Adherence to Antiretroviral Therapy in Sub-Saharan Africa and North America:  A Meta-analysis FREE

Antiretroviral therapy (ART) has improved the health of many human immunodeficiency virus (HIV) positive individuals who otherwise would have died. Treatment efficacy relies, however, on sustained adherence, which constitutes a serious challenge to those receiving ART.^1- 2 The regimens are often complicated and can include varying dosing schedules, dietary restrictions, and adverse effects.³ Consistently high levels of adherence are necessary for reliable viral suppression^4- 5 and prevention of resistance,⁶ disease progression,⁷ and death.⁸

Access to ART is limited for the majority of individuals living with HIV/AIDS in sub-Saharan Africa. The World Health Organization's initiative to have 3 million individuals receiving ART by 2005 (“3 by 5”) has yet to meet its planned provision of care timelines, leaving more than two thirds of the global number of individuals needing care worldwide without access to ART.⁹ Although sub-Saharan Africa represents only 10% of the world's population, it represents 77% of women with HIV, 79% of AIDS deaths, and 92% of the world's AIDS orphans.¹⁰ There has been concern that African patients, many of whom live in poverty and lack formal education, will have suboptimal adherence to ART, which could lead to the development and spread of drug resistence.^11- 12

Are concerns of poor adherence in Africans justified? No study has performed a systematic meta-analysis of adherence levels in Sub-Saharan Africa. We conducted a systematic review of available ART adherence data in both Africa and a resource-rich setting, North America, to determine the level of adherence in emerging African treatment programs relative to the more established North American programs.

Prospective studies assessing adherence rates as a primary or secondary outcome (ie, noninterventional) in general HIV populations in North America and Africa were included. Studies had to have reported a threshold for adherence monitoring (eg, 100%, >95%, >90%, >80%). Studies were excluded if they reported adherence as a mean of all doses taken by the combined group of participants; were conducted in countries outside North America and Africa; contained experimental interventions to promote adherence because these studies do not reflect existing clinical settings¹³; or assessed only specific groups (eg, drug users, children, homeless individuals, homosexuals, men or women only)^14- 15 because the focus was adherence rates in general populations of HIV-positive individuals.

In consultation with an information specialist (Pearl Raju, PhD, Centre for International Health and Human Rights Studies, Toronto, Ontario), 3 of the authors (E.M., B.R., P.W.) developed search strategies. First, search terms that may indicate adherence (eg, adherence, compliance, pill counts, Medication Event Monitoring System [MEMS], directly observed) were identified. The searches combined these terms with Medical Subject Headings for HIV and were conducted from May 2005 to April 18, 2006, independently, in duplicate (B.R., P.W.). MEDLINE via PubMed, EMBASE, Cochrane CENTRAL, AIDSLINE, AMED, CINAHL, TOXNET, Development and Reproductive Toxicology Hazardous Substances Databank, Psych-info, and Web of Science were searched with the inclusion dates of the inception of the individual database up until April 18, 2006, except for AIDSLINE, which was searched from inception up until 2000 when its new citations ended. The Web sites of major HIV conferences also were searched, specifically all International AIDS Society conferences (up to Rio de Janeiro, Brazil, in July 2005) and all Conferences on Retroviruses and Opportunistic Infections (up to Denver in February 2006). Lay publications and Web sites were also searched including the Canadian AIDS Treatment Information Exchange publications, Médecins sans Frontières, AIDS Treatment News, and Google. Individual clinical researchers and AIDS cohort trial groups were contacted via e-mail and telephone (BC Centre for Excellence in HIV/AIDS, International AIDS Society, and Médecins sans Frontières) and asked if they were aware of any unpublished studies.

Using a predefined protocol (available from corresponding author on request), 2 investigators (E.M., P.W.), working independently, scanned all of the abstracts and obtained the full text of articles and reports from nongovernmental organizations that indicated or suggested a measurement of adherence had been achieved. After obtaining the full reports of the candidate studies (either a full peer-reviewed article, conference abstract, or non–peer-reviewed article), the same reviewers independently assessed eligibility. Reviewers were not blinded to study authors, study conclusions, and outcomes because blinding has been shown to have little effect on systematic reviews.¹⁶ To obtain the full information regarding conference abstracts, the studies' authors were contacted via e-mail and telephone. After all potentially relevant full-text articles and abstracts were identified, 2 of the authors (E.M., P.W.) and a member of the study team (Dugald Seely, ND, MSc, University of Toronto, Toronto, Ontario) met to achieve consensus regarding eligibility.

