Collaborative care for the detection and management of depression among adults receiving antiretroviral therapy in South Africa: study protocol for the CobALT randomised controlled trial

The primary objective of the CobALT trial is to compare the effectiveness of a scalable intervention for collaborative care of depression by non-specialist mental health workers with usual services on the mental and physical outcomes in adults on lifelong ART with comorbid depression. The two primary hypotheses are that the intervention will (1) reduce depressive symptoms at 6 months and (2) improve viral load suppression status at 12 months. Secondary objectives are (1) to assess the effect of the intervention on the provision of integrated care for chronic illnesses in the intervention facilities from the perspective of service managers and providers as well as service users and (2) to evaluate the cost effectiveness of the intervention.

This and other administrative information recommended by the SPIRIT checklist [47], including the structure, function and composition of all the trial committees, are summarised in Additional files 1 and 2 of the web-based supplementary files (see Additional file 1 for details, Additional file 2 for the SPIRIT checklist and Fig. 1 for the SPIRIT figure). The trial is supported through grant 5R01MH100470-03 from the National Institutes of Mental Health. The funder had no role in the design of the trial and will not have any role in the conduct of the trial, analyses, interpretation or decision to submit the results for publication.

CobALT is a pragmatic, two-arm, parallel group, stratified and cluster randomised, superiority trial set in two health districts of the North West Province of South Africa. It is a type 2 hybrid design [48], evaluating an implementation strategy comprising clinical, educational and organisational elements (Table 3) in a coherent programme designed in collaboration with operational managers and health providers [42]. It combines characteristics of implementation (e.g. process evaluation and quality of care indicators) and clinical effectiveness research (e.g. primary health outcomes). We randomised primary healthcare clinics with a 1:1 allocation ratio, and will measure outcomes on individual patients. The trial is being implemented in two phases. The site for phase 1 is the Dr Kenneth Kaunda district (20 clinics, 1000 patients), where the intervention started in 10 clinics in April 2015 and patient recruitment commenced in April 2015. We are conducting phase 2 in the adjacent Bojanala district (20 clinics, 1000 patients), where the intervention started in 10 clinics in April 2016 and patient recruitment commenced in July 2016.

The two study districts are typical of rural South Africa, where the public health sector is largely responsible for providing care to an impoverished rural population. The Dr Kenneth Kaunda district, named after the first president of Zambia, has a population of 700,000–800,000 people, and Bojanala has approximately 1,500,000 people, most of whom are Setswana speaking [49]. The main economic activities are mining and agriculture, and the unemployment rate is one of the highest in South Africa at 31.5%, above the national average of 30%. The 2013 National Antenatal Sentinel HIV Prevalence Survey showed that 28.2% of women attending antenatal services in the North West province are HIV positive compared with a national average of 29.7% [50]. ART is provided at 200 nurse-led primary care clinics across the two districts, but resources for mental healthcare are very limited. With the exclusion of a single psychiatric hospital servicing the North West Province in the Dr Kenneth Kaunda district, there are only 1.5 psychiatrists, 8 psychologists and 6 community service psychologists to render services to over 2 million people resident in the two districts. Unlike some provinces of South Africa, there are no dedicated specialised mental health nurses, although general primary care nurses receive some basic training in mental health. There is a regular supply of basic psychotropic medication to primary care, but no depression counselling interventions are routinely available.

Inclusion criteria are age 18 years or older, receiving ART, depressive symptoms as indicated by a total score of nine or more on the Patient Health Questionnaire-9 (PHQ-9) [52], and valid consent. The PHQ-9 is a 9-item measure corresponding to the criteria upon which a diagnosis of depressive disorders is based in the Diagnostic and Statistical Manual of Mental Disorders (DSM-IV). Items are scored based on frequency of response ranging from 0 (‘not at all’) to 3 (‘nearly every day’) and summed assuming equal weighting to yield a follow-up score of between 0 and 27. A score of between 5 and 9 is associated with mild depression, between 10 and 14 moderate depression, between 15 and 19 moderately severe depression, and between 20 and 27 severe depression. It has been widely used for research and clinical purposes in primary care, including in LMICs [53‐55].

