Implementation and Scale-Up of Integrated Depression Care in South Africa: An Observational Implementation Research Protocol

South Africa faces a substantial burden of multimorbid chronic diseases, including HIV, cardiovascular disease, diabetes, common mental disorders (CMDs) (e.g., depression, anxiety), and substance use disorders (1–4). Depression is of particular concern. Depression is the most prevalent disorder in South Africa (5), and people with chronic general medical diseases are more likely to experience depression than the general population (6, 7). Furthermore, depression comorbid with chronic general medical conditions compromises patient self-care and adherence to treatment regimens for general medical diseases (8–10), and this patient population experiences poorer health outcomes and increased mortality (11, 12).

In the context of a 75% treatment gap for CMDs in South Africa (13), the country’s national mental health policy emphasizes the integration of mental health care, including depression care, into primary health care (PHC) via task sharing (14). This focus is in line with health systems reforms toward horizontal integrated programming to respond to the burden of multimorbid diseases in South Africa (14).

PHC-based task sharing as part of pharmacological and psychosocial treatment models for CMDs has proven effective in low- and middle-income countries (LMICs) (15, 16). Furthermore, integrated collaborative stepped care that includes task sharing has been found to be efficient and effective for depression care. Most of this evidence comes from controlled trials in high- and middle-income countries (17–19). Few evidence-based models of collaborative stepped care pertain to integrated depression care in real-world, lower-resource contexts (20–22). Additionally, evidence is lacking on implementation strategies that enable uptake of integrated primary care for depression within routine primary care in LMICs.

A collaborative task-sharing model for integrated primary care for depression was developed through the Program for Improving Mental Health Care (PRIME) in South Africa (PRIME-SA) (23). The model’s efficacy and effectiveness were evaluated through a repeat cross-sectional survey in clinics and a comparison group cohort study, which had good outcomes at the facility and patient level (24), and through a pragmatic cluster randomized controlled trial using study-employed lay counselors (25), which found that the model did not produce inferior outcomes compared with care as usual when psychological treatment was provided by mental health specialists (26). However, uptake of the model—in particular, PHC providers’ identification and referral of chronic care patients with comorbid depression—was poor.

In accordance with the translational research continuum (27), this evidence-based model represents the initial Mental Health Integration (MhINT) model evaluated by this observational implementation science protocol to understand how to promote implementation and broader dissemination of the model. Funding obtained from the Centers for Disease Control and Prevention (U2GGH001197) was used to provide technical support to the KwaZulu-Natal Department of Health (KZN DoH) in the KwaZulu-Natal province of South Africa in order to scale up this model in one district, as a test site. The intervention components of the initial MhINT model involve the provision of psychoeducation, screening, and strengthened clinician assessment and diagnosis of depressive symptoms at the PHC facility level (see Figure A1 in the online supplement). The latter is achieved through the provision of enhanced mental health training in the use of Adult Primary Care (APC), known internationally as the Practical Approach to Care Kit (PACK) (28). APC is a clinical decision-making tool providing integrated evidence-based guidelines for the diagnosis and treatment of chronic conditions. It is intended for use by all health care practitioners working at the primary care level and has been rolled out nationally by the South African National Department of Health.

Referral pathways are also strengthened. Depending on severity of depressive symptoms, patients are referred to either nonspecialist clinic-based staff trained to provide psychosocial counseling using cognitive-behavioral techniques and problem solving, both shown to have an evidence base for delivery within nonspecialized health care settings (15); clinic-based PHC doctors with strengthened training to initiate psychotropic medication; or mental health specialists at the district level. To optimize scale-up of this task-sharing model by the KZN DoH, an initial set of implementation strategies was chosen on the basis of practical experience and process evaluation conducted during the formative and outcome evaluation of the PRIME-SA model (23, 24).

Table 1 describes the roles of various health systems actors in implementing this initial model, including the initial set of implementation strategies employed by MhINT. We have aligned these roles to the implementation strategies specified by the Expert Recommendations for Implementing Change (29). We describe the Southern African Research Consortium for Mental Health Integration (SMhINT) evaluation protocol. Our aim is to observe real-world implementation of the initial MhINT model described above in order to identify opportunities to refine and adapt this initial model and to optimize widespread implementation and dissemination of the model in South Africa.

