Secondary Logo

Journal Logo


Adherence to protease inhibitors, HIV-1 viral load, and development of drug resistance in an indigent population

Bangsberg, David R.a,b; Hecht, Frederick M.b; Charlebois, Edwin D.a; Zolopa, Andrew R.f; Holodniy, Markf,g; Sheiner, Lewisc; Bamberger, Joshua D.h; Chesney, Margaret A.d; Moss, Andrewe

Author Information
  • Free



Recent reports suggest a dramatic improvement in HIV-related mortality since the introduction of combination antiretroviral therapy with protease inhibitors (PI) [1–3]. Failure to adhere to these complex regimens appears to be related to the failure of viral suppression [4–6]. Because it may predict inadequate viral suppression, non-adherence to therapy may also be a risk factor for the development of drug-resistant strains of HIV [7–10]. However, the relationship between adherence, viral suppression and antiretroviral resistance is poorly understood, and in particular, the level of adherence necessary for viral suppression below the limits of detection for current assays has not been established.

The relationship between adherence, viral suppression and genotypic resistance was examined in a cross-sectional analysis of HIV-infected subjects recruited from a representative cohort of homeless and marginally housed people in San Francisco. Because patient self-report probably overestimates adherence [11–14], two objective methods were also studied: electronic medication monitors (container caps containing an electronic microchip to record medication container opening and closing) and pill counts [15,16]. The association between these measures of adherence, HIV-1 viral load, and viral genotypic resistance was assessed.


The Research in Access to Care in the Homeless cohort

We studied true homeless adults living in streets, shelters and other locations not intended for permanent housing, and marginally housed adults living in single-room occupancy (SRO) hotels. Individuals were systematically sampled: (i) from all San Francisco homeless shelters and free meal programs, and (ii) from a random sample of SRO hotels charging less than US$400 a month, in proportions corresponding to the sizes of these two strata in the population. The study is thus representative of the homeless and marginally housed population. The objectives, rationale, and representativeness of the shelter and free meal program sampling methods have previously been described [17]. Hotels were identified from city records, and sampled with a probability proportionate to size. Participants were recruited and HIV tested on a Monday and then returned for results on the following Friday. Participants received a US$10 reimbursement for each visit. Eighty-five per cent of participants agreed to testing and returned for results.

All HIV-infected homeless and marginally housed individuals identified in the screening were invited to join a prospective cohort study, the Research in Access to Care in the Homeless (REACH) study. A total of 89% of those invited agreed to participate in the study. Adherence to PI therapy was examined over 6–10 weeks for each participant between January 1998 and July 1998. All individuals on PI therapy were eligible.

Baseline adherence assessment

After consent was obtained, an appointment for a baseline adherence assessment was arranged at the subject's usual place of residence (SRO hotel, shelter, other). A research assistant collected the following information: usual medication routine, the number of self-reported missed doses, and the number of doses taken over the previous 3 days. Medication instructions, date of previous refill and number of tablets dispensed were obtained from the subject's PI medication container. Pill counts of PI tablets were conducted in duplicate. If the results did not match, a third pill count was performed. After the pill count, all PI tablets were transferred to the electronic medication monitor (Aprex Corp., Menlo Park, CA, USA). Any additional PI tablets and tablets that did not fit in the electronic medication monitor were counted and placed in a sealed plastic bag. Electronic medication monitors were not used for subjects employing weekly medisets in order not to interfere with adherence. Adherence in these individuals was assessed by self-report and pill count only. Subjects using the electronic medication monitor were instructed to take all doses from the electronic medication monitor, close the cap firmly after opening the medication container, keep the cap closed between doses, and place any refill tablets in the electronic medication monitor only after the container was empty.

When a subject's understanding of dosing instructions differed from those on the medication container, the individual was alerted to the discrepancy and instructed to contact his or her healthcare provider to clarify the discrepancy.

