Association between antiretroviral therapy adherence and employment status: systematic review and meta-analysis

Abstract Objective To assess the association between the employment status of human immunodeficiency virus (HIV)-infected individuals and adherence to antiretroviral therapy (ART). Methods We searched the Medline, Embase and Cochrane Central Register of Controlled Trials databases for studies reporting ART adherence and employment status published between January 1980 and September 2014. Information from a wide range of other sources, including the grey literature, was also analysed. Two independent reviewers extracted data on treatment adherence and study characteristics. Study data on the association between being employed and adhering to ART were pooled using a random-effects model. Between-study heterogeneity and sources of bias were evaluated. Findings The meta-analysis included 28 studies published between 1996 and 2014 that together involved 8743 HIV-infected individuals from 14 countries. The overall pooled odds ratio (OR) for the association between being employed and adhering to ART was 1.27 (95% confidence interval, CI: 1.04–1.55). The association was significant for studies from low-income countries (OR: 1.85, 95% CI: 1.58–2.18) and high-income countries (OR: 1.33, 95% CI: 1.02–1.74) but not middle-income countries (OR: 0.94, 95% CI: 0.62–1.42). In addition, studies published after 2011 and larger studies showed less association between employment and adherence than earlier and small studies, respectively. Conclusion Employed HIV-infected individuals, particularly those in low- and high-income countries, were more likely to adhere to ART than unemployed individuals. Further research is needed on the mechanisms by which employment and ART adherence affect each other and on whether employment-creation interventions can positively influence ART adherence, HIV disease progression and quality of life.


Introduction
Recent data from the World Health Organization (WHO) and the Joint United Nations Programme on HIV/AIDS (UNAIDS) indicate that access to human immunodeficiency virus (HIV) treatment and care services in low-and middle-income countries has expanded dramatically. 1 In 2012, over 9.7 million people living with HIV in these countries were receiving antiretroviral therapy (ART). 1 Despite this increase, ensuring adherence to HIV treatment remains challenging in all countries. A meta-analysis of patients in North America (n = 17 573) and Africa (n = 12 116) estimated that only 55% and 77% in these areas, respectively, achieved over 80% adherence 2 and a global meta-analysis, which included 33 199 patients on ART, reported that only 62% achieved over 90% adherence. 3 Common barriers to adherence include medication side-effects, pill burden, the need to disclose HIV serostatus, a perception of feeling well, treatment fatigue and structural and psychosocial factors. 4,5 For individuals, nonadherence can result in virological treatment failure, the development of drug resistance, disease progression and death. [6][7][8][9] At the community level, nonadherence makes HIV transmission more likely, 10,11 can substantially increase health-care costs, particularly for hospitalization to treat opportunistic infections, 12 and can decrease productivity. 13 One factor that has not been explored in depth is whether an individual's employment status influences adherence to ART. 14,15 Adherence to ART may be influenced by factors associated with the disease and its treatment, with the relationship between the patient and the health-care provider and with patients themselves, such as socioeconomic status which is often based on employment or occupational status in addition to educational level and income. 14,16 Moreover, differences in adherence between people employed in the informal and the formal economy have been linked to gender roles and inequalities in employment status. 17,18 A previous meta-analysis by our collaborative research group found that employed HIV-infected patients from low-, middle-and high-income countries were 39% more likely to adhere to ART than unemployed patients. However, the study included very few participants from middle-and low-income countries and therefore did not have sufficient statistical power to determine whether a country's income level had a significant effect on the association between employment status and ART adherence. 19 The aim of this systematic review and meta-analysis was to investigate the relationship between the employment status of HIV-infected individuals and ART adherence using updated data and a larger patient sample, which included more information from middle-and low-income countries.

