Atmiyata, a community-led intervention to address common mental disorders: Study protocol for a stepped wedge cluster randomized controlled trial in rural Gujarat, India

Background While lay-health worker models for mental health care have proven to be effective in controlled trials, there is limited evidence on the effectiveness and scalability of these models in rural communities in low- and middle-income countries (LMICs). Atmiyata is a rural community-led intervention using local community volunteers, called Champions, to identify and provide a package of community-based interventions for mental health, including evidence-based counseling for persons with common mental disorders (CMD). Methods The impact of the Atmiyata intervention is evaluated through a stepped wedge cluster randomized controlled trial (SW-CRCT) with a nested economic evaluation. The trial is implemented across 10 sub-blocks (645 villages) in Mehsana district in the state of Gujarat, with a catchment area of 1.52 million rural adults. There are 56 primary health centers (PHCs) in Mehsana district and villages covered under these PHCs are equally divided into four groups of clusters of 14 PHCs each. The intervention is rolled out in a staggered manner in these groups of villages at an interval of 5 months. The primary outcome is symptomatic improvement measured through the GHQ-12 at a 3-month follow-up. Secondary outcomes include: quality of life using the EURO-QoL (EQ- 5D), symptom improvement measured by the Self-Reporting Questionnaire-20 (SRQ-20), functioning using the World Health Organization’s Disability Assessment Scale (WHO-DAS-12), depression symptoms using the Patient Health Questionnaire (PHQ-9), anxiety symptoms using Generalized Anxiety Disorder Questionnaire (GAD-7), and social participation using the Social Participation Scale (SPS). Generalized linear mixed effects model is employed for binary outcomes and linear mixed effects model for continuous outcomes. A Return on Investment (ROI) analysis of the intervention will be conducted to understand whether the intervention generates any return on financial investments made into the project. Discussion Stepped wedge designs are increasingly used a design to evaluate the real-life effectiveness of interventions. To the best of our knowledge, this is the first SW-CRCT in a low- and middle-income country evaluating the impact of the implementation of a community mental health intervention. The results of this study will contribute to the evidence on scaling-up lay health worker models for mental health interventions and contribute to the SW-CRCT literature in low- and middle-income countries. Trial registration The trial is registered prospectively with the Clinical Trial Registry in India and the Clinical Trial Registry number- CTRI/2017/03/008139. URL http://ctri.nic.in/Clinicaltrials/regtrial.php?modid=1&compid=19&EncHid=70845.17209. Date of registration- 20/03/2017.

3 knowledge, this is the first SW-CRCT in a LMIC evaluating the impact of implementation of a psychosocial mental health intervention. The results of this study will contribute to the evidence on scaling-up lay health worker models for mental health interventions and contribute to the SW-CRCT literature in LMICs.

Background
Mental illness is a substantial public health burden in India, affecting 10.6% of the population (1). There is a shortage of mental health professionals to address the mental health needs of the population, particularly in rural areas (1). This is further compounded by high level of public stigma towards mental illness and the lack of accessible mental health care; resulting in a treatment gap of nearly 80-90% for mental illness in India (1).
In most low-and middle-income countries (LMICs), service delivery models have focused on task-sharing, the process of sharing tasks of mental health care with less specialised health workers, such as community health workers (2). Several programs have been developed to build the capacity of community members and primary care health workers with the aim of increasing their uptake of mental health tasks and enable access to mental health supports in rural areas (3,4).A number of studies in India and other parts of South Asia have shown the efficacy of task-sharing initiatives (5)(6)(7)(8).
A possible reason for the inability to scale-up task-sharing models may be that public health systems in low-and-middle income countries (LMICs) are overburdened with addressing other health needs, and there is limited time and energy to devote to mental health (9)(10)(11)(12). Task sharing approaches may, therefore, need to build community capacity to provide mental health care that complements service provision efforts within the public health system. Atmiyata is a community-led intervention (13) using non-specialised community volunteers for identification, support and referral for persons with common and severe mental disorders. Atmiyata was previously piloted in 41 villages in Nashik district 4 of state of Maharashtra from 2013-2015 (13).
As a mental health program, Atmiyata aims to: (i) reduce the treatment gap for common and severe mental disorders; (ii) improve mental health outcomes for people with common mental disorders; (iii) improve quality of life among people with mental health problems; (iv) improve access to social welfare schemes for people with mental health problems. We hypothesize that the intervention will result in symptomatic improvement in depression and anxiety and well-being as well as narrow the mental health 'care' gap (14).Evaluating both the efficiency as well as the implementation process of this intervention will generate valuable lesions as to what and how we might sustain the intervention's impact when delivered to a large population.

