Data were collected from 3/30/2003 – 5/30/2007 at the adult HIV primary care outpatient clinic at Harborview Medical Center, a public institution affiliated with University of Washington in Seattle, Washington, USA. Advertisements for the study were posted in the clinic waiting room, and all providers were informed of the study. Referrals came from clinic providers, the patients themselves, and a nurse at the clinic dedicated to soliciting research participants. The nurse approached all potentially eligible patients at each clinic session.
Eligible participants were (a) at least 18 years of age, (b) proficient in English, (c) living within the service area of the pager, and (e) initiating or changing at least two medications of a HAART regimen. Some of the changes were prescribed because of poor response, but in other cases new formulations were prescribed in order to reduce pill burden and/or side effects. Individuals who were cognitively impaired, actively psychotic, or had a known history of harming others were excluded. Study staff provided detailed information about the scope of the trial to all eligible patients referred to them and scheduled a baseline appointment to obtain written informed consent and enroll participants. The baseline appointment was scheduled to coincide as closely as possible with the day patients were to initiate their HAART regimen.
At the baseline appointment, participants first completed an approximately 1-hour computer-assisted self-interview assessing adherence as well as other psychosocial and attitudinal variables (i.e., knowledge about treatment, adherence self-efficacy, side effects, substance use, sexual risk behaviors, social support, coping, negative affective states, and quality of life). Next, they were randomly assigned to one of the four study arms: peer support, pager, both peer and pager, or usual care. Allocation concealment involved the use of sequentially numbered, opaque, sealed envelopes containing the study arm assignment, which the study staff opened at the moment of randomization. An external statistician had used a computerized random number generator to select random permuted blocks of four. Due to the nature of the intervention, participants, study staff, and data analysts could not be completely blinded to study arm assignment.
Electronic drug monitors (EDM; i.e., the Medication Event Monitoring System or MEMS®; http://www.aardex.ch
) were provided to all participants for the duration of the trial. EDM technology consists of a plastic pill vial and modified cap containing a microprocessor capable of recording the precise date and time of each vial opening as a presumptive dose. Each participant was given one EDM to use with the most frequently dosed antiretroviral.
All participants continued to receive medical care at the clinic and were asked to return for assessments at 2 weeks (for a brief interview) and 3, 6, and 9 months (for full follow-up assessments). At each full assessment, EDM data were uploaded and blood was drawn to determine HIV-1 RNA viral load (VL) and CD4 lymphocyte count. Participants were reimbursed $60 for the baseline interview, $20 for the 2-week interview, and $35 for the 3-, 6-, and 9-month interviews, with a $25 bonus if they completed all five assessments and returned the EDM. Tracking efforts included unscripted reminder phone calls 1 week and 1 day before each scheduled interview appointment and mailed correspondence in the face of non-response to phone calls.
Peer Support Arm
Clinic patients who were HIV-positive and currently on HAART served as “peers” who provided medication-related social support in the peer support arm. Medical providers in the clinic assisted study staff in identifying appropriate peers from among current patients who reported consistently high levels of adherence, attended clinic appointments regularly, were socially skilled, and were able to participate in an initial training and commit to ongoing supervision. Research staff (i.e., the principal investigator, psychology doctoral candidates, and clinical staff including a pharmacist, medical provider, nutritionist, and case manager) provided approximately 15 hours of training to a total of 17 peers during the course of the study. During the trainings, peers learned how to assess for negative affective states and other barriers to adherence and to sensitively provide emotional, informational, and affirmational social support. Other topics covered in the training included an overview of HIV and HAART, how to maintain appropriate limits on the peer relationship, overcoming potential barriers to the acceptance of support, harm reduction approaches to substance use, appropriate referrals for medical inquiries, and strategies for working with diverse participants. Peers received $25 – $40 twice monthly as an incentive for their involvement based on the number of participants (1 – 3) to whom they were assigned as peers. On average, peers remained actively involved for approximately 1 year.