Periodic adherence assessment

Subjects were told that the research assistant would return ‘every 2 to 4 weeks’ to retrieve information from the electronic medication monitor and count their pills. Subjects were not told the time nor date of the next visit. On a random day between 2 and 4 weeks later, the research assistant would attempt to locate the subject. If the subject was not located on the first attempt, the research assistant would attempt repeat contacts until he or she was located. If the subject did not have the medications on his or her person, a repeat contact was attempted later.

Three periodic adherence assessments were carried out for each participant. The following information was collected at each assessment: whether each dose over the previous 3 days was taken from the electronic medication monitor or from another place such as the subject's pocket; the number of electronic medication monitor cap openings without a corresponding dose over the previous 3 days; a count of all PI tablets in the subject's possession including those in the electronic medication monitor and those stored in the sealed plastic bag; and all items covered in the baseline adherence assessment. At the end of the assessment, the electronic medication monitor cap was exchanged and the electronic medication monitor container was filled with a recorded number of PI tablets. Subjects received a US$10 reimbursement for each assessment.

Adherence measures

Four measures of adherence were defined:

Three day self-report

The percentage of prescribed doses reported taken over the previous 3 day interval. The mean value of three periodic adherence assessment measurements is used in the analysis.

Three day adjusted electronically monitored doses

The number of doses taken directly from the electronically monitored container, plus the number of self-reported ‘pocket’ doses (doses stored somewhere other than the electronically monitored container) combined over the previous 3 day interval. If the electronic record indicated that the electronically monitored container was not opened for a 24 h period, 0% adherence was recorded for that period regardless of self-report. The number of container openings without a corresponding dose was subtracted from the total number of openings. The mean value of the three periodic adherence assessment measurements is used in the analysis. This measure was developed during previous pilot studies in response to the observation that 30% of participants regularly took medications from somewhere other than the pill bottle, such as their pocket.

Pill count

The difference between subsequent pill counts divided by the number of tablets prescribed during the time between counts obtained at each periodic adherence assessment. Percentage adherence over the entire study period is used in the analysis.

Percentage of days doses taken

The percentage of days that the subject opened the electronic medication monitor at least once during the entire study period in the analysis.

Viral load determinations

Two viral load determinations were collected over the study period for each subject. The mean of the two determinations is used in the analysis. Viral load was determined with the Roche polymerase chain reaction (PCR) ultrasensitive assay with a lower limit of quantification of 20 copies/ml. Viral load was treated both as a continuous and a categorical variable (< or ≥ 400 copies/ml) in the analysis.

Genotypic resistance determination

HIV RNA was extracted using a QIAamp Viral RNA Kit (Qiagen, Valencia, CA, USA) according to the manufacturer's recommendations. A 1.3 kb sequence in the polymerase gene of HIV-1 was amplified by reverse transcription and PCR in a one tube reverse transcriptase (RT)–PCR technique using an HIV-1 RT–PCR kit (Visible Genetics Inc., Toronto, Ontario, Canada) according to the manufacturer's recommendations. Cy5.0 and Cy5.5-labelled sequencing ladders were generated from amplified cDNA by CLIP sequencing. Four overlapping sequencing reactions were obtained that span the protease gene from codon 4 to codon 310 in the RT gene. Sequencing ladders were detected in a MicroCell using the MicroGene Clipper. Chromatograms for each reaction were analysed using GeneObjects software (Visible Genetics) and compared with the sequence of HIV-1HXB2. All sequences were manually proofread for accuracy.

Antiretroviral resistance (defined in Table 1) was expressed as: (i) codon substitutions consistent with clinically significant PI resistance to current PI (yes/no); (ii) the number of drugs in the current regimen to which the patient's HIV variant had substitutions associated with clinically significant resistance; (iii) the presence of any primary substitution in the protease gene (yes/no); and (iv) the presence of accessory substitutions in the protease gene (yes/no).

Table 1:
Protease and reverse transcriptase codon substitutions used in the analysis and definitions of drug resistance by codon substitution.