Methods

Learning-Health-Systems Approach

We are guided by a learning-health-systems approach, where policy makers, researchers, service providers, and patients work as a collective to iteratively coproduce new knowledge and engage in shared decision making to strengthen the health system and health outcomes (30). Critically, learning health systems strive to identify interventions and implementation strategies that work in routine contexts: generating new evidence through research when required, solving practical problems of service delivery, and engaging in rigorous evaluation of program effectiveness to improve quality across the health system (30). This study builds on a sustained partnership between the investigators and the South African Departments of Health at the national, provincial, and district levels through the PRIME and MhINT programs, collectively amounting to 6 years of collaboration.

Study Design

We will use an onsite, iterative, observational implementation science design (27) structured around the four phases of intervention scale-up established by Barker et al. (31): an initial “setup” phase, during which the initial MhINT model is introduced into the system; an early assessment and refinement of the initial model into a “scalable unit” for inclusion in routine PHC services; an assessment of the scalable unit across a variety of contexts that would be encountered at scale; and a “going-to-full-scale” phase, during which a larger number of sites adopt the refined MhINT model. The initial setup phase, involving the introduction of the MhINT model into the study site district, was completed between 2016 and 2018. This protocol covers the assessment and refinement of the model into a scalable unit and the assessment of the scalable unit across contexts (renamed as stages 1 and 2, respectively). Barker et al.’s fourth phase is beyond the scope of this study.

We combine the RE-AIM (Reach, Effectiveness, Adoption, Implementation, Maintenance) framework (32) and the Consolidated Framework for Implementation Research (CFIR) (33) with mixed-methods data collection (34) (see Figure A2 in the online supplement). Within the context of our overall aim, the study objectives are described in Box 1.

RE-AIM outlines the critical elements of the population-level effects of health interventions (32). Definitions of the RE-AIM elements are provided in Table 2. Semistructured interviews targeting CFIR domains will be used to explore the implementation component of RE-AIM. CFIR delineates constructs associated with implementation success across several domains: intervention characteristics, such as goodness of fit; inner setting or health system characteristics that may aid or abet integration, including organizational culture; characteristics of the outer setting or external environment, such as policies and community needs; characteristics of individuals, including patient and provider needs; and implementation processes, such as harnessing of support from key stakeholders to improve uptake (33, 35). This study received ethical approval from the University of KwaZulu-Natal Biomedical Research Ethics Committee (reference: BF190/17).

Study Sites

The study site is the Amajuba district in the northwest of KwaZulu-Natal, where the MhINT model was implemented. Amajuba has three diverse subdistricts: Newcastle (urban), Dannhauser (semiurban), and eMadlangeni (rural) (Table 2). Assessment of the initial MhINT setup model (objectives 1.1 and 1.2) will be undertaken in 10 of the 14 PHC facilities in the urban Newcastle subdistrict. Assessment of the refined MhINT scalable unit model (objectives 2.1 and 2.2) is planned to take place across all three subdistricts, providing the opportunity to observe how the refined model and implementation strategies perform in the initial urban site (Newcastle) and whether adaptations are needed for semiurban and rural contexts. All PHC facilities are serviced by full-time PHC nurses, including professional nurses with a 4-year degree or diploma, enrolled nurses who have a 2-year diploma, sessional PHC doctors, and HIV counselors. Limited mental health specialists (N=3) are available at referral facilities.

Study Procedures

Objectives 1.1 and 2.1: RE-AIM assessment.

In order to assess the RE-AIM outcomes, a number of data sources, summarized in Table 3, will be used with accompanying subprotocols for the collection of these data. We provide more details on the subprotocols below.