Design
We employ a Stepped Wedge Cluster Randomized Controlled Trial (SW-CRCT), with a nested health economic evaluation to assess the impact on persons with Common Mental Disorders (CMD) and return on investment of the intervention in the Indian state of Gujarat.
The stepped wedge design is chosen for evaluating the scale-up of Atmiyata intervention as it allows for random allocation of the time at which clusters receive an intervention (15), and all clusters receive the intervention before the trial ends which is ethically appropriate. The design also allows for Atmiyata intervention to be delivered in a staggered manner to account for practical and logistic constraints. Logistically, it is not feasible to start delivering the intervention in all clusters (villages served under groups of PHCs) simultaneously. A stepwise implementation allows the implementation team enough preparation time and is an efficient use of implementation team resources.
Additionally, the staggered implementation of the intervention over time periods allows for more in-depth statistical analysis than a simple pre-post, parallel arm cluster randomized controlled trial design.
There are 56 primary health centers (PHCs) in Mehsana district (where the study takes place), and each PHC serves discrete villages within a geographical area. Each village in the geographical area served by a PHC is a cluster in this study. We created 4 groups of Step' is the order in which a group of clusters switches from control to intervention condition. On the other hand, 'Period' is defined as group of observations by time of measurement. The duration of each period is 5 months to accommodate for baseline and 3 months follow-up data collection [ Figure 1].
This study uses a repeated cross-sectional design with outcome data derived from different participants in each period. All four groups start at baseline in the control condition and are exposed to the intervention at regular time period of five months [ Figure 1].

Setting
Mehsana district, the study site, located in Western India in the state of Gujarat is primarily a rural district (75% rural), with a rural population of 1.52 million people, of which approximately one million are above 18 years of age. The district is divided into 10 blocks/ sub districts with a total of 645 villages and 316,536 rural households (16). Almost half (45.4%) of Mehsana's rural population has a low standard of living as per the Standard of Living Index. Most residents (53%) are employed in the agricultural sector.
The rural population of Mehsana district is economically disadvantaged as agriculture is 6 not always a viable occupation given uncertain climatic events (16). In terms of health services, Mehsana has 56 PHCs, 11 Community Health Centers and 1 District Hospital staffed by 2 psychiatrists, along with District Mental Health Program (DMHP) which provides additional human resources (such as a psychologist and social worker) for mental health at the community level. Mental health care is primarily delivered by psychiatrists at the District Hospital and the psychiatrists also visit Community Health Centers on a fortnightly basis in rotation, as part of DMHP. The district hospital has in-patient and outpatient services for persons with mental illness, and limited psychosocial support services.

Participants
The study sample consists of adult community members with common mental disorders

Primary outcome
The primary outcome is symptomatic improvement of depression and anxiety as measured using a validated Gujarati version of the General Health Questionnaire-12 (GHQ-12) (17) from baseline to 3-month follow-up with an 8-month follow-up to evaluate sustained effects of the intervention. The GHQ is a widely used screening tool with reliable sensitivity for assessing CMD (18). GHQ-12 is a dichotomous 12-item questionnaire with each item rated on a 4-point scale, with possible responses being "less than usual," "no more than usual," "rather more than usual," or "much more than usual." We used a 7 bimodal scoring method, whereby "less than usual" and "no more than usual" is scored as 0 point, and "rather more than usual" and "much more than usual" is scored as1 (17).GHQ-12 scores will be analyzed as both continuous (ranging from 0 to 12) and categorical outcomes (case defined as 3 and above score on GHQ scale; non-case as less than 3 score on GHQ scale).

