JAC Advance Access originally published online on October 5, 2006
Journal of Antimicrobial Chemotherapy 2006 58(5):1036-1043; doi:10.1093/jac/dkl386
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
Explaining variability in the relationship between antiretroviral adherence and HIV mutation accumulation
1 Section of General Internal Medicine, Yale University School of Medicine and VA Connecticut Healthcare System New Haven, CT, USA 2 Department of Industrial Engineering, School of Engineering, University of Pittsburgh Pittsburgh, PA, USA 3 Section of Decision Sciences and Clinical Systems Modeling, University of Pittsburgh School of Medicine Pittsburgh, PA, USA 4 Epidemiology and Prevention Interventions Center, Division of Infectious Diseases, Positive Health Program, University of California at San Francisco San Francisco, CA, USA 5 Department of Medicine, University of British Columbia Vancouver, Canada
*Corresponding author. Tel: +1-203-932-5711; Fax: +1-203-937-4926; E-mail: ronald.braithwaite{at}va.gov
Received 28 April 2006; returned 8 June 2006; revised 29 August 2006; accepted 30 August 2006
 |
Abstract |
|---|
 Top Abstract IntroductionMethodsResultsDiscussionTransparency declarationsSupplementary dataReferences |
|---|
Objectives: Determining the relationship between antiretroviral adherence and resistance accumulation is important for the design and evaluation of adherence interventions. Our objective was to explain heterogeneity observed in this relationship.
Methods: We first conducted a systematic review to locate published reports describing the relationship between adherence and resistance. We then used a validated computer simulation to simulate the patient populations in these reports, exploring the impact of changes in individual patient characteristics (age, CD4, viral load, prior antiretroviral experience) on the shape of the adherence–resistance (A–R) curve.
Results: The search identified 493 titles, of which 3 contained relevant primary data and 2 had sufficient follow-up for inclusion (HOMER and REACH cohorts). When simulating HOMER, the A–R curve had a high peak with a greatly increased hazard ratio (HR) of accumulating mutations at partial compared to complete adherence (simulation, HR 2.9; HOMER, HR 2.7). When simulating REACH, the A–R curve had a shallow peak with a slightly increased hazard of accumulating mutations at partial adherence (simulation, HR 1.2; REACH, HR 1.4). This heterogeneity was primarily attributable to differences in antiretroviral experience between the cohorts.
Conclusions: Our computer simulation was able to explain much of the heterogeneity in observed A–R curves.
Keywords: AIDS , HAART , effectiveness , efficacy
|
Introduction |
|---|
|
TopAbstractIntroductionMethodsResultsDiscussionTransparency declarationsSupplementary dataReferences |
|---|
Combination antiretroviral therapies have transformed HIV to a chronic disease, decreasing mortality rates from 3- to 10-fold.1,2 Because numerous studies have demonstrated that high levels of adherence to antiretroviral therapies are required in order for these therapies to be successful, promoting adherence has become one of the highest priorities in HIV care.3,4 It has been postulated that the relationship between adherence and mutation accumulation has an inverse-u shape (Figure 1),5,6 with minimal rates of mutation accumulation when adherence levels are very low (because selection pressures are low) or very high (because viral replication is suppressed), and a maximal rate of mutation accumulation when adherence levels are intermediate (because selection pressures accompany brisk viral replication).
|
Defining this relationship is important for informing the design and evaluation of adherence interventions. Whereas the short-term impact of adherence interventions can be estimated based upon known associations of adherence with virological and clinical outcomes,7 their long-term impact is more difficult to estimate. Increasing adherence may change the rate at which drug-resistant virus emerges, which consequently may impact clinical outcomes in the long term.
Data describing the relationship of adherence to resistance have recently emerged, and although they broadly support the inverse-u hypothesis, their shapes differ substantially.8,9 Bangsberg et al.8 found a gently descending slope between the peak and the right side of the curve, with little difference in mutation accumulation between perfect adherence and commonly observed levels of adherence (70–90%), whereas Harrigan et al.9 found a steeply descending slope, with large differences in mutation accumulation. Furthermore, these differences occur across a heavily populated region of the curve, magnifying their clinical implications.