Secondary data on project implementation, including project records and routinely collected data, will be gathered from all participating clinics from the start to the end of the cohort studies in stages 1 and 2. Routine service delivery data on screening and treatment rates will be extracted from the provincial District Health Information System. Data that are collected only by facilities (e.g., tracking delivery of some implementation strategies, including morning talks) will be extracted by field-workers. MhINT project monitoring data will include counseling intervention fidelity, measured by using fidelity checklists and records from supervisory visits; referral rates extracted from referral forms; facility and staff characteristics gleaned from facility profiles; patient counseling uptake assessed from patient tracking forms; and implementation of training packages and use of standardized operating procedures, extracted from training records and continuous quality improvement reports.

The cohort study will estimate how depression diagnoses made by nurses and referral for depression management affected participants’ depression outcomes, treatment adherence, stress, disability, income, and employment.

Enrollment of eligible participants for the cohorts will be done by trained research workers with the nine-item Patient Health Questionnaire (PHQ-9) and will occur before consultation with a nurse. Professional nurses will not know participants’ PHQ-9 scores; any diagnosis of depression by professional nurses will be based on APC guidelines and will be independent of enrollment procedures. Whether each participant has been diagnosed as having depression by a nurse is determined with an exit interview conducted after the participant’s consultation with a nurse. Participants will be divided into three groups: patients diagnosed and referred for care by the nurse, patients diagnosed and not referred, and patients not diagnosed. Data will be collected at baseline at patients’ first point of contact with the program and at 3-month and 9-month follow-ups to ascertain stability of effects.

A target sample of 550 participants is set for the cohort study (not diagnosed, N=200; diagnosed and not referred, N=150; diagnosed and referred, N=200). This sample is estimated to provide 86% power to detect a difference in mean PHQ-9 score of 2.0 points (5.3 versus 7.3) between the diagnosed and referred and undiagnosed groups, with 5% significance, assuming SD=5.1, intraclass correlation=0.1, and a 20% loss to follow-up (mean±SD=7.3±5.1 is from the 6-month follow-up in the PRIME trial). The sample of 200 participants in the diagnosed and referred group and 150 in the diagnosed but not referred group will provide 81% power to detect a difference in mean PHQ-9 score of 2.0 points (5.3 versus 7.3) between the groups, with 5% significance, also assuming SD=5.1, intraclass correlation=0.1, and a 20% loss to follow-up. The number of patients recruited in each clinic may vary, depending on the number consecutively enrolled, in order to meet the total sample sizes required. Inclusion criteria are age ≥18 years, having time and ability to complete the full interview, willingness to provide informed consent, and reporting a PHQ-9 score ≥9. Exclusion criteria are reporting a PHQ-9 score <9 and inability to provide informed consent (e.g., presence of severe intellectual disability, currently experiencing an acute medical issue, or lack of private space for the interview). Field-workers will be trained to make these assessments. Patients with severe depression and who have had thoughts of ending their life for 7 days or more in the past 2 weeks will be accompanied by project staff to a clinic’s professional nurses for care; project staff will be instructed not to leave the patient unattended until the patient has been seen by a professional nurse.

After providing written informed consent, participants will be interviewed in the local language (isiZulu) or English. Data will be collected by using digital handheld devices and uploaded onto a secure server at the University of KwaZulu-Natal. Data quality assurance will be managed through daily quality checks of uploaded data and in vivo observations of interviews to ensure fidelity of questionnaire administration. The research team will contact all participants enrolled at baseline for follow-up (irrespective of whether they refused or discontinued recommended clinical care) via telephone, home visits, etc.

Sociodemographic variables, including sex, age, language, ethnic group, educational and employment status, income, and household composition will be recorded at baseline. Questions about employment and income will be repeated at follow-up interviews after 3 and 9 months to track whether the intervention affects change in economic status. Depression symptoms will be measured with the PHQ-9, which is aligned with DSM-IV-TR diagnostic criteria for major depressive disorder (36). The PHQ-9 has been validated for diagnosis of depression, with a cutoff score of ≥9 in the chronic care population in South Africa (37). The PHQ-9 score will be the primary outcome of the cohort study. General disability will be measured with the 12-item World Health Organization Disability Assessment Scale, version 2.0, which has been used across cultures, including in South Africa (38). Treatment adherence to prescribed medications will be measured with the Visual Analog Scale (39), which has been used in South Africa (40) and other settings with limited resources and is a cost-effective alternative to measuring medication levels from biological samples (41, 42). The Perceived Stress Measure is a 10-item self-report measure that assesses the extent to which an individual experiences stress from situations in his or her life; the measure has been used previously in South Africa (43).

