Lower PI adherence and treatment with nelfinavir were each highly statistically significantly associated with a higher probability of demonstrating detectable HIV-1 RNA loads (figure 1). For adherence levels of 95%, 90%, 80%, and 70%, the estimated probabilities of demonstrating detectable HIV-1 RNA loads any time at or after week 24 were 29%, 37%, 55%, and 72%, respectively, for nelfinavir-treated patients. For lopinavir/ritonavir-treated patients, corresponding probabilities were 14%, 20%, 34%, and 53%, respectively. For lopinavir/ritonavir-treated patients, to obtain a probability of >50% of never demonstrating a detectable HIV-1 RNA load, adherence of at least 72% was sufficient, whereas, for nelfinavir-treated patients, >83% adherence was required.
Adherence and resistance among all patients treated for at least 24 weeks.
Overall adherence, as measured by pill count, was quite high in the study, with a mean (median) of 92% (95%) in the lopinavir/ritonavir group and 93% (96%) in the nelfinavir group. Results of resistance testing were available for 51 (69%) of 74 lopinavir/ritonavir-treated and 96 (78%) of 123 nelfinavir-treated patients. For the remaining patients, samples could not be amplified for resistance testing, generally because of low numbers of HIV-1 RNA copies in plasma (median, 819 copies/mL). For the purposes of the present analysis, resistance was assumed to be absent in these patients, as well as in patients whose HIV-1 RNA loads remained <400 copies/mL. Primary PI resistance was observed in 0 of 51 lopinavir/ritonavir-treated patients and in 46 (48%) of 96 nelfinavir-treated patients. Secondary PI mutations and lamivudine resistance were observed in 7 (14%) and 19 (37%) of 51 lopinavir/ritonavir-treated patients, respectively, and in 51 (53%) and 79 (82%) of 96 nelfinavir-treated patients, respectively. Relationships between adherence and lopinavir or stavudine resistance were not assessed, because no patients demonstrated lopinavir resistance and because stavudine resistance was uncommon (0% in the lopinavir/ritonavir group and 9% in the nelfinavir group).
On the basis of the second-order logistic regression models suggested by the local linear regression, the probability of primary PI resistance, secondary PI mutations, and lamivudine resistance was compared between treatment arms. In each comparison, the probability of resistance was statistically significantly higher for nelfinavir-treated patients than for lopinavir/ritonavir-treated patients across all adherence levels (figure 3; P < .001 for each comparison). For nelfinavir-treated patients, the maximum probability of primary PI mutations was 20%, at an adherence level of 85%90%, and the maximum probability of lamivudine resistance was >50%, at an adherence level of 75%85%. For lopinavir/ritonavir-treated patients, the maximum probability of lamivudine resistance was 15%, at an adherence level of 80%85%.
Adherence and resistance among patients with viremia for whom genotype data were available.
When analyses were restricted to patients who had viremia and for whom genotypic data were available, adherence remained similar between treatment groups, with a mean of 89% in the lopinavir/ritonavir group and of 90% in the nelfinavir group. Within each group, these mean values were significantly lower (P < .05 for each group) than the mean adherence values among patients for whom genotypic data were not available (93% for the lopinavir/ritonavir group and 94% for the nelfinavir group). Among patients for whom genotypic data were not available, the mean adherence values for patients with at least 1 detectable HIV-1 RNA value at or after week 24 were similar to those for patients with no such detectable HIV-1 RNA loads.
Among patients for whom genotype data were available, the probability of primary PI resistance increased with increasing adherence for nelfinavir-treated patients, reaching a maximum of 65% in patients who had a viral rebound and 100% adherence (figure 4A). A similar relationship was observed for secondary PI mutations for both drug regimens, although the slope of the relationship appeared to be steeper for the unboosted (nelfinavir) than for the boosted (lopinavir/ritonavir) regimen. In contrast, for lamivudine resistance, patients with the highest adherence levels (97%100%) demonstrated a small decrease in the risk of resistance, compared with the risk in those with 80%85% adherence (figure 4B and 4C). Among nelfinavir-treated patients who developed resistance, patients with both primary PI resistance and lamivudine resistance had higher mean adherence than patients with only lamivudine resistance (93% vs. 85%, respectively; P < .001).
