Interventions and approaches to integrating HIV and mental health services: a systematic review

Abstract Background The frequency in which HIV and AIDS and mental health problems co-exist, and the complex bi-directional relationship between them, highlights the need for effective care models combining services for HIV and mental health. Here, we present a systematic review that synthesizes the literature on interventions and approaches integrating these services. Methods This review was part of a larger systematic review on integration of services for HIV and non-communicable diseases. Eligible studies included those that described or evaluated an intervention or approach aimed at integrating HIV and mental health care. We searched multiple databases from inception until October 2015, independently screened articles identified for inclusion, conducted data extraction, and assessed evaluative papers for risk of bias. Results Forty-five articles were eligible for this review. We identified three models of integration at the meso and micro levels: single-facility integration, multi-facility integration, and integrated care coordinated by a non-physician case manager. Single-site integration enhances multidisciplinary coordination and reduces access barriers for patients. However, the practicality and cost-effectiveness of providing a full continuum of specialized care on-site for patients with complex needs is arguable. Integration based on a collaborative network of specialized agencies may serve those with multiple co-morbidities but fragmented and poorly coordinated care can pose barriers. Integrated care coordinated by a single case manager can enable continuity of care for patients but requires appropriate training and support for case managers. Involving patients as key actors in facilitating integration within their own treatment plan is a promising approach. Conclusion This review identified much diversity in integration models combining HIV and mental health services, which are shown to have potential in yielding positive patient and service delivery outcomes when implemented within appropriate contexts. Our review revealed a lack of research in low- and middle- income countries, and was limited to most studies being descriptive. Overall, studies that seek to evaluate and compare integration models in terms of long-term outcomes and cost-effectiveness are needed, particularly at the health system level and in regions with high HIV and AIDS burden.


Introduction
In comparison with the general population, people living with HIV (PLHIV) are more likely to experience mental health disorders such as depression, anxiety, suicidality, and substance misuse (Chibanda et al. 2016, Hughes et al. 2016, Sherr et al. 2011, Brandt 2009). In low-and middle-income countries (LMICs), the prevalence of these common mental disorders is over 30% among PLHIVs (Chibanda et al. 2014). With estimates of 36.9 million PLHIVs globally, the burden of disease is significant (UNAIDS 2015). In fact, current predictors indicate that both HIV and AIDS, as well as depression will be the first two leading causes of disability globally by 2030 (Pappin et al. 2012, Gupta et al. 2010. The association between mental health problems and HIV and AIDS is complex and bi-directional. HIV virus and opportunistic infections associated with AIDS can cause neurological damage (Dube et al. 2005), while mental health problems can also arise as a side effect of antiretroviral treatment or from the stigma, stress and socioeconomic predicaments associated with the infection and treatment process (Moore et al. 1996, Yi et al. 2015. On the other hand, depression and substance use disorders, which commonly occur together is known to increase the risk of behaviours that promote HIV transmission, such as risky sexual activity and injecting drug-use (van Empelen et al. 2003). International evidence have found that populations with severe mental illness have higher rates of HIV infection (Senn and Carey 2009). Mental illness can also have a detrimental impact on adherence to antiretroviral therapy and progression of AIDS, leading to poorer health outcomes (Buckingham et al. 2013). Collectively, the cluster of diagnoses -HIV, mental illness, and substance abuse disorders -has emerged as a distinct clinical condition wherein patients experience a complex set of medical, psychological and social complications that need to be tackled through integrated care. Against this backdrop, many landmark publications including the UNAIDS Strategy 2016-2021 (UNAIDS 2016) and The Grand Challenges in Global Mental Health Initiative (Kaaya et al. 2013) have called for a stronger commitment towards integration of HIV and non-communicable diseases including mental illness and drug dependency.
Although the need for integrating HIV and mental health services is indisputable, the challenges are evident in implementing service integration that is cost-effective, and of high quality and impact. In LMICs, health systems are commonly overstretched due to poor human and financial resource, and oriented to treating acute conditions, resulting in fragmented care and poor sustainability of healthcare services for long-term disorders like HIV and mental illness (Semrau et al. 2015, Jacob et al. 2007. While high-income countries may have health systems that are better able to deal with a relatively lower overall burden of disease, literature from these countries has shown that initiatives which work initially have a tendency to be less effective when scaled-up (Parry et al. 2013). For these reasons, it is imperative to form an evidence base on what does and does not work in promoting HIV and mental health service integration.
Previous systematic reviews have examined HIV risk behaviours among adults with severe mental illness (Meade and Sikkema 2005); the link between mistreatment in childhood disorders, mental health disorders, and HIV infection (Spies et al. 2012); and literature on HIV and mental illness in low income countries (Collins et al. 2006). Studies have also reviewed intervention trials to improve mental health among PLHIVs in LMICs (Sikkema et al. 2015); as well as interventions using specific approaches like cognitive-behavioural therapy (Crepaz et al. 2008) or that target specific disorders such as depression (Sherr et al. 2011), anxiety ) and suicidality ) among PLHIV. A dearth of evaluated mental health services in HIV care is still evident, particularly in LMICs (Kaaya et al. 2013). We are unaware of any systematic review of the existing systemic approaches to the integration of mental health and HIV and AIDS services, and their effectiveness in enhancing patient identification, engagement in care, retention in care programs, treatment adherence, and clinical outcomes. Such a synthesis is needed, given the complexity of implementing models of care delivery that integrate HIV and mental health services as this requires multidisciplinary and inter-professional collaboration, coordination and communication. To address this gap, we systematically reviewed quantitative and qualitative studies describing and evaluating programs or services that seek to integrate HIV and mental health services in adult populations, reporting outcomes where available, and concluding with recommendations for future research.