The primary analyses will compare mean PHQ-9 scores measured at 3 months and the proportion of participants whose PHQ-9 scores decreased by at least 50% from baseline to 3 months. Comparisons will be made among the diagnosed but not referred group, the diagnosed and referred group, and the undiagnosed group, adjusting for propensity scores using the t-effects package in Stata, version 16. The propensity scores will be the predicted probabilities of receiving a diagnosis of depression and being referred, respectively, at the baseline visit. These propensity scores will be drawn from logistic regression models, with baseline PHQ-9 scores, age, sex, and socioeconomic indicators as explanatory variables. These propensity scores will also be used to compare PHQ-9 scores and all other outcomes at 3 and 9 months. The comparisons between groups will account for intraclass correlation of outcomes between clinics by using robust adjustment. We will also use multilevel mixed models to analyze pooled panel data from baseline and 3 and 9 months, with group, time, and sociodemographic variables as covariates and with patient and clinic as random effects. Finally, linear and logistic regression analyses will investigate additional predictors, moderators, and mediating factors of the outcomes (44).

A costing analysis will be performed in stage 2 of the study for each subdistrict. We will estimate the incremental cost per outcome (change in PHQ-9 score, adherence, etc.), estimate the cost to scale up, and perform a budget impact analysis from the program/payer perspective (i.e., KZN DoH). We will estimate economic costs, including actual financial outlays and costs of donated time and volunteer time (see Table A1 in the online supplement).

We will estimate the incremental costs of implementing the intervention compared with the current standard for each subdistrict. Intervention costs will include costs associated with start-up, personnel, transport, communication, consumables, and overhead costs. Cost data will be collected from the study budget, clinic expense reports, published information on labor costs, and staff interviews. To assess staff time spent on the intervention, we will conduct time and motion studies in all subdistricts while the intervention is running at full capacity. Together, the microcosting data, time and motion studies, and clinical outcomes will be used to estimate the average cost of providing services.

The costing results will be reported separately for each subdistrict. For all key inputs and outputs, we will follow the standard guidelines of the Second Panel of Cost-Effectiveness in Health and Medicine (45). We will conduct sensitivity analyses around key cost inputs to account for the uncertainty in our results. For budget impact analysis, we will consider direct program costs in the different subdistricts. Direct medical costs will be measured to ensure that KZN DoH costs reflect the opportunity cost of the resources used in delivering services. Furthermore, the top-down approach of expense report collection will be compared with the bottom-up microcosting approach in order to triangulate and refine our cost estimates.

Objectives 1.2 and 2.2: CFIR interviews.

After 12 months of data collection assessing the RE-AIM outcomes, qualitative and quantitative interviews targeting CFIR domains will be performed to understand these RE-AIM outcomes in both stages 1 and 2. Structured quantitative interviews will comprise the Organizational Readiness for Implementing Change, a 12-item measure of change commitment and change efficacy based on Weiner’s theory of organizational readiness for change (46), and the Mental Illness: Clinicians’ Attitudes Scale, a 16-item mental health–related stigma scale (47). Interview guides will be translated into isiZulu and “back-translated” to ensure accuracy.

All facility managers and service providers from the facilities across the subdistricts will be requested to complete the quantitative interviews. Semistructured qualitative interviews will be conducted with facility-level managers, district managers, service providers, and patients (stage 1, N=78; stage 2, N=282) by trained interviewers (for a breakdown of sample sizes, see Table A2 in the online supplement). With the exception of the counselors (all of whom will be interviewed), participants will be sampled purposively according to the following strata: facility-level managers according to high- and low-performing clinics, individual professional nurses according to high and low diagnoses rates, and patients according to high and low uptake rates of counseling sessions. Interviews will be audiotaped and conducted in isiZulu or English, depending on participant preference and following informed consent procedures. When necessary, audiotaped interviews will be translated from isiZulu into English and transcribed, and back-translation checks will be performed by multilingual members of the research team. Any discrepancies will be reconsidered in consultation with a third multilingual researcher (48). Transcripts will be analyzed with the framework analysis typically used in health policy research. A thematic framework will first be developed by using the CFIR domains. Transcript content will then be coded by using the thematic framework while allowing inductive themes to emerge. Cases will be grouped by the aforementioned participant strata to identify multilevel barriers and enabling factors (49).