The 3-month peer support intervention consisted of two parts: (a) six twice-monthly one-hour gatherings (i.e., “peer meetings”) held at the clinic, consisting of all peers and actively enrolled participants in the peer support arm and (b) weekly phone calls from peers to participants. At the time of enrollment, participants were assigned to individual peers by study staff based on peer availability and presumed compatibility (i.e., whenever possible, an effort was made to match peers to participants based on race/ethnicity, sex, and sexual orientation). In the group setting, participants had the opportunity to interact face-to-face with their assigned peer as well as meet the other peers and participants, with the goal of benefiting from the discussion of the shared experiences of the group. The meetings were designed to identify barriers to HAART adherence and generate problem-solving strategies to overcome them. Other emerging themes included life issues that impact adherence, including HIV status disclosure, dating, substance use, and struggles with mental health issues. One of several research staff members (each with graduate training in psychology) coordinated the groups by making reminder phone calls, preparing the meeting room, and providing refreshments. With assistance from the peers, they facilitated the meetings by refocusing the discussion on adherence-related topics when appropriate. Otherwise, they refrained from interfering with the group process and the exchange of support among peers and participants, resulting in predominantly peer-led groups. Participants received a $15 incentive for attending each of the six sessions. All were welcomed to continue attending thereafter but were provided neither reminder phone calls nor monetary incentives for additional meetings.
Between group meetings, peers were instructed to call each of their study participants weekly to provide more in-depth one-on-one attention and feedback. Phone calls also were better suited for participants with confidentiality concerns and those who had difficulty traveling to the clinic or had scheduling conflicts with the set meeting times. Peers were instructed not to initiate contact with participants following the 3-month intervention period, but they were allowed to respond to requests for contact from the participants at their own discretion.
In order to preserve and monitor intervention fidelity, peers were evaluated at the end of the training to ensure they acquired competency in their respective tasks. Peers also received ongoing supervision from the group facilitator after each group meeting and at twice-monthly telephone check-ins. Additionally, they completed a one-page log for each telephone and face-to-face contact with a participant. These logs were reviewed with project staff during supervision.
Pager Messaging Arm
The 3-month pager intervention consisted of a customized pager system that combined the communicating abilities of the World Wide Web and two-way pagers. A computer program allowed for the timing of all text messages to be specified up front and then sent automatically without any further intervention from study staff. At randomization, the study coordinator distributed a two-way pager and customized a message schedule to the participant’s daily medication regimen, which was confirmed by the clinic pharmacist. In addition to dose reminders (which included medication names familiar to the patient and number of pills to be taken), three other types of text messages were sent: (1) educational (referring to side effects and their management, medication benefits, understanding laboratory values, the importance of adherence, drug interactions, proper medication storage, resistance, and self-advocacy); (2) entertainment (jokes or thoughts for the day), and (3) adherence assessments. There was some flexibility in the frequency and content of messages to accommodate participants’ differing needs and schedules, but there was a minimum of three pager messages sent daily to participants for the first two months. Pages gradually tapered in the last month of the 3-month intervention to avoid a rebound in nonadherence. A confirmation return page was requested for every message sent. Project staff periodically reviewed the return page log and if no responses were received, the participant could be called to determine if they were having any problems with the pager.
Between the first contact and the end of the 3-month intervention, the participants were asked to wear the pager at all waking moments. Participants had to use the study pagers and could not substitute with their own pager or cell phone. They were instructed not to use the devices for personal use. Participants could select from a variety of auditory alarms or a vibrate-only option if desired. If the pager was lost or failed it was promptly replaced at no cost to the subject.
Before initiation of HAART, all clinic patients who are naïve patients or off HAART for more than six months are required to complete the HAART Protocol, a clinic-based program designed to provide education regarding HAART and adherence as well as to identify and correct adherence barriers prior to HAART initiation. The HAART Protocol involves a series of three separate appointments with a pharmacist, nutritionist, and case manager after an initial meeting with a provider and before a final meeting when the provider decides whether to prescribe HAART. During these preliminary interviews, patients are given social and mental health referrals as appropriate, a daily medication schedule, and information on medication side effects and techniques to help manage them. All participants in each of the study’s four arms participated in the HAART Protocol if they had not received it in the last six months. There were no time or attention control activities.
The primary outcome measures were self-reported and EDM-based medication adherence at 3 months (immediate post-intervention) and at 6 and 9 months (post intervention follow-ups). The secondary outcomes were the clinical laboratory biomarkers of CD4 lymphocyte count and VL at 3, 6, and 9 months.