Because the factors which determine the shape of the A–R curve are important yet are not fully understood, we have used a previously validated computer simulation of adherence and resistance accumulation in HIV10,11 to explore why this curve may be shaped differently for different patient populations.
|
Methods |
|---|
|
TopAbstractIntroductionMethodsResultsDiscussionTransparency declarationsSupplementary dataReferences |
|---|
We first describe our literature review, and then how we used our computer simulation to explore what factors may have caused the variation in observed results.
Literature review
We conducted an electronic literature search using MEDLINE, starting from 1996 (when combination antiretroviral therapies first came into common use) and ending in March, 2005. Our search strategy required at least one of the following terms to identify adherence (text words adherence or compliance, or subject heading patient compliance), at least one of the following terms to identify resistance (text words mutation, mutations, genotype, genotypic, phenotype, phenotypic and resistance; or subject headings mutation, genotype, phenotype and drug resistance including its viral subheadings), and at least one of the following terms to identify antiretroviral medications (text words antiretroviral therapy, combination therapy, or HAART, or subject headings drug therapy, combination and antiretroviral therapy, highly active). Titles and abstracts were then manually reviewed, and manuscripts were obtained if they included primary data delineating an A–R curve that were based on prospective and longitudinal measurements of genotypic mutations, or if they included models that aimed to estimate these curves. Letters to the editor, comments, editorials and review articles were excluded during manual review of titles and abstracts. Research that was only published in abstract form was excluded. The search was not limited by language of publication, as all abstracts were available in English.
Computer simulation
We have developed a probabilistic simulation of the natural history of HIV disease in the current treatment era of combination antiretroviral therapies specifically designed to model HIV drug-resistance mutation accumulation by level of adherence.10,11 Our simulation generates HIV mutations at a rate proportional to viral replication, and models the accumulation of these mutations in the viral population based on the presence of selection pressure produced by antiretroviral drug regimens. It separately tracks the number of accumulated mutations that may confer resistance to each of the three constituent drug categories in antiretroviral therapy: nucleoside reverse transcriptase inhibitors (NRTIs), non-nucleoside reverse transcriptase inhibitors (NNRTIs) and protease inhibitors. It then uses this information to determine the likelihood of phenotypic resistance to combination therapies (Figure 2).
|
Each simulated patient proceeds through the model separately. The chance that each of these events can occur is based on a random draw from a probability distribution that reflects the characteristics of the particular patient. For example, the overall probability of adhering to medications is a specified characteristic; however the adherence at any one particular time is probabilistic and therefore may vary greatly. Similarly, the observed viral load at any one particular time may vary greatly. Because each hypothetical patient may have a distinct clinical trajectory, the simulation can mirror the heterogeneity that is present in true populations.
This simulation has been calibrated by ensuring that it could reproduce Kaplan–Meier curves describing survival and the time to failure of rounds 1, 2 and 3 of combination antiretroviral therapy in a large multi-site observational cohort study,10 and has been validated by demonstrating that it could predict mortality for a separate clinical cohort,10 and by verifying that the mutation accumulation and non-adherence rates producing the best model fit were well within the ranges of clinical observations.11 Additional details of the simulation are described elsewhere10,11 and in the Appendix [available as Supplementary data at JAC Online (http://jac.oxfordjournals.org/)]. It is important to emphasize that the model does not directly represent the relationship between adherence and resistance accumulation. Rather, it models directly the relationship of adherence to HAART efficacy, which, in turn, affects viral load and accumulated mutations. Therefore, A–R curves presented in this paper are ‘emergent properties’ of the model, and are not derived from direct statistical modelling of the relationship between adherence and resistance.