As part of the preparatory work for the trial, we validated a Setswana version of the PHQ-9 among 676 chronic care patients in the Dr Kenneth Kaunda District, comparing the performance of a localised version of the PHQ-9 administered by fieldworkers to that of the Structured Clinical Interview for DSM-IV administered by a clinical psychologist. This showed that it is a valid tool for measuring depressive symptoms in the trial population [56]. However, we showed that a threshold of 9, as opposed to the more widely used threshold of 10, was more appropriate for detecting depression in this population and so applied it to the trial eligibility criteria. The PHQ-9 area under the receiver operator curve (AUROC) was 0.85 (95% CI 0.82–0.88). The PHQ-9 AUROC for the sub-samples of patients with HIV or with hypertension were comparable (0.85 and 0.86, respectively). The original and localised versions are shown in Table 2. During the localisation and lay panel testing process the category labels were changed from phrases ‘not at all’ to numbers (0 days) to facilitate response by local Setswana speakers.

In keeping with the pragmatic orientation of this trial [57, 58], we intentionally have few exclusion criteria. We only excluded individuals if they planned to move away from the vicinity of the clinic in the next year and would therefore be unable to participate in follow-up data collection, or were otherwise judged unable to provide informed consent because of acute medical illness or psychosis. We did not exclude participants with active suicidal ideation, but urgently referred them for clinical review in both intervention and control arms.

The co-primary patient mental and physical outcomes are (1) PHQ-9 response at 6 months, defined as at least a 50% improvement in PHQ-9 score compared with baseline; and (2) viral load suppression 12 months after patient enrolment, defined as a viral load value of < 1000 copies/mL in line with the third ‘90-90-90’ target [1].

There are multiple ways of managing the participant level analysis metric for the PHQ-9, which we are using both to define eligibility in this trial (score of 9 or more) and to assess response to treatments. We have selected as the primary outcome measure the response at 6 months, defined as a 50% improvement from baseline score. Secondary outcome measures related to PHQ-9 include response at 12 months; mean scores at 6 and 12 months; remission, defined as a score of 5 or less at 12 months; and category of severity (mild, moderate, severe) at 6 and 12 months. We record the treatment for depression, including initiation or intensification of antidepressant medication, referral to a counsellor for depression counselling, number of contacts with a counsellor, and referral to a mental health specialist (clinical psychologist, psychiatrist or secondary care services). The Perceived Stress Scale is included to measure the degree to which life situations are considered stressful [59].

Secondary HIV measures are viral suppression at 12 months defined as a viral load of < 400 copies/mL, virologic failure defined as two consecutive viral load values > 1000 copies/mL, and change in viral load values over time. Other ART-linked outcomes include programme retention, appropriate maintenance on enrolment ART regimen and ART regimen switched to second line, and the Internalised AIDS Related Stigma Scale [60], which we include to explore the relationships between stigma, retention in care and viral load suppression.

Viral loads will be measured using venous samples in participants at baseline and at the 12-month follow-up interview, and the samples submitted to the National Health Laboratory Service responsible for routine viral load monitoring, which current ART programme guidelines recommend should occur annually [61]. Thus, the majority of patients are expected to undergo both routine programme and research viral load monitoring during the course of the trial. Funds available for viral load monitoring as part of the trial are limited and therefore only patients with no documented programme results (in the past 6 months) will be asked to submit a baseline research viral load. Research-funded 12-month viral loads will be collected on all patients as this is the primary physical outcome.

We will also aim to collate all programme and research-viral loads into a single longitudinal record for every patient enrolled in the trial, through linkage with the viral load repository at the National Health Laboratory Service and application of algorithms for this linkage previously applied to province-wide ART datasets [27]. In South Africa, detectable viremia is most commonly caused by non-adherence, and only rarely due to resistant virus [62]. Thus, the proportion of participants with a suppressed viral load provides a reasonable estimate of ART adherence. In addition, we have included a Visual Analogue Scale for self-reported adherence in the last month as a secondary outcome measure at 12 months [63‐65]. The integrated care nature of the intervention, with a focus on the detection and management of chronic diseases, and the surge in multi-morbidity in the general population, and ART populations in particular [66], are the reasons for monitoring risk factors for cardiovascular diseases (blood pressure, weight, smoking status) and the detection and treatment of other chronic diseases.