Methods
This review was developed according to the PRISMA guidelines (Moher et al. 2009) and is one element of a larger systematic review on integration of HIV and non-communicable diseases. Drawing on the definitions proposed by Briggs, Atun,

Key Messages
• Available literature on interventions integrating HIV and mental health services reveal that there is much diversity in the approaches adopted in combining treatment modalities; ranging from integration within a single facility, to multi-facility integration, and integrated care coordinated by non-physician case managers. • Existing evidence, although limited, suggest that integrating HIV and mental health services may be linked to improved patient and service delivery outcomes in diverse settings. • There is a need for higher quality and robustly designed studies to evaluate and compare integration models at different levels of service delivery in terms of long-term impact on patient outcomes and cost-effectiveness, particularly in lowand middle-income countries with high HIV and AIDS burden.
Garner 2006), the concept of integration and its key attributes is described in Box 1 (WHO 2008, Atun et al. 2010b).

Inclusion criteria
We included all quantitative and qualitative studies describing or evaluating a management or organizational change policy or intervention implemented within an existing health system, aiming to integrate HIV and chronic disease care at the service delivery level. To be considered for inclusion for this paper, the studies had to integrate services for one or more mental disorders (e.g. depressive, anxiety, substance-related and psychiatric disorders) with HIV, which includes both the integration of mental health services into HIV services, as well as the integration of HIV services into existing mental health services. Services could be provided in health facilities or in the community and include any adult population. We did not exclude reports based on study design; nor did we require them to include outcome measures. We imposed no language, publication date, or publication status restrictions. Conference abstracts were included as this is an important source of unpublished studies.

Search strategy
The search strategy and terms were developed collaboratively with an information specialist, and were consistent with methods adopted by other authors who have conducted systematic reviews on health services integration (Groene andGarcia-Barbero 2001, Briggs andGarner 2006

Search and retrieval of studies
Two reviewers independently screened the list of articles obtained following the electronic database search based on title or title and abstract, to identify those meeting the inclusion criteria. If either of the two reviewers considered a study potentially eligible, we retrieved the full text for further assessment. For articles in languages other than English, a reviewer who could read and understand the article assessed it. The reviewers were able to read in Spanish and French. The two reviewers assessed the retrieved full texts independently to assess whether they met the inclusion criteria. Any disagreements were resolved by discussion with a third reviewer.

Data synthesis
Five reviewers (HLQ, DB, LG, NW and LO) independently extracted data from included studies using standardized forms. Differences in data extraction or interpretation of the studies were resolved by discussion and consensus among the five reviewers and with additional revisions by FLHC, VEH, SEO and FC when there were disagreements among the different pair of reviewers. We extracted data from the results and discussion sections of both quantitative and qualitative studies including information on: (1) study characteristics including study design, setting and sample size, (2) participants characteristics including age, gender, ethnicity and country of origin, (3) integration activities of the intervention, (4) results and type of outcome measure including process and patient outcomes, and (5) the advantages and disadvantages of integration activities as discussed in each study. We conducted a narrative synthesis of the findings.

Levels of integration
Valentijn's taxonomy of integration which is organized as the dimensions of the Rainbow Model of Integrated Care (Valentijn et al. 2013) was used as a framework to categorize papers in the data extraction and synthesis process. Drawing on this analytical framework, we consider integration at the macro level to involve the integration of delivery systems within the HIV, mental health and primary care sectors. We categorised integration at the meso level on two dimensions, i.e. organizational integration and professional integration. Organizational integration involves collaborative networks and relationships between agencies providing HIV, mental health and/or substance abuse services. Professional integration constitutes inter-professional partnerships of a multidisciplinary HIV, mental health and/or substance abuse team based on shared roles, responsibility and accountability reflecting the treatment plans of patients with multiple co-morbidities. At the micro level, clinical integration refers to the coordinated person-centred care in a single process across time, place and discipline, wherein all components of a patient's care in HIV, mental health and substance abuse are merged into one treatment plan. (Valentijn et al. 2015) Risk of bias assessment First, four independent reviewers (LA, NW, DB, LO) assessed risk of bias for papers assigned. Then, a fifth independent reviewer (HLQ) was involved to compare the results and resolve the differences in assessment. The Cochrane risk of bias tool was used to assess randomized control trials (RCT) (Higgins et al. 2011) while observational studies was assessed using a proforma with three domains: selection bias, information bias (differential misclassification and non-differential misclassification) and confounding. Each domain was assessed as low, unclear or high. We classified studies that had a low risk of bias in all domains as having a low overall risk of bias. Studies that had a high or a unclear risk of bias in one or more domains were classified as having an overall high or a unclear risk of bias. We evaluated qualitative studies using an adapted version of a checklist used in a previous series of mixed methods systematic reviews (Rees et al. 2006, Oliver et al. 2008.

11
,057 records were identified during the initial database searches. 7,616 articles, remaining after exclusion of duplicates, were screened by title and abstract for inclusion. 340 full-texts and abstracts were assessed for eligibility and 155 studies were found to include one or more non-communicable disease. For the purpose of this review, we then selected studies addressing HIV and mental health. Forty-five articles met the eligibility criteria for this review (See Figure 1), including 39 full papers and six conference abstracts. All papers reviewed were in English. Due to the heterogeneity in study design, intervention types, participants, and outcomes, we did not conduct a meta-analysis but instead present a summary of the articles, and a synthesis of their results and outcomes where available.

Characteristics of included studies
Of the 45 included studies, 26 of the articles were quantitative, two were qualitative, three were mixed-method studies and 14 were program or model descriptions. Of the 26 quantitative studies, seven were RCTs, five were non randomized intervention studies, five were cohort studies, three were case-series studies, three were crosssectional studies, and three were retrospective record reviews. Based on the World Bank's classification of income status, 38 of the 45 studies (84%) were carried out in high-income countries, 32 of which were in the USA, three in the UK, one in Canada, one in Australia and one in France. Two were carried out in an upper middle-income country, South Africa; and five in low-income countries, of which three were in Uganda, one in Zimbabwe and one in Tanzania (See Figure 2 for a geographical representation of the studies by integration models that are described in the following sections). Five of the 45 papers provided a definition of integration (Table 1). Of the 45 papers, only two studies described integrating HIV services within existing mental health services (Rosenberg et al. 2010, Lemmon andShuff 2001) while in the remaining papers, mental health and/or substance abuse services were integrated within existing HIV services. In 10 of these papers, these services were integrated in primary care settings (Farber et   Williams 1995, Winiarski et al. 2005, Feingold and Slammon 1993, Wolfe et al. 2003, Zaller et al. 2007, Wright and Shuff 1995, Esposito-Smythers et al. 2014, Nebelkopf and Penagos 2005, Dodds et al. 2004).

