Sub-optimal adherence to combination anti-retroviral therapy and its associated factors according to self-report, clinician-recorded and pharmacy-refill assessment methods among HIV-infected adults in Addis Ababa

ABSTRACT

Adherence to combination antiretroviral therapy (cART) is generally high in most resource-limited settings. However, sub-optimal adherence occurs in a sizable proportion of patients, and is independently predictive of detectable viremia. We investigated sub-optimal adherence according to self-report, clinician-recorded, and pharmacy-refill assessment methods, and their associated factors among HIV-infected adults receiving cART in Addis Ababa, Ethiopia. Eight-hundred seventy patients who initiated cART between May 2009 and April 2012 were randomly selected, and 664 patients who were alive, had remained in clinical care and were receiving cART for at least six-months were included. Sub-optimal adherence was defined as patients’ response of less than “all-of the time” to the self-report adherence question, or any clinician-recorded poor adherence during the six most recent clinic visits, or a pharmacy-refill of <95% medication possession ratio (MPR). Logistic regression models were fitted to identify factors associated with sub-optimal adherence. The average adherence level to cART, expressed as MPR, was nearly 97%. However, sub-optimal adherence occurred in 12%, 4%, and 27% of patients according to self-report, clinician-recorded, and pharmacy-refill measures, respectively. More satisfaction with social support was significantly associated with less sub-optimal adherence according to self-report and clinician-record. Younger age, lower educational level, and lower CD4 cell count at cART initiation were significantly associated with sub-optimal refill-based adherence. Findings from our large multi-center study suggest that sub-optimal adherence was present in up to a quarter of the patients, despite a high degree of average adherence to cART. Interventions aimed at preventing sub-optimal adherence should focus on improving social support, on younger patients, on patients with lower educational level, and on those who started cART at a lower CD4 cell count.

KEYWORDS:

• Sub-optimal adherence
• self-report
• clinician-recorded
• pharmacy-refill
• HIV/AIDS
• cART
• resource-limited setting

Introduction

Adherence to combination antiretroviral therapy (cART) is essential for HIV treatment success (Friedland & Williams, 1999; Paterson et al., 2000). In resource-limited settings without routine plasma viral load monitoring in the daily clinical care (Federal Democratic Republic of Ethiopia, Ministry of Health [FMoH], 2007; World Health Organization [WHO], 2013), treating clinicians need to rely on the assessment of medication non-adherence to screen patients for treatment failure. Therefore, identifying the predictors of sub-optimal adherence would be the initial step to design interventions aiming at reducing non-adherence and preventing the subsequent development of detectable viral load.

Previously conducted studies in resource-limited settings investigated the predictors of adherence using primarily one method to assess adherence (Amberbir, Woldemichael, Getachew, Girma, & Deribe, 2008; Wakibi, Nganga, & Mbugua, 2011). Given the absence of a single “gold standard” adherence measure (Berg & Arnsten, 2006; Chesney, 2006), and the wide range of variables that might influence adherence (Ammassari et al., 2002; Langebeek et al., 2014), it would be commendable to identify the variables associated with sub-optimal adherence based on different adherence assessment methods. This could help to comprehensively target all relevant factors associated with adherence.

In a previous study, we compared adherence according to self-report, clinician-recorded, and pharmacy-refill assessment methods among a representative sample of HIV-infected adults who are receiving cART in the public ART-program in Addis Ababa (Mekuria, Prins, Yalew, Sprangers, & Nieuwkerk, 2016). The present study was conducted to identify factors associated with sub-optimal adherence as assessed with each of these methods.

Methods

Study setting and participants

This study was conducted between September 2012 and April 2013 in 10 randomly selected health-care facilities located in Addis Ababa, Ethiopia. Selection of the health facilities has been described in detail previously (Mekuria, Nieuwkerk, Yalew, Sprangers, & Prins, 2016). At the time of the study, HIV-infected adults who fulfilled the following WHO criteria were automatically eligible to initiate cART once their readiness for treatment was assured: WHO clinical stage-IV, irrespective of CD4 cell count; WHO clinical stage-III with a CD4 cell count of ≤350/mm³; or all WHO clinical stages with CD4 cell counts ≤200/mm³ (FMoH, 2007). Study participants were selected in a two-step procedure using the ART register as a sampling frame. First, we identified the 8016 treatment naive HIV-infected persons who initiated cART in Addis Ababa between May 2009 and April 2012; 558 patients who had been formally transferred-out after initiating cART or were aged <18 years were excluded. Second, out of these 7458 patients, we selected 870 using a systematic random sampling procedure. Patients who were still alive, were retained in HIV care and using cART for at least six-months were eligible to participate in the present study.