DISCUSSION
Across all adherence levels in our study, the probability of detectable viremia and of PI or lamivudine resistance was significantly lower for lopinavir/ritonavir-treated patients than for nelfinavir-treated patients, which is consistent with higher antiviral activity and lower risk of resistance with the lopinavir/ritonavir-based regimen. Paterson et al. [1] demonstrated a significant association between adherence and the risk of virologic failure, with the lowest risk of virologic failure observed among patients with adherence >95%. These results are sometimes flatly interpreted to mean that 95% adherence is a minimum threshold for reliable viral suppression [16, 17]. However, the results were based on single PIs administered without boosting doses of ritonavir. Our results indicate that, with a more potent PI-based regimen, viral suppression may be reliably achieved across a broader range of adherence. Thus, although increasing adherence was associated with a better virologic response in both regimens, 10%12% higher adherence to nelfinavir was required to achieve a virologic response equivalent to that of lopinavir/ritonavir. Moreover, at adherence levels >75%, the relative risk of detectable HIV-1 RNA loads was 50%110% higher for nelfinavir-treated patients than for lopinavir/ritonavir-treated patients. As a consequence, although adherence remains a crucial factor in maximizing viral suppression, it is important to recognize that different PI-based regimens may offer different levels of "forgiveness" to nonadherence.
We observed a bell-shaped relationship between adherence and the probability of primary PI resistance among nelfinavir-treated patients, on the basis of statistical methods that made no assumptions about the form of the relationship. Our results are consistent with those of the model proposed by Friedland and Williams [7] and observed in a prospective study by Bangsberg et al. [8]. Ours is the first randomized, controlled study to confirm such a relationship. Among patients with detectable viral load anytime at or after week 24, we observed that, among nelfinavir-treated patients, the risk of primary PI resistance increased with increasing adherence, which is consistent with several previous reports [8, 9, 18, 19]. The nature of the relationship between adherence and resistance to the lopinavir/ritonavir regimen could not be assessed, because no lopinavir/ritonavir-treated patients demonstrated primary PI resistance. Notably, the absence of resistance to lopinavir across a range of 65%100% adherence suggests that this regimen exerts a high genetic barrier to resistance that is tolerant of variable degrees of nonadherence.
Our study had several limitations. The results may not apply to other classes of drugs. NNRTIs, in particular, are likely to have different adherence-resistance relationships [18], although results from Harrigan et al. [10] and Sethi et al. [20] are suggestive of a skewed bell-shaped model among patients receiving various PI- and/or NNRTI-based regimens. Because no PI resistance was observed among lopinavir/ritonavir-treated patients in the present study, we were not able to assess whether the relationship for boosted PIs was the same as that for unboosted PIs, but this observation suggests that the overall risk during treatment with lopinavir/ritonavir is quite low. Correspondingly, in other clinical trials of lopinavir/ritonavir in ART-naive patients, no PI resistance was observed for up to 4 years of treatment [21, 22], and a similar observation has been reported for one trial of ritonavir-boosted fosamprenavir that lasted 48 weeks [23]. The bell-shaped relationship that we observed for lamivudine is unlike that reported by Bangsberg et al. [18] but is consistent with the bell-shaped relationship observed by Harrigan et al. [10].
Differences in the adherence-resistance relationships within and between individual drug classes may reflect differences in the pharmacokinetic and genetic barriers provided by each specific agent. The NNRTI class generally provides high drug levels relative to wt susceptibility, but a single mutation can provide levels of resistance that overcome these exposures (high pharmacokinetic barrier but low genetic barrier to resistance). Unboosted PIs generally have a low pharmacokinetic barrier but a higher genetic barrier, because trough drug levels are rarely >24-fold more than the IC50 of wt HIV-1, but single mutations tend to lead to more-limited changes in viral susceptibility. Correspondingly, although resistance to PIs is not common at low levels of adherence, NNRTI mutations have been reported to arise after only single doses [24], which indicates that resistance to NNRTIs is facile at low adherence levels, likely because of long drug half-lives. For ritonavir-boosted PIs, both the pharmacokinetic and genetic barriers to resistance are generally high. In addition, lopinavir levels have been shown to decay rapidly during missed doses, which theoretically reduces the amount of time during which the selection of mutant virus is most likely [25]. Thus, the absence of resistance to lopinavir/ritonavir, irrespective of adherence level, is likely explained by the pharmacokinetic and genetic barriers in combination with the pharmacokinetic profile during missed doses of lopinavir/ritonavir.