Risk of bias assessment
We conducted risk of bias assessments only for papers that evaluated integration of services and reported outcome measures or qualitative results. These included 15 quantitative studies, one mixed-methods study and one qualitative study. Nine studies were assessed to have an overall high risk of bias while seven studies were assessed to have an overall unclear risk of bias, and the qualitative study was assessed as unclear due to missing information. The risk of bias assessment ratings for the 17 studies by domain is shown in Table 5.

Levels of integration
Of the 45 papers, only two involved integration at the macro level (Wright andShuff 1995, Lemmon andShuff 2001). 31 papers involved integration at both the meso and micro level of which two integration models were identified, while the remaining 12 papers involved integration at the micro level only, representing a third integration model in this review ( Figure 3 represents the three models graphically by level of integration).

Macro-level integration
Both of the macro-level papers were written on the Indiana Integration of Care Project (IICP), a federally-funded project in the USA that integrated mental health services with Indiana's existing HIV and AIDS service delivery system at the state level (Wright andShuff 1995, Lemmon andShuff 2001). One of the papers described the program and the theoretical foundation underlying its conception, and included a cross-sectional baseline analysis of the linkages between community mental health providers with primary care and HIV providers (Wright and Shuff 1995). The other study sought to investigate the effect of mental health centre staff turnover on HIV and AIDS service delivery integration (Lemmon and Shuff 2001).
Meso-and micro-level integration 31 papers involved interventions in which integration occurred both at the meso and micro levels. From these papers, two distinct integration models were identified involving integration in a singlefacility and integration across multiple facilities. Twelve other  Winiarski et al. 2005 Integrated care defined as mental health services provided on-site at the medical clinic Coleman et al. 2012 Collaborative care defined by: (1) its guiding principles as described in The Chronic Care Model (CCM) which includes taking a team-based, patient-centered, collaborative approach that incorporates elements of patient care such as patient registries, patient education, screening or assessment tools, adherence monitoring, and evidence-based treatment guidelines; and (2) the degree of collaboration described as a continuum from less to more collaborative Weaver et al. 2009 The merging of health and medical services conceptualized on a continuum of care ranging from coordinated, meaning that care is delivered in different settings with information sharing among programs; to co-located, meaning that services are delivered at one location; to integrated, meaning that medical and behavioral healthcare components are merged in one treatment plan Dodds et al. 2004 Integrated service systems defined as multifaceted approaches to providing services for patients with complex needs, whereby two or more entities develop linkages to improve outcomes for their clients and combine efforts to serve clients more responsively. This means that providers from multiple disciplines share referrals, collaborate on case planning, and activate the resources of multiple agencies rather than constraining clients to a single agency or program  papers described interventions that integrated services exclusively at the micro level through the use of case managers, serving as the 3 rd distinct model of integration identified. The three models are described below to provide a sense of how HIV and mental health services are integrated at the meso and micro levels.
In terms of treatment modalities, 6 out of the 20 studies involved interventions that integrated HIV and mental health services (Farber et al. 2014, Feldman et al. 2012, Harris and Williams 1995, Hyam et al. 2012, Feingold and Slammon 1993, Nakimuli-Mpungu et al. 2014. In five other studies, the process was part of a larger package of integration with other services, including general primary health care (PHC) (Coleman et al. 2012, Winiarski et al. 2005 (Cohen et al. 2011, Dillard et al. 2010, Wood 2008), Hepatitis C treatment (Tetrault et al. 2012), risk reduction interventions (Nebelkopf and Penagos 2005) and specialist services (Kobayashi and Standridge 2000) in a single site. Table 2 lists the papers describing this model presented according to treatment modality and setting.
The single-facility integration model, otherwise known as 'onestop shopping', allows patients to access a variety of services at a single site. Four studies described that care coordination was implemented through regular case conferences bringing together members of the multidisciplinary team (Nebelkopf and Penagos 2005, Winiarski et al. 2005, Wood 2008, Kobayashi and Standridge 2000, while in one case, individual discussions, voicemails and shared medical notes were used as additional means to coordinate care (Winiarski et al. 2005). One study described an internal referral system to facilitate interdepartmental care coordination (Feldman et al. 2012). In another study conducted in the USA, there were also joint consultations involving HIV primary care, and mental health providers, in addition to case discussions and referrals. In this study, the degree of collaboration varied according to the patients' needs along the care continuum (Feingold and Slammon 1993). The single-facility integration model involved activities both at the meso-   and micro-levels, with professional integration based on multidisciplinary inter-professional partnerships and clinical integration based on patient-centered case conferencing and joint consultations.