Methods
This meta-analysis was reported in accordance with the Preferred reporting items for systematic reviews and meta-Analyses: Objective To assess the association between the employment status of human immunodeficiency virus (HIV)-infected individuals and adherence to antiretroviral therapy (ART). Methods We searched the Medline, Embase and Cochrane Central Register of Controlled Trials databases for studies reporting ART adherence and employment status published between January 1980 and September 2014. Information from a wide range of other sources, including the grey literature, was also analysed. Two independent reviewers extracted data on treatment adherence and study characteristics. Study data on the association between being employed and adhering to ART were pooled using a random-effects model. Between-study heterogeneity and sources of bias were evaluated. Findings The meta-analysis included 28 studies published between 1996 and 2014 that together involved 8743 HIV-infected individuals from 14 countries. The overall pooled odds ratio (OR) for the association between being employed and adhering to ART was 1.27 (95% confidence interval, CI: 1.04-1.55). The association was significant for studies from low-income countries (OR: 1.85, 95% CI: 1.58-2.18) and high-income countries (OR: 1.33, 95% CI: 1.02-1.74) but not middle-income countries (OR: 0.94, 95% CI: 0.62-1.42). In addition, studies published after 2011 and larger studies showed less association between employment and adherence than earlier and small studies, respectively. Conclusion Employed HIV-infected individuals, particularly those in low-and high-income countries, were more likely to adhere to ART than unemployed individuals. Further research is needed on the mechanisms by which employment and ART adherence affect each other and on whether employment-creation interventions can positively influence ART adherence, HIV disease progression and quality of life.
the PRISMA statement. 20 Studies of any design were included if they satisfied the following criteria: (i) the study involved people living with HIV; (ii) participants were receiving highly active antiretroviral therapy; (iii) treatment adherence was assessed using objective or selfreported measures; and (iv) employment was considered a possible factor influencing adherence.
We searched the Medline, Embase and Cochrane Central Register of Controlled Trials databases for the period January 1980 to September 2014 (Box 1). In addition, we used a narrative literature review approach to analyse and summarize information on HIV treatment, particularly on adherence, from a range of sources including UNAIDS secretariat reports, scientific conference abstracts and other grey literature. We contacted individual researchers for details of unpublished studies.
Two reviewers evaluated the eligibility of the studies identified and a third reviewer provided arbitration if there was a discrepancy. One reviewer extracted data, which were checked by others. The quality of the studies included was assessed using the Risk of Bias Assessment tool for Non-randomized Studies (RoBANS; details available from the corresponding author on request). 21 The risk of bias in a study was graded as low, high or unclear on the basis of study features including the selection of participants (selection bias), consideration of confounding variables (selection bias), outcome measurement (detection bias), incomplete outcome data (attrition bias) and selective outcome reporting (reporting bias).

Data extraction
For each study included, we recorded: the first author's last name; the year of publication; the country where the study was performed; details of the study design; the years when data were collected; the type of controls in case-control studies; the duration of follow-up in cohort studies; sample size; details of exposure measures, such as indicators of occupation or employment; age; sex; the odds that an employed patient versus an unemployed patient would adhere to ART; and the variables controlled for. Study countries were classified by geographical area and categorized as low-, middle-or high-income, as defined by the World Bank for 2014. 22 Study participants were defined as being of working age either by the study investigators or using the International Labour Organization's standard definition. 23 In addition, the International Labour Organization's definition of employment was applied: Persons in employment comprise all persons above a specified age who during a specified brief period, either one week or one day, were in the following categories: paid employment and selfemployment. 24 Consequently, studies were considered eligible if their authors defined employed people as those who, during a specified brief period such as one week or one day: (i) performed some work for wage or salary in cash or in kind; (ii) had a formal attachment to a job but were temporarily not at work during the reference period; (iii) performed some work for profit or family gain in cash or in kind; or (iv) were with an enterprise such as a business, farm or service but who were temporarily not at work during the reference period for any specific reason. 24

