A systematic review of the literature was conducted, and this report follows the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) Statement (Appendix A) (Moher et al., 2015). The Centers for Disease Control and Prevention’s (CDC) Prevention Research Synthesis (PRS) Project’s cumulative HIV, acquired immunodeficiency syndrome (AIDS), and sexually transmitted disease (STD) research database was used for identifying relevant reports. The PRS database comprises published literature located through automated and manual searches developed by librarians with expertise in developing and conducting comprehensive systematic searches (DeLuca et al., 2011). The searches included MEDLINE (OVID), EMBASE (OVID), CINHAL (EBSCOhost), PsycINFO (OVID), and Sociological Abstracts (ProQuest) databases. Each of the automated searches were developed in the MEDLINE database with indexing and keyword terms cross-referenced by using Boolean logic with no language limits. The finalized search was translated into the other databases to adhere to each proprietary indexing system. The PRS database comprises search strategies focused on (1) behavioral risk reduction, (2) medication adherence, and (3) linkage to, retention in, and re-engagement in HIV care (see Appendix B for the three searches as implemented in MEDLINE [OVID] with any restrictions applied). Other searches are available from the corresponding author.

The search of the PRS database for this review consisted of two separate queries, (1) a title, abstract, and keyword search using MH terms (e.g., bipolar, depression, mental disorder, mood states, posttraumatic stress disorder, psychotic disorder) (see Appendix C for the full list of MH terms); and (2) a search of citations coded with an indexing term for linkage or retention in HIV care (Appendix C). The two queries were initially conducted in July 2015 (search period: January 2002–July 2015) and were repeated in March 2016 and August 2017. Additionally, a separate hand-search was conducted to identify articles in non-HIV–related literature that might have been missing in the PRS database including checking the reference lists of included citations (Appendix D).

Studies were included in the review if they (1) were conducted in the United States; (2) assessed persons with HIV; (3) contained a measurement of MH status and a measure of retention in HIV care; (4) measured any of the following MH variables: Diagnostic and Statistical Manual of Mental Disorders (DSM) psychiatric diagnosis (American Psychiatric Association, 2013), MH-related symptoms, or MHSU (e.g., visit to a health professional for MH treatment or psychological counseling); (5) included children, adolescents, or adults with diagnosed HIV/AIDS; and (6) assessed MH symptoms or diagnosed MH disorders by self-report measures, interviews, or medical record abstraction. This review examined the entire range of MH diagnoses and symptoms that PLWH can experience by including the full continuum of MH diagnoses and symptoms that might affect retention in HIV care.

This review exclusively focused on MH diagnoses or symptoms and excluded studies that only reported a measure of substance abuse and did not assess MH status. Additionally, to ensure that the association between MH and retention in HIV care was not confounded by any ongoing HIV− or mental health–related interventions, intervention reports were excluded. Lastly, studies whose target population was an HIV provider or caregiver were excluded because PLWH were the target population for this review.

Citations and full articles from the search were reviewed by a varying combination of two independent reviewers from a pool of 4 authors (RR, MW, AA, CP) at the title and abstract level, followed by each full text by using DistillerSR v.2 (Evidence Partners, Ottawa, Canada) (Evidence Partners, 2016). Data were abstracted from each paper that met the inclusion criteria by using standardized forms developed by study staff (MW, AA). Information abstracted included study characteristics (i.e., study location, study design, target population, and inclusion criteria), sample characteristics (i.e., analytic sample size, age, sex, race/ethnicity, sexual orientation, substance use status, and health insurance status), MH variable information (i.e., type of MH diagnosis or symptoms, MH measurement tool, type of MH service, and provider type), retention outcome, and study quality. Effect estimates were abstracted that assessed the association between MH diagnoses or symptoms or MHSU as the exposure variable and retention in HIV care as the outcome variable. Additionally, adjusted effect sizes that accounted for potential confounders were abstracted for analysis when available.