Between May 1, 2005, and April 23, 2006, data extraction was conducted independently, in duplicate, using a standardized form. Data abstractors collected information about the study country, study populations (age and sex), participant ethnicity (as classified in the original study), sample size, methods of adherence measurement, and outcomes. When more than 1 adherence measurement was used, data on all measures used were extracted and the most objective method was chosen for this analysis (eg, MEMS). Adherence measurements were defined in the studies as primary adherence thresholds. No studies reported exclusions of patients due to adherence levels prior to the study. However, in 8 studies, data are only reported for patients who completed the adherence assessments. In this case, only the data for study completers were used. Data on study settings (eg, nongovernmental organization clinic, specialist clinic) and predictors of ART adherence (eg, adherence threshold, use of MEMS, quality of assessment) were abstracted. Finally, where available, data were abstracted regarding whether patients receiving ART paid for it or received the therapy for free. Data on populations' disease state were not abstracted due to large heterogeneity of each study population. The data were entered into an electronic database such that duplicate entries existed for each study; when the 2 entries did not match, we reached consensus through discussion and if necessary requested third-party arbitration. We considered study quality according to whether studies used multiple measurement tools to assess adherence (n = 9) and when applicable considered loss to follow-up of greater than 20% of the study sample as being poor (n = 10).

To assess interrater reliability on inclusion of the articles and abstracts, the φ statistic, which provides a measure of interobserver agreement independent of chance, was used.¹⁷ Descriptive analyses were used to compare the number of full-text articles published over time. To determine if complexity of regimens might yield differing adherence rates, a z test was used for the pooled North American adherence rates prior to and after 2002, the year when the African studies first appeared in print. Thresholds for adherence were considered as greater than or equal to the cutoff levels. To determine pooled proportions of study participants adherent to individual study measurement thresholds, the variances of the raw proportions (r/n) were stabilized using a Freeman-Tukey–type arcsine square root transformation^18- 19: y = arcsine[√(r/(n +1)]+arcsine [√(r +1)/(n +1)], with a variance of 1/(n +1), where n is the denominator for population size. The I² statistic was calculated as a measure of the proportion of the overall variation in adherence that was attributable to between-study heterogeneity.²⁰ We anticipated large heterogeneity considering the varied populations, health care delivery systems, and course of the epidemic. The DerSimonian-Laird random-effects method was used to pool the transformed proportions,^21- 22 which recognizes and anchors studies as a sample of all potential studies, and incorporates an additional between-study component to the estimate of variability. Random-effects logistic regression was used to explore this heterogeneity and to compare continents after adjusting for the potential confounding due to adherence thresholds (100%, >95%, >90%, >80%); assessment criteria (self-reported, pharmacy refills); whether the study participant paid for ART; loss to follow-up (>20%); and clinic settings (outpatient, nongovernmental organization). For any missing data on covariates, the authors of the articles were contacted and when there was no response, these studies were considered negative. A sensitivity analysis was also conducted to examine the impact of these negative studies on the meta-regression. The unadjusted and regression-adjusted estimates and odds ratios were calculated by continent. Separate pooled analyses were conducted of all studies using the same cutoff thresholds for adherence. For sensitivity analysis, we confirmed that a Mann-Whitney U test gave results consistent with those of the other methods used to compare continents. The sensitivity analysis was conducted using a Bayesian random-effects model with an alternative logit transformation in addition to Monte Carlo Markov Chain simulations of variability.²³ Forest plots were created for each continent, showing individual study proportions with Clopper-Pearson confidence intervals (CIs) and the overall DerSimmonian-Laird pooled estimate. Each individual study in the forest plot represents the proportion of the study population meeting the threshold for appropriate adherence, as defined in the original studies. Results are reported as combined adherence proportions with 95% CIs. All P values are exact and P<.05 was considered significant. Analyses were conducted using StatsDirect version 2.5.2 (StatsDirect Ltd, Cheshire, England), which was developed by 1 of us (I.B.), Stata version 9.0 (StataCorp, College Station, Tex), and OpenBUGS version 2.1 (http://www.mathstat.helsinki.fi/openbugs/).

From the initial searches (May-November 2005), 136 relevant abstracts of full text articles were identified. Of these, 60 studies passed the first screening. There was near perfect agreement between the reviewers on the inclusion of 28 full-text articles and 2 abstracts addressing North American studies and 7 full-text articles and 15 abstracts addressing African studies (φ = 0.91). In a search update (April 18, 2006), 1 abstract addressing adherence in North American settings, 2 full-text African studies, and 3 abstracts addressing adherence in African settings were identified. Agreement on abstract inclusion was perfect. A flow diagram of studies included in the analysis is detailed in Figure 1.