MESO-LEVEL
The heterogeneity of the study locations indicates that this model of integration has been implemented in a wide range of different settings; although most commonly, services were integrated within primary care clinics (Farber et al. 2014, Harris and Williams 1995, Feingold and Slammon 1993, Winiarski et al. 2005, Dodds et al. 2004 or in a HIV clinic (Coleman et al. 2012, Namata Mbogga Mukasa et al. 2014, Jonsson et al. 2011, Esposito-Smythers et al. 2014, Surah 2013, Vergara-Rodriguez et al. 2012, Wood 2008, Tetrault et al. 2012, Kobayashi and Standridge 2000. In six of the studies in these settings, mental health services comprised of specialized psychiatric liaison services or consultative treatment (Coleman et al. 2012, Harris and Williams 1995, Kobayashi and Standridge 2000, Surah 2013, Hyam et al. 2012, Vergara-Rodriguez et al. 2012. In some of the studies, integration was implemented through a specific treatment program. Examples include: a measurementbased approach to depression care (Coleman et al. 2012), and cognitive behavioural therapy and contingency management measures (Esposito-Smythers et al. 2014, Nakimuli-Mpungu et al. 2014. The advantages of the single-facility integration model were discussed in some of the papers. From a provider's perspective, singlesite integration of services is perceived to enhance communication between providers, and reduce scheduling and coordination time (Coleman et al. 2012, Dillard et al. 2010. The involvement of a multidisciplinary team on site also increases the likelihood that the overall needs of a patient with dual or triple-diagnoses are considered within the treatment plan and competing priorities are addressed and minimised, reducing the occurrence of contradictory treatment demands (Dillard et al. 2010). From a patient's perspective, this model of integration reduced physical barriers to access, including transportation which often hampers continuous access to care, and other practical challenges facing those with mental or physical impairment (Dillard et al. 2010). Integration with primary health care or with other services, was also reported to improve confidentiality that might be breached when someone is seen attending a specialist mental health or HIV facility, reducing stigma and alleviating some of the anxiety among patients seeking care. (Coleman et al. 2012, Harris and Williams 1995, Wood 2008, Dillard et al. 2010). On the contrary however, it may be more difficult to implement single-site integration in smaller cities or rural areas where there is a lack of resources. Providing a full continuum of care within one facility may not be practical or cost-effective for patients with multiple co-morbidities, as they may need a more comprehensive or specialised range of healthcare services (Wood 2008).
Model 2: multi-facility integration. In 10 of the studies, services were integrated via inter-agency collaborations or mechanisms for external referrals to an intermediary: a collaborating agency or a collaborative network of providers. Nine of the studies were conducted in a high-income country, six of which in the USA (Curran et al. 2011, Daughters et al. 2010, Woods et al. 1998, Wood 2008, Taylor 2005, Rosenberg et al. 2010, one in Australia (Sternhell et al. 2012), one in France (Leclerc et al. 2005) and one in UK (McCarthy et al. 1992); one study was conducted in a low-income country, Zimbabwe (Duffy et al. 2014). Nine studies were reported in full papers and one was a conference abstract (Duffy et al. 2014). Of these, there were four descriptive studies (Woods et al. 1998, Wood and Austin 2009, Sternhell et al. 2012, Taylor 2005, two case-series (Daughters et al. 2010, Leclerc et al. 2005, two RCTs (Curran et al. 2011, Rosenberg et al. 2010, one non-randomized intervention study (McCarthy et al. 1992) and one mixed-methods study (Duffy et al. 2014).
In terms of treatment modalities, 2 of the 10 studies involved interventions that integrated HIV and mental health services (Curran et al. 2011, Duffy et al. 2014) while in one other study, these services were also integrated with Hepatitis C treatment (Sternhell et al. 2012). Three studies involved interventions that integrated HIV, mental health and substance abuse services (Daughters et al. 2010, Leclerc et al. 2005, Woods et al. 1998) while three other study interventions integrated these services along with primary  (Taylor 2005), and risk reduction services (Rosenberg et al. 2010). Table 3 lists the papers in which integration involved multiple facilities, presented according to treatment modality and description of referral channels.
In most of the studies, integration of services generally occurred via established referral systems between facilities or agencies that provide separate services (Duffy et al. 2014, Sternhell et al. 2012, Rosenberg et al. 2010, Daughters et al. 2010, McCarthy et al. 1992, Taylor 2005. In four of these studies however, off-site referrals were made only when the patient required more specialized mental health or HIV services (Daughters et al. 2010, Curran et al. 2011, McCarthy et al. 1992, Sternhell et al. 2012. For example, in one of the interventions that combined a brief behavioural activation approach and cognitive behavioural approach to treat depression and improve HIV medication adherence, patients were only referred for psychiatric treatment at a different facility when they were diagnosed with a psychiatric condition (Daughters et al. 2010). In two other studies, providers communicated through a network of agencies, and referrals were conducted via linkages between agencies within the established network (Woods et al. 1998, Wood andAustin 2009). In one of these studies, regular inter-agency case-conferences were also organized to coordinate patient care (Wood and Austin 2009). The multi-facility integration model involves integration at both meso-and micro-levels. Professional and organizational integration is achieved through collaboration of different specialized agencies mediated via collaborative networks and referral mechanisms, while clinical integration occurred through inter-agency case conferences and joint consultations.
