Current treatment strategies for patients infected with HIV-1 involve the use of multiple drugs due to the rapid emergence of resistance to antiretroviral monotherapy and development of opportunistic diseases . Consequently, individuals participating in AIDS clinical trials may be required to adhere to complex dosing schedules involving a combination of active drugs/placebos, together with other medications, including prophylaxis for opportunistic infections and other routinely prescribed medications. The complexity of the dosing regimen may be further increased if the patient is enrolled in more than one protocol.
Several studies have demonstrated that adherence to prescribed medications decreases with increasing dosing frequency [2–4]. The complexity of the regimen with which study participants must comply is therefore an important issue in the design of AIDS clinical trials since substantial non-adherence, if undetected, may compromise interpretation of study results especially when intent-to-treat analysis is to be performed .
Using plasma drug concentrations of prescribed study medications as a marker for adherence in AIDS Clinical Trial Group (ACTG) protocol 175, a comparative safety and efficacy study of monotherapy versus combination dideoxynucleoside therapy, we found that a substantial fraction of study participants (approximately 25%) deviated appreciably from the assigned treatment regimen . Since drug assays provide only a point estimate of adherence, we conducted a second substudy of ACTG 175 in which electronic monitoring devices were used to examine drug-taking behavior in a subset of patients.
The primary aim of this second study was to characterize drug-taking behavior in order to determine the differences in adherence to study medications amongst the treatment arms. To this end, we calculated four indices of drug-taking behavior and examined their distributions and relationship to the prescribed regimen. In addition, we compared adherence indices with other routinely-used measures of adherence, including pill counts of the unused portion of returned medications, and nurses' assessments, in order to assess the value of the latter as markers of adherence in AIDS clinical trials. A future study will relate quantitative adherence to patient covariates using a model-based analysis.
ACTG 175 was a large, multicenter, double-blind, randomized, comparative safety and efficacy study of dideoxynucleoside monotherapy versus combination therapy to prevent disease progression in asymptomatic HIV-infected individuals (details of the study have been published elsewhere ). Accrual commenced in December 1991 and was completed in October 1992. Clinic visits were scheduled at 12-week intervals and the study closed in March 1995, a period of 3 years after it had opened. A total of 2467 individuals with CD4 cells 200–500 × 106/l were accrued and analyzed in one of four treatment arms: arm I, zidovudine (ZDV) 600 mg daily plus zalcitabine (ddC) placebo plus didanosine (ddI) placebo; arm II, ZDV 600 mg daily plus ddC 2.25 mg daily plus ddI placebo; arm III, ZDV 600 mg daily plus ddI 400 mg daily plus ddC placebo; arm IV, ddI 400 mg daily plus ZDV placebo plus ddC placebo.
Dosing instructions and advice on scheduling of doses were provided by the study nurse during routine assessments at scheduled clinic visit. ZDV (two 100 mg capsules per dose) and ddC (two 0.375 mg tablets), or their matching placebos, were prescribed to be taken three times daily, and patients were instructed to take ddI (two 100 mg tablets) or matching placebo on an empty stomach (30 min before or 2 h after eating) twice daily. The total daily pill count was 16, with placebos accounting for from one- to two-thirds of daily doses.
At routine clinic visits during the period October 1992 to November 1993, study participants enrolled at Stanford and San Francisco Kaiser AIDS Clinical Trials Units were approached to co-enroll in a substudy to examine drug-taking behavior. Participants were ineligible if they had discontinued study medications. Informed consent was obtained from patients at the time of accrual into the substudy, and human experimentation guidelines of the US Department of Health and Human Services were followed in the conduct of this research.
ZDV, ddC, ddI, and their matching placebos, were dispensed in vials fitted with electronic monitoring devices [Medication Event Monitoring Systems (MEMS), Aprex Corporation, Fremont, California, USA]. The device consists of a microchip housed in a plastic cap that fits standard-sized medication vials. It records the date and time of each opening and closure of the medication vial, coincident with the removal of a dose for ingestion. In addition to the medication containers, each participant was offered a ‘portable’ container, similarly fitted with a MEMS device. In a pilot study, participants noted that the large size of medication vials limited their portability. The ‘portable’ container (60 ml vial) was supplied to circumvent this problem.