Ethical approval

The Institutional Review Boards of the College of Health Sciences, Addis Ababa University, and Millennium Medical College, St. Paul Hospital, approved this study. Also, patients gave informed consent to participate in a face-to-face interview, to retrieve drug dispensing information from their pharmacy-refill records, and to check their medical files for socio-demographic, clinical, and treatment-related variables. Patient data were abstracted anonymously from medical records using the unique ART number or medical record number.

Dependent variable

The primary outcome was sub-optimal adherence to cART according to self-report, clinician-recorded, and pharmacy-refill measurements. We previously found that each of the adherence measures described below had a likelihood of more than 80% to discriminate correctly between the HIV-infected patients with a plasma viral load below or above 400 RNA copies/mL (Mekuria, Prins, Yalew, Sprangers, & Nieuwkerk, 2016).

Self-reported adherence

A self-report adherence question was adopted from the Adult AIDS Clinical Trial Group (AACTG) questionnaire (Chesney et al., 2000). Patients were asked how often during the past four weeks they took all the antiretroviral (ARV) medications prescribed (from “1 = none-of the time” to “5 = all-of the time”). Any self-report of less than all-of the time was considered as sub-optimal.

Clinician-recorded adherence

Clinician-recorded adherence refers to the routine assessment of patients’ medication taking as Good, Fair, or Poor during each scheduled regular follow-up visit to the ART-clinic, usually every one to three months. It is based on patients’ responses to questions about the number of missed ARV-drug doses during the past month and/or the clinicians’ judgment based on missed clinic follow-up visits. Patients with at least one clinician-recorded poor adherence in the six most recent follow-up clinic visits were considered as having sub-optimal adherence.

Pharmacy-refill adherence

Refill data over the last six most recent drug dispensings were abstracted from the pharmacy-refill card and/or electronic pharmacy-refill record in the ART-pharmacy. The pharmacy-refill adherence in each refill-period was assessed by calculating the medication possession ratio (MPR) as {[(number of pills dispensed at refill/number of pills prescribed per day)/number of days between refills] multiplied by hundred} (de Boer, Prins, Sprangers, & Nieuwkerk, 2010). The date of the interview served as the index date.

For patients in whom the dates of the “pharmacy-refill visit” were missing but the corresponding “clinic visit dates” were available, we looked for the original prescription-refill paper in the ART-pharmacy. When this prescription-refill paper was found, we considered the missing refill data due to non-recording. When the prescription-refill paper was not found, we used the corresponding “clinic visit date” to re-calculate the MPR. When the pharmacy-refill/prescription-refill and the corresponding clinic visit dates were both missing, we assumed the patient had not collected the drug for this specific refill-period.

Because most patients were prescribed a triple combination of antiretroviral drugs, we first calculated the average MPR for each ARV drug over the five most recent refill periods. When the supply of pills was in excess of the period between two refill dates, we carried forward those leftover pills to the next refill-period, except when the patient changed his or her regimen. Next, we truncated the MPR of each refill-period to 100%. Then, we calculated the average MPR for all the drugs in the regimen combined. Because the MPRs for the separate components of the regimen were almost similar to each other (data not shown), we decided to use the average MPR for all the drugs in the regimen combined. Patients with <95% MPR were considered as having sub-optimal adherence.

Independent variables

Socio-demographic variables

Age, sex, educational level, marital status, religion, and HIV-status disclosure [yes (to family/friend/neighbor)/no] were collected at the time of the interview. One question from the AACTG questionnaire was used to measure the frequency patients had a drink containing alcohol in the last 30 days (Chesney et al., 2000). It was scored on a response scale ranging from “1 = daily” to “7 = never”.

Clinical variables

Dates of HIV diagnosis, enrollment in HIV care and cART initiation, duration since HIV diagnosis or cART initiation at the time of the study, WHO clinical stage, functional status (assessed by the treating clinician as working, ambulatory, or bedridden), CD4 cell count, hemoglobin level, evidence of tuberculosis at or after start cART (yes/no), any fixed-dose combination (FDC) in the regimen (double or triple FDC), and type of health-care facility (hospital versus health center) were abstracted from the pre-ART/ART register, patient follow-up card, and intake form using a Case Report Form.