In this model of integration, a facility may offer a range of integrated services co-located at one site and coordinate with other agencies and professionals for more specialized services. From a provider's perspective, the advantage of a multi-facility integration model such as this lies in the practicality and cost-effectiveness of offering a comprehensive range of services to patients with complex needs. One study described a community-based multiservice organization in the USA, which had a HIV and AIDS intensive case management and coordination unit, but reported that it was not feasible to provide the entire continuum of care on-site as the complexity of the patients' medical and social problems demanded a more comprehensive package of services. In this case, it seemed more practical to create a collaborative network of agencies (Wood and Austin 2009). In another study, however, splitting services over different sites was presumed to create barriers, as patients accessing different medical providers received fragmented, inconsistent, and poorly coordinated care (Daughters et al. 2010).
Model 3: integration through care-coordination using case managers. In 12 of the studies, integration of services involved the use of a non-physician, such as a nurse or a social worker, acting as a case manager responsible for developing an integrated treatment care plan and facilitating referrals. Nine of the studies were conducted in a high-income country, of which eight were in the US (Andersen et al. 2003, Sullivan et al. 2015, Adams et al. 2011, Wolfe et al. 2003, Zaller et al. 2007, Adams et al. 2012b, Sacks et al. 2011, Bouis et al. 2007) and one in Canada (Husbands et al. 2007); one was conducted in a middle-income country, South Africa (Andersen 2012); and two were conducted in low-income countries, in Uganda (Odokonyero et al. 2015) and in Tanzania (Adams et al. 2012a). Eleven studies were reported in full papers and one was a conference abstract (Andersen 2012). Of these, there were two descriptive studies (Andersen et al. 2003, Zaller et al. 2007, three RCTs (Adams et al. 2012b, Husbands et al. 2007, Sacks et al. 2011, two nonrandomized intervention studies (Bouis et al. 2007, Adams et al. 2011, one cohort study (Adams et al. 2012a), one case-series (Odokonyero et al. 2015), one cross-sectional study (Wolfe et al. 2003), one mixed-methods study (Andersen 2012) and one qualitative study (Sullivan et al. 2015).
In terms of treatment modalities, eight out of the 12 studies involved interventions that integrated HIV and mental health services (Andersen et al. 2003, Sullivan et al. 2015, Odokonyero et al. 2015, Adams et al. 2012a, Andersen 2012, Husbands et al. 2007, Adams et al. 2011, Adams et al. 2012b) while four studies involved interventions that integrated HIV, mental health and substance abuse services (Wolfe et al. 2003, Zaller et al. 2007, Sacks et al. 2011, Bouis et al. 2007. Table 4 lists the papers that described interventions which had case managers who integrated services for patients through a care plan. Out of the 12 studies, four described integrated care led by a nurse (Andersen et al. 2003, Sullivan et al. 2015, Odokonyero et al. 2015, Adams et al. 2012a, three described integrated care led by primary care staff (Andersen 2012, Wolfe et al. 2003, Zaller et al. 2007, three led by a social worker (Husbands et al. 2007, Adams et al. 2011, Bouis et al. 2007, one led by a depression-care manager (Adams et al. 2012b) and one that was integrated by the patient (Sacks et al. 2011). In most of the studies, the case manager was responsible for providing or facilitating integrated care by linking patients and assisting them to access necessary services as part of an integrated treatment plan (Sullivan et al. 2015, Andersen et al. 2003, Husbands et al. 2007, Zaller et al. 2007, Bouis et al. 2007). In some instances, the development of the treatment plan involved the collaboration between the care coordinator and patient or care providers (Andersen et al. 2003, Zaller et al. 2007, Bouis et al. 2007). In two studies, the nurse or primary care staff was also responsible for conducting screening for depression (Odokonyero et al. 2015), other mental health issues or substance abuse (Wolfe et al. 2003). In one study, the patients themselves were taught to coordinate service components of a modified therapeutic community aftercare program and integrate their own treatment. Through various self-help strategies and support groups, patients were educated on how to navigate services and were provided tools to manage and monitor vital elements of their treatment progress. Such client-level integration was perceived to be effective in bridging the gaps in care coordination and empowering clients to track and adhere to the key elements of their treatment plan (Sacks et al. 2011). The use of an algorithm-based tool for prescription and medication management by a nurse or depression-care manager was described in three studies (Odokonyero et al. 2015, Adams et al. 2012a, Adams et al. 2012b, of which two discussed it as part of a measurement-based approach to depression care involving the use of routine symptom measurement to inform treatment planning (Adams et al. 2012a, Adams et al. 2012b. In all three studies, the care manager was supported or supervised by a psychiatrist. It was propounded that this model of integration could help address the problem of under-diagnosis of depression in PLHIVs, account for antidepressant-antiretroviral interactions, and facilitate quality antidepressant management within HIV care (Adams et al. 2012b).
As described in one study, the nurse coordinating the care played a key role in helping patients access resources and providing psychosocial support and education on how to interact with doctors, and served as a source for patients to seek clarification when they were unsure about the information given by providers (Sullivan et al. 2015). Another perceived advantage of this integration model was its ability to promote continuity of care for patients as they relate to a single case manager. Yet to achieve these advantages, much effort is required on the part of the case manager to initiate collaborations between providers, which can be hindered by the competing priorities of the various providers with a different disciplinary orientation. As such, appropriate professional training of case managers is essential (Bouis et al. 2007).