Cohorts simulated
Based on the results of our literature search, we simulated two cohorts with contrasting A–R curves, the HOMER and REACH cohorts (Table 1).8,9 The HOMER cohort consists of 1191 antiretroviral-naive individuals initiating antiretroviral therapy. Genotype testing was a prospectively defined outcome measure that was performed approximately four times per year on any individual with a detectable viral load. Adherence was estimated by identifying the ratio of filled-to-prescribed medication orders. Adherence–resistance (A–R) analyses in REACH cohort consists of 148 individuals without permanent housing, the majority of who were antiretroviral experienced. Genotype testing was also a prospectively defined outcome measure, and was performed at least twice over an interval of at least 6 months. Adherence was estimated by study personnel, who interviewed enrollees on an unscheduled and unannounced basis, and conducted pill counts during the interview.
|
To approximate the HOMER cohort, we simulated an antiretroviral-naive cohort of individuals who were starting antiretroviral therapy, using baseline (pre-treatment) values for median age, viral load and CD4 count based on that cohort (Table 2). To approximate the patient population of REACH, we simulated an antiretroviral-experienced cohort of individuals who were starting a new round of combination therapy, using values for age and CD4 count based on that cohort (Table 2). Because the population of REACH was identified while on treatment, it was difficult to estimate a baseline viral load value, and to avoid falsely attributing differences in simulation results to differences in patient characteristics, we used the same baseline viral load value as the HOMER cohort.
|
Outcome measures
To summarize simulation results, we chose outcome measures that resembled those used by the corresponding clinical cohort. When simulating the HOMER cohort, we examined hazard ratios (HRs) for mutation accumulation at each adherence stratum relative to the accumulation rate in the highest adherence stratum, and we examined the proportion of individuals having new resistance to any, 
2 or 3 antiretroviral drug classes at the end of 1 year, irrespective of adherence. HOMER divided NRTI mutations into two separate groups (M184V versus other) whereas our simulation does not make this distinction. For this reason, our definitions for 2 classes and 3 classes do not correspond exactly.
We performed analyses both with and without an adjustment factor that compensated for pharmacy refill data's overestimation of adherence (Odds Ratio for missed doses, actual adherence compared to pharmacy refill estimate, 3.0). This adjustment factor was based on two studies that compared the adherence estimated by pharmacy refill or other medication inventories with a more resource-intensive but precise method of measuring medication doses based on recording whether pill bottles were opened at the appropriate times (Medication Event Monitoring System, or ‘MEMs caps’). Choo et al.12 compared MEMs caps with pharmacy refill measurements among 286 patients receiving antihypertensive therapy, and the corresponding Odds Ratio for non-adherence according to MEMs caps versus medication inventories was 3.1. Liu et al.13 compared MEMs caps with clinic-based pill inventories among 108 HIV-infected adults receiving protease inhibitors, and found that the odds ratio for non-adherence according to MEMs caps versus medication inventories was 2.9.
When simulating the REACH cohort, we examined the mean number of new mutations in each adherence stratum. Because strata were defined on the basis of quintiles, we chose to simulate adherence levels that corresponded to the midpoint of each adherence quintile. As REACH measured adherence in the context of unscheduled and unanticipated interviews, overestimation of adherence was less likely, and we did not use the adjustment factor.
We tested the goodness of fit between A–R relationships reported by each study and A–R relationships specified by the corresponding simulation runs using the 
2 goodness-of-fit statistic.
Sensitivity analyses
We performed analyses to determine how our results would vary with different assumptions regarding patient and treatment characteristics.
Patient characteristics.. To determine whether differences in clinical characteristics would impact our results, we varied age, viral load, CD4 counts and extent of antiretroviral experience (as indicated by numbers of resistant mutations at baseline) across clinically plausible ranges, and examined the effect of these changes on the overall risks of accumulating mutations and on the shapes of the A–R curves.