Depressive symptoms

The Amharic version of the Kessler-6 scale (Tesfaye, Hanlon, Wondimagegn, & Alem, 2010), which consists of six questions each containing 5-point Likert scales (from “never = 0” to “all the time = 4”) was used to assess depressive symptoms. Patients with sum scores >12 in the Kessler-6 scale were considered to have depressive symptoms (Kessler et al., 2010). The scale yields high content and criterion validity, and is one of the most frequently used scales in developing and developed countries.

HIV-stigma

Stigma was measured with the six-item Internalized AIDS-Related Stigma Scale which was found to yield high levels of reliability and validity in Africa (Kalichman et al., 2005, 2009; Tsai et al., 2013). A dichotomous response scale was used, that is, “agree” versus “disagree”. The scale scores were summed to obtain a composite personalized stigma score ranging from 0 to 6, with higher scores indicating more internalized stigma.

Social support

A single item from the AACTG questionnaire was used to measure the patients’ general satisfaction with social support (from “1 = very dissatisfied” to “4 = very satisfied”) (Chesney et al., 2000). Responses to this item referred to the past month, and a higher score indicates more satisfaction with social support.

General and specific quality of life

We used a single item from the WHOQoL-HIV BREF questionnaire (Deribew et al., 2009; O’Connell & Skevington, 2012) to assess the overall quality of life (QoL) of the patients. It was scored from “1 = very poor” to “5 = very good”. Two additional items assessed transportation problems and financial constraints using a response scale ranging from “1 = not at all” to “5 = extremely/completely”. Moreover, one question assessed whether respondents and their family could obtain sufficiently nutritious food. This item was adopted from a previously conducted study in Ethiopia (Olsen, Jensen, Tesfaye, & Holm, 2013). It was scored on a response scale ranging from “1 = not at all” to “4 = very much”. Responses to the items referred to the past month, and a higher score indicates a higher value on the corresponding item.

Holy Water use

Previous studies in Addis Ababa reported that some HIV-infected patients renounce cART in favor of Holy Water use (Berhanu, 2010). Therefore, we included a single item about the frequency of Holy Water use as part of HIV/AIDS treatment to see if there is a relationship with sub-optimal adherence. It was scored on a response scale ranging from “1 = all-of the time” to “6 = none-of the time”.

Interviews

Clinicians or case-managers asked patients to participate following a scheduled regular clinic consultation. Patients willing to participate completed the interviewer-administered questions. To reduce socially desirable responses, introductory statements that acknowledge the use of Holy Water or the difficulties of maintaining optimal adherence were read before the interview began. In addition, interviewers were trained to be non-judgmental to responses of non-adherence.

Statistical analyses

Three separate logistic regression models were fitted for each of the dependent variables: for the self-report of less than all-of the time, for the clinician-recorded poor adherence during the six most recent clinic visits, and for the refill-adherence level of <95% MPR. Patients’ socio-demographic, clinical, and treatment characteristics, depressive symptoms, HIV-stigma, social support, QoL, and Holy Water use were the independent variables.

Independent variables associated with sub-optimal adherence were first assessed in a bivariate logistic regression analysis. Next, all the independent variables with a p-value of <.2 were fitted in a multivariate logistic regression model. Then, variables with a larger p-value were successively removed and kept aside one-by-one in a backward elimination procedure until only variables significantly (p < .05) associated with sub-optimal adherence or variables that hold potential clinical importance were remaining in the final model. During the course of variable elimination, we examined whether the primary coefficients for the remaining variables were significantly changed. In addition, variables not selected at a previous step, including those kept aside, were added back to the multivariate model (one at-a-time) throughout the model building process. Likelihood ratio tests were used to compare between the fit of various models. Data were first entered into EPI-data version 3.1 and later exported to STATA version 11 for statistical analyses.

Results

Participants

Of the 870 patients, 101 (11.6%) were lost to follow-up and 70 (8.0%) were dead by the time the study was conducted. An additional 34 patients declined participation and one patient had “stopped” taking ARVs while remaining in clinical care. Therefore, 664 (76.3%) patients who remained in HIV-care and were receiving cART were included in the present study.

At enrollment in HIV-care, the median (inter-quartile range, IQR) CD4 cell count was 152 (82–251) cells/µL. Table 1 describes patients’ characteristics.

Table 1. Patients’ socio-demographic, clinical, and treatment characteristics (n = 664).