The characteristics of the North American studies^{3- 4,24- 52} appear in Table 1. All full-text articles and abstracts were published in English. Of the 28 full-text articles reporting on studies conducted in North American settings,^{3- 4,24- 47} 26 were from the United States and 2 were from Canada.^48- 49 Of the 3 abstracts reporting on studies conducted in North American settings, 2 assessed US populations^50- 51 and 1 assessed a Canadian population.⁵² The characteristics of the African studies^53- 79 appear in Table 2. Of the 9 full-text articles reporting on studies conducted in Africa, 2 were from South Africa,^53- 54 2 were from Nigeria,^55- 56 1 was from Uganda,⁵⁷ 1 was from Senegal,⁵⁸ 1 was from Cameroon,⁵⁹ 1 was from Botswana,⁶⁰ and 1 was from Malawi.⁶¹ Of the 18 abstracts reporting on studies conducted in African settings, 2 were from Nigeria,^62- 63 4 were from South Africa,^64- 67 5 were from Uganda,^68- 72 1 was from Malawi,⁷³ 1 was from Rwanda,⁷⁴ 1 was from the Democratic Republic of Congo,⁷⁵ 2 were from Burkina Faso,^76- 77 1 was from Cote d’Ivoire,⁷⁸ and 1 was from Tanzania.⁷⁹

Full-text studies conducted in North America enrolled a median of 220 patients (interquartile range [IQR], 130-683). This was largely unchanged when combined with the 3 North American abstracts (median, 219 patients; IQR, 116-683). Full-text studies conducted in Africa enrolled a median of 109 patients (IQR, 60-263) and when the 18 African abstracts were included, the median was 100 patients (IQR, 60-270). Four abstracts were missing data^65,67,71,77; 3 on loss to follow-up and 4 on payment of ART.

Twenty-two (71%) of the North American studies assessed adherence using patient-reported adherence levels over a specified period. Three studies (10%) used pharmacy claims. Three studies (10%) in North America used the MEMS to assess adherence. Two studies (6.5%) used a mix of patient reporting and clinician observations and 1 study (3%) used clinic staff reporting. Eighteen (66%) of the 27 studies assessing adherence in African settings used patient-reported adherence over a specified period. Three studies (11%) used pharmacy claims and 6 studies (22%) used a mix of patient reporting of pill counts and clinician observations.

Fifteen North American studies defined adherence as 100% during the measurement period.^{3,24,26,30,32,36,38,40,42- 47,51} Nine North American studies assessed adherence as greater than 95%,^{4,27,33- 34,39,41,49- 50,52} 4 as greater than 90%,^29,35,37,48 and 3 as greater than 80%.^25,28,31 Eleven African studies assessed adherence as 100%,^{56,59,61,64,71- 72,74- 77,79} 11 as greater than 95%,^{54,57,60,62- 63,65,67- 70,73} 2 as greater than 90%,^53,78 and 3 as greater than 80%.^55,58,66 Thresholds defined to assess adherence were not systematically different across settings.

Included studies assessing ART adherence in North America began in 1998 and continued to the present. There was no difference in the pooled effect size for North American studies between studies published prior to 2002 (n = 4) and studies published since 2002 (n = 27) (P = .52). We were unable to identify studies assessing ART adherence in Africa prior to 2002 that met our inclusion criteria.

Our primary meta-analyses aimed to determine the overall proportion of patients in each continent meeting the threshold measurements used in the specific studies. The combined continent analysis indicates adherence of 64% (95% CI, 59%-70%; I², 98.7%). Figure 2 displays a forest plot of all North American studies and Figure 3 displays a forest plot of all African studies. When we pooled abstracts with full-text articles, North American studies (31 studies; 17 573 patients total) yielded a pooled estimate of 55% (95% CI, 49%-62%; I², 98.6%) and African studies (27 studies; 12 116 patients total) yielded a pooled estimate of 77% (95% CI, 68%-85%; I², 98.4%), indicating a significantly (P<.001) higher level of ART adherence in Africa. North American full-text articles (28 studies) had a pooled proportion of 57% (95% CI, 49%-64%; I², 98.9%) and African full-text articles (9 studies) had a pooled proportion of 71% (95% CI, 62%-79%; I², 91.6%); the superior ART adherence rate in Africa remained (P = .02). The Bayesian sensitivity analysis provided an alternative statistical manner to evaluate pooled proportions and were largely similar to the pooled random-effects analysis of 55% (95% CI, 49%-62%) adherence in North America and 81% (95% CI, 72%-87%) adherence in Africa. The Mann-Whitney U analysis in case of scale transformation problems with arcsine remained significant (P<.001). The adjusted odds ratio of ART adherence in African studies in relationship to ART adherence in North America, independent of the thresholds used, is 3.0 (95% CI, 2.6-3.6). The odds ratio, adjusted for all other potential covariates, is 2.5 (95% CI, 1.9-3.3).