Statistical methods

All data were analysed using the SAS statistical analysis software (SAS Institute, Cary, NC, USA) or software supplied with the electronic medication monitor. Fisher's exact test or chi square test for trend (Mantel–Haenszel) were used to test associations between categorical variables. The Wilcoxon two-sample test or non-parametric analysis-of-variance was used to test for differences in continuous variables by categorical variables. A paired t-test was used to evaluate temporal differences in the two viral load measurements obtained. Least squares linear regression (bivariate or multivariate) was use to test associations between continuous variables.


A total of 2058 subjects were screened, 188 were identified as HIV positive and 153 (81%) of these were recruited in the REACH cohort. Thirty-six subjects received PI therapy during the period January 1998 to June 1998; 34 of the 36 subjects (94%) consented to the study.

The population was characterized by high rates of hospitalization for mental illness, drug use, unstable housing and incarceration. When we compared them with REACH participants not using highly active antiretroviral therapy (HAART), the 34 patients in the study receiving HAART were more likely to be non-injecting-men who have sex with men (MSM), and less likely to report ever using illegal drugs (Table 2). Subjects on HAART were also marginally less likely to describe themselves as ever having been homeless. The two groups were similar with respect to age, race, sex, CD4 cell count, and history of psychiatric hospitalization.

Table 2:
Subject characteristics.

Antiretroviral regimen

Thirty-three subjects were receiving one PI and two RT inhibitors (RTI), and one subject was receiving one PI and only one RTI. The median duration of PI combination therapy was 12 months at the study midpoint. Twenty out of 34 (59%) subjects had received RTI before PI therapy and 14 out of 34 (41%) were treatment-naïve at the time the PI was initiated. Of those subjects with previous RTI therapy, eight (40%) had at least one RTI changed when their PI was started, but 12 (60%) had a PI added to an existing RTI regimen. The median duration of RTI therapy before initiating a PI was 5 months for those with previous RTI exposure. Seven (21%) subjects were using mediset containers to organize their medication.

Adherence assessment completion rates

A total of 134 out of 136 (98%) scheduled self-report assessments were completed. The median duration of observation was 66 days (9.4 weeks).

Adherence by self-report, pill count, adjusted electronically monitored doses, and percentage of days doses taken

The median proportion of prescribed doses taken over the study period was 89% by self-report (range: 16–100%), 73% by pill count (range: 3–100%), 67% by adjusted electronically monitored doses (AEMD; range 0–100%), and 100% by percentage of days doses taken (PDDT; range: 20–100%). No difference was observed in adherence at baseline and follow-up visits (Table 3). Thirty-eight per cent of the subjects had over 90% adherence by pill count.

Table 3:
Median adherence by each measure and visit number.

Adherence and viral suppression

Strong linear relationships were found between the adherence measures and concurrent log viral load during the study period (Fig. 1). AEMD and pill count were most strongly related to log viral load (r = 0.81, P < 0.0001 and r = 0.67, P < 0.0001, respectively). Self-reported adherence and PDDT (not shown in Fig. 1) were somewhat more weakly related to HIV RNA (r = 0.60, P = 0.0002 and r = 0.56, P = 0.004, respectively). The slopes of the regression lines in Fig. 1 were similar for self-report, pill count and AEMD (0.36, 0.34, and 0.37, respectively).

Fig. 1.:
a–c.  Adherence and viral suppression by type of adherence measure. Least squares linear regression estimate designated by solid lines. Lines represent the least squares regression equation with 95% confidence intervals. r, correlation coefficient. See text for definitions of adherence measures.

The proportion of subjects with HIV RNA of less than 400 copies/ml was examined by adherence level for the first three measures (Table 4). Adherence categories were designated by quartiles of pill count adherence. The prevalence of HIV RNA of less than 400 copies/ml was 56–67% for those with 98–100% adherence, depending on the measure of adherence.