Patient characteristics	Value^d
Age – years^a, mean (SD) (n = 661)	37.6	(9.3)
Sex – male^a	242	(36.4%)
Educational level^a (n = 661)
No education	81	(12.3%)
Primary	210	(31.8%)
Secondary	271	(41.0%)
Tertiary	99	(15.0%)
Marital status^a
Never married	115	(17.3%)
Married or cohabiting	346	(52.1%)
Divorced or separated	125	(18.8%)
Widowed	78	(11.7%)
Religion^a
Christian	618	(93.1%)
Muslim/other	46	(6.9%)
Functional status^b (n = 654)
Working	558	(85.3%)
Ambulatory	79	(12.1%)
Bedridden	17	(2.6%)
WHO clinical stage^b,c (n = 661)
Stages I/II	357	(54.0%)
Stages III/IV	304	(46.0%)
Evidence of tuberculosis at or after start cART – yes (n = 653)	119	(18.2%)
First line ARV-drug regimen^b
TDF-containing	339	(51.1%)
AZT-containing	266	(40.1%)
d4T-containing	59	(8.9%)
First-line ARV-drug regimen^b,
EFV-based	434	(65.4%)
NVP-based	230	(34.6%)
HIV-status disclosure – yes^b (n = 661)	599	(90.6%)
CD4 cells/µL^b, mean (SD) (n = 661)	150	(89.7)
Months since start cART^a, mean (SD)	26	(10.7)
Months since start cART^a, median (IQR)	26	(17–35)
Months since HIV diagnosis^b, median (IQR) (n = 653)	1	(0–8)
Self-reported alcohol intake in the last 30 days^a
Daily	2	(0.3%)
Nearly everyday	6	(0.9%)
Three to four times a week	10	(1.5%)
Once or twice a week	53	(8.0%)
Two or three times a month	66	(9.9%)
Once a month	136	(20.5%)
Never	391	(58.9%)

Notes: TDF, tenofovir; AZT, zidovudine; d4T, stavudine; EFV, efavirenz; NVP, nevirapine; IQR, inter-quartile range; SD, standard deviation.

^a at the time of the study.

^b at the start of cART.

^c according to the revised World Health Organization clinical staging of HIV/AIDS for adults and adolescents (WHO, 2005).

^dValues are n (%) unless otherwise indicated.

Self-reported sub-optimal adherence

Based on the response of less than all-of the time to the self-report adherence question, 82 of the 664 patients (12.3%) were classified as having sub-optimal adherence to cART.

Clinician-recorded sub-optimal adherence

Looking into individual patients’ follow-up cards, treating clinicians had recorded adherence for 621 (93.5%) of the patients during their most recent clinic follow-up visit. Seven patients had poor adherence recorded in their most recent clinic follow-up visit while 4, 8, 9, 5, and 6 patients had a poor adherence record during their second, third, fourth, fifth, and sixth most recent follow-up clinic visits, respectively. Overall, 27 (4.1%) patients were found to be poorly adherent based on at least one clinician-recorded assessment during the six most recent clinic follow-up visits.

Pharmacy-refill sub-optimal adherence

Medication adherence according to the pharmacy-refill measurement was calculated for 664 patients. Initially, refill data were unavailable for 10 patients and 111 patients had no refill data for at least 1 refill-period. Hence, the median (IQR) refill-adherence level was 97% (89–99%). After the “missed pharmacy-refill dates” were replaced with the corresponding “prescription-refill dates” for these 121 patients (if available – in total 286 refill periods), the median (IQR) refill-adherence level increased to 98% (94–99%). With this estimate, 180 (27.1%) patients were initially classified as sub-optimally adherent using the <95% MPR threshold. For 14 patients, whose “prescription-refill paper” was not found, we used the corresponding “clinic visit dates” and re-calculated the MPR; however, this did not change the average MPR appreciably (results not shown). Adding leftover pills from the previous refills slightly increased the median (IQR) MPR to 98.7% (96–100%).

Depressive symptoms, HIV-stigma, social support, QoL, and Holy Water use

Patients’ responses to questions about depressive symptoms, HIV-stigma, social support, QoL, and Holy Water use in the past month are described in Table 2.

Table 2. Patients’ responses to questions about depressive symptoms, HIV-stigma, social support, QoL, and Holy Water use (n = 664).