Data synthesis
The meta-analysis was performed using the DerSimonian and Laird randomeffects model to obtain a pooled estimate for the odds ratio (OR) and the associated 95% confidence interval (CI). 25 The model was chosen since it takes into account both within-and between-study variability, as betweenstudy heterogeneity was anticipated. Heterogeneity among studies was assessed by inspecting forest plots of the odds ratios from each study and by using the χ 2 test for heterogeneity, with a 10% Box 1. Search terms used for studies on the association between employment status and adherence to antiretroviral therapy level of statistical significance, and the I 2 statistic, with which a value of 50% represented moderate heterogeneity. We used leave-one-study-out sensitivity analysis to evaluate the stability of the results and to test whether any one study had an excessive influence on the meta-analysis. 26 In addition, we performed subgroup analyses to assess the influence of study design (i.e. crosssectional versus prospective cohort), the country's income group (i.e. low, middle or high), the adherence threshold and adherence measures. We used the χ 2 test to subgroup differences and reported the P-value for interaction between pooled OR and study-level characteristics. We used a variance-weighted, least-squares regression approach to estimate the effect of the year of publication (i.e. trend analysis) and the study sample size on the association between employment status and ART adherence. For all tests, a probability less than 0.05 was considered significant. All statistical tests were twosided. Analyses were performed using Stata version 12 (StataCorp. LP, College Station, United States of America).

Results
A flow diagram of study selection is shown in Fig. 1. Twenty-eight studies, involving a total of 8743 patients from 14 countries, met criteria for inclusion in the systematic review (Table 1). 4, The studies were carried out between 1996 and 2012 and publication took place between 1996 and 2014. Overall, 24 studies, involving 7484 of 8743 patients (86%), were cross-sectional, whereas the other four were prospective cohort studies. Eight studies, involving 1775 patients (20%), were carried out in the United States, three each were carried out in Ethiopia, India and South Africa and two were carried out in Uganda.
Twenty-seven studies, involving 8008 patients (92%), included participants who were already receiving ART. Twenty-three studies, involving 8171 patients (94%), defined adherence as receiving 95% or more of prescribed doses in a given period. In addition, 22 studies, involving 7755 patients (89%), used self-report questionnaires to assess adherence, whereas three used pharmacy refill data, one used the pill count, one used the blood drug concentration and one used a combination of methods. The median sample size was 300 participants (range: 46-1016). When reported, the percentage of males ranged from 21.3% to 88.0% between studies. The median percentage of unemployed participants was 41% (range: 16-79%) and the percentage of participants who adhered to ART ranged from 14.9% to 93.1%.

Risk of bias
The results of our assessment of the risk of bias in all studies included in the meta-analysis are shown in Fig. 2. The risk of bias in the selection of participants was low in all studies. However, the risk of selection bias due to inadequate confirmation or consideration of confounding variables was low in 16 studies but high in 12: 16 studies adjusted for major confounding variables during the analysis phase, whereas the remaining 12 reported unadjusted associations between employment status and adherence to ART. The risk of detection bias due to inadequate outcome assessment was low in the six studies that assessed adherence using objective measures and high in the remaining 22, which used self-report questionnaires. The risk of attrition bias due to inadequate outcome data handling was low in 24 studies, unclear in two and high in the two studies in which more than 20% of patients were lost to follow-up. The risk of selective reporting bias was low in all studies.

Association between adherence and employment
The strength of the association between being employed and adhering to ART reported in each study in low-, middle-and high-incomes countries is shown in Fig. 3, Fig. 4 and Fig. 5, respectively. The figures also give pooled estimates for the association in each income group. The pooled estimate for the OR in all countries was 1.27 (95% confidence interval, CI: 1.04-1.55). There was evidence of substantial statistical heterogeneity between the results of all studies (I 2 : 77%). The leave-one-studyout sensitivity analysis showed that no single study had an undue influence on the pooled estimate.