Table 4:
Percentage HIV RNA below 400 copies/ml by quartile adherence.

Failure to take medications for more than 24 h was correlated with HIV RNA of over 400 copies/ml. Eight out of 15 people (53%) who opened their pill bottle at least once every day had a HIV RNA of less than 400 copies/ml, whereas none of 12 people who had at least one episode of failing to open the pill bottle for a day had a HIV RNA of less than 400 copies/ml (P = 0.02).

To evaluate for an intervention effect of the adherence measurements on viral load, the first and second viral loads were compared with a paired t-test. In order not to confound the analysis by those patients who had a declining viral load due to the recent initiation of therapy, two patients with less than 3 months of therapy were excluded. There was a small but statistically significant decrease between the first and second viral load measured over the observation period (log viral load no. 1 = 3.6, SD = 1.6; log viral load no. 2 = 3.3, SD = 1.5;P = 0.03).

Genotypic resistance

HIV-1 genotypic analysis was attempted in all 34 subjects. A complete genotype was obtained in 32 patients.

According to our genotypic definitions of resistance, three patients (9%) were resistant to their current PI. Nine (28%) patients had a primary protease gene substitution, and an additional 18 patients (total 27; 84%) had at least one secondary substitution. Fourteen (44%) patients had viral resistance to their current RTI. Of these 14 patients, three had primary protease gene substitutions. Twenty-two (69%) had one or more primary resistance substitutions in the RT gene.

Looking at resistance to both PI and RTI, nine (31%) patients were resistant to one current drug; four (13%) were resistant to two current drugs, and one (3%) was resistant to all three current drugs. All subjects with protease gene substitutions also had RT substitutions. Many patients had substitutions associated with drugs taken at a previous stage of therapy rather than with current antiretroviral drugs.

Drug resistance substitutions in the protease gene were observed more frequently in adherent than in non-adherent subjects (Fig. 2a,b). The mean pill count adherence was 88% in subjects with any primary protease gene substitution (n = 9) and 59% in those without a protease gene substitution (n = 23;P = 0.013). Clinically significant PI resistance (defined in Table 1) was not seen in subjects with less than 50% adherence. The 50% most adherent subjects were resistant to a mean 0.93 drugs in their current regimen, and the 50% least adherent subjects were resistant to a mean 0.31 drugs in their current regimen (P = 0.03).

Fig. 2.:
  Drug resistance by adherence and viral suppression. a. Protease inhibitor resistance by pill count adherence and viral suppression. Solid dot indicates significant resistant to current protease inhibitor, star indicates presence of any primary protease substitution without clinically significant resistance, circle indicates no primary protease gene substitution. b. Number of currently prescribed drugs to which subject's HIV isolate demonstrated significant resistance by PC adherence and viral load. Arabic numeral represents the number of current antiretroviral agents with genotypically defined resistance for each patient (see Table 1); +, genotype not available. Lines represent the least squares regression equation with 95% confidence intervals.

However, individuals with protease gene substitutions were significantly more likely to have had their PI added without changing one or more of their RTI. Of those who had at least one RTI changed (or started) when their PI was started, three out of 20 (15%) had a primary resistance substitution. Of those who did not have their RTI changed, six out of 12 (50%) developed a primary resistance substitution (P = 0.04).

Multivariate analysis of adherence, drug resistance, duration of protease inhibitor use, CD4 cell count and HIV RNA

A multivariate linear regression was carried out to examine the relationship between pill count adherence and HIV RNA, controlling for resistance to two or more current drugs, the duration of PI use, and CD4 cell count. Pill count adherence was a highly significant predictor of current log viral HIV RNA level (r = 0.73, P = 0.001). In the full model, every 10% difference in adherence was associated with a doubling of HIV RNA. The presence of resistance to two or more antiretroviral drugs, the duration of PI use, and CD4 cell count did not independently predict HIV RNA when pill count adherence was used in the model.