Secondary outcomes
Secondary outcome measures are assessed at 3 months, and 8 months after the start of the intervention [ Table 1].
Quality of life (QOL): Improvement in quality of life of persons with CMDs is assessed using the validated Gujarati version of EURO Quality of life 5D (EQ-5D) (19).The EQ-5D's descriptive system is a preference-based Health Related Quality of Life measure with one question for each of the five dimensions that include mobility, self-care, usual activities, pain/discomfort, and anxiety/depression measured at 5 levels: no problems, slight problems, moderate problems, severe problems, and extreme problems. Lower score indicates better quality of life (19).
Psychiatric symptoms: Improvement in psychiatric symptoms is assessed using a validated Gujarati version of Self Reporting Questionnaire (SRQ). SRQ is a scale developed by the World Health Organization to screen for psychiatric disturbances for low-and middleincome countries consisting of 20 questions which are scored 1= yes and 0=no indicating presence or absence of a particular symptom over the past month. SRQ is a continuous scale; responses are calculated as total score ranging from 0 to 20 with lower scores indicating recovery of symptoms (20).
Disability: Reduction of disability and reduction in number of days unable to work and improvement in productivity is assessed using validated Gujarati version of WHO-DAS-12.
WHODAS-12 is useful for brief assessment of overall functioning as it assesses difficulties 8 due to health conditions. The scale uses12 items, 5-point rating scale ranging from none, mild, moderate to severe and extreme. Responses are calculated as a total score ranging from to 12 to 60 (21).
Depression and anxiety symptoms: Improvements in depression symptoms is assessed using a validated Gujarati version of Patient Health Questionnaire (PHQ-9). PHQ-9 total scores range from 0 to 27greater score indicating greater symptoms (22). Improvements in anxiety scores is assessed by using a validated Gujarati version of Generalized Anxiety Disorder (GAD-7) with total scores ranging from 0 -21 (23).

Economic Evaluation
The most common measure of efficiency in the health sector is cost-effectiveness analysis (CEA) which measures only health related benefits and expresses these in natural unit such as lives saved, or symptoms reduced. However, return of investment (ROI) analysis expresses all the benefits in monetary terms. Expressing both the costs and the full range of benefits of an intervention in the same units (money) has the distinct advantage of making investment decisions very straightforward (26). If the money value of benefits of an intervention is larger than the cost of the intervention, it may be regarded as a sound investment. Hence, we have chosen to do a Return of Investment (ROI) analysis for the Atmiyata intervention. We will follow the reporting guideline of the ROI in global mental health innovation. (26).
Costs are calculated using both government and societal perspective (26 Cost data is collected at baseline, 3 months and 8 months from all study participants.
Time spent by the champion will be obtained from the programme implementation data.
Minimum wage rate of Gujarat will be used to value their time. Total hours spent for the programme will be multiplied by hourly wage (obtained from minimum wage rate) to get the time cost of the Champions. Benefits are considered in terms of improved health, functioning, participation, productivity, increased saving and investment, reduced informal care giving and health and welfare services.

Intervention condition
The Atmiyata intervention has been described extensively elsewhere (13). Briefly, Atmiyata is a complex psychosocial intervention involving two-tiers of community volunteers for identification and support to people in distress and with symptoms of

Adverse events
We considered adverse events as attempted suicide, self-harm or death by suicide. The EUC also has provision for providing active support to participants in crisis. A crisis is 14 defined as the participant revealing a recent self-harm attempt or expressing thoughts of self-harm during data collection. Such participants are encouraged to seek help immediately and the data collectors seek participant consent to inform their family member or a friend about the crisis and thus, mobilize social support to deal with the crisis.

Sample size and power calculations
A trained lay health worker led intervention study conducted in India reported a risk difference of 12% at follow up between intervention and control condition for recovery of CMD patients (33,34). The sample size for this SW-CRCT is calculated to detect a 13% difference in CMD cases at 3-month follow-up using GHQ-12 as a categorical measure between intervention (58% improved) and control condition (45% improved). Assuming an intra-cluster correlation coefficient (ICC=0.1), number of steps (t=4), number of clusters randomized in each step (k=14), average cluster size (m=4), power (80%) and alpha of 0.05, translates to 1120 participants, i.e., approximately 56 individuals per cluster per period. [ Figure 1, 3] The sample size was calculated using "stepped wedge" function of STATA version 14 (35).