Self-reported medication adherence was assessed with one item from the Simplified Medication Adherence Questionnaire (SMAQ)14
assessing the number of doses missed during the previous week (none of the time
, 1–2 times
, 3–5 times
, or 6–10 times
). A dichotomous variable for past-week 100% adherence was created with participants classified as perfectly adherent if they missed zero doses and non-adherent if they missed 1 or more doses.10, 15
We computed a continuous EDM adherence variable, defined as the number of bottle openings recorded by the MEMS cap during the 7 days before each assessment date divided by the number of prescribed doses. The EDM adherence outcome was analyzed as a continuous variable and as dichotomous variables with cut-offs at 80, 85, 90, 95, and 100% at each assessment point and across all three follow-up assessments.
Study staff abstracted VL and CD4 lymphocyte counts from patient medical records if they were available within 30 days of the baseline, 3, 6, and 9-month interviews; otherwise, blood draws for VL and CD4 counts were conducted the day of the interview. As VL data were not normally distributed, we conducted a log transformation and used the transformed values in all further analyses. Clinical outcomes were analyzed as continuous variables and as dichotomous variables with cut-offs for VL of undetectable, 1,000, and 10,000 and for CD4 of 350 at each assessment point and across all three follow-up assessments.
Initial power analyses were based on a dichotomous measure of adherence as the outcome and .80 power to detect a 15–20% difference in the percentage of 100% adherent patients. An analysis “at the margins” was planned whereby all patients treated with one intervention are compared to all patients not treated with the intervention. For example, in evaluating the peer support intervention, all participants in the peer support arm and those in the peer support and pager arm would be compared with those in the pager only and usual care arms. This method maximizes the statistical power for detecting intervention effects when analyzing a factorial trial.16
These power calculations indicated that 120 patients were required to receive each intervention, resulting in a total sample of 240.
All participants who were randomly assigned were included in the analyses (i.e., an intent-to-treat approach), with adjustments made only in the case of two participants who were dropped from the study when they were determined during the baseline interview to be ineligible. Missing data were set equal to zero for the primary adherence outcomes and the detection limits for the secondary biomarker outcomes.
To evaluate efficacy, repeated-measures analyses were conducted using generalized estimating equations (GEE).17
In these analyses, each outcome was regressed on pager arm, peer support arm, pager × peer support arm, time (assessment point), all arm × time interactions, and previous experience with HAART as a control variable (i.e., HAART naïve or not). Time was partitioned into contrasts of 2-week assessment point (or, in the case of the biomedical outcomes, baseline) against each follow-up (i.e., 2 weeks vs. 3, 6, and 9 months). The statistical tests of the intervention effects were the peer support × time and pager support × time interactions. A logit link function was utilized for the binary outcomes and Gaussian link function for continuous outcomes. An unstructured working correlation structure was specified to account for correlated outcome data across assessments.
To assess whether an analysis “at the margins” was appropriate, the full factorial models were compared to models that excluded any interaction effects between the pager and peer support interventions. Wald tests were performed to evaluate whether the pager × peer support and pager × peer support × time terms were collectively significant for each primary and secondary outcome. Analyses conducted at the margins were comparable to the full factorial models for 3 out of the 4 outcomes, so results from the analyses “at the margins” are reported. All trend-level findings at p < .12 are reported for interpretive purposes.
Additional post hoc analyses evaluated whether dichotomous versions of the primary (EDM adherence) and secondary (VL, CD4 count) outcomes would be associated with the interventions. In the first set of analyses, we substituted dichotomous version of EDM adherence, VL, and CD4 count in the GEE analyses. In the second set of analyses, we used a stricter dichotomous outcome that required meeting the dichotomous outcome across all follow-up assessments (e.g., EDM adherence > 80% at 3, 6, and 9 months). In these analyses, the strict dichotomous outcome for (i.e., meeting the cutoff at all follow-up assessments), was regressed on peer support, pager support, and previous experience with HAART and baseline levels as control variables.
In further post hoc analyses, we examined whether the number of mandatory peer support meetings attended or pager response rate was associated with intervention effectiveness among those in the experimental arms. In these quasi dose response analyses, each outcome was regressed on dose, time (assessment point), dose × time, and previous experience with HAART and baseline levels as a control variable. The statistical test of the dose response effect for each outcome was the dose × time interaction.