A strong relationship was found between adherence and concurrent HIV-1 viral load in homeless and marginally housed patients on PI combination therapy. A decrease of 10% in adherence was associated with a doubling of the HIV RNA level, suggesting that small differences in adherence may result in major differences in virological control. Current adherence explained between 36 and 65% of the variation in HIV RNA level, depending on the adherence measure. Strong relationships were seen for pill count and AEMD measures of adherence, and a slightly less strong relationship with 3 day self-report. The three measures of adherence led to similar estimates of the relationship between adherence and viral suppression.

Adherence was a much stronger predictor of viral load than was drug resistance in this study. The relative strength of adherence vis-à-vis drug resistance as a predictor of virological response may vary according to the clinical setting, drug regimen and study population. In this study, however, it was observed that subjects with drug resistance substitutions but good adherence still achieved relatively low viral loads. These subjects may have had low viral loads because they adhered to treatment that was still effective despite the presence of drug resistance substitutions.

Because this study is cross-sectional, information on pre-treatment viral load levels is lacking. This information would allow us to answer more definitively the issue of whether adherence resulted in sustained viral load reductions despite resistance substitutions. The lack of pre-treatment viral load data is also a potential limitation in the analysis linking adherence to viral load. It is possible that providers started therapy in presumed non-adherent patients only in the setting of late-stage disease and a high viral load. The high viral loads observed in non-adherent patients may thus be confounded by the stage of disease when they started therapy. The adjustment for current CD4 cell count in the multivariate model combined with the observation that providers have difficulty predicting adherence [18,19] reduces, but does not exclude, the potential for bias caused by this issue.

A close relationship was found between adherence and viral load after a median of 12 months of therapy. The relationship between adherence and viral load may not be the same during the initiation of HAART and at 12 months of therapy. Adherence during the initiation of therapy may be especially important, because people who achieve an undetectable viral load but then rebound have slower rates of disease progression than those whose viral load remains detectable [20].

A straightforward relationship between adherence and the development of resistance was not observed. Resistance to RTI was seen both in adherent and non-adherent subjects. RTI resistance probably reflected exposure to previous drug regimens in the 20 subjects (62%) who had a median of 5 months of RTI exposure before starting PI therapy. A greater level of resistance to RTI than PI was observed. Previous RTI resistance may lead to a less than optimal response to a PI-based regimen [6,21,22]. PI resistance was closely related to failure to change at least one RTI when initiating the PI, thus effectively adding monotherapy in those subjects with previous RTI resistance. This was common practice when PI were first introduced. Paradoxically, these subjects may have developed PI resistance because they were thought to be adherent and good candidates for early PI therapy.

A high prevalence of PI resistance was not observed in non-adherent subjects. Only one out of 20 (5%) subjects who had their PI therapy initiated together with new RTI had significant resistance to his or her PI after a median of 12 months of PI therapy. Therefore, in subjects for whom PI were initiated according to current recommendations, PI resistance was uncommon, despite a very wide range of adherence characterizing this indigent population. Primary protease gene substitutions were not seen in any of eight subjects with less than 50% adherence. This finding has several limitations. The possibility that minority populations of drug-resistant HIV variants were not detected by our genotypic testing cannot be excluded. Also, accessory protease gene substitutions were common and were seen in non-adherent subjects. These substitutions may allow for the development of more significant resistance [23].

Current thinking is that non-adherence will lead to rapid antiretroviral resistance as a result of viral replication in the presence of drug pressure. However, the level of adherence that creates the most efficient combination of drug pressure and viral replication to select for resistant virus is unknown. On the basis of these data, we hypothesize that extremely poor adherence (< 50%) does not often lead to the rapid development of PI resistance because of inadequate selective pressure for resistant virus.