Measures of effectiveness of integration
Seventeen studies involved evaluation of one or more measures of effectiveness of an integrated program, intervention, model or approach. We define patient outcomes as changes in the health status of the patients or their knowledge, attitudes and behaviours, while service delivery outcomes are defined as measures that reflect the effectiveness of the processes involved and delivery of integrated services. The 17 studies described at least one measure of effectiveness in either of these types of outcome, none of which reported longterm impacts on morbidity or mortality indicators (See Table 5 for the results of the studies that evaluated integration including a summary of the patient and process outcomes).

Macro-level integration
Of the 17 studies, one study evaluated integration at the macro level, investigating the effect of staff turnover on HIV and AIDS service delivery integration across three service components comprising of primary health care, mental health services, and HIV and AIDS dedicated care coordination. This cross-sectional study surveyed a sample of 51 staff from 17 mental health centres and found that staff turnover rates did not negatively impact integration, except for within-centre services, i.e. when HIV was integrated within the mental health system itself [t(15) ¼ þ0.05, P > 0.05]. The overall risk of bias was unclear, although the study identified some important challenges in the implementation of integration relating to poor communication and information sharing within centre, which can lead to a breakdown of referral patterns and limit access to quality patient care (Lemmon and Shuff 2001).

Meso-and micro-level integration
Among the 15 studies that reported one or more measures of effectiveness of integration at the meso and micro levels, seven studies involved single-site integration, three studies involved multi-facility integration and five studies involved integrating services through a case-manager. One study in particular, involved all three models of integration. This was an RCT in the US that assessed the costeffectiveness of integrated HIV primary care, mental health and substance abuse services for triply diagnosed patients where integration was across four different sites using single-site multidisciplinary case management, off-site referrals, and care coordinated by an adherence counsellor or nurse. Patients were randomly assigned to the intervention group (n ¼ 232) receiving integrated care, or the control group (n ¼ 199) who received care-as-usual. At the end of the 12month trial, the total average monthly cost of health services decreased from US$3,235 to US$3,052 in the intervention group group: 32% to 29%, P < 0.001 Unclear risk of selection and performance bias; High risk of detection bias; Low risk of attrition and reporting bias and US$3,556 to US$3,271 in the control group, but the decreases were not statistically significant. The percentage attributable to hospital care in both groups decreased, but there were no significant differences between them in annual cost of health services and quality of life. The overall risk of bias for this study was unclear (Weaver et al. 2009).
Model 1: single-facility integration. Among the seven studies, some assessed specific approaches like the measurement-based approach to depression care (Coleman et al. 2012) while others evaluated operating systems to facilitate inter-organizational referrals (Feldman et al. 2012). Four studies compared outcomes before and after intervention (Coleman et al. 2012, Cohen et al. 2011, Farber et al. 2014, Vergara-Rodriguez et al. 2012) and one retrospectively reviewed clinic data of a patient cohort on completion of referrals (Feldman et al. 2012).
Collectively, these studies reported improvements in clinical outcomes of HIV and mental health disorders, reduction in substance use behaviours and stigma, improvements in social functioning, and higher patient engagement in care, although the overall risks of bias of the studies were high or unclear. The evidence substantiating these reported outcomes are specified in Table 5. Two other studies of integration within a single-site were nonrandomized intervention studies (Winiarski et al. 2005, Surah 2013). In a study conducted in the US, 47 PLHIVs in the treatment group who received integrated mental health, HIV and primary care services designed to be culturally responsive and co-located within a single site; were compared to a control group of 100 PLHIVs who had access only to usual care, which included mental health services that were non HIV-specific and not co-located with primary care. Utilization rates were higher among the treatment group and this was associated with fewer mental health problems [F (1, 58) ¼ 8.22, P < 0.01], HIV-related physical symptoms [F (1, 34) ¼ 8.67, P < 0.01], alcohol [F (1, 37) ¼ 15.21, P < 0.01] and cocaine use [F (1, 79) ¼ 7.03, P < 0.01], and improvements in social functioning [F (1, 83) ¼ 4.35, P < 0.05]. The overall risk of bias for this study was unclear (Winiarski et al. 2005). The other non-randomized intervention study evaluated integrated care versus standard care among HIV-infected intravenous drug users seeking services at a HIV clinic with psychiatry-led addiction services in Ireland. Thirty clients were recruited to the intervention group and 26 to the control group. Clinical outcomes improved significantly among the intervention group, although there were no significant differences in healthrelated quality of life (HRQOL), anxiety, depression and substance misuse between the groups (Surah 2013). The risk of bias for this study was unclear as information was presented in an abstract only.
Model 2: multi-facility integration. Three studies assessing programs involving multi-facility integration reported outcomes reflecting one or more measures of effectiveness. One study examined the integration of a combined depression and HIV medication adherence program of three case series which reported improvements in depression rates, initiation of HAART and medication adherence (Daughters et al. 2010). Another study using mixed-methodology sought to determine the feasibility of a Stepped-Care Model integrating services between community, traditional medicine practitioners and health facilities using standard operating procedures and trainer manuals. The survey in this study presented a high percentage of successful referrals (80-100%), as well as increased awareness and reduced stigma among healthcare personnel in treating patients with co-morbidities (Duffy et al. 2014). These were not assessed for risks of bias.
The third study is an RCT that assessed the STIRR intervention (Screening and Testing for HIV, Immunization against hepatitis A and B, Risk-reduction counselling, and Referral and support for medical care). This intervention sought to facilitate integrated infectious disease programs in mental health settings and increase acceptance of such services among clients. The trial recruited 236 dually diagnosed clients receiving services at a community mental health centre and randomly assigned them to the STIRR intervention (n ¼ 118) or the control group (n ¼ 118). The control group received enhanced usual treatment, which included information on blood-borne diseases, information on local community health sources for blood testing, immunization against hepatitis A and B, and treatment as needed. Subjects randomized to STIRR had high levels (over 80%) of participation and acceptance of core services and were more likely to be tested for hepatitis B and C (88% vs. 14% at 6 months); immunized for hepatitis A and B (76% vs. 5% at 6 months); have an increase in their hepatitis knowledge (F ¼ 15.68, P <0.001) and reduce their substance abuse (F ¼ 4.54, P ¼ 0.34). However, they had no reduction in risk behaviour, were no more likely to be referred to care (81 vs. 75%) and gained no increase in HIV knowledge. The risk of bias was generally low in all regards except for the potential performance bias as subjects and researchers were not blinded. (Rosenberg et al. 2010).
Model 3: integration through care-coordination using case managers. Five studies assessing programs involving integration by a case manager had reported outcomes reflecting one or more measures of effectiveness, of which three were feasibility studies. One adopted a qualitative design in assessing a cognitive behavioural based intervention in an integrated program and reported reductions in depressive symptoms, global distress and level of impairments, although risks of bias could not be assessed (Andersen 2012). The other cohort study sought to test the feasibility of a task-shifting model of measurement-based depression care, reporting a reduction in depression score measured with PHQ-9 from 19.76 at baseline to 8.12 at week-12 (t ¼ 19.62, df ¼ 16, P < 0.001) (Adams et al. 2012a). Finally, the third study-a non-randomized intervention study, evaluated the feasibility of a collaborative depression care model using social workers to coordinate care and found a decrease in depression scores measured with PHQ-9 from 18.33 6 6.06 to 11.44 6 7.91 (t-2.73, df ¼ 8, P ¼ 0.03) (Adams et al. 2011). However, the risk of selection and non-differential bias were rated as high.
There were two RCTs associated with this integration model. One sought to evaluate an integrated therapeutic community aftercare program for triply diagnosed individuals in Philadelphia, US. Forty-two (55%) subjects were assigned to the intervention group who received integrated care and 34 (45%) to the control group who received standard aftercare services. The intervention consisted of health and self-management groups, peer-support groups, selfhelp groups, individual case assistance and family support groups designed to ensure treatment continuity and to assist patients' transition to more independent functioning in the community. Among the group of participants who had greater psychological and physical health at baseline, those in the intervention group had greater overall improvements in their mental health and substance use than those in the control group (Sacks et al. 2011).
The other study was an RCT in Canada that assessed a case management approach used to support integrated services implemented in a service organization located in Toronto to support PLHIVs. The study sample comprised 79 patients randomized to either the intervention or control group, although the potential for selection bias was evident. The intervention group undertook self-directed use of services facilitated by a social worker who would assist the patient in accessing support, while the control group received care-as-usual comprising only self-directed use of services. Those with more severe depression benefited the most from case management which had a positive effect on their physical, social and mental well-being, as well as on their risk behaviours. Additionally, participants' use of community services was associated with a lower expenditure for all direct health and social services (Husbands et al. 2007).