Self-report questions	Value^b
Depressive symptoms^a (n = 664)
Sum score ≤12	565	(85%)
Sum score >12	99	(15%)
HIV-stigma^a, mean (SD) (n = 656)	3.16	(1.7)
Satisfaction with social support^a (n = 664)
Very dissatisfied	127	(19.1%)
Somewhat dissatisfied	39	(5.9%)
Somewhat satisfied	203	(30.6%)
Very satisfied	295	(44.4%)
Overall QoL^a (n = 664)
Very poor	20	(3.0%)
Poor	80	(12.1%)
Neither poor nor good	215	(32.4%)
Good	200	(30.1%)
Very Good	149	(22.4%)
Patients have enough money to meet their needs^a (n = 657)
Not at all	178	(27.1%)
A little	189	(28.8%)
Moderately	234	(35.6%)
Mostly	34	(5.2%)
Completely	22	(3.4%)
Satisfaction with the transport^a (n = 662)
Very dissatisfied	57	(8.6%)
Dissatisfied	118	(17.8%)
Neither dissatisfied nor satisfied	259	(39.1%)
Satisfied	142	(21.5%)
Very satisfied	86	(13.0%)
Extent to which patients and their families obtain sufficient food with nutritional value^a (n = 664)
Not at all	29	(4.4%)
A little	140	(21.1%)
Moderately	422	(63.6%)
Very much	73	(11.0%)
Frequency of Holy Water use^a (n = 664)
All-of the time	110	(16.6%)
Most of the time	67	(10.1%)
A good bit of the time	27	(4.0%)
Some of the time	95	(14.3%)
A little bit of the time	76	(11.5%)
None-of the time	289	(43.5%)

^aat the time of the study.

^bValues are n (%) unless otherwise indicated.

Factors associated with sub-optimal adherence

Supplementary tables 3–5 present the results of the univariate and multivariate logistic regression analyses showing the relationship between sub-optimal adherence according to the different assessment methods and the independent variables.

Self-report: In the univariate analyses, a higher frequency of drinking alcohol was associated with more sub-optimal adherence, while having more satisfaction with social support was associated with less sub-optimal adherence. In the multivariate logistic regression model, these factors remained significant (Supplementary table 3).

Clinician-record: In the univariate analyses, attending follow-up care in a health center was significantly associated with a higher likelihood of sub-optimal adherence, while more satisfaction with social support was significantly associated with a lower likelihood of sub-optimal adherence. In the multivariate analyses, attending follow-up care in a health center and a high level of overall QoL were significantly associated with a higher likelihood of having sub-optimal adherence, while more satisfaction with social support remained significantly associated with a lower likelihood of having sub-optimal adherence (Supplementary table 4).

Pharmacy-refill: According to the pharmacy-refill adherence measure, younger age, lower educational level, attending follow-up care in a health center, and receiving d4T-containing cART regimens at cART initiation were significantly associated with a higher likelihood of having <95% MPR in the univariate analyses. In the multivariate logistic regression model, younger age, lower educational level, and lower CD4 cell count at cART initiation were significantly associated with a higher likelihood of refill-based sub-optimal adherence (Supplementary table 5).

Discussion

In this study, we investigated sub-optimal adherence using different adherence assessment methods among HIV-infected adults receiving cART in Addis Ababa. Despite an overall high degree of adherence to cART, that is, 97%, sub-optimal adherence as measured with self-report, clinician-record, and pharmacy-refill occurred in up to a quarter of the patients. We previously reported that non-adherence according to these adherence assessment methods was significantly predictive of having detectable viremia (Mekuria, Nieuwkerk, Yalew, Sprangers, & Prins, 2016). The present findings suggest that ART-programs in similar settings should not be lenient with the high level of adherence they may find in reports or published articles (Gill, Hamer, Simon, Thea, & Sabin, 2005; Mills et al., 2006; Oyugi et al., 2004). Instead, interventions aimed at maintaining high adherence levels and/or avoiding sub-optimal adherence are needed to prevent the subsequent development of virological treatment failure.

The degree of sub-optimal adherence appeared lower when based on the clinician-recorded adherence measure. One possible explanation is that clinicians may not have always recorded adherence, as patients’ follow-up cards are sometimes missing temporarily, especially in high patient load health-care facilities and during busy clinic days. Moreover, clinicians might not always be accurate to assessing and predicting their patients’ medication taking behavior (Bangsberg et al., 2001; Gross, Bilker, Friedman, Coyne, & Strom, 2002). Each sub-optimal adherence measurement was significantly associated with a unique set of independent variables. The only exception was that more satisfaction with social support was associated with a lower likelihood of sub-optimal adherence as measured with both the self-report and clinician-recorded measures. Our study results suggest that ART-programs should consider using various adherence assessment methods to explore factors potentially predictive of sub-optimal adherence at different levels, and from a different perspective.

