Introduction

Depression is common among people living with HIV (PLHIV). A meta-analysis in 2014 reported that the prevalence of depression among PLHIV was 39% (95% confidence interval [CI]: 33–45%) [1]. Depression is a persistent mood disorder, characterized by feelings of sadness, loss, anger, and frustration that interfere with daily living. Depression can impair individuals’ social and cognitive functioning, and lead to disability and suicide at its’ worst cases. Although cost-effective treatments are available for depression, a substantial proportion of depressed individuals are never been diagnosed and treated [2]. Research has shown that depression has a negative impact on the uptake of HIV testing and care [3,4,5,6]. Individuals with depression can experience self-abasement, cognitive impairment, and self-isolation [7], which might explain why some depressed PLHIV are less likely to engage in HIV care.

Antiretroviral therapy (ART) is beneficial for both HIV-infected individuals and public health. For individuals, ART can boost their immune system, reduce the likelihood of opportunistic infections, slow progression to AIDS, and improve AIDS survival [8, 9]. Research has also shown that ART can reduce chances of both AIDS and Non-AIDS-related death [10]. From a public health perspective, ART can reduce HIV secondary transmission by lowering viral load among PLHIV. HIV testing and care are the key components of the strategy of “Treatment as Prevention” for preventing new infections [11]. HIV testing is the gateway to HIV care, and only those who are linked to and retained in care have the opportunity of starting ART. However, coverage of ART is less than 50% in most countries [12,13,14].

Depression could be one of barriers for ART use. Depressed individuals may feel desperate and have social functioning impairment, including weakening HIV care seeking behavior. However, the effect of depression on ART use is inconclusive in the literature. Some studies have shown that individuals with depression were less likely to initiate ART [15,16,17,18,19,20,21], while other studies have indicated that depression could increase the likelihood of ART use [22,23,24]. An argument is that depression might be a reflection of health concern, and can motivate PLHIV to seek HIV care and use ART. Turner and Fleishman reported that the effect of depression on ART use differed by ethnic groups [25]. The prevalence of depression and access to HIV care can vary among different ethnic groups. In addition, study design, measurement of depression and ART use, criteria for ART initiation, and settings where studies were conducted may have impacts on the association between depression and ART use. Given these disparate reports, we conducted a meta-analysis to summarize the evidence on the association between depression and ART use among PLHIV.

Methods

Literature Search and Study Selection

A systematic literature search was performed according to the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA). We aimed to identify studies that evaluated the relationship between depression and ART use, published between 1996 advent of combination ART and December 15th, 2015. Ten electronic databases were searched: BIOSIS Previews (Biological Abstracts & Biological Abstracts/RRM, Thomson Scientific Technical Support, New York), Embase (Elsevier, Amsterdam, The Netherlands), Ovid Medline (Ovid Technologies, Inc., New York), Pubmed (National Center for Biotechnology Information, Bethesda MD), PsycINFO (American Psychological Association, Washington), Scopus (Elsevier, Amsterdam, The Netherlands), Web of Science (Thomson Scientific Technical Support, New York), CNKI (Tongfang Knowledge Network Technology Co. Ltd., Beijing, China), CQVIP (Chongqing VIP Information Co. Ltd., Chongqing, China), and Wanfang Data (Chinese Ministry of Science & Technology, Beijing, China). CNKI, CQVIP and Wanfang Data were Chinese Databases. We also searched conference abstracts from 2001 to 2015 of the International AIDS Society (IAS) Conference (http://www.abstract-archive.org/) and Conference on Retroviruses and Opportunistic Infections (CROI), and searched dissertations and theses through ProQuest (1996–2015).

We used four main search terms based on the search strings developed by the Cochrane Collaboration. The search strategy for PubMed is attached in the appendix.

Study Criteria and Selection

The criteria for eligible studies are as follows: (1) used any study design; (2) targeted PLHIV; (3) measured depression as an exposure variable; (4) measured ART use as an outcome variable; and (5) provided sufficient information to calculate effect size (ES) estimates. We originally intended a separate assessment of anxiety, but found only one eligible paper.

Studies searched from multiple electronic databases were imported into EndNote 6.0 (Thomson Reuters Corporation, New York). Title screening was conducted after removing duplicates. Most irrelevant studies were excluded in this phase. The remaining studies were further screened by reading abstracts. Only studies that mentioned the association between depression and ART use were chosen for full text screening.

Data Extraction

For eligible studies, the following information was extracted in a standardized manner: first author’s name, year of publication, study period, study country, study design, sample size, study population (transmission route), percentage of Caucasians in study population, percentage of male participants, prevalence of depression, method of measuring depression, and method of measuring ART use.

ART use was a self-reported binary outcome in most of the eligible studies. Hence, odds ratios [OR] were used to quantify the association between depression and ART use among PLHIV. People without depression were used as the reference group for the calculation of OR. Several studies also reported the association using risk ratio or hazard ratio; we calculated OR and 95% confidence interval [CI] based on available data for synthesis. If data were not available in the paper, we contacted corresponding authors for related information that could be used for adjusted OR, or raw data that we could use to calculate crude OR.