We will assess the provision of integrated care from a patient perspective using the Patient Assessment of Care for Chronic Conditions [67] at 12 months. Economic outcomes include healthcare utilisation, productivity and the World Health Organization’s Disability Assessment Scale [68]. We will monitor potential harms, including hospitalisation, death and suicide, the last two through follow-up at clinics, linkage with the national mortality register, and interviews with clinic staff and family members. Table 4 provides a full list of outcomes, time of measurement and source of data. In the section on ‘Adverse events and monitoring of possible harms’ and in Table 5, we describe in more detail outcomes reflecting possible harms and methods used for their detection.

We shall assess the provision of integrated care from a health system perspective using the Assessment of Chronic Illness Care, which covers six areas of system change to promote optimal chronic care (linkages to community resources, self-management support, decision support, delivery system design, clinical information systems and organisation of the health system) [69]. We will complete the Assessment of Chronic Illness Care on a sample of clinic staff from intervention and control clinics at 12 months. In addition, we shall conduct qualitative process evaluation interviews with one facility manager and one service provider from each of the intervention clinics at 12 months to understand their perspectives on how integrating mental health services for depression has affected overall HIV clinical care.

We will conduct clinic recruitment in two phases. We deem this phased approach necessary because of the scope of the intervention support and fieldwork operation required for the trial, both of which place considerable demands on the trial co-ordinating centre operating with limited human resources and a limited budget. The trial will start in the Dr Kenneth Kaunda district because the intervention has been piloted in that district and because the comparison intervention, the Integrated Care Services Model, has been implemented in the district before Bojanala because of its inclusion as one of the NHI pilot districts. We will conduct clinic randomisation on two separate occasions to allow for changes in clinic operation (smaller clinics are closed or relocated on a regular basis in South Africa) and the volume of attendances. We will factor in an embedding period of 2 months between the start of the intervention and recruitment of patient participants. In determining the duration of the embedding period, we balanced the time required to ensure the intervention was functioning within the clinics with the likelihood that patient participants had not yet been exposed to the intervention by the time of enrolment. Figure 3 shows the recruitment of clinics and patient participants into the trial as well as follow-up evaluations.

We calculated the sample size for a superiority trial because we hypothesized that the intervention would lead to reduced depressive symptoms for patients on ART with comorbid depression, which would in turn positively influence their adherence to ART and care, and improve their HIV outcomes. We based our estimates of viral load suppression on data from our previous RCT on task-shifting HIV care in which 70% of ART patients had suppressed viral loads at follow-up with an intra-clinic correlation coefficient (ICC) of 0.046 [40]. We assumed that rates of viral load suppression will be slightly lower in a group of patients suffering from depression (65%) as it is a known risk factor for poor adherence [18, 70]. A sample size of 2000 participants from 40 clinics (50 patients per clinic, 1000 per group) provides 80% power to detect a 10% difference in 1-year viral load suppression at the 5% significance level assuming an ICC of 0.04 and a 20% loss to follow-up based on a previous trial in this setting [38]. To determine the appropriate parameters for the PHQ-9 sample size calculation we searched PubMed for all RCTs which had used the PHQ-9 as an outcome. Of the 126 trials screened, we identified nine as relevant to the design of our trial [71‐79], including four cluster trials and five individually randomised studies. The trials used a mix of metrics, including difference in means, response and remission rates, measured at a range of time points. We disregarded several of the trials because their baseline scores were far higher than those we had observed in our population of interest during the validation of the Setswana localisation of the PHQ-9. We instead based our sample size estimates on three cluster trials completed in the Netherlands [71], Italy [77] and the UK [79], where PHQ-9 response rates at 6 months were reported and where baseline scores were similar to what we anticipated. Only two of the five cluster trials reported an ICC [71, 79] and, in one of these, it was uninterpretable because of very small cluster sizes [71]. We then completed a series of calculations to determine power to detect a difference of between 10% and 12% in response rates at 6 months, with the number of clusters and cluster size fixed by the viral load calculation at 40 and 50, respectively (Table 6). We allowed for the proportion of those who responded in the control group to vary between 30% and 35%, for the ICC to vary between 0.02 and 0.04 (we assumed lower ICCs were likely given that the PHQ-9 is a health outcome) and for a loss to follow-up rate of 20%. This showed that a sample of 50 patients in 40 clusters (1000 per group) yielded a power of between 73% and 96% to detect a difference of between 10% and 12% in response rates at 6 months.