Treatment characteristics.. We also varied assumptions regarding the relationship between adherence to antiretroviral therapy and viral load suppression, alternatively assuming a relationship that may be typical of NNRTI-based regimens14 (low levels of adherence producing little or no viral load suppression and moderate levels of adherence producing complete viral load suppression) and a relationship that may be typical of unboosted protease inhibitor-based regimens (viral load suppression incomplete even at high levels of adherence). Our baseline assumption regarding this relationship, a linear correlation between adherence and the logarithm of viral load suppression, was that which was most consistent with the simulation's previous calibration and validation.10
|
Results |
|---|
|
TopAbstractIntroductionMethodsResultsDiscussionTransparency declarationsSupplementary dataReferences |
|---|
We first conducted a literature review for studies that might further elucidate the characteristics of A–R curves. We then used a validated computer simulation of adherence and resistance accumulation in HIV to explore how the shape of this curve would vary with differing patient characteristics, and a preliminary exploration of how it would vary with different antiretroviral regimens.
Literature review
The electronic search identified 493 titles, of which 8 met our inclusion criteria during manual review.6,8,9,15–19 Of these 8 articles, 3 contained primary data that allowed construction of an A–R curve,8,9,15 2 had additional primary data but with insufficient detail to allow construction of an A–R curve16,17 and 3 employed statistical or mathematical modelling to gain insights about the A–R curve.6,18,19 Of the three articles that constructed A–R curves from primary data, one (Hopkins)15 estimated adherence over a brief timeframe (3 days), whereas two (HOMER and REACH)8,9 estimated adherence over the duration of the study. HOMER studied an antiretroviral-naive cohort and used pharmacy refill data to estimate adherence, whereas REACH studied an antiretroviral-experienced cohort and employed unannounced pill counts to estimate adherence. The HOMER and REACH groups had contrasting A–R curves, with the HOMER cohort finding a steep descent in mutation accumulation rates between commonly observed levels of adherence and complete adherence, and the REACH cohort finding only a slight descent. The A–R curve constructed from Hopkins data was more similar in shape to the HOMER curve than to the REACH curve. Because its estimations of adherence were based on very brief patient-reported timeframes (3 days) and very few individuals reported <60% adherence, which suggests that underreporting may have been frequent, we decided to focus our simulations on the HOMER and REACH cohorts.
Of the three modelling papers, two were published by the group of REACH investigators,6,18 and used statistical methods to explore the impact on the A–R curve of the type of antiretroviral therapy and the assumed probability of viral load suppression with complete adherence. These papers concluded that the A–R curve would have a more prominent peak if individuals were on a regimen with greater likelihood of viral load suppression with perfect adherence. The third modelling paper was performed before any clinical data were available,19 and its predictions were sufficiently discordant from subsequent clinical results that it will not be described further (>60% adherence would prevent any mutations from accumulating).
Computer simulation
Because our model allows us to define the simulated population by specifying a wide range of characteristics such as age and CD4 count, we were able to simulate an antiretroviral-naive cohort with characteristics similar to HOMER cohort as well as an antiretroviral-experienced cohort with summary characteristics similar to the REACH cohort.
Mutation accumulation rate.. The simulation was able to closely reproduce the overall mutation accumulation rates of each cohort. In HOMER, 17%, 6% and 1% of individuals had new mutations to any, 2 or more and 3 or more drug classes at the end of 1 year, respectively. The corresponding estimates for the simulation were 21%, 6% and 1%, respectively. HOMER did not analyse the mean number of mutations accumulated per year; the simulation estimated this value to be 0.28 mutations per year.
REACH did not stratify mutation accumulation by number of drug classes; however the simulation estimated that 58%, 27% and 11% of hypothetical individuals with characteristics similar to the REACH cohort would have new mutations to 1, 2 and 3 or more drug classes at the end of 1 year. REACH determined that the mean accumulation rate was 0.96 mutations per year. The corresponding estimate from the simulation was 1.00 mutations per year.
Adherence–resistance curve..
