The SMART Mental Health intervention combined an electronic decision support system (EDSS) for primary care doctors and CHWs, and a community-based antistigma campaign to manage depression, anxiety, and self-harm risks among adults in rural India. It was implemented across 44 primary health center (PHC) clusters in 2 states—Haryana and Andhra Pradesh—with 22 clusters randomized to the intervention arm and 22 to the control arm.

The intervention had 2 components. The first was an mHealth component. An EDSS was developed for PHC doctors and village-based CHWs known as ASHAs to identify and manage people at high risk of CMDs, referred to in this paper as “high-risk” people.

Following randomization of PHC clusters, individuals identified as “high-risk” in the intervention clusters were regularly followed up by ASHAs and encouraged to seek medical advice. PHC doctors used tablet‑based decision support apps adapted from the World Health Organization’s Mental Health Gap Action Programme Intervention Guide [8] to support clinical decision‑making. The SMART Mental Health platform maintained a registry of high-risk patients, accessible to both ASHAs and doctors within their respective catchment areas. A traffic light–based feature in the app alerted ASHAs about patients who had not visited the doctor or needed additional follow-ups. ASHAs received automated prompts and reminders to support follow‑up on their phones through an interactive voice recording system. PHC doctors attended to patients either at the PHC or at health camps organized in the villages. Those requiring specialist care were referred to psychiatrists at government facilities.

The second component of the intervention was a community-based campaign that used multimedia information, education, and communication (IEC) strategies to enhance knowledge of mental health and reduce stigma, negative attitudes, and improve behaviors toward people needing mental health support. The campaign printed materials, video narratives of people with lived experience, animated and promotional videos featuring local influencers, and live or recorded street‑theater performances in local languages. A detailed description of the intervention has been published elsewhere [6].

The control arm/usual care arm clusters received information on CMDs through pamphlets. Those identified as high-risk were advised to seek care from the PHC doctor or psychiatrist by the ASHA. The antistigma and the mHealth components were not used in the control/usual care arm.

A detailed protocol for the process evaluation is available [9]. The UK Medical Research Council (MRC) guidance on process evaluation [10] and the Reach, Effectiveness, Adoption, Implementation, and Maintenance (RE-AIM) framework [11] were used to develop key areas of enquiry. The MRC guidance [10] suggests that process evaluations should answer questions related to 3 components: “Context” (how does context affect implementation and outcomes?), “Implementation” (what is delivered and how?), and “Mechanisms of impact” (how does the delivered intervention produce change?). Thus, these 3 components (implementation, mechanism of impact, and context) formed the broad areas of inquiry during the process evaluation. The RE-AIM framework [11] makes use of 5 key areas to evaluate intervention: reach, effectiveness, adoption, implementation, and maintenance. RE-AIM was used to evaluate the “implementation” component of the program. A brief snapshot of how theory was used to guide our line of enquiry is presented in Table 1.

The process evaluation used a mix of quantitative and qualitative data. The quantitative data consisted of program data recorded by the implementation team and stored at the backend of the SMART Mental Health server. These included some key metrics, including follow-up visits made by the ASHAs, doctor consultations made by the study participants, and exposure to various antistigma content.

The qualitative component consisted of focus group discussions (FGDs) and interviews with key stakeholders. Qualitative data were collected in 8 purposively selected clusters out of the 44 clusters. These included 6 intervention clusters and 2 control clusters, equally distributed across the 2 regions. Since the study aimed to focus in-depth on each cluster and a large number of interviews and FGDs were planned for each selected cluster, 8 clusters were deemed feasible and enough to provide rich contextual insights. After discussion with the implementation team, PHC clusters were purposively selected using the principle of maximum divergence, or selecting cases with the most variation to capture a broad spectrum of perspectives and possible reasons behind variations. The following criteria were used for selecting the clusters: (1) clusters where challenges were faced during implementation, (2) clusters where a large proportion of high-risk patients had sought care, and (3) the staff had faced relatively limited challenges in implementation. The purposive overrepresentation of intervention clusters was done to enable a deeper understanding of the implementation of the intervention. This approach necessarily excludes experiences from many clusters and limits comparative inference between intervention and control sites.