It is unclear why we observed a different resistance-adherence relationship for lamivudine than have other studies; however, because different lamivudine-containing regimens can have very different risks of lamivudine resistance [12], it is plausible that differences in regimen may also affect the resistance-adherence relationship. Notably, although the risk of lamivudine resistance was higher than the risk of primary PI resistance in each group, the overall relationship with adherence was similar. In the nelfinavir-treated group, patients with only lamivudine resistance had lower mean adherence than those with both lamivudine and PI resistance. Thus, higher adherence to regimens with unboosted PIs may, among patients with detectable viral load, increase the likelihood of multiclass resistance. Because of the generally high adherence values that we observed, our results do not address the likelihood of resistance emergence at very low adherence levels, which may be more common outside the setting of a clinical trial. Other studies have suggested that PI-treated patients with very low adherence (<40%) are unlikely to develop PI resistance [9, 18].
The use of overall adherence for the study period limits our ability to differentiate between different adherence patterns resulting in the same overall adherence level; for example, 2 patients may demonstrate 80% adherence, but 1 may do so by missing every fifth dose and the other by perfect adherence except for a drug holiday of 20% of the study period, and the resistance implications may be quite different. Adherence was assessed by pill counts at regular study visits; this potentially resulted in an overestimation of the adherence rate, but, because pill counts are generally correlated with electronic adherence measures [26], the relationships that we observed would likely have held but would have been shifted toward lower adherence values if more-sensitive adherence monitoring tools had been used.
In summary, our results match theoretical and empirical results that have shown a bell-shaped relationship between adherence and resistance, as well as those that have shown that, among patients with detectable viremia, higher adherence increases the chance of PI resistance. Our study is the first to demonstrate these relationships in a randomized, controlled trial. Antiretroviral agents with high pharmacokinetic and genetic barriers to resistance, such lopinavir/ritonavir, appear to minimize the development of resistance by reducing both the risk of detectable viremia and the consequences thereofthat is, as the likelihood of detectable viremia is reduced, so is the probability of resistance from replicating virus during periods of poor adherence. Prospective studies that address the relationship between adherence and resistance may benefit from the use of electronic adherence measures, given that diverse patterns of nonadherence are likely to have different effects on the risk of the development of resistance.
Acknowledgments
We gratefully acknowledge Kathleen Sheehan, Guang Yang, Eric Bauer, and Jennifer Moseley, for their contributions to study and database management.
References
1. Paterson DL, Swindells, S, Mohr J, et al. Adherence to protease inhibitor therapy and outcomes in patients with HIV infection. Ann Intern Med 2000; 133:2130. First citation in article
2. Wainberg MA, Friedland G. Public health implications of antiretroviral therapy and HIV drug resistance. JAMA 1998; 279:197783. First citation in article
3. Mayers DL. Drug-resistant HIV-1: the virus strikes back. JAMA 1998; 279:20002. First citation in article
4. Bangsberg DR, Perry S, Charlebois E, et al. Non-adherence to highly active antiretroviral therapy predicts progression to AIDS. AIDS 2001; 15:11813. First citation in article
5. Hogg RS, Heath K, Bangsberg D, et al. Intermittent use of triple-combination therapy is predictive of mortality at baseline and after 1 year of follow-up. AIDS 2002; 16:10518. First citation in article
6. Wood E, Hogg RS, Yip B, Harrigan PR, O'Shaughnessy MV, Montaner JS. Effect of medication adherence on survival of HIV-infected adults who start highly active antiretroviral therapy when the CD4+ cell count is 0.200 to 0.350 × 10(9) cells/L. Ann Intern Med 2003; 139:8106. First citation in article