Randomization and treatment allocation
The unit of randomization in stepped wedge trials is a cluster or group of clusters,

Recruitment
For control condition, a screening list is generated from district electoral roll using systematic random sampling method with pre-decided random start and random interval, with every nth number from the pool being selected. We used electoral roll as it is the most complete, comprehensive and accessible national frame of residential addresses in India and are extensively used for drawing random sample for general population (36). For A different recruitment procedure is used in the intervention condition as using structured questionnaires for identification (e.g. GHQ, PHQ) was perceived as impractical when implementing the intervention at scale and seen as stigmatizing in a community setting.
Champions are trained to identify a person in their catchment area (i.e. villages) with CMD based on symptoms described by the participant during an unstructured interview. As described earlier champions received detailed training in identifying symptoms of distress as well as depression and anxiety.When a Champion identifies a person, who in their opinion has CMD and who they intend to provide 4-6 counselling sessions, they are asked to inform their CF who in turn informs the project manager who creates a caseload list for each Champion. All these caseload lists are then merged to create a master list. The sample for the intervention condition is drawn from this master caseload list using computer generated random method. The drawn sample is screened by data collection staff using GHQ-12 (score of 3 and above) for recruiting intervention participants. This screening and subsequent baseline data collection for participants meeting the inclusion criteria is done prior to the Champion starting psychosocial counselling sessions with the participant. This process is continued till the target sample size of 56 cases is reached for each cluster and period. Different recruitment procedure for control and intervention condition was chosen as one of the secondary outcomes of the study is to check the specificity of identification of CMD cases by champions. The recruitment procedure in the intervention group allows us to answer this question of specificity of identification by Champions. Using similar recruitment procedure in both control and intervention condition (using electoral rolls) would have answered the sensitivity question (how many people with common mental disorders were identified by the Champion) but not allow us to address the specificity question. The sensitivity question relates to coverage (how many persons in the population with CMD are identified and receive the intervention), but given that we are using volunteers, we first want to establish whether these Champions are accurately able to identify persons with CMD, before addressing the coverage question.
Furthermore, we have other data to estimate the population coverage of the intervention.
To account for recruitment as a potential confounder, the statistical analysis plan includes adjusting for baseline covariates (baseline data for all the scales), assuming baseline differences between control and intervention participants.

Data collection
Written Informed consent is sought from all participants. A thumb impression and signature of a witness is taken for illiterate participants (37). Data is collected by trained researchers in two stages using paper pencil method. In the first stage, demographic data along with GHQ-12 is collected. The data collectors score the GHQ using a scoring sheet and if the participant has a score of 3 or more, secondary outcome data is collected at same time. The questionnaire for secondary outcome data takes 40-45 minutes to complete. Participants are not compensated for their time, as this is a volunteer-led intervention. Data collection staff travel to participant's home for data collection to avoid any travel costs for participants. Data collectors are recruited from the intervention district but recruited from different villages. Data collectors are not paired with participants and they are not assigned participants from their own villages for data collection, thus reducing the likelihood of study staff having prior acquaintance with the participants. Furthermore, if any study staff had any prior personal acquaintance with a participant, they were replaced with another data collector who was not acquainted with the participant.

Data management
Atmiyata uses a comprehensive data management system that aids in collecting high quality data by maintaining on-going-on-site and off-site quality assurance and quality control checks. Research staff (data collectors) handling data are thoroughly trained in interview techniques and procedures for sensitive data handling. Several measures to control for quality of data collected are implemented, including weekly checks, field monitoring visits twice a month by project managers and spot checks (once a month).
Additionally, refresher training sessions are provided once in 4 months for quality assurance purposes. The project manager ensures completeness and legibility of the data prior to data entry and is responsible for storing all the data. Designated data entry person is trained for specific entry guidelines to avoid erroneous data entry. The deidentified data is entered in password protected Excel sheets. Personally identifiable information is not entered in the database. Raw data is not uploaded on internet; instead all entered data is shared with statistician through offline electronic data transfer from the site by the project manager on monthly basis. Statistician collates the data, maintains the database and reviews data quality in terms of numbers, consistency and completeness. Measurement of percentage agreement among the data collectors is obtained once a year, to ensure reliability of the data collected. Several strategies are adopted to achieveadequate participant enrolmentincluding three telephonic follow-up calls to participants not available during in-person visits, two reminders for follow-up visits and rescheduling visits as per participant convenience. Recruitment, follow-up rates and missing data are discussed at monthly team review meetings between project manager and data collectors.