Different categories of informants (PHC doctors, ASHAs, high-risk and non–high-risk study cohort, and local key informants) who could provide insights and feedback related to the intervention were identified. In each cluster, interviews or FGDs were conducted with all categories of informants we identified. The high-risk and non–high-risk cohort participants were selected based on their availability at the time of data collection.

Based on prior qualitative research experience, we planned a large volume of interviews and FGDs that were expected to yield sufficient informational redundancy. Data collection was completed as planned, followed by thematic analysis. During analysis, we observed substantial repetition of themes and no emergence of new insights relevant to the study objectives, indicating that the dataset was sufficient. Accordingly, no additional data collection was undertaken.

Data were collected within 2 months after the end of the intervention and before the trial results were known. In Haryana, the intervention ended in September 2021, and data collection was done (by AM, AK, and MD) in December 2021. In Andhra Pradesh, the intervention ended in December 2021, and data collection was done (by AM, S Kallakuri, and SD) in January 2022. All data were collected face-to-face. The data collection team consisted of 2 male and 4 female researchers with a PhD or a master’s degree and prior experience in collecting qualitative data. All members of the data collection team (except AM) were involved in training and oversight of trial activities but were not engaged in routine field-level implementation and had limited prior interaction with study communities. This partial insider position presented both constraints and opportunities. Familiarity with intervention logic and operational challenges supported contextually informed probing and interpretation of implementation experiences. At the same time, participants may have perceived researchers as institutionally affiliated outsiders, potentially encouraging socially desirable responses. However, community members and ASHAs openly articulated implementation challenges, including difficulty recalling IEC content and perceptions that repeated visits were unnecessary. In a small number of ASHA FGDs, the individual involved in delivering training also facilitated data collection; this dual role was recognized as potentially shaping responses and was considered reflexively during analysis. Data collection was intentionally gender-matched where feasible, with male researchers facilitating discussions with male community members, and female researchers conducting FGDs with female community members and ASHAs, to enhance comfort and openness, particularly among women participants.

Topic guides with probes were prepared in Hindi and Telugu. The objective of data collection was explained, and a participant information sheet was provided to all participants before starting the interview or FGD. Data were collected by team members proficient in the local language. In Andhra Pradesh, a few of the interviews and FDGs required translation support from a field supervisor, as one of the researchers (AM) could understand the local language but was not proficient in speaking it. Data were collected at clinics, community halls, and the homes of participants. All data were audio-recorded after taking consent from participants. The duration of FGDs ranged from 20 to 180 minutes. The interview duration ranged from 10 to 30 minutes. On average, 7 participants attended the FGD. The range was from 4 to 12. There were 2 FGDs with ASHAs that had only 4 participants. This was because in the selected PHC cluster, in all 6-7 ASHAs were part of the intervention, and when we conducted the FGDs with ASHAs, not all of them could attend.

A total of 33 interviews and 38 FGDs were conducted (Table 2). FGDs for men and women were done separately to honor social norms. A total of 288 participants (120 male and 168 female) were part of the qualitative study.

A detailed COREQ (Consolidated Criteria for Reporting Qualitative Research) checklist has been provided in Multimedia Appendix 1.

Data analysis was done using the framework analysis approach. This approach uses an existing framework consisting of key thematic areas of inquiry. Under each thematic area, there are codes (which loosely correspond to subthemes). The qualitative data is coded or indexed within these broad themes and codes. In the final analysis, the data for each theme is summarized.

The conceptual framework of the study (Table 1) informed the initial framework for analysis and guided the development of a set of a priori codes aligned with key areas of inquiry. Select transcripts were read by the research team members (MD, AM, S Kallakuri, and SKY) to come up with additional codes emerging from the data. The codebook developed after this process was used by the team to code the transcripts. The codebook was iteratively refined during the data coding process, and additions to the codebook were made (Multimedia Appendix 2). Disagreements were resolved through discussion. In case of a lack of consensus, the final decision was taken by the process evaluation lead (AM). All researchers had real-time access to the codebook and data files. The NVivo (version 12; Lumivero) collaboration server was used for data coding. For each code, the codebook included a clear definition specifying its meaning and illustrative examples of the type of data to be coded under that code. The codebook was stored on a shared drive accessible to all members of the research team, and any revisions were documented and visible to all coders. This process supported the consistent application of codes across transcripts.