Associations in subgroups
The magnitude and strength of the association between being employed and adhering to ART varied by country income group: it was highest in the seven studies from low-income countries, for which the pooled estimate for the OR was 1.85 (95% CI: 1.58-2.18). In the 11 studies from middle-income countries, the association was not significant as the pooled estimate for the OR was 0.94 (95% CI: 0.62-1.42). However, the association was significant in the 10 studies from high-income countries: the pooled estimate for the OR was 1.33 (95% CI: 1.02-1.74). In addition, the strength of the association was significantly higher in prospective cohort studies than in cross-sectional studies: the pooled OR was 2.05 (95% CI: 1.50-2.81) and 1.17 (95% CI: 0.95-1.44) for the two study types, respectively (P-value for interaction = 0.003). The association was significant in studies that used an adherence threshold less than 100% (OR: 1.59, 95% CI: 1.35-1.87) but not

Fig. 1. Flowchart showing the selection of studies for the meta-analysis of the association between employment status and adherence to antiretroviral therapy
Records remaining after duplicates removed (n = 8322) Records identified in the literature search (n = 11 537) Records screened (n = 8322) Records excluded as they did not meet inclusion criteria (n = 8150) Articles whose full text was assessed for eligibility (n = 172) Articles excluded on full text assessment -none reported a relevant outcome (n = 144) Studies included in the meta-analysis (n = 28)  Participants' ages are given as a range or as a mean (standard deviation).
c Whether or not the measure of the association between employment status and adherence to antiretroviral treatment was adjusted for major confounding variables.

Period and study size effects
The association between being employed and adhering to ART was observed to change over time, i.e. there was a period effect. The magnitude of the association declined with the publication year -the pooled OR for studies published between 2011 and 2013 was lower than that for those reported before 2011 (Fig. 6).
Similarly, we found that study size also significantly influenced the association, which was more pronounced in those with small samples (Fig. 7). For each additional 100 study participants, the magnitude of the association decreased by 9% (P = 0.001).

Discussion
Overall, we found that patients with HIV infections who were employed were 27% more likely to adhere to ART than those who were unemployed. This is in agreement with the results of our previous meta-analysis 19 and of others 5 who reported that one of the barriers to ART adherence in both developed and developing countries was financial constraints, which may be considered a proxy for unemployment. One difference from our previous report is that the magnitude and strength of the association between being employed and adhering to ART were highest in studies from low-income countries. Hence, unemployment may have a greater effect on ART adherence in these settings. It is possible that employment facilitates adherence to HIV treatment because it is associated with, for example, increased social support, better structuring of time and improved psychosocial well-being -these associations have all been documented in the general population. 54,55 A review of 16 longitudinal studies showed evidence that unemployment has a negative effect on mental health 56 and other studies have found that depression and impaired psychosocial well-being are associated with poor ART adherence. 57 Employment may also promote increased material well-being, for example, by improving food security and housing quality and by reducing poverty -all three are known to be associated with adherence to HIV treatment. 5,58,59 In addition, employment may promote adherence to ART by improving access to medical services through employer-sponsored health programmes that specifically  5,60 Conversely, the association may, in part, be due to the positive effect of ART on an individual's ability to find and retain employment, i.e. reverse causation. For example, it was found that HIV-infected workers who were not receiving ART were almost twice as likely to report being unable to work in the previous week than those who recently began and were adherent to ART. 61 Another study on more than 2000 HIV-infected adults enrolled in an ART programme estimated that four years after ART initiation, employment had returned to about 90% of the rates observed in the same patients three to five years before ART initiation. 62 Further prospective, longitudinal, cohort studies of employment status, ART adherence and other factors could help tease out the relative effects of ART adherence and employment on each other.
Despite mutual reinforcement between employment and ART adherence, it can be difficult for individuals to maintain adherence. For example, it was found that the main reasons for the discontinuation of antiretroviral therapy in the workplace in South Africa were: individuals being uncertain about their own HIV status and about the value of ART; poor relationships between patients and health-care providers; and discrimination in the workplace. 63 Furthermore, workplace participants also felt that the follow-up visits required by ART clinics created problems for them with their employers. The most frequently cited reason for treatment discontinuation among these individuals was harassment and discrimination by line managers who refused to grant time off from work for clinic attendance. For public sector employees, the main reasons included relocation away from HIV care providers and having insufficient money for transportation to clinical facilities. 63 These findings suggest there is a considerable need to promote awareness of the importance of ART adherence and to develop employment arrangements among both employers and employees that encourage adherence.