The simulation was able to reproduce a large amount of the variation in A–R curves between these two cohorts, estimating mutation accumulation rates that were often within the confidence intervals of the HOMER cohort (Figure 3), and were always within the confidence intervals of clinical observations in the REACH cohort (Figure 4). In particular, when the simulation reproduced the HOMER cohort, the A–R curve had a high amplitude peak with a greatly increased hazard of accumulating mutations with partial adherence compared with the highest adherence category (simulation, HR 2.91; HOMER, HR 2.73). Without adjusting for the tendency of pharmacy refill data to overestimate adherence, the simulation curve was shifted leftward compared with the clinical cohort curve, with the maximum risk of mutation accumulation occurring at 
60% adherence rather than at 85% adherence. However, after adjusting for this overestimation, this shift was no longer substantial, with both peaks occurring in the range of 60–70%, and the two curves were not statistically distinguishable (P = 0.84).
|
|
In contrast, when the simulation reproduced the REACH cohort, the A–R curve had a low amplitude peak, with only a small increased hazard of accumulating mutations with partial adherence compared with the highest adherence category (simulation, HR 1.15, REACH, HR 1.42). There was no apparent shift of the simulation curve compared with the clinical cohort curve, with the maximum risk of mutation accumulation occurring at
80% adherence for both curves. The two curves were not statistically distinguishable (P = 0.99). Variation of A–R curve with prior antiretroviral exposure.. Prior antiretroviral exposure impacts HIV outcomes by leading to genotypic mutations. We explored how the shape of the A–R curve varies with increasing antiretroviral exposure by increasing the number of genotypic mutations at the start of the simulation, and holding all other patient characteristics constant (Figure 5). When a cohort starts with no genotypic mutations (corresponding to a fully antiretroviral-naive population), the curve has a high amplitude peak and then a relatively steep downward slope over commonly observed adherence levels, and the overall rates of mutation accumulation are low. Therefore, its characteristics generally resemble those from the HOMER cohort. Prior antiretroviral exposure and pre-existing drug resistance lead to a lower amplitude peak and a more shallow downward slope over commonly observed adherence levels. The net effect suggests greater population levels of drug resistance, and resembles the A–R characteristics from the REACH cohort.
|
Variation of A–R curve with other patient characteristics.. Other patient characteristics had a much smaller impact on the A–R curve. Variations in age and CD4 count did not change its shape appreciably. However, baseline viral load had a significant impact on its magnitude, with higher viral loads amplifying the peak of the curve. When we assumed that an antiretroviral-naive population started the simulation with a low baseline viral load (10 000 copies/mL), the peak mutation accumulation rate (0.132 mutations per year) was only 1.6 times as large as the rate with complete adherence (0.085 mutations per year). In contrast, when we assumed that an antiretroviral-naive population started with a high baseline viral load (1 000 000 copies/mL), the peak accumulation rate (0.533 mutations per year) was 4.4 times as large as rate with complete adherence (0.120 mutations per year).
Another important factor influencing the shape of the A–R curve was the presumed relationship of adherence to viral load suppression. For our base case analyses, we assumed that adherence had a linear relationship with the logarithm of viral load suppression, as this assumption was most compatible with our previous calibration and validation of the simulation. However, we also explored the impact of two contrasting but plausible alternative forms for this relationship, based on previously observed variations with the type of antiretroviral regimen. When we assumed that low levels of adherence produced little or no viral load suppression and moderate levels of adherence were sufficient to produce complete viral load suppression, as may be typical of NNRTI-based antiretroviral regimens, the peak of the A–R curve shifted to the left, moving from 60% adherence to 40% adherence. When we assumed that viral load suppression was not complete even at high levels of adherence, as may be typical of unboosted protease inhibitor-based antiretroviral regimens, the peak of the A–R curve shifted to the right, moving from 60% adherence to 80% adherence.
|
Discussion |
|---|
|
TopAbstractIntroductionMethodsResultsDiscussionTransparency declarationsSupplementary dataReferences |
|---|
We used a previously validated computer simulation of HIV disease to gain insights about which factors lead to the variation in A–R curves that has been observed clinically. We were able to replicate many of the contrasting features of the A–R curves between populations by varying the level of exposure to prior antiretroviral therapy.