Objectives 1.3 and 2.3: participatory concept mapping.

Purposively selected key stakeholders will be engaged in a participatory concept mapping exercise to reflect on the results of objectives 1.1, 1.2, 2.1, and 2.2. Collaborative learning sessions—specifically, process mapping and nominal group techniques for developing priorities—will be used to brainstorm potential strategies to overcome identified barriers and optimize facilitating factors. Identified strategies will be sorted and rated according to perceived importance and feasibility for participants. Approximately 22 stakeholders, grouped according to their designation as managers, providers, or patients, will be involved in workshops at each stage (see Table A3 in the online supplement). We will include a national representative responsible for community mental health services, all members of the provincial mental health directorate, key district managers responsible for PHC and mental health, operational managers from all participating facilities, a range of high- and low-referring nurses purposively selected, and outreach team leaders and community health workers purposively selected on the basis of high and low number of referrals from the community level. Patients will also be purposively selected in order to provide a range of individuals with both high and low uptake of the counseling intervention. Informed by this process, the initial MhINT model will be refined in stage 1 to develop a MhINT scalable unit that will then be assessed across the diverse subdistricts in stage 2. Subgroup analysis for each site will use the methods outlined above for the RE-AIM framework and CFIR interviews. The same process will be repeated in stage 2 to further refine the model for broader scale-up.

Results

Given that strengthening health systems requires adaptability, collaborative mechanisms, and routine data gathering (50), the adoption of a learning-health-systems approach reflected in the two-stage design will facilitate the iterative examination of findings of this observational research design, in collaboration with patients, providers, health managers, and researchers. Shared decisions, identified in stage 1, on how to address bottlenecks to uptake of the initial setup MhINT model, will lead to the coproduction of novel and localized interventions and strategies for optimizing implementation and scale-up (i.e., the scalable unit), which will be tested in stage 2 across three diverse contexts (urban, semiurban, and rural). Assessment of the varying strengths and limitations of the refined model in these three contexts will allow us to identify distinct implementation strategies needed to strengthen integration in urban, semiurban, and rural contexts. The findings should help strengthen the MhINT model with interventions and implementation strategies to assist the South African government in scaling up to “full scale.” The findings will also build on previous research from other counties, which suggests a need to integrate mental health care into PHC with an approach that strengthens the whole health system (51). This study will also provide knowledge on strategies to optimize scale-up of collaborative care in integrated care for depression comorbid with chronic general medical conditions in other settings with limited resources.

Discussion and Next Steps

Global mental health stakeholders have increasingly recognized the need for systems change to enable the integration of mental health, inclusive of depression, into PHC (52); the primacy of open and flexible models that allow for cross-context adaptations (53); and the paucity of studies testing optimal implementation strategies to promote scale-up of integrated depression care in routine primary care in LMICs (22). Against this backdrop, and while the results of this study protocol remain to be seen, the design of this study may be helpful for other implementation research collaborations that aim to promote uptake of depression care integrated into routine primary care services. Notably, the iterative learning-health-systems approach is important for ensuring the uptake and sustainability of interventions and implementation strategies (54) (Box 2).

However, compared with controlled implementation research designs, this approach provides less control over the research process and requires greater investment of researchers’ time and resources in negotiations and capacity building activities (55). Furthermore, in light of a higher probability of sustaining mental health integration in the presence of systems reforms that enable chronic care (51), investment of time and resources into such systems strengthening initiatives may also be necessary to create this enabling environment.

Next steps for the researchers engaged in the SMhINT study include sustaining ongoing engagements with policy makers, planners, and managers; providing technical support for health systems strengthening to facilitate implementation; and building the capacity of policy makers and managers in implementation science to promote wider dissemination and sustainment of the intervention.