Table 3 displays the pooled proportions of each continent adjusted for thresholds. As anticipated, we found large heterogeneity across study analyses. In multivariable analyses (Table 4), heterogeneity was examined by continent, adherence threshold (100%, >95%, >90%), and use of more than 1 adherence measure. North American studies using MEMS to assess ART adherence had a reduced but nonsignificant level (P = .08) compared with other North American studies that did not use MEMS. Free access to care was not associated with higher ART adherence in Africa (16 studies; 74% [95% CI, 64%-82%]) than North America (24 studies; 82% [95% CI, 67%-93%]) (P = .33).

The findings from this systematic review and meta-analysis suggest that ART adherence among sub-Saharan African patients in early treatment programs are favorable, although it should be noted that the complexity of treatment regimens is potentially greater in North America, which should be considered when interpreting the data. This suggests that concerns about suboptimal adherence are not supported by the data and such concerns should not contribute to delayed access to treatment. While these data are promising, these relatively high levels of adherence may decline as treatment access expands. The African studies in these analyses were conducted in patients with early access to limited therapy and are possibly not generalizable to the larger HIV epidemic in Africa. Furthermore, most African studies include patients during early therapy when they are experiencing dramatic increases in health status and before they develop long-term adverse effects of therapy. While expectations of poor levels of adherence in Africa appear to be thus far unwarranted, efforts to sustain adherence in Africa and elsewhere remain important goals to optimize outcomes for individuals and global HIV treatment.

Strengths of this systematic review include explicit eligibility criteria and conduct of a comprehensive search that identified a number of eligible articles not published or available on electronic databases. We attempted to contact all of the articles' authors, however only 36 provided appropriate contact information. Thirty-two authors (89%) responded and clarified that adherence rates were correct as abstracted. Independent reviewers assessed eligibility and agreement was high. In keeping with the large heterogeneity across studies, we used the random-effects model to pool proportions. We used random-effects logistic regression to account for the extent of differences between populations and examined a priori defined variables to explain the heterogeneity.

The main limitation of our review is in the quality of the studies. There is no gold standard for evaluating adherence to medication.⁸⁰ Patient recall and pill counts have inherent biases in their measurement.⁸¹ However, proportionately, both continents used a similar number of patient recall and pill counts to evaluate adherence, thereby making our analysis as valid as the policies to provide care in North America. There were only 3 studies that assessed adherence using the MEMS and these were in North America.^4,33,37 These studies reported a nonsignificant pooled reduction in ART adherence compared with self-reported studies of 19%, which is in keeping with expectations that self-report may exaggerate adherence. We found large heterogeneity across the meta-analyses as we had expected. We were able to explain heterogeneity across several outcomes. There are unidentified factors either related to populations, drug regimens, or methodological features that may have a large impact on either apparent or real adherence. Setting aside the pooled estimates, all but 2 African studies had more than 50% of all patients meeting the threshold for appropriate ART adherence. Furthermore, despite our extensive searching and contact with organizations providing care in North America and Africa, as well as reviewing other review articles,^82- 83 which have addressed several of our included studies,^{54,57- 60,63- 65,67- 68,70,73- 76,84} it is possible that we have missed some unpublished studies assessing ART adherence. We cannot know the extent to this limitation because funnel plots do not apply in nonintervention-based studies. Finally, our analysis was based on studies across a great spectrum of geographical and economic locations. It is possible that political or sociodemographic status can affect adherence rates. These data were rarely available in the included studies and while a more detailed analysis comparing regions would be informative in determining which factors within Africa influence ART adherence, it would not change the primary finding that treatment programs in Africa have thus far had relatively high levels of adherence even accounting for this heterogeneity.