The final analysis was done by the first author (AM), who synthesized coded data within each theme. The findings are organized under the RE-AIM domains. Qualitative data were intentionally combined with quantitative data to explain cluster-level differences. In addition, qualitative and quantitative data were used to explain variation in service uptake across different levels of care.

This study was approved by the ethics committees of the George Institute for Global Health India (009/2018) and All India Institute of Medical Sciences, New Delhi, India (IEC-315/01.06.2018, RP-58/2018, RP-16/2018). The study was approved by the Health Ministry’s Screening Committee, Indian Council of Medical Research. Written informed consent was obtained from all participants prior to the data collection. No compensation was provided to study participants for participating in the trial. All data management processes were compliant with national privacy law. Data were securely stored and analyzed on servers located at the George Institute India office.

The program achieved a broad reach across geographically diverse rural settings in 2 states. In total, 9928 adults across 44 rural PHC clusters were enrolled, and 1697 people identified as high risk were allocated to the intervention arm. Of these, 1585 people were followed up at the end of 1 year to assess the outcomes [7]. Intervention components reached the majority of high-risk individuals in the intervention clusters. Table 3 presents cluster‑level follow‑up, and consultation indicators summarized as medians and IQRs across the 22 intervention clusters. The median proportion of high‑risk individuals who received at least 1 ASHA follow‑up was 98% (IQR 96.6%-100%), and the median number of ASHA visits per individual was 11.5 (IQR 6.2-19.0), with notable variation across clusters. A median of 87.6% (IQR 51.9-93.7) of high‑risk individuals were both seen by a PHC doctor and followed up by ASHAs at least 6 times.

The antistigma component used multiple delivery methods and was implemented both individually for the study cohort, comprising high-risk and non–high-risk participants, and at the community level. Community meetings were organized in common village spaces, during which awareness videos were screened using a projector. These meetings were attended by members of the study cohort as well as community members who were not recruited into the study.

In addition, project field staff conducted regular home visits to high-risk participants to ensure exposure to antistigma materials, particularly for those who may not have attended community meetings. Printed materials were distributed across all villages within each intervention cluster.

Live drama performances were conducted in 1 village per intervention cluster, followed by interactive discussions with audience members. Villages with relatively larger populations and adequate space to conduct public performances were selected for these live shows. In remaining villages where live performances were not feasible due to logistical constraints and cost considerations, video recordings of the drama were screened, accompanied by facilitated discussions around the content.

Exposure of the study cohort to each antistigma campaign component was systematically tracked throughout the intervention period. Table 4 presents cluster-level exposure to individual antistigma campaign components among high-risk and non–high-risk participants, summarized as median percentages with IQRs across PHCs. Overall, the intervention achieved high exposure to most antistigma materials. Exposure to live drama was lower compared to other campaign modalities, due to the infeasibility of conducting live performances in every village; however, recorded drama clips were used to ensure broader reach. Across intervention clusters, a median of 84% (IQR 65.7-95.9) of the study cohort was exposed to at least 1 audiovisual campaign component.

Exposure and dosage metrics were tracked using backend data and program monitoring data.

Overall, the community-level screening and referral by ASHAs was seen as an effective way to ensure care for people with CMDs. A PHC doctor who organized regular weekly psychiatry clinics at his PHC told us that the footfall in his weekly clinic was lower than what was seen in village health camps conducted as part of the intervention.