We conducted clinic recruitment in two phases. We obtained an updated list of clinics in each district with recent headcount data. We scrutinized the GPS co-ordinates of the clinics to identify any clinics that might be located next to each other and thus be at risk of contamination. We then forwarded a cleaned-up list of eligible clinics to the statistician, who selected the largest facilities and completed randomisation within sub-district strata. Because clinics in South Africa sometimes close or are amalgamated with others, we repeated the process for the Bojanala clinics a few months before starting work in that district. We split clinic recruitment evenly between the two districts with 20 clinics from each.

Patient recruitment occurred independently of clinical care or participation in the group psychosocial intervention for depression. We adopted a pragmatic approach, enrolling all patients who met the eligibility criteria regardless of exposure to the intervention, recognising that the detection of mental disorders and linkages to care are pressing challenges for integration of mental health services into primary care and that failure to account for this in the design of the trial would limit the generalisability of its findings to real-world primary care settings. Patient participant recruitment started approximately 2 months after the training of nurses and counsellors in the intervention clinics. Trained fieldworkers invited, screened and enrolled patient participants in the trial using a three-stage process as outlined in Additional file 4. First, trained fieldwork recruiters explained the study to patients in the waiting room, inviting those attending for chronic care services to come forward for more information if they were interested. Depression, HIV or ART were not be mentioned during this general information sharing so as not to single out those who wished to participate. The recruiter completed a brief verbal screen of patients who came forward to establish whether or not they were on ART and whether they planned to stay in the study area for the duration of the following year; this was done on a one-to-one basis. Eligible patients were then walked across to a fieldwork interviewer based in a private section of the clinic where the emotional content of the PHQ-9 instrument was explained and opt out consent sought before proceeding with the full screen. We then sought formal consent from patients who screened positive using the PHQ-9. This covered permission to be interviewed at enrolment, at the two subsequent time points, 6 and 12 months later, determination of the participant’s viral load at enrolment and 12 months later, permission to review their clinical folders and hospital records to assess quality of care provided and healthcare utilisation, and to send reminders and confirmation of follow-up interviews by text message and phone calls. We requested participants to supply their national identity numbers to permit linkage with the national population register, which records more than 90% of all deaths in South Africa. We only collected patient identifiers once written informed consent was provided (Additional file 5). The recruitment process for the PRIME and CobALT trials was conducted as a combined process in the Dr Kenneth Kaunda district and patient recruitment continued until the target of 50 patients per clinic was met for both trials. We invited patients who were on both ART and antihypertensive medications, and who qualified for enrolment in the trial, to participate in both studies and complete two sets of consent forms. Fieldwork interviewers conduct interviews in one of the three local languages (Setswana, English, Afrikaans), chosen by the participant (see Additional file 6 for the full questionnaire). Fieldwork interviewers collect data using handheld electronic devices, which we have previously used in similar field work data collection exercises in South Africa [80]. We ask participants to return to the clinic for follow-up interviews. We send text reminders in advance of follow-up visits, and contact participants by telephone to confirm their appointments. At follow-up, participants undergo an interview to determine the severity of their depressive symptoms, the nature of the care they received at the clinic, including care for depression and HIV, and other healthcare utilisation and economic outcomes. Follow-up viral loads will be drawn from all patients at 12 months.