For antiretroviral-naive individuals, simulated and observed results both suggest that the likelihood of accumulating new mutations will increase sharply with even small departures from perfect adherence, with a rise to 1.9 times higher for individuals with 90% adherence and to 2.4 times higher for individuals with 80% adherence. Indeed, the maximum likelihoods of accumulating mutations occur at some of the most commonly observed adherence rates (60–80%). This implies that many antiretroviral-naive individuals may benefit substantially from adherence interventions not only because of the short-term benefit that accrues from greater viral load suppression, but also because a long-term benefit would accrue due to preservation of future drug options. Therefore adherence interventions may be particularly effective at preventing drug resistance over this range of adherence. In contrast, for antiretroviral-experienced individuals, simulated and observed results both suggest that the likelihood of accumulating new mutations will increase only slightly with small departures from perfect adherence. Therefore, the main benefit of adherence interventions in this group may stem from the short-term gain of greater viral load suppression, rather than the long-term gain of preserving future drug options.
There are plausible biological reasons for why the A–R relationship should vary with prior antiretroviral exposure. Pre-existing mutations confer partial resistance to combination therapies, and therefore viral replication continues at a brisk pace even if adherence is perfect. Because this replication occurs under circumstances of great selection pressure, mutations will accrue rapidly even with perfect adherence. In contrast, if there is no resistance at baseline, perfect adherence will strongly suppress viral replication, so even though selection pressures are strong, mutations will accrue slowly.
Our results expand upon but are broadly consistent with other models of adherence and resistance among individuals with HIV. Bangsberg et al.6 used statistical modelling to explore varying assumptions about how effectively antiretroviral therapy would suppress viral load with complete adherence. They found that if suppression was assumed to be rare, the A–R curve would rise monotonically and lose its inverse-u shape, whereas if suppression was assumed to be guaranteed, the curve would have a nearly symmetric inverse-u shape with steep ascending and descending slopes. Our simulation produced similar changes when we varied the amount of antiretroviral experience, a major determinate of viral load suppression with complete adherence. Kagay et al.20 estimated the cost-effectiveness of directly observed therapy (DOT), and estimated that it would be ineffective at preventing resistance because many individuals were on the ascending slope of the A–R curve. In our simulation, comparably adherent individuals would not benefit from DOT unless the effect size was extremely large. Bangsberg et al.14 modelled the impact of different types of antiretroviral regimens on the A–R curve, and found that NNRTI-based regimens were associated with a leftward shift of the peak of the A–R curve, also consistent with our results. Lastly, a manuscript that was published after we performed our systematic review, King et al.,21 described the relationship between mutation accumulation and adherence in 653 antiretroviral naive HIV patients. They also found that the A–R curve had a high amplitude peak in naive patients, with a greatly increased risk of accumulating mutations at partial compared with complete adherence.
Our work has important limitations. Adherence is characterized only by the proportion of doses consumed, and pattern of non-adherence is not considered even though adherence pattern may be an important modulator of the A–R curve. Modelling was based on unboosted protease inhibitor regimens rather than the boosted protease inhibitor regimens that have largely replaced them and that may protect from the emergence of drug-resistance mutations more effectively. Treatment experience is only likely to explain variance in the A–R curve if there is cross-resistance between HAART regimens; if newer HAART regimens were developed with entirely distinct resistance patterns, our results would no longer be generalizable. Finally, although we explored many possible factors that may account for the heterogeneity in clinical data, it is always possible that unexamined factors may have played an important role. However, this analysis has the unique strength of being based on a simulation that has been extensively validated,10,11 and that was able to estimate overall mutation accumulation rates from cohorts with greatly differing characteristics.