The purpose of the MEMS device was explained and demonstrated to study participants. Participants were instructed to open medication containers only to withdraw a dose at the time of ingestion or to transfer doses to the ‘portable’ container. Participants were asked not to transfer medications to any other container for storage. The pattern of usage of the portable container was recorded in a calendar, in which study participants were also asked to note any events relevant to their drug-taking behavior, for example, adverse effects, or deviations from their usual pattern of behavior. Because we were interested in examining drug-taking behavior, participants were encouraged not to change their regular pattern of dosing.
At the next scheduled clinic visit after accrual into the substudy, MEMS devices and calendars were collected from study participants and returned to the Stanford AIDS Clinical Trials Clinical Pharmacology Core laboratory for data retrieval and analysis. MEMS devices were downloaded to a desktop computer via a communicator unit (Aprex Corporation). Individual dosing histories for all study drugs were compiled using both the electronically recorded data obtained from the MEMS devices (including the portable container, if it was used) and patient-reported information in the personal calendars provided. If an entry in the calendar indicated the date and time of a dose that was not recorded electronically, then it was included in the individual's dosing history as a dosing event. Similarly, if the patient reported that more than one dose was commonly removed at one time to be taken later, and a consistent pattern of behavior was established, then the doses were included in the dosing history.
Four indices of drug-taking behavior were computed. The fraction of doses taken (FR) was the frequency of dosing during the monitoring period, analogous to a pill count. FR was calculated from MEMS dosing records (FRM) as follows: FRM = (no. doses taken)/ (no. doses prescribed in the analysis period).
Daily count adherence (DC) was the fraction of days on which the patient complied with the prescribed dosing frequency . In this analysis, a day was defined as the 24 h period between 0300 h and 0259 h. Adherence was calculated as follows: DC = (no. days on which prescribed no. doses were taken)/(no. days in the analysis period).
The third index, therapeutic coverage (TC) was defined as that period of time during which it may be presumed the drug was exerting a therapeutic effect . The dideoxynucleoside drugs are prodrugs, which are anabolized in cells to the therapeutically active triphosphorylated moieties . Although the relationship between intracellular dideoxynucleotide concentrations and pharmacological response has not been fully elucidated, it is reasonable to assume that the time-course of therapeutic effect is more closely related to the concentration of dideoxynucleotide at the site of action, than the plasma dideoxynucleoside concentrations. We therefore based our calculations of TC on the assumption that the duration of therapeutic effect is equal to the approximate intracellular dideoxynucleotide half-life after a dose, which roughly corresponds to the prescribed dosing intervals employed in ACTG 175 (8 h for ZDV and ddC, and 12 h for ddI). The fraction of the total analysis period during which the patient was assumed to have therapeutic coverage was therefore calculated as follows: TC = (no. hours with therapeutic coverage)/(total no. hours in analysis period).
We calculated the number of days on which no doses (NDD) were taken as a fraction of the total analysis period as an index of periods of missed doses, or ‘drug holidays’ . This fourth index was calculated as follows: NDD = (no. days on which no doses were taken)/(no. days in the analysis period).
According to the definitions of FRM, DC and TC, a perfect complier should have a value of 1 for all of these indices, whereas the NDD index should have a value of 0.
Patterns of drug-taking behavior over time were examined by comparing average daily dosing frequencies for the periods comprising the entire monitoring period, the first 28 days of monitoring, all remaining days of the monitoring period (after the first 28 days), and the day before study participants returned to the clinic.
During the clinic visit, study participants returned the unused portions of all medications to the dispensing pharmacy where pill counts were performed. In addition, the study nurse at each study site assessed adherence to study medications based on patients' reports. The total fraction of doses taken during the analysis period, as determined from pill counts (FRPC) and nurses' assessments (FRN), were determined. The fraction of doses taken using each of these methods was compared with FRM.