If the development of resistance among poorly adherent subjects is not rapid, it may be safer than currently assumed to give patients a brief (e.g. one month) ‘trial’ of therapy in order to assess their adherence. Such an adherence trial, with monitoring, could be useful to assess candidates for longer-term therapy. As medical providers have significant difficulty assessing adherence in individual patients [18,19], it is often difficult to distinguish good from poor candidates for therapy (on the basis of predicted adherence) before starting therapy. Because of the limitations discussed above, the suggestion of an adherence trial period is exploratory, and its safety requires confirmation by further studies before its use in clinical practice.

Although adherence varied considerably, with many of the subjects taking less than 50% of their medication, overall, it was better than commonly anticipated. Fifty per cent of the population reported over 89% adherence, although the more reliable pill count median adherence was 73%. A substantial fraction of the cohort (38%) had better than 90% pill count adherence. Haubrich et al. [24] reported that 11% of patients in a clinical trial had less than 80% adherence. This level of adherence is higher than in the patients in this study. However, clinical trial populations are highly selected and not representative of the general HIV-infected population in the USA. Very little adherence data have been reported to date among observational rather than clinical trial cohorts. Furthermore, there are no standard measures of adherence to make valid comparisons. To the extent that adherence is related to viral suppression, the subjects in this study had comparable rates of viral suppression after one year on therapy (30% with < 400 copies/ml) to patients followed at the Johns Hopkins Moore Clinic (37% with < 500 copies/ml) [25], but substantially less than patients followed in the Swiss Cohort (81% with < 400 copies) [20]. Neither study, however, included formal adherence assessments. We believe that until formal and comparable adherence assessments are reported, it is unclear whether the level of adherence measured in homeless and marginally housed people is substantially different from that in other populations outside of clinical trials [4,26]. This level of adherence, however, is notable considering the high rates of comorbidity and competing life needs in these individuals, and suggests that combination PI therapy can be successful in many people in this population.

Only 24% of the REACH cohort of HIV-infected homeless and marginally housed people had been put on HAART at the time of this study. This treatment rate is low compared with stably housed HIV-infected populations [1], but is comparable to treatment rates seen in observational cohort studies of injection drug users [27,28]. As adherence was measured in 34 out of 36 people on HAART, these findings reflect adherence in the subset of homeless and marginally housed people currently receiving therapy in San Francisco.

Although the sample studied reflects those persons currently on therapy, those on therapy clearly do not reflect the population as a whole, but rather physicians’ decisions about which patients to prescribe to. These decisions may well have been based on assumptions about adherence. Participants receiving HAART were more likely to be non-injecting MSM and to report never using illegal drugs, and were marginally less likely to describe themselves as homeless. These factors may reflect the belief of many providers that drug use and homelessness are barriers to adherence. Because decisions about who should get therapy and when to start therapy are complex in this population [29–31], additional factors associated with receiving HAART probably exist that we were not able to measure. These adherence findings may thus not be generalizable to those homeless and marginally housed individuals who receive therapy in the future.

These data should be regarded as preliminary, given the limitations discussed above, the small number of subjects, and cross-sectional design of this study. Although the sample size is small, it was systematically obtained and few of those eligible refused to participate. There was not sufficient power to analyse adherence by subgroups or to estimate the precise level of adherence associated with a high probability of viral suppression. Definitive inferences about the causal relationship between adherence, viral load, and resistance require a prospective rather than a cross-sectional study, including pretreatment viral load, CD4 cell count and drug resistance substitutions.

The intensive measurement of adherence may have altered adherence itself. A statistically significant (but not biologically significant) decrease in viral load was observed over the study period. It is possible that intensive adherence measurement improved adherence and reduced viral load. However, it was not possible to detect an improvement in adherence during the study interval. There was no difference between self-reported adherence at baseline (without concomitant objective measures) and during follow-up (with concomitant objective measures). Although no improvement in the objective measures were observed during the observation period, it is possible that the adherence measurements led to improved adherence relative to pre-observation levels, subsequently leading to a declining viral load over the observation period (Table 3).