We randomised clinics and their patients to one of two parallel groups, with an allocation ratio of 1:1. Within each district, we considered the distribution of clinics within sub-districts to ensure that we did not introduce potential confounding through geographically determined management of clinics. Within the Dr Kenneth Kaunda district, 12 clinics fall within the Matlosana sub-district in the area, including and surrounding the major town of Klerksdorp, and have been randomised as one stratum. The remaining eight clinics are situated in more rural areas and randomised together. Within the Bojanala district, 10 clinics are in the central district of Rustenburg in and around the town of the same name and are randomised in a 1:1 allocation. The remaining 10 clinics are distributed over a far wider area than in Dr Kenneth Kaunda district and spread across the three sub-districts of Moretele, Kgetleng River and Madibeng. We combined two of the smaller sub-districts (Moretele and Kgetleng River) into one, so that we have two strata each with five clinics available for randomisation. We randomly selected one of these strata to have three intervention clinics to two control clinics, and vice versa in the other stratum, so that we have 10 intervention clinics and 10 control clinics in the Bojanala district. The trial statistician carried out randomisation using n-Query advisor prior to the intervention or screening of participants for recruitment. Blinding of participants and clinicians is not possible because it will be clear which clinics will be implementing what interventions. The trial statistician who carried out randomisation did not know the characteristics of the clinics being randomised, and the primary statistical analysis will also be blinded to allocation status.

Interview data is uploaded as soon as questionnaires are completed, internet connectivity permitting, and stored in a secure SQL server at the Trial Co-ordinating Centre at the University of Cape Town Lung Institute. We integrate data from interviews with those from other sources to create a longitudinal record for every patient participant (Fig. 5). These other sources include blood results for the viral load samples drawn as part of the research process as well as routinely monitored viral loads which are collated in a central national repository. We search hospitalisation records quarterly and integrate linkages into the research dataset. We maintain a separate database of patients referred to facility-based counsellors and link their attendance of sessions with the trial patient database. The research version of the national population register, curated by the South African Medical Research Council, will be linked once at the end of the study to identify any deaths which had not already been identified through follow-up of patients who did not return for their 6- and 12-month interviews.

Quality control measures include supervision of fieldworkers, electronic alert messages for fieldworkers if unusually high or low values are entered into the electronic questionnaire, automated skips, and monitoring of the data to identify unusual values or trends. The handheld devices used for data collection are password protected and interview data is uploaded at the end of each interview without being stored on the device. Only a small number of data managers have access to personal identifiers. Anonymised data (without names or national identity numbers) will only be provided to selected members of the research team. Members of the research team, including all fieldworkers and fieldwork supervisors, are required to sign a confidentiality agreement to ensure the protection of confidential data.

We will estimate the effects of the intervention by comparing the outcomes of intervention and control group patients using multiple mixed effects regression models, with clinics as a random effect to account for the intra-cluster correlation of outcomes. We will include the stratum as a covariate in the regression models, and will use Stata statistical software for the analysis. All clinics and patients will be analysed in the treatment group to which they were randomly assigned. We will ascertain the completeness of follow-up by group to evaluate the integrity of the planned analysis. We will use binomial regression to estimate differences in proportions of patients with suppressed viral loads and who have responded in terms of their depressive symptoms, measured using the PHQ-9. We will analyse secondary outcomes as follows: binomial regression for binary outcomes, linear regression for changes in depression scores, blood pressure, weight, and waist circumference by comparing values at follow-up adjusted for baseline values, and Poisson regression for count outcomes such as clinic visits and inpatient days, accounting for individuals’ duration of follow-up.

Planned secondary analyses include analysis of primary outcomes stratified and adjusted for treatment severity, treatment status and comorbidities at baseline. We will also explore whether there is a differential effect among those patients who reported receiving the counselling intervention.

The economic evaluation will include descriptive statistics describing the health system and societal costs of living with HIV and depression, a cost effectiveness analysis, and modelling of the intervention beyond the timeframe of the study.