Adherence indices obtained from analyses of MEMS data were summarized using descriptive statistics for each of the active study medications and their matching placebos. Comparisons of adherence indices amongst the three active drugs (ZDV, ddC and ddI) and the three placebos, and between active drugs and their matching placebos were performed. Data were analyzed overall and within and between strata when stratified by treatment arm using Kruskal–Wallis rank-sum test or Wilcoxon rank-sum test for two sample data. Adherence indices amongst the three active drugs were also analyzed accounting for repeated measurements from each subject, and average daily dosing frequencies during various segments of the analysis period were compared using Friedman rank-sum test. P < 0.05 was considered significant.
Stepwise multiple regression analyses were performed to identify demographic correlates of adherence for each of the three study drugs. Adherence indices were log-transformed so that they were approximately normally distributed. All patient characteristics were included as potential covariates in a generalized additive model (GAM) in S-Plus (Mathsoft Inc, Seattle, Washington, USA). Patient characteristics were selected based on the Akaike Information Criterion (reflecting improvement in goodness of fit corrected for model complexity) and the significance of their contribution to the fit.
Of 54 eligible ACTG 175 participants at the Stanford and San Francisco Kaiser sites who were approached, 49 (91%) consented to participate in the adherence substudy. At the completion of the monitoring period, one person was lost to follow-up, and data from seven subjects could not be analyzed due to incorrect use or failure of MEMS devices or non-adherence to the instruction to remove doses directly from the issued containers.
The remaining 41 datasets (85% of the original sets) were analyzed to evaluate adherence to study medications. Patient-reported dosing events were included in 24, 22 and 17 of the individual dosing histories for ZDV, ddC and ddI, respectively. Up to 10% of records were included in 10 individuals' dosing histories for each drug, and greater than 50% of dosing records were generated based on established patterns of dosing and patient reports for each of four, three, and one individuals for ZDV, ddC and ddI, respectively. The median duration of monitoring was 84 days (range, 47–106 days). Patient demographics are shown in Table 1.
Active study drugs
Adherence index statistics for the three study medications are presented in Table 2. Although FRM was greater than 80% for all drugs, this afforded median durations of TC of only 67–75% of the monitoring period. Whereas all individuals taking active ddC achieved greater than 50% therapeutic coverage, therapeutic coverage attained with ZDV was more variable, with 26% of this sample of patients demonstrating less than 50% therapeutic coverage. Therapeutic coverage for ddI was substantially more variable, varying eightfold, although a larger proportion of patients demonstrated >90% adherence to ddI (17%), compared with 4 and 0% for ZDV and ddC, respectively. Lapses in dosing for a 24 h period (NDD), accounted for a median of 3% (range, 0–46%), 2% (range, 0–28%) and 5% (range, 0–77%) of the monitoring period for ZDV, ddC and ddI, respectively.
Greater interindividual variability was observed in adherence to prescribed daily dosing frequencies (DC). Marginally higher adherence to the prescribed twice daily dosing frequency was observed for ddI (median DC, 76%) than for ZDV doses (55%) and for ddC doses (66%), which were prescribed at a three times daily dosing frequency.
No significant differences in adherence indices with each of the study medications were found overall or when stratified by treatment arm. However, when the analysis was performed accounting for repeated measures, the differences in NDD and DC adherence amongst the three drugs achieved marginal significance (P = 0.054 and 0.076, respectively).
Active drug versus placebo
There were no significant differences in indices of adherence between the active study medications and their matching placebos both overall and when stratified by treatment arm (Table 2). However, although not statistically significant because the numbers of participants are small, there appeared to be a trend towards lower adherence to placebo compared with active drug. This was most evident for ddC, for which placebo adherence indices were spread over a wider range of values; for example, all subjects receiving active ddC demonstrated greater than 60% FR adherence (range, 0.64–0.97), whereas for placebo FR, scores ranged from 0.16 to 1.0. Adherence indices were not significantly different amongst the matching placebos for ZDV, ddC and ddI.