Regardless of these challenges, accurate methods to assess and monitor adherence may be essential for determining whether there is a range of non-adherence that puts a patient at high risk of the development of resistant virus and virological failure. They will also be required to resolve the controversy of whether reported examples of early virological failure with wild-type virus arises from non-adherence or other mechanisms (such as increased target cell availability) [32]. They will be important clinically to identify patients in need of more intensive assistance. And finally, accurate methods to measure adherence will be essential to evaluate interventions designed to improve adherence, which may be crucial to the long-term success of antiretroviral therapies.


The authors would like to thank Drs Marjorie Robertson, Jacqueline Tulsky, and Linda Wolfenden for designing the sampling methodology and assisting with subject recruitment; Jay Jankowski, Nelia Dela Cruz, Paula Zenti, John Day, and Richard Clark for tracking the cohort and performing data collection; Marta Schulte for performing the genotyping; Joanna Morales for assistance in preparing the interview material; and Dr Steven Deeks for comments on the manuscript.


1. Palella FJ Jr, Delaney KM, Moorman AC, et al. Declining morbidity and mortality among patients with advanced human immunodeficiency virus infection. HIV Outpatient Study Investigators.N Engl J Med 1998, 338 :853–860.
2. Ventura SJ, Anderson RN, Martin JA, Smith BL. Births and deaths: preliminary data for 1997. Natl Vital Stat Rep 1998, 47: 1 –41.
3. Centers for Disease Control and Prevention. Update: trends in AIDS incidence – United States, 1996.MMWR 1997, 46 : 861–867.
4. Hecht FM, Colfax G, Swanson M, Chesney MA. Adherence and effectiveness of protease inhibitors in clinical practice.Fifth Conference on Retroviruses and Opportunistic Infections, Chicago, Illinois, 1998, [abstract 151].
5. Vanhove GF, Schapiro JM, Winters MA, Merigan TC, Blaschke TF. Patient compliance and drug failure in protease inhibitor monotherapy. JAMA 1996, 276: 1955 –1956.
6. Montaner JS, Reiss P, Cooper D. et al. A randomized, double-blind trial comparing combinations of nevirapine, didanosine, and zidovudine for HIV-infected patients: the INCAS Trial. :Italy, The Netherlands, Canada and Australia Study. JAMA 1998, 279: 930 –937.
7. Wainberg MA, Friedland G. Public health implications of antiretroviral therapy and HIV drug resistance. JAMA 1998, 279: 1977 –1983.
8. Altice FL, Friedland GH. The era of adherence to HIV therapy [Editorial]. Ann Intern Med 1998, 129: 503 –505.
9. Cohen OJ, Fauci AS. Transmission of multidrug-resistant human immunodeficiency virus – the wake-up call [Editorial; Comment]. N Engl J Med 1998, 339: 341 –343.
10. Mayers DL. Drug-resistant HIV-1: the virus strikes back [Editorial; Comment]. JAMA 1998, 279: 2000 –2002.
11. Paes AH, Bakker A, Soe-Agnie CJ. Measurement of patient compliance. Pharm World Sci 1998, 20: 73 –77.
12. Straka RJ, Fish JT, Benson SR, Suh JT. Patient self-reporting of compliance does not correspond with electronic monitoring: an evaluation using isosorbide dinitrate as a model drug. Pharmacotherapy 1997, 17: 126 –132.
13. Hales G, Mitchel J, Smith DE, Kippex S. Validity of patient questioning versus pill count as an assessment of compliance.12th World AIDS Conference, Geneva, 1998, [abstract, 32373].
14. Melbournem K, Geketko S, Brown S, Wiley C, Chase S, Fisher A. Electronic adherence assessment versus self-report in HIV-infected individuals.38th Interscience Conference on Antimicrobial Agents and Chemotherapy, San Diego, 1998, [abstract 175].