In summary, we have used a validated computer simulation of mutation accumulation in HIV disease to replicate the differences in A–R curves that have been observed clinically. Because current clinical data on A–R curves is limited to very few sources, future research is needed to define these curves in additional populations, as well as the degree of virological and clinical benefit of effective adherence interventions.
|
Transparency declarations |
|---|
|
TopAbstractIntroductionMethodsResultsDiscussionTransparency declarationsSupplementary dataReferences |
|---|
Drs R. S. B., S. S., M. S. R., A. S. and A. C. J. have none to declare. Dr D. R. B. has received honoraria from Abbott, Pfizer and GlaxoSmithKline as well as grants from Abbott and Pfizer. Dr P. R. H. has worked in the pharmaceutical industry (Glaxo Wellcome) and HIV diagnostic industries (Virco). He has received fees for consulting, honoraria, owned stock, acted as a consultant and received grants from a range of companies including Abbott Laboratories, Boehringer Ingelheim, Bristol-Myers Squibb, GlaxoSmithKline, Roche, Northern Lipids, Merck, Pfizer and Virco.
|
Supplementary data |
|---|
|
TopAbstractIntroductionMethodsResultsDiscussionTransparency declarationsSupplementary dataReferences |
|---|
The Appendix is available as Supplementary data at JAC Online (http://jac.oxfordjournals.org/).
|
Acknowledgements |
|---|
Supported by grant K23 AA14483-01, National Institute of Alcohol Abuse and Alcoholism, National Institutes of Health.
|
References |
|---|
|
TopAbstractIntroductionMethodsResultsDiscussionTransparency declarationsSupplementary dataReferences |
|---|
1 Palella FJ Jr, Delaney KM, Moorman AC, et al. (1998) Declining morbidity and mortality among patients with advanced human immunodeficiency virus infection. HIV Outpatient Study Investigators. N Engl J Med 338:853–60.
2 Karon JM, Fleming PL, Steketee RW, et al. (2001) HIV in the United States at the turn of the century: an epidemic in transition. Am J Public Health 91:1060–8.[Abstract]
3 Phillips AN, Pradier C, Lazzarin A, et al. (2001) Viral load outcome of non-nucleoside reverse transcriptase inhibitor regimens for 2203 mainly antiretroviral-experienced patients. AIDS 15:2385–95.[CrossRef][ISI][Medline]
4 Paris D, Ledergerber B, Weber R, et al. (1999) Incidence and predictors of virologic failure of antiretroviral triple-drug therapy in a community-based cohort. AIDS Res Hum Retroviruses 15:1631–8.[CrossRef][ISI][Medline]
5 Friedland GH and Williams A. (1999) Attaining higher goals in HIV treatment: the central importance of adherence. AIDS 13:S61–72.
6 Bangsberg DR, Porco TC, Kagay C, et al. (2004) Modeling the HIV protease inhibitor adherence-resistance curve by use of empirically derived estimates. J Infect Dis 190:162–5.[CrossRef][ISI][Medline]
7
Paterson DL, Swindells S, Mohr J, et al. (2000) Adherence to protease inhibitor therapy and outcomes in patients with HIV infection. Ann Intern Med 133:21–30.
8 Bangsberg DR, Charlebois ED, Grant RM, et al. (2003) High levels of adherence do not prevent accumulation of HIV drug resistance mutations. AIDS 17:1925–32.[CrossRef][ISI][Medline]
9 Harrigan PR, Hogg RS, Dong WW, et al. (2005) Predictors of HIV drug-resistance mutations in a large antiretroviral-naive cohort initiating triple antiretroviral therapy. J Infect Dis 191:339–47.[CrossRef][ISI][Medline]
10 Braithwaite RS, Justice AC, Chang CC, et al. (2005) Estimating the proportion of patients infected with HIV who will die of comorbid diseases. Am J Med 118:890–8.[CrossRef][ISI][Medline]
11 Braithwaite RS, Schechter S, Chung C, et al. (2006) Estimating the rate of accumulating drug resistance mutations in the HIV genome. Value Health in press.
12 Choo PW, Rand CS, Inui TS, et al. (1999) Validation of patient reports, automated pharmacy records, and pill counts with electronic monitoring of adherence to antihypertensive therapy. Med Care 37:846–57.[CrossRef][ISI][Medline]
13
Liu H, Golin CE, Miller LG, et al. (2001) A comparison study of multiple measures of adherence to HIV protease inhibitors. Ann Intern Med 134:968–77.