The finding that more satisfaction with social support was significantly associated with a lower likelihood of having sub-optimal adherence (based on self-report and clinician-recorded adherence) is in line with several other studies (Amberbir et al., 2008; Ammassari et al., 2002; Langebeek et al., 2014). Drinking alcohol was significantly associated with sub-optimal adherence according to self-report. This is in agreement with previous study findings in both resource-limited and resource-rich settings (Conen et al., 2009; Medley et al., 2014), and it might be related to forgetfulness to take ARV drugs on time or inability to strictly follow the refill appointment schedules as a result of drinking. Therefore, interventions aimed at tackling sub-optimal adherence should consider improving social support and limiting alcohol intake as potential program targets.

Another important finding is that younger age was significantly associated with sub-optimal adherence based on the pharmacy-refill measure. These same patients were previously found to have a significantly higher likelihood of having a detectable viral load (Mekuria, Nieuwkerk, Yalew, Sprangers, & Prins, 2016). Several other studies have also reported that younger patients have lower adherence levels compared to older patients (Auld et al., 2014; Hadland et al., 2012; Nachega et al., 2009). Therefore, a well-designed and more comprehensive package of services is needed, specifically for young people, to reduce HIV infection, to promote testing for HIV, to enroll patients in HIV care immediately after being tested positive, to improve treatment adherence and retention in care, and finally to prevent detectable viral load.

Lower CD4 cell count at cART initiation was significantly associated with more sub-optimal adherence based on pharmacy-refill counts. This finding is in line with some other studies (Bitew, Berhane, Getahun, & Abyu, 2014; Seguy, Diaz, Campos, Veloso, & Grinsztejin, 2007). Patients with a lower educational level had a higher likelihood of sub-optimal refill-based adherence (Iliyasu, Kabir, Abubakar, Babashani, & Zubair, 2005). This might be related to poor understanding of taking the complex cART regimens or not following the scheduled appointments strictly (Tymejczyk et al., 2016). Therefore, these patients should be prioritized for targeted adherence support and close monitoring. The finding that a high level of QoL was associated with more clinician-recorded sub-optimal adherence was unexpected. This might be related to patients’ reluctance to strictly attend their clinic appointments once their QoL improves and they start feeling healthy (Assefa, van Damme, Hailemariam, & Kloos, 2010). Another explanation might be that patients who have a better QoL may also have a better relationship with their clinician, for example, have more trust, and are therefore more likely to disclose non-adherence (Hillen, de Haes, & Smets, 2011; Langebeek et al., 2014). However, cautious interpretation is required because of the different time frame both variables were referring to. While the self-reported QoL referred to the past month, the clinician-recorded adherence referred to the past six clinic visits (that is, about 6–18 months on average).

Our study has several strengths. First, the large sample size and the random selection of patients render our conclusions representative for all HIV-infected adults receiving cART in Addis Ababa. Second, we assessed sub-optimal adherence based on data from multiple sources, that is, patients themselves, patient follow-up cards, prescription-refill papers, manually-recorded pharmacy-refill cards, and electronic refill databases, making our data set more reliable and the study results more credible. Third, the involvement of second-year masters students receiving specific training for this study helped to collect high-quality data.

However, results from our study should be interpreted cautiously. First, the time frame for most of the psychosocial variables and the primary outcome variables was not the same. This may have affected the results. Second, unmeasured confounders, such as patient–clinician interaction, cannot be ruled out. Third, there were missing data for clinician-recorded adherence.

To conclude, our study results indicated that sub-optimal adherence was present in up to a quarter of the patients despite a high degree of adherence according to self-report, clinician-recorded, and pharmacy-refill assessment methods. Younger age, lower educational level, and lower CD4 cell count at start of cART were associated with sub-optimal adherence, which is also consistent with our previous study result that these variables were significantly associated with a detectable viral load. More satisfaction with social support, however, was associated with less sub-optimal adherence. Therefore, interventions aimed at improving adherence should focus on improving satisfaction with social support, and on younger patients, patients with lower educational level, and those who initiated cART at a lower CD4 cell count.

Acknowledgements

We would like to thank the patients, ART-clinic, ART-pharmacy, and Laboratory staff in all the participating health-care facilities who took part in the study.

Disclosure statement

No potential conflict of interest was reported by the authors.