7. Friedland GH, Williams A. Attaining higher goals in HIV treatment: the central importance of adherence. AIDS 1999; 13(Suppl 1):S6172. First citation in article
8. Bangsberg DR, Porco TC, Kagay C, et al. Modeling the HIV protease inhibitor adherence-resistance curve by use of empirically derived estimates. J Infect Dis 2004; 190:1625. First citation in article
9. Walsh JC, Pozniak AL, Nelson MR, Mandalia S, Gazzard BG. Virologic rebound on HAART in the context of low treatment adherence is associated with a low prevalence of antiretroviral drug resistance. J Acquir Immune Defic Syndr 2002; 30:27887. First citation in article
10. Harrigan PR, Hogg RS, Dong WWY, et al. Predictors of HIV drug-resistance mutations in a large antiretroviral-naive cohort initiating triple antiretroviral therapy. J Infect Dis 2005; 191:33947. First citation in article
11. Walmsley S, Bernstein B, King M, et al. Lopinavir-ritonavir versus nelfinavir for the initial treatment of HIV infection. N Engl J Med 2002; 346:203946. First citation in article
12. Kempf DJ, King MS, Bernstein B, et al. Incidence of resistance in a double-blind study comparing lopinavir/ritonavir plus stavudine and lamivudine to nelfinavir plus stavudine and lamivudine. J Infect Dis 2004; 189:5160. First citation in article
13. King MS, Bernstein BM, Walmsley SL, et al. Baseline HIV-1 RNA and CD4 cell count predict loss of virologic response to nelfinavir, but not lopinavir/ritonavir, in antiretroviral therapynaive patients. J Infect Dis 2004; 190:2804. First citation in article
14. Hirsch MS, Brun-Vezinet F, Clotet B, et al. Antiretroviral drug resistance testing in adults infected with human immunodeficiency virus type 1: 2003 recommendations of an International AIDS SocietyUSA Panel. Clin Infect Dis 2003; 37:11328. First citation in article
15. Fan J, Gijbels I. Local polynomial modelling and its applications. London: Chapman & Hall, 1996. First citation in article
16. Lesho EP, Gey DC. Managing issues related to antiretroviral therapy. Am Fam Physician 2003; 68:67586. First citation in article
17. Carrieri MP, Chesney MA, Spire B, et al. Failure to maintain adherence to HAART in a cohort of French HIV-positive injecting drug users. Int J Behav Med 2003; 10:114. First citation in article
18. Bangsberg DR, Charlebois ED, Grant RM, et al. High levels of adherence do not prevent accumulation of HIV drug resistance mutations. AIDS 2003; 17:192532. First citation in article
19. Gallego O, de Mendoza C, Perez-Elias MJ, et al. Drug resistance in patients experiencing early virological failure under a triple combination including indinavir. AIDS 2001; 15:17016. First citation in article
20. Sethi AK, Celentano DD, Gange SJ, Moore RD, Gallant JE. Association between adherence to antiretroviral therapy and human immunodeficiency virus drug resistance. Clin Infect Dis 2003; 37:11128. First citation in article
21. Hicks C, King MS, Gulick, RM, et al. Long-term safety and durable antiretroviral activity of lopinavir/ritonavir in treatment-naive patients: 4-year follow-up study. AIDS 2004; 18:7759. First citation in article
22. Eron JJ, Feinberg J, Kessler HA, et al. Once-daily versus twice-daily lopinavir/ritonavir in antiretroviral-naive HIV-positive patients: a 48-week randomized clinical trial. J Infect Dis 2004; 189:26572. First citation in article
23. MacManus S, Yates PJ, Elston RC, White S, Richards N, Snowden W. GW433908/ritonavir once daily in antiretroviral therapy-naive HIV-infected patients: absence of protease resistance at 48 weeks. AIDS 2004; 18:6515. First citation in article
24. Jourdain G, Ngo-Giang-Huong N, Le Coeur S, et al. Intrapartum exposure to nevirapine and subsequent maternal responses to nevirapine-based antiretroviral therapy. N Engl J Med 2004; 351:22940. First citation in article
25. Bertz R, Chiu Y-L, Foit C, et al. Estimation of selective pressure by lopinavir/ritonavir vs. nelfinavir by examination of terminal-phase pharmacokinetics at steady-state [abstract 6.1]. In: Programme and abstracts of the 5th International Workshop on Clinical Pharmacology of HIV Therapy (Rome). Utrecht: Virology Education, 2004:31. First citation in article
26. Liu H, Golin CE, Miller LG, et al. A comparison study of multiple measures of adherence to HIV protease inhibitors. Ann Intern Med 2001; 134:96877. First citation in article