14 Bangsberg DR, Acosta EP, Gupta R, et al. (2006) Adherence-resistance relationships for protease and non-nucleoside reverse transcriptase inhibitors explained by virological fitness. AIDS 20:223–31.[ISI][Medline]
15 Sethi AK, Celentano DD, Gange SJ, et al. (2003) Association between adherence to antiretroviral therapy and human immunodeficiency virus drug resistance. Clin Infect Dis 37:1112–18.[CrossRef][ISI][Medline]
16 Roge BT, Barfod TS, Kirk O, et al. (2004) Resistance profiles and adherence at primary virological failure in three different highly active antiretroviral therapy regimens: analysis of failure rates in a randomized study. HIV Med 5:344–51.[CrossRef][ISI][Medline]
17
de la Rosa R, Ruiz-Mateos E, Rubio A, et al. (2004) Long-term virological outcome and resistance mutations at virological rebound in HIV-infected adults on protease inhibitor-sparing highly active antiretroviral therapy. J Antimicrob Chemother 53:95–101.
18
Bangsberg DR., Moss AR, Deeks SG. (2004) Paradoxes of adherence and drug resistance to HIV antiretroviral therapy. J Antimicrob Chemother 53:696–9.
19 Wahl LM and Nowak MA. (2000) Adherence and drug resistance: predictions for therapy outcome. Proc Biol Sci 267:835–43.
20 Kagay CR, Porco TC, Liechty CA, et al. (2004) Modeling the impact of modified directly observed antiretroviral therapy on HIV suppression and resistance, disease progression, and death. Clin Infect Dis 38:S414–20.
21 King MS, Brun SC, Kempf DJ. (2005) Relationship between adherence and the development of resistance in antiretroviral-naive, HIV-1-infected patients receiving lopinavir/ritonavir or nelfinavir. J Infect Dis 191:2046–52.[CrossRef][ISI][Medline]
CiteULike 
Connotea 
Del.icio.us What's this?
This article has been cited by other articles:
|
|
 |
|
  J. Shuter Forgiveness of non-adherence to HIV-1 antiretroviral therapy J. Antimicrob. Chemother., April 1, 2008; 61(4): 769 - 773. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 S. Willard and C. Flexner Introduction: The Changing Face of HIV J Int Assoc Physicians AIDS Care (Chic Ill), March 1, 2008; 7(1_suppl): S3 - S4. [PDF]
|
|
|
|
 |
|
 R. S. Braithwaite, M. S. Roberts, C. C. H. Chang, M. B. Goetz, C. L. Gibert, M. C. Rodriguez-Barradas, S. Shechter, A. Schaefer, K. Nucifora, R. Koppenhaver, et al. Influence of Alternative Thresholds for Initiating HIV Treatment on Quality-Adjusted Life Expectancy: A Decision Model Ann Intern Med, February 5, 2008; 148(3): 178 - 185. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 R. S. Braithwaite, J. Concato, C. C. Chang, M. S. Roberts, and A. C. Justice A Framework for Tailoring Clinical Guidelines to Comorbidity at the Point of Care Arch Intern Med, November 26, 2007; 167(21): 2361 - 2365. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 S. J Gange, M. M Kitahata, M. S Saag, D. R Bangsberg, R. J Bosch, J. T Brooks, L. Calzavara, S. G Deeks, J. J Eron, K. A Gebo, et al. Cohort Profile: The North American AIDS Cohort Collaboration on Research and Design (NA-ACCORD) Int. J. Epidemiol., April 1, 2007; 36(2): 294 - 301. [Full Text] [PDF]
|
|
|
|
|
|
R. S. Braithwaite, M. S. Roberts, and A. C. Justice Incorporating Quality of Evidence into Decision Analytic Modeling Ann Intern Med, January 16, 2007; 146(2): 133 - 141. [Abstract] [Full Text] [PDF]
|
|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||