Despite a trend towards higher adherence to active study drugs in the monotherapy treatment arms, median adherence scores were not significantly different from those observed in combination therapy arms. Median ZDV TC was 76% in the ZDV alone arm, compared with 66 and 69% in the ZDV–ddI and ZDV–ddC arms, respectively. Similarly, median TC for ddI was slightly higher in the monotherapy arm compared with the combination (ZDV–ddI) treatment arm (80 versus 64%), although this difference was not statistically significant.
In addition to substantial inter-individual variability in adherence to prescribed drug therapy, marked variability was observed in individual patterns of drug-taking behavior over the monitoring period. Dosing patterns of ddI for four individuals in the ddI monotherapy treatment arm are illustrated in Fig. 1. In contrast to near-perfect adherence shown in Fig. 1a, the other patterns illustrate varying patterns of partial adherence, with sporadic dosing illustrated in Fig. 1d. In Fig. 1b and c, there is a consistent pattern of dosing at the prescribed dosing frequency, with a notable drug holiday during which no doses were taken for a period of approximately 5 days (Fig. 1b) and occasional extra doses following missed doses (Fig. 1c).
Pill counts and nurse assessments
For individuals with less than perfect adherence (<80%), as determined by FRM for all three drugs, FRPC and FRN both yielded overestimates of adherence compared with FRM (Fig. 2). Interestingly, three individuals who demonstrated poor but measurable adherence by MEMS monitoring had not removed doses from the pill containers (FRPC = 0). This is a known limitation of the use of electronic monitoring devices since patients may occasionally (perhaps due to curiosity) open the container and not take the drug, upwardly biasing the MEMS estimate of adherence. Without these three datapoints, a much greater overestimation of FR by FRPC would have been seen.
Trends over time
There was a small but statistically significantly higher average daily dosing frequency during the first 28 days of the monitoring period compared with the average over the entire monitoring period (P < 0.002 for all drugs), with significantly less than average adherence observed during the remainder of the monitoring period (P < 0.001 for ZDV and ddC, P = 0.004 for ddI). On both the day immediately prior to the return clinic visit and the morning of the clinic visit day, the average number of doses taken was less than that averaged over the entire monitoring period, although the difference was not significant (P = 0.1 for ZDV, P > 0.2 for ddC and ddI).
Demographic correlates of adherence
For each of ZDV, ddC and ddI, there was no evidence of a non-linear relationship between covariates and adherence indices in the GAM analysis (Table 3). Consequently, the model used was a simple multiple linear one. In this analysis, older patients were more compliant with the prescribed dosing frequency and prescribed dosing intervals and had higher therapeutic coverages for ZDV and ddC. For ddI, increasing age also correlated with higher adherence to the prescribed dosing frequency. In addition, study participants with low quality of life assessments were poorer compliers, as indicated by the higher NDD adherence index. However, in general, patient characteristics were poorly predictive of adherence to study medications.
In addition to interindividual variability in pharmacokinetics and pharmacodynamics, variability in medication adherence is believed to cause considerable variability in response to drug treatment amongst individuals in controlled clinical trials [11,12]. Electronic monitoring devices have provided insight into the multidimensionality of adherence, including aspects of the quantity, frequency and duration of dose-taking [13,14].
Electronic adherence monitoring, as for all indirect methods of adherence monitoring (with the exception of directly observed therapy), requires the assumption that a dose is ingested immediately after the study participant opens the container and the MEMS device records the event. Although some individuals may attempt to deceive the device, this behavior would be difficult to maintain over long periods of time, as we observed for three individuals in this study. Alternatives to continuous adherence monitoring in clinical trials, such as pill counts and nurses' assessments, provide single-point estimates of adherence. Although all these methods of adherence assessment gave similar estimates of FR adherence when the fraction of doses taken was high (>80%), pill counts and nurses' assessments tended to overestimate adherence overall, in agreement with the findings of other investigators [15–17].