Several patients were referred to psychiatric care from the district hospitals. Mental health professionals in these hospitals found the SMART Mental Health approach to be useful. Prior to the intervention, the psychiatrists at the district level received very few patients referred to them by PHC providers. Most came through self-referral. However, during the intervention, psychiatrists in the district hospitals found that many more people with CMDs were referred by the PHCs' doctors. The providers found that involving CHWs for screening, identifying high-risk people, and having a referral and follow-up mechanism in place were useful strategies to reach a larger number of patients in need of psychiatric care.

High-risk people who consulted a doctor and experienced improvement acknowledged the importance of consulting the primary care doctor. In some cases, the process of sharing problems with the doctor and receiving positive advice led to perceived improvement, and no medication was required.

The antistigma campaign received mixed responses, with better recall for video content compared to print material. One of the factors behind poor recall could be that majority of the study cohort in both sites have education only up to the primary level, with a large proportion being nonliterate. We found that while participants did not always recall the exact message delivered by the antistigma campaign, they acknowledged that the program had provided them with new information. Participants said that it was important to seek help if someone was undergoing mental stress, and there were treatments available for mental illnesses.

The SMART Mental Health platform, including the integrated EDSS, was viewed by doctors as a useful tool for the diagnosis and management of CMDs. None of the interviewed PHC doctors had previously used a decision support tool for mental health assessment. Doctors provided positive feedback on the availability of patient-specific diagnostic and follow-up information on the platform, which supported ongoing patient management.

The ASHAs used standardized tools—Patient Health Questionnaire-9 [12] and Generalized Anxiety Disorder-7 [13]—integrated in handheld tablets to screen people at high risk of depression, anxiety, and self-harm. The SMART Mental Health platform also provided them with a traffic light–based priority listing app to follow up patients. This was a novel experience for ASHAs as they had never used a digital device for community-level screening or follow-up. Some of them had never used a smartphone and initially were not confident about using a tablet device. ASHAs reported that good training and individual support to clarify their queries were crucial in helping them overcome initial reservations and use the mHealth platform with ease.

Following initial support during the early phase of implementation, ASHAs reported feeling comfortable using the tool.

The PHC doctors felt that, given the low levels of training in psychiatry, they usually lacked the ability to diagnose and manage CMDs. However, capacity building, combined with the use of the EDSS, had provided them with the skills to diagnose and manage people with CMDs. One positive impact was that some of them had started asking for mental health–related questions in their regular history taking during outpatient clinics, even with patients not part of the intervention, something they had not done earlier.

The ASHAs provided positive feedback about transitioning to a digital system for recording data and patient follow-up. They felt that the skills they had learnt in the project were useful, given the government’s push on digitalization. There was also acknowledgment that the training had helped to ASHAs to gain skills in identifying people with mental health care needs in the community and that this skill would be useful for them in any future programs on mental health.

Context can refer to a range of factors that affect or can be affected by the intervention, including but not limited to social, cultural, organizational, or policy-related factors. Identifying and reporting relevant contextual factors that led to adaptations or impacted outcomes can be useful for designing and scaling up similar interventions [14,15].

The local context presented several cultural barriers. Efforts were made to address concerns of Muslim families in Haryana regarding audiovisual content and interaction with male staff. Similarly, additional efforts were made to reach out to groups that are difficult to reach, such as migrants, the fishing and truck driver communities. Our experience shows that additional time, planning, and resources are needed to ensure vulnerable groups are included in a mental health intervention.

COVID-19 led to certain adaptations, such as introducing teleconsultation in Andhra Pradesh. The program also needed to adapt and put in place strategies that could address barriers to accessing care at the PHC due to a lack of transport and the cost of travel. Bringing PHC doctors to villages to conduct health camps was helpful. These adaptations were made based on what was feasible and acceptable during implementation, rather than following a predefined adaptation framework, in order to respond to contextual constraints. While core intervention components—such as screening and clinical decision support—remained standardized, delivery strategies were adapted to local conditions. Similar context-specific adaptations, such as adjusting meeting times and conducting health camps for difficult-to-reach communities in the vicinity of these communities, have been highlighted as a useful strategy by health functionaries in another state in India [16].