Recognizing these limitations, the consistent difference of ART adherence in North America and Africa raises the question as to why early opinions may have underestimated adherence among Africans. This sentiment was expressed at high levels of international agency decision making.⁸⁵ Reports indicate that individuals living in poverty in North America had suboptimal adherence in the range of 56% to 67%.^86- 87 These data may have been interpreted to mean that poverty is a risk factor for incomplete adherence and that individuals living in extreme poverty would then have lower levels of adherence. The barriers to adherence among impoverished individuals in North America appear, however, to be due to poor patient-clinician relationships, untreated depression, substance abuse, and other factors that are common among poor individuals in the North American setting rather than poverty itself.⁵ It appears that the interactions between such factors and ART adherence in Africa may be quite different. Specific factors that can correlate with poverty are at play in reduced adherence and poverty itself is not the only determinant.

To date, the most important and prevalent factors that have been reported to negatively affect adherence in sub-Saharan Africa are cost,^{54,59,88- 89} not disclosing HIV status to a loved one or fear of being stigmatized,^54,60 alcohol abuse,⁹⁰ and difficulty in following complex drug regimens.^59,91 Studies report that the majority of patients receiving ART have disclosed their HIV status to family or friends^92- 93 and that those who have not appear to do worse with therapy.^54,60 Such patients are likely to have frequent treatment interruptions due to the fact that tablets must be hidden and therefore not taken in the presence of others. Encouraging voluntary HIV status disclosure in a community with access to ART may result in improved uptake of voluntary counseling and testing, help decrease the stigma, and improve adherence.

The findings of this analysis have implications for clinicians in both continents. We have shown that there are patients in both settings that have suboptimal adherence and that factors beyond poverty play an important role. Clinicians should therefore proactively inquire with patients about current barriers or facilitators of adherence to HIV medications. We have previously identified that important barriers to ART adherence in both the developed and developing world included forgetfulness, a lack of understanding of treatment benefits, severity of adverse events, and the level of complexity of the drug regimen.⁹⁰ Although the success of interventions to improve adherence is modest to date,⁹⁴ the use of patient-directed interventions and innovative reminder systems may be desirable for some patients.

Although the World Health Organization's “3 by 5” initiative aims to increase access to ART throughout sub-Saharan Africa, the goal is far from being achieved. In all of the sub-Saharan African countries included in our analyses, estimates of access to ART remain severely limited. For example, the World Health Organization estimates indicate that as of June 2005, the proportion of patients requiring urgent access to ART and currently receiving therapy was 56% in Botswana, 10% in Burkina Faso, 15.8% in Cameroon, 5.4% in Cote d’Ivoire, 3.2% in the Democratic Republic of Congo, 13.6% in Malawi, 8% in Nigeria, 26.5% in Rwanda, 12.5% in South Africa, 3.2% in Tanzania, and 56% in Uganda.^95- 96

Given the apparent relatively high level of ART adherence in Africa, one of the most controversial program components in developing countries is whether there is a need for intensive interventions such as directly observed therapy of ART (DOT-ART). Through observation of a patient actually taking a dose, by a close family member or friend to whom the patient has voluntarily disclosed their HIV status, DOT-ART is proposed to influence patient outcomes. Indeed, in settings with high HIV status disclosure rates, community-based DOT-ART with a patient-nominated treatment accompagnateur or supporter^97- 99 has been reported to be feasible and helps to improve or maintain high levels of ART adherence. However, because patient outcomes may be confounded by a constellation of services provided by the program, community support, or other factors, the effectiveness of community-based DOT-ART still needs to be confirmed in well-conducted randomized trials, which are under way in several African countries. Alternatively, long-term, clinic-based DOT-ART is not likely to be feasible due to the lifelong nature of HIV treatment and long transportation distances in rural settings. Patients who do not disclose their HIV status due to fear of stigma, discrimination, or violence will need other adapted and culturally sensitive innovative adherence support if they prove to be poorly adherent to ART.

The findings of this analysis have important policy implications. First, the expectation of poor adherence in Africa is not an evidence-based rationale for delaying the expansion of ART programs in resource-poor settings. Second, given the average relatively high levels of adherence in resource-poor settings documented in this analysis, the focus (or priority) must now be to maintain these ART adherence rates by increasing access to affordable ART and establishing reliable drug supply and distribution networks from the pharmacy to the individual patient. Third, understanding culturally specific barriers to adherence will be important in developing evidence-based interventions targeted at the individuals with poor ART adherence.

HIV/AIDS is the most difficult public health challenge the world currently faces. However, adherence to ART in Africa may not be the challenge that many predicted. Policies designed to combat this pandemic must be based on sound and timely evidence. Formulating policies based on assumptions, without seeking evidence, may leave millions of individuals without effective interventions.