Adherence, employment and gender
Unexpectedly, we found no evidence that gender had a significant influence on the association between employment status and ART adherence. Several factors could account for this finding. There is a gender gap in access to antiretroviral therapy. In most areas of the world, and especially in settings with a high burden of HIV infection, women are more likely than men to access both ART and supportive HIV services, such as targeted counselling and programmes for the prevention of mother-to-child HIV transmission offered by antenatal services. 64 Conversely, men may have better

Fig. 3. Association between being employed and adhering to antiretroviral therapy in studies from low-income countries, 2009-2014
Study reference OR (95% CI) Beyene

Period effect
Our finding that the association between being employed and adhering to ART declined over time may reflect changes in ART regimens, such as the trend to-wards once-daily dosing, reductions in the pill burden and side-effects, 67,68 better access to treatment for both employed and unemployed individuals or a drop in the cost of HIV medications. 1 Further research is needed to determine whether this finding is specific to the studies we included or whether it signifies a true decrease in the association over time.

Limitations
The findings of this meta-analysis should be interpreted with caution. The observational nature of the data -85% of the studies included had a crosssectional design -limited our ability to draw causal inferences and our findings may be affected by reverse causation bias or by other unknown confounding factors that were not adjusted for. Additionally, we found significant heterogeneity across the studies, which suggests that a substantial percentage of the variability in effect estimates was due to heterogeneity rather than to sampling errors, i.e. to chance. Much of the heterogeneity observed may be explained by differences in the adherence threshold, study sample size and study design. Nevertheless, even when there is substantial heterogeneity, metaanalysis is regarded as preferable for data synthesis to qualitative or narrative interpretation since these approaches can lead to misleading conclusions that should not be generalized beyond the scope of the analysis. 69 The ability to draw conclusions from quantitative data is an important feature of metaanalyses and its absence is one reason for avoiding narrative interpretations without a data synthesis. It is worth noting that the heterogeneity observed in the current study appears to be the norm rather than the exception in meta-analyses of ART adherence. 14,15 An additional potential limitation is that none of the studies included compared treatment adherence across different types of occupation. Further, we found evidence for a small study effect and it is possible that the observed magnitude of the association between being employed and adhering to ART could have been inflated by studies with small sample sizes. Despite these limitations, our study had important strengths. We conducted comprehensive searches of databases to ensure that all relevant, published studies were identified. We also carried out meta-regression analyses to determine whether any particular study-level factor explained the results or could account for the observed variations between studies. In performing this comprehensive and robust review of the existing literature, we identified gaps in the current literature on determinants of ART adherence.  Best fit Proportion to odds ratio variance Proportion to variance b a The odds ratio represents the odds that an employed patient would adhere to treatment compared with an unemployed patient. b The area of each circle is in inverse proportion to the precision of the odds ratio. Systematic reviews Antiretroviral therapy adherence and employment Jean B Nachega et al.

Areas for future research
Our review was unable to address all pertinent questions on the association between employment status and optimal adherence to HIV treatment. Our study findings indicate a need for a range of further research: (i) to investigate mechanisms by which employment and ART adherence affect each other; (ii) to determine how different types of employment can differentially influence ART adherence; (iii) to study the association and interaction between employment and adherence to medications for chronic comorbid diseases, e.g. hypertension, diabetes and asthma in HIV-infected or uninfected individuals; (iv) to conduct interventional studies of how employment-creation programmes can positively influence HIV treatment adherence, disease progression and quality of life; and (v) to carry out costbenefit and cost-effectiveness analyses of selected employment-creation interventions for people living with HIV. ■

Sample size
Best fit Proportion to odds ratio variance Proportion to variance b a The odds ratio represents the odds that an employed patient would adhere to treatment compared with an unemployed patient. b The area of each circle is in inverse proportion to the precision of the odds ratio.

Systematic reviews
Antiretroviral therapy adherence and employment Jean B Nachega et al.