Stigma, particularly perceived stigma, was a barrier in seeking care that was highlighted during formative work and by other research in the study area [3,4,17]. The antistigma campaign used previously piloted as well as newly developed evidence-based strategies in the community [17-19]. The study found that the audiovisual medium was more acceptable compared to printed content. The intervention ensured that the high-risk cohort was exposed to the content multiple times. This was, however, a less acceptable strategy and did not increase recall. In most cases, the videos were shown individually. Showing the material in small community groups can be an alternate strategy. This will ensure that people do not feel singled out for mental health messaging. It can also provide opportunities for group interaction and engagement.

The coverage or ‘reach’ of the intervention was ensured by addressing key barriers discussed earlier. Exposure to the content of the antistigma campaign and follow-up of high-risk cohort by ASHAs was high. This was achieved through meticulous monitoring of individual-level data and planning follow-up visits and patient consultation with ASHAs and doctors to ensure all high-risk people were reached. Program‑monitored data provided a reliable measure of exposure and follow‑up, capturing coverage rather than the depth or quality of engagement, which was examined through qualitative analysis. While reach is a core component of the RE‑AIM framework, it does not capture heterogeneity in engagement among those reached. Maximizing reach required intensive delivery and repeated exposure to intervention components; however, this intensity was not uniformly beneficial. Qualitative findings indicate that repeated exposure to antistigma content led to diminishing returns for some participants, including irritation and limited recall, suggesting potential saturation effects. Together, these findings highlight a key tension between maximizing coverage and sustaining engagement and underscore the importance of considering both dose and quality of delivery when interpreting implementation fidelity and success.

The “effectiveness” of an intervention refers to the extent to which an intervention achieves its intended outcome. This has been discussed in detailed trial results [7]. The process evaluation captured perceptions of stakeholders on why they felt the intervention was effective or not. A community-based screening model for CMDs was perceived as an effective strategy by PHC doctors as well as specialist mental health professionals providing care at the district hospital. Doctors at PHCs rarely diagnosed or referred patients with CMDs prior to the intervention. Limited training and experience in diagnosing symptoms of CMDs were reasons cited by PHC doctors. The training, EDSS, and consultation with high-risk patients helped to address this gap. There was an increase in the number of patients referred by PHC doctors in the intervention to the district psychiatrist.

“Adoption” to the extent to which users and implementers are willing to take up the intervention. We found that the digital component was a novel experience for both ASHAs and doctors. The EDSS was seen as a useful tool for diagnosis and treatment by PHC doctors. Our findings show that to enable the adoption of digital health interventions among CHWs and doctors, 1-time training is not enough. It needs to be supplemented with continuous handholding, support, and feedback in the early implementation phase. Some intangible aspects, such as good rapport building and respectful communication by project staff, were important in ensuring buy-in from CHWs. Other studies in low- and middle-income countries (LMICs) contexts have highlighted the importance of understanding the variation in experience, education, and exposure to digital technologies among CHWs before starting the training. Effective training of CHWs has been found to be a key element in successful mHealth interventions [20].

There were several positive spillover effects that ASHAs reported. Many felt that they had personally benefited from participating in SMART Mental Health. Some ASHAs felt that they had overcome the inhibition of speaking with male members in the community. Others reported that improved understanding of mental health and the practice of asking questions during screening had provided them with an important skill. It had helped them to connect better with the community, to listen to them and discuss their problems.

Within the RE‑AIM framework, “maintenance” is conceptualized as the long‑term sustainment of intervention effects and continued delivery of intervention components beyond the active trial period. Direct assessment of posttrial maintenance lies beyond the scope of this process evaluation. A separate study conducted 3 months into the posttrial phase specifically examines issues related to intervention maintenance and will be reported elsewhere. However, this process evaluation provides important insights into early indicators and conditions likely to influence maintenance, particularly in the context of scale‑up within the public health system.

One such area was related to human resource training. We found that intensive handholding support, including village-level supervision in the initial months, was an important facilitator for the adoption of the EDSS for the ASHAs. While scaling up such interventions, legitimate concerns related to who can provide this support, how much time and resources are needed, and whether it is sustainable in the existing health system need to be thought through. Midlevel health workers with whom ASHAs have a direct reporting relationship can be trained to provide this initial handholding and supportive supervision, as it complements their existing responsibilities.