Study quality was evaluated by using the National Heart, Lung, and Blood Institute (NHLBI) quality-assessment tool for observational cohort and cross-sectional studies (National Heart Lung Blood Institute, 2014). Responses to each of the 13 items included “Yes,” “No,” “Not Applicable,” or “Not Reported.” Items with a “No” response were considered a limitation of the study. We removed a single question from the NHLBI tool on sample power description, since it did not apply to the included studies. Citations were given an overall quality rating of “Good,” “Fair,” or “Poor,” based on a priori criteria. Prospective and retrospective cohort studies that assessed the MH variable before measuring the outcome with sufficient time between measurements and controlled for confounders were scored as “good.” Cross-sectional studies were categorized as “fair.” Cross-sectional and retrospective studies that failed to account for confounders and did not describe the exposure or outcome variable were categorized as “poor.” Responses were reconciled to resolve mismatches.

Publication bias was assessed by visual inspection of funnel plot asymmetry (Egger, Davey Smith, Schneider, & Minder, 1997) and validated by Begg and Mazumdar rank correlation (Begg & Mazumdar, 1994). Additionally, the random effects fail-safe N was estimated to determine what number of missing studies would be required to nullify the overall mean effect estimate (Rosenberg, 2005).

The primary outcome of interest was retention in primary HIV care. Retention in care was defined as continuity in HIV care appointments over a duration of time (Mugavero et al., 2012) and is operationalized as multiple, completed outpatient HIV care visits with a health care professional. Retention in HIV care include gaps in care (e.g., discontinuities in care), visit constancy (e.g., consecutive time intervals with one or more kept medical appointments), and appointment adherence (e.g., percentage of appointments attended or missed out of the number scheduled).

Effect sizes were converted to log odds ratios (ORs), weighted by the inverse variance, and then aggregated by using a random effects model (Hedges & Olkin, 2014). Effect sizes were then converted back to an OR for interpretation. Effect sizes that were reported in the opposite direction (association between MH and poor retention in care) were transformed so that an OR ≥1.0 indicated an improvement in retention in HIV care and a value <1.0 was indicative of a lower odds of retention in HIV care. The summary effect estimate, reported as an OR, was calculated by using Comprehensive Meta-Analysis v.2 (Biostat, Englewood, New Jersey) (Borenstein, Hedges, Higgins, & Rothstein, 2005). The likelihood of heterogeneity across studies was determined by the Q statistic, and the percentage of variability caused by heterogeneity was calculated by I² (Higgins, Thompson, Deeks, & Altman, 2003). I² values of 25%, 50%, and 75% were categorized as low, medium, and high heterogeneity, respectively.

A moderator analysis was then conducted by examining whether factors identified a priori might modify the effect estimate or explain any heterogeneity included: age (mean age), sex (percentage male), race/ethnicity (percentage black race), sexual orientation (percentage men who have sex with men [MSM]), substance abuse status (percentage currently or within the previous year using illegal substances), health insurance status (percentage receiving public health insurance), education status (percentage with a high school education or less), type of MH diagnosis or symptoms (depression diagnosis or symptoms versus other MH diagnosis or symptoms), measurement of MH status (MH clinical diagnosis versus self-reported symptoms), assessment period for retention in HIV care, retention in care measurement, and study design (prospective cohort, cross-sectional, or retrospective). Studies that did not report a specific population or study level characteristic were not included in the respective moderator analysis. Meta-regression was used to assess how continuous population level variables modify the overall effect estimate. Mixed model analysis was used to assess between group (Q_B) differences in dichotomous or categorical study level variables.

Studies that examined the association between more than one form of a MH diagnosis or symptoms (e.g., both depression and posttraumatic stress disorder [PTSD]) and retention in care outcome contributed multiple effect estimates to the overall summary effect. A sensitivity analysis was conducted to determine whether including multiple dependent effects from a single study substantially changed the overall results. The analysis was repeated with only one effect per study by using a pooled MH outcome (e.g., psychosis, psychiatric diagnoses, or mood disorders). If no pooled MH outcome variable was reported, the association between depression and retention in care was used as the single effect size, because depression was the most commonly assessed MH outcome. Only one effect size was included for studies that examined MHSU and retention in care.