Data storage, security and confidentiality
Study data is anonymized using unique study identification (ID) codes for participants, which is matched to the physical consent form and then entered in the study database.
Only the consent form includes personally identifiable details. A code sheet linking participant's personal identifiable information is linked to the unique study ID code. Data is stored on a password-protected external hard drive periodically as a back-up. All consent forms and data forms are stored in a locked cabinet at the site office in Mehsana, accessible only to the principal investigator (PI) and project manager. After the study is over, the data will be stored in the sealed cabinet as required by Indian regulations.

Data monitoring
An advisory committee consisting of 4 experts in medical ethics, public health administration and public health and social science research was formed to monitor the implementation and research. The Committee meets every 6 months with the research team and also makes periodic site visits to personally interact with a few participants. All adverse events will be reported to the committee.

Statistical analysis
Baseline characteristics will be summarized using counts (percentages) for categorical variables and means (SD) for continuous variables. Analysis will be based on intention to treat and participants will be analyzed in the group that the cluster was assigned to at each time point.
The analysis plan is based on the Hussey and Hughes model for analysis of cross-sectional SW-CRCT designs (15). Generalized Linear Mixed Model (GLMM) will be used to determine the size and direction of the difference between the control and intervention conditions for primary and secondary outcomes. The estimated intervention effect will be reported as the mean outcome difference for continuous variables and Odds Ratio for categorical variables between intervention and control condition assuming a constant treatment effect over time. Estimates of the difference and 95% CIs will be calculated. To take time effect into account, all analysis will be adjusted for time (periods) of the intervention and for clusters. Period (time) and intervention (counseling sessions) will be specified as fixed effects and clusters as random effect. Analysis will be adjusted for baseline covariates to account for potential imbalance arising due to different recruitment procedures and regional differences across control and intervention condition. The analytical plan does not include any interim analyses.
Two broad model extensions (38), random cluster by period effect and random cluster by treatment effect will be used for secondary analysis. The secondary analysis will investigate an interaction effect between intervention and time and interaction effect between cluster and time. Additional analyses of the primary outcome will be conducted controlling for demographic variables if required. Statistical analyses will be carried out using STATA version 14 (35).

Economic evaluation analysis
All data will be analysed in Microsoft excel. The ratio of costs and benefits will be calculated and will be presented as ROI. This will inform whether Atmiyata intervention is a sound investment. Apart from this, the study will also provide information on economic burden of CMD in Mehsana district, which is of value to funders, policy makers and can be used for advocacy purposes.

Discussion
Atmiyata is a community-led intervention focused on reducing distress, particularly depression and anxiety symptoms in rural communities in India. The evaluation of the intervention through a stepped-wedge cluster randomised controlled design, offers several unique opportunities. First, there is limited literature of SW-CRCTs in low and middleincome countries, particularly involving mental health interventions mental health care.
This study will contribute to this sparse evidence base and be able to use implementation lessons to inform further scale-up of the intervention to other districts and states, as well as to inform potential intervention scale-up in other settings.
Several challenges exist with using a SW-CRCT design, including a lack of consensus on the model of analysis (38). Although the Ottawa Statement (39)  Goldberg et al. (42) suggested the best threshold for GHQ-12 scores varied from 1/2 to 6/7, with the most common cut-off score being 2/3. A recent Indian study confirmed an optimal cut-off score of 2 for community studies based on receiver operating characteristic curve (ROC) analysis (43). Taking all the above into account, we conservatively chose 3 as the cutoff score for our community study as it ensures inclusion of community participants with mild to moderate CMD.
We anticipate challenges during data collection of the trial in a large community setting.
Due to stigma and silence around mental ill health, there is a likely reluctance to participate in community mental health studies. Refusal to participate may also be due to the caste, gender, religion and other social attributes. To address these challenges, data collectors are trained and re-trained to build rapport with participants, and to maintain privacy and confidentiality. We also aim to recruit data collectors diverse on a variety of social attributes to represent Mehsana's heterogeneous population.
Community members often change residence, or the address is incorrectly entered in the electoral register, which increases the time for identification of control condition participants. Data collection timelines also have to accommodate for community events such as farming season, religious festivities. Safety of data collection staff is an equally 22 important concern given the large geographical are being covered. The team conducts regular meetings to troubleshoot challenges on the field to maintain the protocol.

Consent for publication
Not applicable

Availability of data and material
The datasets will be available to appropriate academic parties on request from the principal investigator in accordance with the data sharing policies of the institute within