ASHAs perceived increased community respect and recognition of their role as health workers and attributed it to technology use and its association with competence and credibility. With the government introducing digital data systems at the same time, ASHAs in the project saw themselves as early adopters and better prepared than their peers. The alignment of the intervention with government digital initiatives suggests strong maintenance potential. ASHA’s perceptions of increased preparedness and relevance indicate that core intervention components were compatible with routine system requirements, supporting the likelihood of continued use beyond the intervention period. At the end of the intervention, the PHC doctors were provided with a hard copy of the questions used in the EDSS for future reference. Doctors reported including questions related to mental health in their regular history taking for patients, indicating partial maintenance of some clinical practices even in the absence of the EDSS.

Mechanism of change refers to a theory of why change occurs or how an intervention leads to particular outcomes. This is of relevance for scaling and designing future programs. The assumptions and causal mechanisms articulated at the design stage may evolve as we accumulate more evidence on outcomes, implementation processes, and context.

SMART Mental Health articulated several hypothesized pathways to change at inception. Stigma is a major demand-side barrier for mental health help-seeking; hence, it was theorized that the antistigma campaign would improve knowledge and attitudes related to mental health stigma, thereby increasing demand for mental health services. In parallel, the EDSS combined with training of PHC doctors and ASHAs was intended to improve their capacity to identify, treat, and refer individuals at high-risk of CMDs.

Our results show clear improvements in stigma‑related knowledge, attitudes, and perceptions about help‑seeking. However, these improvements did not translate into corresponding changes in stigma‑related behavior. Consistent with the stigma literature, behavioral outcomes are typically less sensitive to intervention effects than knowledge or attitudes, with several reviews reporting smaller and less consistent effects on stigma‑related behaviors even when attitudinal outcomes improve. Similar patterns have been seen in outcomes related to mental health help-seeking behavior [21-23]. This signals the well‑recognized intention-behavior gap, where positive attitudes and intentions do not always lead to action [24]. Literature to explain the intention-behavior gap has often highlighted individual-level factors, such as the role of intentions, attitudes, subjective norms, or perceived behavioral control, and, to a lesser extent, highlighted possible structural factors such as access to services [25,26]. Other factors that may explain this gap include temporal lag. Behavior change may take longer to emerge than attitudinal change, and short follow‑up periods may fail to capture delayed behavioral effects [24,27]. Notably, despite low stigma‑related behavioral scores, a substantial proportion of individuals in the intervention arm consulted primary care providers, suggesting that the intervention was successful in increasing demand for mental health care at the PHC level. In contrast, a markedly smaller proportion of individuals attended specialist services following referral. This steep attrition along the referral cascade represents one of the most policy‑relevant findings of the process evaluation and points to structural supply‑side constraints that may cap the impact of primary care digital interventions, a pattern widely documented in global mental health task‑sharing initiatives [28,29]. While the intervention actively reduced access barriers to PHC-based care by facilitating village-level visits by PHC doctors along with follow-up support by ASHAs, similar facilitation was not available for specialist care. Barriers such as distance to facilities, travel and treatment costs, and limited availability or perceived quality of specialist services reduced referral completion. This reflects broader evidence from LMICs showing that weaknesses in the health system often outweigh individual-level changes in shaping mental health service use [28].

Differences between states further highlight the importance of the health-system context. In Andhra Pradesh, where district mental health services were relatively better established and the psychiatrist was mandated by the local health administration to make regular visits to PHCs, a higher proportion of referred individuals accessed specialist care compared to Haryana, where the psychiatrist was unable to make such visits due to a high patient load at the hospital and a vacant post for an additional psychiatrist. This suggests that the benefits of primary care digital interventions may be capped by supply-side constraints such as limited specialist availability, high clinical workloads, and weak referral systems. This is consistent with findings from global reviews of primary mental health care integration [29].