15. Matsuyama JR, Mason BJ, Jue SG. Pharmacists, interventions using an electronic medication-event monitoring device's adherence data versus pill counts. Ann Pharmacother 1993, 27: 851 –855.
16. Olivieri NF, Matsui D, Hermann C, Koren G. Compliance assessed by the Medication Event Monitoring System. Arch Dis Child 1991, 66: 1399 –1402.
17. Zolopa AR, Hahn JA, Gorter R. et al. HIV and tuberculosis infection in San Francisco's homeless adults. :Prevalence and risk factors in a representative sample. JAMA 1994, 272: 455 –461.
18. Gilbert JR, Evans CE, Haynes RB, Tugwell P. Predicting compliance with a regimen of digoxin therapy in family practice. Can Med Assoc J 1980, 123: 119 –122.
19. Paterson D, Swindells S, Mohr J, et al. How much adherence is enough? A prospective study of adherence to protease inhibitor therapy using MEMSCaps.Sixth Conference on Retroviruses and Opportunistic Infections, Chicago, 1999.
20. Ledergerber B, Egger M, Opravil M. et al. Clinical progression and virological failure on highly active antiretroviral therapy in HIV-1 patients: a prospective cohort study. :Swiss HIV Cohort Study. Lancet 1999, 353: 863 –868.
21. Zolopa AR, Warford A, Montoya J, Hsu P, Katzenstein D, Merigan T, Efron B. HIV-1 genotypic resistance patterns predict response to saquinavir/ritonavir in patients who have failed prior protease inhibitors. Ann Intern Med 1999, 131: 813 –821.
22. Cunningham P, Kaufman GR, Sayer D, Smith D, French M, Cooper D. A. Pre-existent mutations of the HIV protease gene were associated with therapeutic failures in patients on saquinavire/ritonavir therapy.12th World AIDS Conference, Geneva, 1998, [abstract 32294].
23. Craig C, Race E, Sheldon J. et al. HIV protease genotype and viral sensitivity to HIV protease inhibitors following saquinavir therapy. AIDS 1998, 12: 1611 –1618.
24. Haubrich RH, Little SJ, Currier JS. et al. The value of patient-reported adherence to antiretroviral therapy in predicting virologic and immunologic response. :California Collaborative Treatment Group. AIDS 1999, 13: 1099 –1107.
25. Lucas GM, Chaisson RE, Moore RD. Highly active antiretroviral therapy in a large urban clinic: risk factors for virologic failure and adverse drug reactions. Ann Intern Med 1999, 131: 81 –87.
26. Gallant J, Block D. Adherence to antiretroviral regimens in HIV-infected patients: results of a survey among patients and physicians. J Int Assoc Physicians AIDS Care 1998, 4: 32 –35.
27. Celentano DD, Vlahov D, Cohn S, Shadle VM, Obasanjo O, Moore RD. Self-reported antiretroviral therapy in injection drug users. JAMA 1998, 280: 544 –546.
28. Strathdee SA, Palepu A, Cornelisse PG. et al. Barriers to use of free antiretroviral therapy in injection drug users. JAMA 1998, 280: 547 –549.
29. Bangsberg D, Moss A. When should we delay highly active antiretroviral therapy? J Gen Intern Med 1999, 14: 446 –448.
30. Bangsberg D, Tulsky JP, Hecht FM, Moss AR. Protease inhibitors in the homeless. JAMA 1997, 278: 63 –65.
31. Lerner BH, Gulick RM, Dubler NN. Rethinking nonadherence: historical perspectives on triple-drug therapy for HIV disease. Ann Intern Med 1998, 129: 573 –578.
32. Havlir DV, Marschner IC, Hirsch MS. et al. Maintenance antiretroviral therapies in HIV infected patients with undetectable plasma HIV RNA after triple-drug therapy. :AIDS Clinical Trials Group Study 343 Team [see Comments]. N Engl J Med 1998, 339: 1261 –1268.

access to therapy; adherence; HIV; highly active antiretroviral therapy; homeless; injection drug use; protease inhibitor; resistance; viral load

© 2000 Lippincott Williams & Wilkins, Inc.