Discussion
This review brings together evidence on different models that have sought to integrate HIV and mental health services, ranging from integration within a single facility to multi-facility integration and integrated care coordinated by a non-physician case manager. The treatment modalities integrated within each model differed; some were more complex than others, especially those that included substance abuse and other types of services. This, coupled with the difference in settinge.g. rich and poor countries, varied packages of care, and the broad spectrum of degree of integration, i.e. less to more integrated -affirms that it is not possible to draw firm conclusions about the effectiveness of models. However, some tentative deductions can be drawn on the potential advantages and disadvantages of the differing integration activities and strategies within each model from a patient and provider perspective ( Table 6).
As expected, single-site integration is advantageous where there are already different providers working under one roof. The heterogeneity of the study locations indicates that this model of integration can be implemented in a wide range of settings. Single-site integration can increase access to services by reducing the inconvenience, additional costs and physical barriers that (often vulnerable) patients may encounter (Coleman et al. 2012, Dillard et al. 2010, Wood 2008. However, providing a full continuum of care for patients with dual or triple diagnoses who may need a more comprehensive range of services, can be costly and impractical. Multi-facility integration through a collaborative network of specialized agencies may be more effective when the treatment needs of a patient with multiple co-morbidities are beyond what can be provided within a single facility (Wood and Austin 2009). In some instances, integration via a system that facilitates rapid referrals may be more appropriate, particularly when a patient requires very specialized care or when few mental health specialists are available (Wood 2008). However, the degree to which services are integrated via referral mechanisms can be examined, as integration activities can range from mere referrals between services (least integrated), to having more formalized referral systems and linkages organized within a pre-established network of agencies that coordinate care via inter-agency case conferencing (more integrated). Effective referral systems supported by appropriate coordination mechanisms may be needed to prevent fragmented and poorly coordinated care in multi-facility integration.
Integrated care coordinated by a case manager can enable continuity of care for patients. However, this requires that these cadres have adequate training in the separate areas of HIV, mental health and substance abuse and are well supported, if they are to coordinate care effectively. This model of integration may be adapted in LMICs given the limited resources, scarcity and poor distribution of mental health specialists. While studies from LMICs is limited, the available evidence seem to show that task-shifting mechanisms may be feasible and beneficial through the use of less specialised personnel such as nurses, medical assistants and 'expert clients' who can be trained in detecting, screening and managing psychological conditions under the supervision of a psychiatrist (Odokonyero et al. 2015, Adams et al. 2012a. In LMICs, the integration of services may also need to consider alternative providers such as traditional medicine practitioners (Duffy et al. 2014). Active screening can be possible via the innovative use of data collection tools implemented within existing HIV facilities to effectively identify patients with potential symptoms of mental illness (Namata Mbogga Mukasa et al. 2014). Additionally, consideration of the social and cultural context in which patients conceptualize their beliefs and understanding of mental illness and treatment are likewise important in the development of integrated services of HIV and mental health (Nakimuli-Mpungu et al. 2014).
We also identified some novel approaches to integration, wherein patients were taught how to coordinate service components within their own treatment plan via self-management and support groups designed to educate patients on how to navigate services and use self-help tools to monitor vital elements of their treatment progress (Sacks et al. 2011). This model not only bridges the gaps in care coordination, but also engages with patients and enables them to take personal responsibility for decision-making and management of their own care. There is evidence that empowering the patients can increase the likelihood of positive treatment outcomes while reducing the burden on healthcare resources and capacities (Swendeman et al. 2009). However, regardless of which model is adopted, the context in which it is implemented must be taken into account, including factors such as resource availability and distribution, as well as the patient's specialized needs and where they are on the continuum of care: diagnosis, initiation of treatment, care for additional morbidities etc. Culture, institutional and social norms, as well as patient and family preferences are likely to be important in determining whether the patient will be motivated to play an active role in their own treatment (Martin et al. 2005).
Very few papers in this review defined integration. When defined, the term was commonly used interchangeably with collaborative or coordinated care to describe similar models of service delivery. Definitions varied greatly, from describing the term as simply as a co-location of services to more comprehensive descriptions of coordinated care along a continuum that included referrals and linkages of services via inter-agency collaborations. This is expected considering the complexity and multi-dimensionality of integrating multiple treatment modalities in striving to deliver quality and costeffective care to patients with dual and triple diagnoses. A previous systematic review on measurements of integrated healthcare delivery supports this notion that despite the vast literature on the subject, there is no consistent definition or fully-developed concept of service integration (Strandberg-Larsen and Krasnik 2009). The lack of conceptual clarity challenges the systematic understanding of integrated care and its attributes, which could hamper the design, delivery, management and evaluation of integrated programs (Valentijn et al. 2013). A clearer construct of the complex phenomenon of integrated care at the outset can help to guide empirical research and validate the evaluation outcomes of integration, thus allowing an accurate assessment of whether activities designed truly reflect an integration of services that is cost-effective, and that ultimately improves patient outcomes. Additionally, conceptual clarity on what integration should or should not be, and the attributes that underlie the integrating activities could help interpret evidence better on the value of the various integration models.