Beyond formal service use, participation in the intervention appeared to enable relational and supportive processes that may have contributed to symptom improvement. Qualitative accounts indicate that high‑risk individuals valued opportunities to share their concerns with PHC doctors, ASHAs, and project staff. Regular follow-ups often involved asking about the well-being of the high-risk individuals. Some ASHAs took an interest in spending time and listening to the problems of the high-risk people in their community. Drawing on counseling and psychotherapy literature as an interpretive lens, processes such as feeling heard, understood, and accepted are widely discussed as potentially therapeutic. We invoke this literature to help interpret participants’ experiences, rather than as confirmatory evidence of effect, and to suggest that these relational interactions may represent an additional, unanticipated mechanism through which the intervention could have supported symptom improvement [30].

EDSS was perceived as an important aid to reach a diagnosis by PHC doctors. It was, however, emphasized that individuals with mental health–related problems needed separate dedicated clinic timings as more time was needed for taking history, making the patient comfortable about discussing their problems, and counseling them, which was difficult to provide in the PHC setup. Another challenge related to making psychotropic medications available at the PHCs was also recognized by the PHC doctors. Due to limited demand in the past, these medicines were rarely indicated at the PHCs. Some doctors preferred to refrain from prescribing psychotropic medications themselves due to fear of possible side effects and referred any patients in need of medication to the psychiatrist. Together, these findings underscore the complexity of mechanisms underpinning the intervention: technological decision support alone was insufficient without concurrent alignment of clinical time, medication supply, provider confidence, and referral systems.

The data collection for the process evaluation was done by researchers who were purposively blinded to results. This was done to reduce possible bias resulting from knowledge of results. However, a disadvantage of this was that specific questions that explain some of the results could not be asked during data collection. For example, we could not specifically enquire why improvement in knowledge and attitude related to stigma did not lead to a change in behavior.

The process evaluation relied largely on self-reported data from participants and providers, which may be subject to social desirability bias. Some respondents may have overreported positive attitudes, knowledge, or perceptions about the intervention.

In our protocol [9], we had planned to capture detailed differences in some key metrics across the clusters. However, some of the data, such as time taken to administer the screening tool by ASHAs and the EDSS tool by doctors, could not be used in this paper due to difficulty in authenticating the actual start time and end times taken during administration of the tools due to technical issues in the design of the tools. In some cases, the app was not switched off till much after the end of the patient consultation. In other cases, doctors asked questions from memory first and consulted the tablet later. Similar challenges were faced by ASHAs.

The process evaluation was conducted in a specific context with particular levels of health system constraints, human resource capacities, and digital readiness. Therefore, these findings should not be generalized to other LMICs without due consideration to the local context.

The findings from the SMART Mental Health trial have important implications for mental health–related service delivery in India and other LMIC settings. The intervention was delivered with the existing cadre of primary health workers, including PHC doctors and ASHAs. It demonstrated the ability of primary care doctors to integrate EDSS in their practice with the right training and support. The Indian government has put forth a plan for comprehensive primary health care in which mental, neurological, and substance use disorders will be integrated into primary care [31]; entitled, the national mental health policy of the Government of India [32], which also aligns with the World Health Organization’s Comprehensive Mental Health Action Plan 2013-30 [33]. A system for community-level screening for CMDs by ASHAs, as well as referral and management of CMDs are part of these policy guidelines. The policy envisages patient-level digital health records at the primary care level, but leaves much scope to explore possibilities of integrating digital technology. There is potential for service delivery using digital platforms. The SMART Mental Health platform has been successfully used to manage referrals, ensure follow-up, and integrate an EDSS for improved diagnosis and management of mental, neurological, and substance use disorders. Any such initiative needs to ensure adequate support for training, supervision, and incentives from health care workers to ensure sustainability. There is also a need to ensure that the overall digital infrastructure is strengthened, particularly in remote regions, to avoid inequities in access to digital services. Future research should focus on scaling this model and further research to understand how improved knowledge and attitudes related to mental health stigma can translate to positive behavior change.