These are enumerated and defined time frames for monitoring and reporting as specified in Table 5. Most guidance would recognise hospitalisation, or prolongation of hospitalisation, and death as serious adverse events, requiring prompt reporting to regulatory bodies. However, both of these events are expected to occur frequently among adults on ART. Based on follow-up from our previous trial in a similar ART population, we expect a mortality rate as high as 5% during the 12-month follow-up period, and a hospitalisation rate of approximately 10% [40]. We expect much of this mortality and morbidity to be due to the underlying HIV infection, advanced enough to warrant treatment with ART prior to the introduction of universal treatment. Systematic monitoring of hospitalisations and deaths is planned, but will be episodic, occurring at the 6- and 12-month follow-up participant interviews, and with quarterly linkage of hospitalisation and once-off linkage with the national population register data. Given that these events, while serious, are common in this population, are likely to be related to the underlying HIV disease progression, and that monitoring is episodic, we have proposed that the frequency of these events be reviewed 6 monthly by the Data and Safety Monitoring Board (DSMB), and then reported to the relevant Institutional Review Boards (IRBs) (University of Cape Town, University of Kwazulu-Natal and King’s College London), with the DSMB Letter of Recommendation. Because suicide may well be related to a mental health intervention, and is likely to be rare, we report any such events within 7 days of the knowledge of the event in accordance with the recommended Standard Operating Procedures of the University of Cape Town’s IRB. All deaths will be followed up with clinic staff and family members to determine whether or not they were the result of suicide.

A further serious adverse event that is unlikely to be related to the intervention, but which may occur as a result of participation in the data collection component of the trial, is a markedly elevated viral load in the context of pregnancy, placing the foetus at risk of transmission. For this reason, it will be managed as a serious adverse event and reported to the IRB and DSMB within 7 days. We follow-up all viral loads in excess of 1000 copies/mL to see whether or not they appear to be reported during pregnancy and we urgently pursue all missing viral loads in women who report themselves to be pregnant. Once we confirm a high viral load in a pregnant woman, we immediately notify the clinic staff and advise them to recall the women urgently for clinical care and intervention to limit transmission of HIV to the baby.

We report non-serious adverse events to the DSMB 6 monthly. We ask the participants who respond positively to the ninth item of the PHQ-9 “Thoughts that you would be better off dead or of hurting yourself in some way” to confirm their answer and, if they indicate having such thoughts between 1 and 7 days in the last 2 weeks, we provide them with written educational material on suicide prevention. If they confirm having such thoughts for 8 days or more in the past 2 weeks, we immediately refer them to clinic staff for review. This protocol applied to the pre-screening process even for patients who responded positively to the ninth item of the PHQ-9 but were otherwise not eligible for enrolment in the trial. Likewise, we immediately refer participants with a blood pressure of ≥ 180/110 to clinic staff for review of their blood pressure. Once data collection is complete, we will present a summary of adverse events to clinics together with recommendations for further treatment. These adverse events are a PHQ-9 score of ≥ 20, suggestive of severe depression, raised blood pressure at follow-up representing undiagnosed or uncontrolled hypertension, and detectable viral load at follow-up representing possible adherence problems or treatment failure.

We will use the adapted Service Utilisation Questionnaire to systematically collect resource-use data, including any inpatient care, consultations with health providers, use of medications and laboratory tests, and time and travel costs associated with service uptake. We will collect information on the financing sources for each of the service categories to allow for an estimation of the extent of private, out-of-pocket expenditures incurred by recruited participants and their families. The unit costs or prices of these various resource inputs will be taken from previously conducted costing studies in South Africa [81] and from WHO-CHOICE data [82]. We will obtain medication costs from the Primary Care Guideline App published by the National Department of Health in South Africa [83]. Time loss from work may be affected by the interventions and we will measure this at baseline and follow-up and will value it using average wage rates. We will add the above costs to those of the actual intervention, which will be determined using an ‘Active Ingredients’ approach and allocated to trial participants using responses derived from the Service Utilization Questionnaire and linkage to related databases (e.g. counselling records). We will derive these from activity data (session provided per patient) and therapy unit costs (based on salaries, overheads and working hours). We will then make cost comparisons using bootstrapping methods to account for likely skewness in the data distributions.

To examine issues of cost-effectiveness, we will first relate estimated service costs per case in the different arms of the trial to primary study outcomes (viral loads and PHQ-9 scores). Based on earlier studies of stepped care for depression in India [84], a decrease in overall resource use can be expected due to a drop in ineffective but financially burdensome visits to providers unqualified to offer appropriate depression care. Since health outcomes for the stepped care intervention are also expected to improve significantly, the intervention will ‘dominate’ usual care (i.e. better outcomes, less cost); such a hypothesis negates the need for a power calculation. If, however, costs are found to be higher in the intervention group, bootstrapped incremental cost-effectiveness ratios for PHQ-9 depression and viral load scores will be derived. Irrespective of whether point estimates demonstrate dominance, results will be plotted on a cost-effectiveness plane and presented as cost-effectiveness acceptability curves in order to show the probability of the intervention being cost-effective at a range of ‘willingness-to-pay’ threshold levels. We will conduct sensitivity analysis to take account of uncertainty and imprecision in the measurements, including multiple imputation models for missing values.