After de-duplication, a total of 5,827 studies were identified that were published during 2002–2017. No relevant articles were identified in any of the hand-searched non-HIV–related journals. A total of 5,354 articles were excluded at title and abstract level, and 428 articles were excluded after full text review. After excluding irrelevant studies, 45 studies, involving approximately 57,334 participants in total, met the inclusion criteria (Figure 1). Data were abstracted from the included studies and analyzed (Appendix E). A total of 39 (52 effects) studies reported on MH diagnoses or symptoms, and 14 (12 effects) studies reported on MHSU.

The quality scores for each study are provided in Appendix F. The most common limitations among studies were exposure being assessed before the outcome measurement; lack of a sufficient time frame to measure an effect; and repeated exposure of assessment. Sample size justification and blinding of the outcome measurement were not determined to apply to the majority of the included studies. Similarly, because the majority of studies were cross-sectional or retrospective, loss to follow-up was also not applicable. Overall, we determined that the quality of evidence was fair because the majority of studies were cross-sectional or retrospective cohort design that did not account for confounders. In the subgroup analysis for MH diagnoses or symptoms and retention in HIV care (Table 2), studies that were rated poor had a larger effect estimate, compared with good and fair studies. However, study quality was not statistically significantly different across levels (Q_B = 4.55; p = 0.10).

The results of this study demonstrate that having MH diagnoses or symptoms is statistically significantly associated with lower odds of being retained in HIV care and are consistent with multiple quantitative and qualitative primary studies (Holtzman, Shea, et al., 2015; Sprague & Simon, 2014) that report poor MH is a salient barrier to retention in care. The present study attempted to examine specific types of MH diagnoses or symptoms and identify statistically significant modifiers of the association between MH and retention in care. Unfortunately, few studies reported on specific MH diagnoses or symptoms (e.g., anxiety or bipolar disorder) other than depression. However, depression was identified as statistically related to lower odds of being retained in HIV care. Although MH diagnoses or symptoms were a barrier to retention in HIV care, these results also reveal that increased MHSU was associated with higher odds of being retained in HIV care. The clinical implications of this study are that persons with HIV who are treated for their MH diagnoses or symptoms might have better HIV clinical outcomes than those whose MH diagnoses or symptoms remain untreated. Current HIV prevention strategies, such as ‘test-and-treat’ (Cambiano, Rodger, & Phillips, 2011; Cohen et al., 2016) and ‘treatment as prevention’(Centers for Disease Control and Prevention, 2017a), emphasize ART medication as key components. The results of this study suggest that the clinical treatment of HIV may need also to address mental health for patients to achieve optimal HIV outcomes, including retention in care. If persons living with HIV are retained in care, they more likely to be virally suppressed (Centers for Disease Control and Prevention, 2017c).

Use of ancillary services (e.g., MH treatment) has been previously reported to improve retention in HIV care by addressing the complex health care needs of persons who have multiple diagnoses (Conover & Whetten-Goldstein, 2002). This is the first systematic review to quantitatively examine the association between MHSU and retention in HIV care. The results of this review examining correlational studies demonstrated a larger association between MHSU and retention in HIV care than reports of intervention studies examining MH interventions (e.g., cognitive behavior therapy or motivational interviewing) (Crepaz et al., 2008; Mbuagbaw, Ye, & Thabane, 2012). Previous studies have demonstrated that the benefits of MHSU in PLWH include improvements in CD4⁺ count and mortality outcomes (Blashill, Perry, & Safren, 2011). Recent evidence indicates the need for MH services continues to be unmet (DeGroote, Korhonen, Shouse, Valleroy, & Bradley, 2016; Korhonen et al., 2016), and in the case of HIV care, this can affect retention. Thus, more methods are needed for improving retention in HIV care among these populations, including provision of access to and support for MH services.

A small, negative association was observed between MH diagnoses and symptoms and retention in HIV care. Sensitivity analysis revealed that the association was no longer statistically significant after removing one study (Ashman et al., 2002) that had a larger effect size and the largest population of all included studies. However, the effect estimate remained fairly stable thus did not change the interpretation of the results. Multiple studies have demonstrated that patients with a MH diagnosis who are receiving MH care are 10%–40% more likely to be retained in HIV care, compared with patients whose need for MH services remains unmet (Himelhoch, Brown, et al., 2009; Himelhoch, Josephs, et al., 2009; Meade et al., 2009; Sherer et al., 2002; Weiser et al., 2006). Therefore, the small negative association between having MH diagnoses or symptoms and retention in HIV care might have been larger if the populations studied only included patients not receiving any MH services. The patients in the included studies who were suffering from MH diagnoses in this review might have received other ancillary services (e.g., substance abuse treatment) that are also likely to improve retention in HIV care (Lucas et al., 2010). Future research examining the association between MH diagnoses or symptoms and retention in HIV care can explore this association by accounting for MHSU.