Statistical Methods

A random effects meta-analysis using inverse variance weights was used to pool effect sizes across eligible studies. The natural logarithm of OR and 95% CI were used in the synthesis. We then transformed the overall effect size back into OR and 95% CI for ease of interpretation. The meta-analysis results were displayed with a forest plot.

To assess heterogeneity among eligible studies, the Q statistic, I 2 statistic, and τ2 were used [26, 27]. The Q statistic is underpowered to detect true heterogeneity, given that a small number of studies are included. Hence, we used a 10% significance level to increase our power to detect heterogeneity using the Q statistic.

To control for the effects of potential moderators and to explore the source of heterogeneity between studies, subgroup analyses and a random effects meta-regression model were conducted. Sensitivity analyses were performed to find influential studies, and we removed one study analysis at a time (total 11 populations in nine studies). We used a funnel plot, the Egger test, and the trim and fill analysis to assess publication bias. All statistical analyses were conducted in Stata 12.0 (Stata Corporation, College Station, TX).

Results

Results of Literature Search

Figure 1 shows the process of our study selection. The initial searches in 10 individual electronic databases yielded 3116 records, and in conference websites, we found 1042 conference abstracts, of which 1324 were duplicates and were excluded, leaving 2834 records for title and abstract screening. Thirty-two publications were identified for full text screening of which 13 reported associations between depression and ART use, and were eligible for our meta-analysis. Two studies did not report enough information to calculate effect sizes [28, 29], and two studies had overlapping samples [30, 31]. Hence, nine unique studies were included in our analyses. Two of these studies contributed two estimates of OR, as heterogeneous effects were observed across different study populations (e.g., drug users vs. non-users, and Caucasian vs. Black/Hispanic). Hence, we had 11 effect size estimates in the quantitative synthesis.

Fig. 1
figure 1

Flow chart of study selection

Full size image

Description of Studies

Table 1 summarizes the characteristics of the studies. Of 7375 pooled study participants, 68% were women, Caucasians represented 30%, and 15% had a CD4+ T-lymphocyte (CD4) count less than 350 cells/μL. Seven of nine studies used a scale for measuring depression, one study was based on physician’s diagnosis [16], and one study used self-report of depressive symptoms [22]. For ART use, six studies used self-report, and the other three studies were based on medical records. Seven studies used a prospective cohort design, and two were cross-sectional. Six of nine studies were conducted in the continental United States, with the other three from Puerto Rico, Uganda, and Russia. All of them were published between 2004 and 2014.

Table 1 Studies on association between depression and use of antiretroviral therapy (ART) among people living with HIV (PLHIV)
Full size table

Effect of Depression on ART Use

The prevalence of depression ranged from 19% to 77% in the included studies, and the pooled depression prevalence was 41% (95% CI 29–53%). ART use rate varied from 11% to 71%, and the pooled estimation was 52% (95% CI 37–67%). The pooled synthesis showed that depressed PLHIV were 14% less likely to use ART than those without depression (OR 0.86; 95% CI 0.71–1.05; Fig. 2). The Q statistic rejected the null hypothesis that the true heterogeneity was due to chance (chis-square: 32.20, degree of freedom = 10, P < 0.001) at a 10% significance level. The I 2 statistic indicated that 68.9% of variance could be explained by true heterogeneity. The variability of effect sizes (τ2) across studies was 0.061.

Fig. 2
figure 2

Forest plot of the effect of depression on antiretroviral therapy use among people living with HIV

Full size image

Moderator Analyses

Table 2 presents results of pooled OR from the subgroup analyses. The synthesized point estimates from studies with a prospective design, studies conducted in the continental United States, and studies using scale screening for depression showed that depression was associated with a lower likelihood of ART use. The pooled estimate from studies with a prospective study design showed a statistically significant association between depression and ART use (pooled OR 0.84; 95% CI 0.71–0.99). The pooled results from studies conducted in the continental United States showed a marginally significant effect on ART use (pooled OR 0.85; 95% CI 0.71–1.01). In the subgroup analyses by the measurement methods of ART use, the pooled point estimates were different (0.91 for scale screening vs. 0.62 for medical records), and their 95% confidence intervals did not substantially overlap. The pooled point estimate in studies without limitation on CD4 count was 0.91 (95% CI 0.71–1.17), and the synthesized estimate from studies targeting patients with a CD4 count less than 350 cells/μL was 0.75 (95% CI 0.55–1.03). However, none of these differences was statistically significant in the random effects meta-regression model. Publication year and percentage of Caucasian participants were not significant factors, either.

Table 2 Subgroup analyses of association between depression and antiretroviral therapy (ART) use
Full size table

Sensitivity Analyses and Publication Bias

Sensitivity analyses were conducted to detect influential studies in our meta-analysis. Table 3 lists the pooled OR, with one study at a time removed from the meta-analysis. We found that two studies caused a relative large change on point estimation, and both confidence intervals excluded 1.00, when either was removed [22, 25]. This indicated that the two studies were influential studies compared to the rest of the studies included in this synthesis. The study by Turner and colleagues showed depressed PLHIV were more likely to be on ART, and this conclusion was contrary to most other studies in our analysis. This study was conducted among Caucasian whites who might have more access to care for both depression and ART, and the association between depression and ART use among them might differ from other populations [22, 25]. The study by Feliciano and colleagues also had the similar finding. Depression was measured by self-report in response to a study question “are you depressed?” This measurement might introduce information bias [22, 25]. These two studies were influential in our meta-analysis.