Among the four treatment arms, adherence to all three active study medications deviated substantially from ideal, with a trend towards lower adherence in the combination therapy arms compared with those assigned to receive monotherapy. In a separate substudy of ACTG 175 in which we used plasma drug concentration measurements as a marker of adherence to study medications, adherence was estimated to be approximately 75% , in close agreement with the extent of adherence determined using MEMS. However, although there was some suggestion of differential adherence to ddI observed using plasma drug concentrations, we did not observe evidence of differential adherence (or non-uniform adherence between the monotherapy versus combination therapy arms) with electronic monitoring.
Patterns of adherence similar to those observed here have been observed in a variety of other studies of ambulatory patients , perhaps suggesting that the dynamics of medication adherence are related to behavioral phenomena, independent of drug indication. For example, remarkably similar adherence index scores to those seen here were reported in a study of 21 hypertensive patients treated with a twice daily regimen of a single antihypertensive drug . The lower FRM in our study may reflect the greater complexity of the regimen in our group of patients enrolled in the AIDS clinical trial. Indeed, the results of previous studies indicate that dosing regimen complexity is inversely related to adherence [2–4]. Since overall adherence scores in our study were not substantially less than those observed in other patient populations who were required to comply with simpler dosing regimens, other factors unique to the HIV-infected study participants may have motivated drug-taking behavior, countering the effect of regimen complexity on medication adherence .
Furthermore, in contrast to common beliefs regarding drug-taking behavior, there was no evidence of a ‘toothbrush’ effect, or a period of increased adherence preceding a clinic visit, in this group of HIV-infected patients . It was not clear whether the overall trend towards reduced adherence over time that we observed was a function of the drug therapy or an artifact due to the presence of the monitoring devices.
Participants in this substudy of ACTG 175 demonstrated imperfect adherence to prescribed dosing schedules of active study medications. Although more than 80% of prescribed doses of active medications were taken by study participants, less than three-quarters were taken at the prescribed daily dosing frequency and fewer than one-third were taken at prescribed dosing intervals (data not shown). Study participants took no doses on a median of 2–5% of days in the analysis period. As a result of this deviation from the prescribed drug regimens, we estimate that study participants may have experienced suboptimal therapeutic effects for more than 25% of their time on therapy. However, the consequences of drug holidays may be greater than a temporary loss in antiretroviral activity; the results of a recent study provide evidence that these individuals may be at risk of treatment failure. In a study in which adherence to the protease inhibitor, saquinavir, was monitored simultaneously with virologic markers of response in HIV-positive patients, it was observed that the emergence of viral resistance to the drug was frequently preceded by drug holidays . This observation has broad implications in terms of the economic and health costs of drug holidays with antiretroviral drug regimens. Consequently, it has been recommended that only those patients who are ready and willing to commit to adhering to combination antiretroviral drug regimens involving protease inhibitors should be treated with them . The role of the prescribing physician in attempting to minimize the difficulties of complying with complex regimens for HIV-infected individuals must also be considered .
Understanding the nature and extent of the inter- and intraindividual variability in patterns of drug-taking behavior should allow us to better understand how patterns of partial adherence influence response to drug therapy. Recent advances in data analysis techniques have enabled a model-based analysis to be developed and validated that successfully captures the inter- and intraindividual variability in adherence patterns over time and provides quantitative estimates of various aspects of drug-taking behaviors . The influence of variability in adherence on drug exposure–response relationships in ACTG 175 is currently being investigated.
The authors appreciate the assistance and cooperation of the Stanford and San Francisco Kaiser ACTG 175 study sites and personnel, and the study participants themselves.
1. NIH Panel: Report of the NIH Panel to Define Principles of Therapy of HIV Infection. Ann Intern Med 1998, 128:1057–1078.
2. Cramer JA, Mattson RH, Prevey ML, Scheyer RD, Ouellette VL: How often is medication taken as prescribed? A novel assessment technique. JAMA 1989, 261:3273–3277.
3. Eldred LJ, Wu AW, Chaisson RE, Moore RD: Adherence to antiretroviral and Pneumocystis prophylaxis in HIV disease. J Acquir Immune Defic Syndr Hum Retrovirol 1998, 18:117–125.