Study strengths and limitations
A strength of this review was the use of a wide range of databases and conference archives to increase the number of papers from LMICs for inclusion, although studies identified from the search Table 6. Potential advantages and disadvantages of each model. were mostly from high income countries, particularly the USA which could be due to publication bias. There are few real world initiatives that are evaluated, and it is also possible that studies with null findings are less likely to be published. Although conference archives were searched as a source of unpublished studies, conclusions could not be drawn on effect sizes and risks of biases of these interventions due to the limited information provided in these abstracts. Similarly, a majority of the papers included in this review were descriptive. While these provided useful insights on the approaches and strategies adopted in integrating HIV and mental health services, we could not infer the effectiveness of the various interventions described. In total, there were 17 studies that reported measures of effectiveness on integration, of which only four were RCTs. These studies were of variable methodological quality, a majority of which had an overall high or unclear risk of bias.

Implications for research
This review reveals that much of the research on integrated HIV and mental health care has described small-scale interventions or specific treatment approaches that involve some degree of integration activities at the meso and micro levels. Evidence on the effectiveness of systemic approaches to the integration of HIV, mental health and substance abuse services at the macro-level is clearly lacking. Further research is necessary to evaluate functional approaches to integration that engage with the financing, information systems, and management modalities of service delivery within health systems. There is also a need for evidence on strategies that could facilitate the normative underpinnings of integrated care, including sharedvalues, culture and goals across individuals, professionals, organizations and systems (Valentijn et al. 2013). Additionally, none of the papers reviewed reported long-term outcomes or impacts relevant to HIV or substance abuse, such as mortality. The longest period over which outcomes were measured was 6 months. Also, none of the papers compared outcomes or cost between different models of integration. This exemplifies the need for higher quality and robustly designed studies that seek to evaluate and compare integration models in terms of their long-term impact on patient outcomes and system-level outcomes. These may include mortality and morbidity indicators relevant to the disease progression of HIV and mental disorders; as well as the reporting of service coverage outcomes, institutional-based outcomes, and cost-effectiveness of realworld interventions. The incorporation of evaluative elements in study designs is also necessary to identify stronger causal linkages between intervention components and desired outcomes.
Given the varying needs of patients with HIV along the care continuum, there is a need for more evaluation of interventions that seek to integrate services at the pre-antiretroviral and end-of-life phases in HIV care. In this review, we found no studies that explored interventions at these periods. Additionally, very few studies described the integration of HIV screening or care into existing mental health services. As described at the beginning of this paper, mental health conditions are known to precipitate HIV transmission behaviours and affect antiretroviral therapy adherence. Therefore, further research is needed to address the under-diagnosis and undertreatment of HIV infection among patients with serious mental illnesses. Additionally, none of the studies reviewed involved integration of HIV and mental health services within antenatal care programs. Further research is necessary since previous studies have identified psychiatric symptoms -particularly depression, as a common condition among pregnant women with HIV globally (Kapetanovic et al. 2014).
While this review sought to include papers in languages other than English and studies conducted in different geographical regions, only seven papers were identified from LMIC countries within Eastern and Southern Africa, which has the greatest burden of the AIDS epidemic. A previous systematic review revealed that the majority of HIV and AIDS and mental health studies in sub-Saharan Africa focused on mental health-related HIV risk behaviours, HIV in psychiatric populations, and mental illness in HIVpositive populations (Breuer et al. 2011). As such, more research is needed on how best to integrate HIV and mental health services in this region. Importantly, there were no studies from Asia or Latin America, signifying the need for more research in these regions too.
Findings from the intervention studies provide some evidence on the effectiveness of integration activities in yielding positive patient outcomes, particularly on improvements in mental health, HIV symptoms, social well-being and substance misuse. However, differences between intervention and control groups were not statistically significant for some of these measures in a number of the studies, especially in regards to patient's improvement in quality-of-life and in one study, the annual cost-savings of health services (Weaver et al. 2009). It is nevertheless imperative to be cognizant of the diversity in integration approaches adopted and varying methodologies across the studies. Overall, the heterogeneity in integration activities, patient populations, study designs and analysis strategies make it difficult to draw any firm conclusions for policy, beyond the finding that integration, which a priori seems a sensible goal to pursue, has been shown to be associated with some improved outcomes in diverse settings. However, given the scope for publication bias noted above, the implementation should, where possible, be accompanied by rigorous evaluation methodologies. While it is highly beneficial to measure process outcomes to identify strategies in overcoming integration barriers and the contextual drivers for successful integration, evaluation should move beyond the mere measurement of process indicators to address more importantly, the short and longterm patient outcomes, which is fundamentally the primary aim of integration itself.

Conclusions
This review identified a diversity of integration models combining HIV and mental health services at the meso and micro levels, each with its respective advantages and disadvantages from the patient and providers' perspective. These provide insight into the principles that could underpin the development and implementation of integrated care models for HIV and mental health services. Firstly, single-site integration augments multidisciplinary coordination while reducing access barriers, but can be difficult to implement when a fuller continuum of specialized care involving multiple treatment modalities is needed particularly in low-resource settings. Secondly, multi-facility integration may comprehensively serve multi-morbid patients, but appropriate coordination and referral mechanisms are crucial to prevent fragmented care. Thirdly, active case management by non-clinicians offers considerable potential especially in low resource settings with shortages of mental health specialists, although appropriate training and support is essential. Finally, involving the patients not just as service users but also as active partners in improving integration within the treatment process, is a promising approach. While the current body of evidence on integration of HIV and mental health services from this review presents several benefits encompassing a myriad of positive patient and service delivery outcomes, the imperative for higher quality and robustly designed evaluative studies is evident, particularly in LMICs. As national planners and policy makers consider new ways of financing, implementing, managing and evaluating integrated care for HIV and mental health services, the evidence reviewed here can contribute to this process.

Funding
Funding was received from The Joint United Nations Programme on HIV/ AIDS (UNAIDS) grant number ADDEVH48, which covered for HL-Q's salary and payment for accessing papers.
Conflict of interest statement. None declared.