Table 3 Sensitivity analyses of association between depression and antiretroviral therapy use
Full size table

For publication bias, the funnel plot was asymmetric (Fig. 3). However, the Egger test failed to reject the null hypothesis that no small study effects exist at a 5% significance level. The trim and fill analysis did not yield any trimmed or filled studies. There is no significant publication bias.

Fig. 3
figure 3

Funnel plot for assessing publication bias

Full size image

Discussion

Our meta-analysis of 11 populations in nine studies including 7375 PLHIV suggested that depression was associated with a lower likelihood of ART use. Though the overall finding was not significant at a 5% significance level, subgroup analyses showed a statistically significant reduction in ART use among depressed patients in the subset of studies with a prospective study design, or using a scale for depression measurement, or utilizing medical records to assure ART use. Prospective studies could provide more convincing evidence due to the ascertained temporal relationship. Studies with more rigorous measurements of depression and ART use could reduce reporting bias. Hence, pooled results from these subgroup analyses should be more reliable. Of nine included studies, six were conducted in the continental United States, and three studies (four populations) were from the Women’s Interagency HIV Study (WIHS) that contributed significantly to the overall effect [18, 19, 21]. WIHS is a multi-site prospective cohort study to understand how HIV impacts women’s lives, in which depression and anxiety were evaluated. More studies in different regions and populations are needed to provide more evidence for the effect of depression on ART use among PLHIV.

Depression measured among PLHIV is less likely to be a reflection of the immediate emotional reaction to HIV positive status and health concerns [32], as these studies included participants who had typically known their positive status for years or months. Depressed individuals can experience impairment of cognitive and social functioning [33]. In such cases, they might be less likely to seek HIV care and initiate ART use. Intervention for depression among PLHIV is important to improve their mental health in the long-run, and may have an indirect effect on increasing ART use.

Both depression and ART use rates may vary in different ethnic groups. Some studies found that HIV-infected Caucasians had a higher prevalence of both depression and ART use compared to those of Hispanics and African Americans [25]. However, few studies have treated ethnicity as a potential effect modifier. Ethnicity was only considered as a confounding variable for adjustment in most of included studies. Given the potential effect modification of ethnic group, failure to consider this may lead to a misleading combined result, given that heterogeneous effects exist. Illicit drug use might be an effect modifier for the association between depression and ART use [18]. Most studies in our meta-analysis did not report on this potential interaction. Hence, the pooled effect size on depression and ART use can be biased, due to ignoring the potential heterogeneous effects across different groups.

Research has shown that the prevalence of depression among PLHIV can vary among countries with varied income levels [1]. Some studies suggested that PLHIV from high- and middle-income countries had higher prevalence of depression than those from low-income countries [1]. The coverage of ART also varied in countries with different income levels [34]. Six of nine studies included in this analysis were conducted in a high-income country- the United States, so we could not perform a subgroup analysis by country income level. Thus, our study findings may not be generalized to the entire HIV-infected global population, as we did not have a representative sample from countries with varied income levels. We excluded two eligible studies due to incomplete data to calculate effect sizes, and we failed to elicit responses from the authors [28, 29]. One study was conducted in Nepal, and the other was in the continental United States. If we could add these two studies to our meta-analysis, our results might change.

Heterogeneity assessment suggested that the majority of variance was due to the true heterogeneity. We did not find a significant moderator effect in either the subgroup analyses or Meta regression between different groups (Table 2). A small number of studies in our meta-analysis might limit our power to detect the true difference. In addition, we cannot evaluate the variance within groups, as studies within groups used different methods to measure a certain variables. Take the measurement of depression as an example. Within the group of studies using scales for screening depression, some studies used the Center for Epidemiologic Studies Depression Scale (CES-D), and other used the Composite International Diagnostic Interview short-form (CIDI-SF) (Table 1). Even though studies used the same scale for depression, they might choose different thresholds for depression (CES-D: 16 vs. 23 as the cutoff value). Self-reported depression is a potentially biased measure. Hence, further studies on depression and ART use with a common measurement method are necessary to provide a more convincing pool estimate.

Publication bias was not noted in this meta-analysis. The Egger test failed to reject small study effects, and the trim and fill analysis did not trim or add a single study. As studies included in our meta-analysis were mostly from the continental United States, studies conducted in other regions might have different results. In conclusion, our study was the first meta-analysis to evaluate the association between depression and ART use among PLHIV. The evidence suggests that depression is associated with lower uptake of ART, particularly noted in the higher quality studies, and integrating mental health service into HIV care may improve ART use among PLHIV. As most of studies related to depression and ART use were implemented in the continental United States, more research is needed to provide evidences from other countries with varied ethnicities and income levels, especially low- and middle-income countries.