4. Kruse W, Eggert-Kruse W, Rampmaier J, Runnebaum B, Weber E: Dosage frequency and drug-compliance behaviour: a comparative study on compliance with a medication to be taken twice or four times daily. Eur J Clin Pharmacol 1991, 41:589–592.
5. Vander Stichele R: Measurement of patient compliance and the interpretation of randomized clinical trials. Eur J Clin Pharmacol 1991, 41:27–35.
6. Kastrissios H, Suarez JR, Hammer S, Katzenstein D, Blaschke TF: The extent of non-adherence in a large AIDS clinical trial using plasma dideoxynucleoside concentrations as a marker. AIDS 1998, 12:2305–2311.
7. Hammer SM, Katzenstein DA, Hughes MD, et al.: A trial comparing nucleoside monotherapy with combination therapy in HIV-infected adults with CD4 cell counts from 200 to 500 per cubic millimeter. N Engl J Med 1996, 335:1081–1090.
8. Urquhart J: Ascertaining how much compliance is enough with outpatient antibiotic regimens. Postgrad Med J 1992, 68(suppl 3):S49–S59.
9. Sommadossi JP: Nucleoside analogs: similarities and differences. Clin Infect Dis 1993, 16(suppl 1):S7–S15.
10. Urquhart J: Role of patient compliance in clinical pharmacokinetics. A review of recent research. Clin Pharmacokinet 1994, 27:202–215.
11. Peck CC, Barr WH, Benet LZ, et al.: Opportunities for integration of pharmacokinetics, pharmacodynamics, and toxicokinetics in rational drug development. Clin Pharmacol Ther 1992, 51:465–473.
12. Urquhart J, de Klerk E: Contending paradigms for the interpretation of data on patient compliance with therapeutic drug regimens. Stat Med 1998, 17:251–267.
13. Eisen SA, Miller DK, Woodward RS, Spitznagel E, Przybeck TR: The effect of prescribed daily dose frequency on patient medication compliance. Arch Intern Med 1990, 150:1881–1884.
14. Urquhart J: Real-time compliance data to help define optimal drug regimens. Ann NY Acad Sci 1991, 618:522–532.
15. Rudd P, Ahmed S, Zachary V, Barton C, Bonduelle D: Improved compliance measures: applications in an ambulatory hypertensive drug trial. Clin Pharmacol Ther 1990, 48:676–685.
16. Waterhouse DM, Calzone KA, Mele C, Brenner DE: Adherence to oral tamoxifen: a comparison of patient self-report, pill counts, and microelectronic monitoring. J Clin Oncol 1993, 11:1189–1197.
17. Guerrero D, Rudd P, Bryant-Kosling C, Middleton BF: Antihypertensive medication-taking. Investigation of a simple regimen. Am J Hypertens 1993, 6:586–592.
18. Kastrissios H, Blaschke TF: Medication compliance as a feature in drug development. Annu Rev Pharmacol Toxicol 1997, 37:451–475.
19. Bond WS, Hussar DA: Detection methods and strategies for improving medication compliance. Am J Hosp Pharm 1991, 48:1978–1988.
20. Vanhove GF, Schapiro JM, Winters MM, Merigan TC, Blaschke TF: Patient compliance and drug failure in protease inhibitor monotherapy. JAMA 1996, 276:1955–1956.
21. Katzenstein DA, Lyons C, Molaghan JP, Ungvarski P, Wolfe GS, Williams A: HIV therapeutics: confronting adherence. J Assoc Nurses AIDS Care 1997, 8(suppl):46–58.
22. Ickovics JR, Meisler A: Adherence in AIDS clinical trials: a framework for clinical research and clinical care. J Clin Epidemiol 1997, 50:385–391.
23. Girard P, Blaschke TF, Kastrissios H, Sheiner LB: A Markov mixed effect regression model for drug compliance. Stat Med 1998, 17:2313–2333.
© 1998 Lippincott Williams & Wilkins, Inc.