As a second stage, we will move beyond the time and space constraints of the trial and consider longer-term costs and effects of the trial intervention at the population level. For this, we will construct and populate a quantitative economic model of expected lifetime costs and health gains, based on the earlier Markov modelling studies of Bachmann [85] and Cleary [81], but adapted to better capture other health states and outcomes of interest (depression and other NCDs, in particular cardiovascular disease and diabetes). We will convert health improvements observed in the trial into years of healthy life gained at the population level in the newly developed mental, neurological and substance use module of the inter-UN OneHealth strategic planning and costing tool [86], specifically by multiplying observed effect size for the intervention by the target rate of treated prevalence in the population. We will similarly compute population-level costs in OneHealth by multiplying the resource use and cost per case by the number of cases treated in the population at target coverage levels.

Given the complex nature of the intervention, which involves changing clinical flow and diversifying clinical functions, we have planned an intense process evaluation that is informed by the Medical Research Council framework for process evaluation of complex interventions [87]. We will use mixed methods using a combination of quantitative process variables collected across the clusters as well as in-depth qualitative process evaluation interviews.

In order to understand factors that may influence variations in the main health outcome findings between the arms and across the clusters, we will assess both possible contextual factors as well as factors associated with the fidelity and quality of implementation of the intervention. We will gather robust indicators on fidelity and quality of implementation of the intervention across the clusters through the collection of process indicators on the following:

To clarify causal mechanisms and identify contextual factors associated with variation in outcomes of the trial in the implementation and control arms as well as across the clusters, we will collect the following data:

We obtained permission to enrol and randomise the clinics in the trial from the relevant district and provincial health department. We ask patients to provide consent for data collection and not randomisation as the clinic to which they are assigned is determined by where they live. We use a clearly defined and phased recruitment process to enrol patients in clinics (Additional file 4) to minimise the chance of inadvertent identification of patient participants as being HIV positive within clinic waiting rooms, and to manage the risk of discomfort posed by the more sensitive content of the PHQ-9 questionnaire. Fieldworkers will explain the trial information sheets in detail to prospective patients (Additional file 5). We specifically require patients to declare that they understand the consent to be voluntary and that it can be withdrawn at any time, to agree to the use of their personal information for study purposes, to undergo three interviews (baseline, 6 and 12 months), to allow research staff to access their medical records, and to permit future analysis of their de-identified data provided such analyses are reviewed and approved by a research ethics committee. Patient level data will be made available from the corresponding author at lara.fairall@uct.ac.za at the time of publication of the main trial results. Consent for data sharing will not be explicitly specified in the consent process but the presented data will be anonymised and risk of identification is low.

Ethical approval for the trial was obtained from the University of Cape Town Human Research Ethics Committee (reference number 211/2013), King’s College London Research Ethics Office (reference number PNM/12/13-159), the University of Kwazulu-Natal Biomedical Research Ethics Committee (reference 211/2013), and the North West Provincial Department of Health.

We will first submit the results of the trial for peer-review publication within 1 year of the last patient being followed-up. We will acknowledge substantive contributions to the design, conduct, analysis and interpretation of the trial through co-authorship and criteria for contributorship specified by the International Committee of Medical Journal Editors will be applied [90]. We have planned separate publications to document the intervention, and the economic and process evaluations. We will feed back the results to the district and provincial health departments through a series of local workshops at district and provincial level. They will be disseminated and debated with the South African National Department of Health through the Project Steering Committee (for composition see Additional file 1). Policy briefs and a video summary will be used to disseminate results to a wider audience.

Follow-up interviews are complete in the Dr Kenneth Kaunda district and are in progress in the Bojanala district.