In the moderator analysis, public health insurance was determined to modify the association between MH diagnoses or symptoms and retention in HIV care. Studies with a lower proportion of participants with public health insurance had lower rates of retention in HIV care, compared with studies with a higher percentage of public health–insured persons. Moreover, one study (Byrd et al., 2015) that examined only Medicaid-insured participants reported that 61% of the population was retained in HIV care and a MH diagnosis was related to fewer gaps in HIV care. The higher odds of being retained in care among publically insured populations with a MH diagnosis might be attributable to their ability to receive more follow-up, compared with uninsured populations. In addition to health insurance, study design was identified as a substantial moderator in that prospective studies observed a larger effect estimate, compared with retrospective and cross-sectional studies. When assessing the long-term effects of a MH diagnoses or symptoms on retention in HIV care, prospective study designs are able to assess this association temporally. Because retention is defined as extending over a period, examination of variables affecting retention in care longitudinally is needed (Horstmann et al., 2010). Lastly, in the stratified analysis, no difference in effect size occurred when comparing studies that measured MH status by using diagnostic interview versus self-report assessments, which indicates that the association between retention in HIV care and MH is similar, regardless if it is a MH diagnosis or self-reported symptom. Furthermore, it might also indicate that self-reported assessment of MH symptoms with valid tools might be as useful as clinician assessments in identifying those in need of MH services while in HIV care.

This review had certain limitations. First, serious mental illness (e.g., schizophrenia or bipolar disorder) was not well-represented in the included studies; therefore, the results are not generalizable to all MH conditions. Further, the moderator analysis did not identify any factors that substantially reduced the heterogeneity. Therefore, other unidentified factors might account for the variation in the overall effect estimate. Few studies reported on a specific MH diagnoses other than depression, and the non-significant association between other types of MH diagnoses and retention may be due to lack of power. Moreover, the lack of power made it difficult to detect significant differences in the size of the relationship between retention and other types of mental health diagnoses or symptoms. Additionally, few studies reported subgroup data; therefore, this review was limited in assessing the association of MH and retention in HIV care across different populations and settings.

This review identified gaps in the literature, which limits generalizability to certain populations. Multiple studies examined black populations primarily; however, few studies examined other minority populations (e.g., Latinos or Hispanics). In addition, few studies specified the type of MH professionals providing the services, and only three studies were conducted in rural settings. Furthermore, although youth living with HIV suffer from high rates of mental illness (Mellins et al., 2009), few studies reported on that population. This review, however, did not find an effect of age on the association between MH diagnoses or symptoms or MHSU and retention in HIV care. Future research examining effects of MH on retention in HIV care and the benefit of MHSU on retention is needed among various MH diagnoses (e.g., generalized anxiety or bipolar disorder) and among specific populations (e.g., youth, Latinos, or Hispanics). Also needed are studies in rural settings and among different types of MH service deliverer (e.g., psychologist or social worker) because the associations might differ on the basis of these factors. Lastly, studies among general populations can provide stratified analyses to understand differences in association among certain subgroups.

The results of our study indicate that depression is statistically related to retention in care and that other types of MH diagnoses or symptoms need further examination (e.g., anxiety, bipolar disorder, and PTSD) in PLWH. To our knowledge, this is first meta-analysis to specifically identify public health insurance as a significant moderator of the association between mental health and retention in care. Although the effect of MH on retention in care was limited, few studies accounted for whether persons were concurrently receiving MH services. Future studies that investigate the role MH plays on retaining populations whose need for MH services remain unmet (i.e., have underutilization of MH services) will aid in identifying the importance of MH care for HIV outcomes. Moreover, results of this study demonstrate the need for addressing barriers to MH treatment in interventions focused on improving retention in care because mental health treatment is a strong facilitator of retention in HIV care.