The global burden of disease due to tobacco, alcohol, and other substance use is increasing and substantial. Tobacco use, alcohol use, and substance use ranked as the 2nd, 9th, and 14th highest causes of disability-adjusted life years (DALYs) worldwide, with increases in age-standardized DALY rate of 24.3% for tobacco and 37.1% for alcohol from 1990 and 2019 (Murray et al., 2020). The calculated disease burden does not include the impact of alcohol and substance use on a wide variety of social, financial, legal, and relationship problems for individuals, families, and communities (Humeniuk et al., 2010a; Ndetei et al., 2009). As such, alcohol and substance use disorders (AUD and SUD) put an enormous burden on the healthcare system and impact economic development (Patel et al., 2007). Health expenditures and access to care for mental illnesses and substance use disorders are disproportionately low compared to their associated disease burden (Vigo et al., 2016; World Health Organization, 2018). Only 10.3% of those with past-year SUDs received minimally adequate treatment in high-income countries, and only 1% in low- and middle-income countries (LMICs) (Degenhardt et al., 2017). This can change. Other sources of disease burden have decreased greatly due to concerted efforts worldwide (Murray et al., 2020). Similar concerted and strategic actions to target harm from substance use are long-overdue and have been deemed a priority by the World Health Organization (WHO) in the Mental Health Global Action Plan (mhGAP) (World Health Organization, 2010).

Integrating effective interventions such as screening and brief interventions (BIs) in primary care is needed to address the treatment gap in LMICs (Patel et al., 2015; World Health Organization, 2019). There is strong evidence supporting the use of BIs to decrease alcohol use and its associated complications and comorbidities and some evidence for other substances (Humeniuk et al., 2008, 2012; Kaner et al., 2018). However, most of these studies have been conducted in high-income countries (HICs) (Babor et al., 2013; Kaner et al., 2018; O'Donnell et al., 2013). LMICs differ greatly from HICs in terms of availability and level of training, access to primary care, general population awareness of substance use’s impacts, and in cultural norms (Awenva et al., 2010; Babor et al., 2013; Hill et al., 2021; O'Donnell et al., 2013). Some of the most rigorous studies in LMICs have been based on the alcohol, smoking, and substance involvement screening test (ASSIST)-linked BI (Humeniuk et al., 2008, 2010, 2012). However, there have been few replications studies or studies on addressing the health worker training gap identified by the WHO (World Health Organization, 2019). This highlights the need for research on effective and scalable training in LMICs and the impact and sustainability of services ensuing from those training.

Blended-eLearning (in-person plus digital) could provide effective and cost-effective training options, even in LMICs, especially as technological access improves (Bahia & Suardi, 2019; George et al., 2014; Gomez, 2014; Lewis et al., 2014; Liu et al., 2016; Maloney et al., 2012, 2015; Marrinan et al., 2015; Sandars, 2010; Sissine et al., 2014; The World Bank, 2020; Walsh et al., 2010). However, it has not been studied extensively in LMICs.

The computer-based drug and alcohol training and assessment in Kenya (eDATA-K) mixed-method research program dynamically adapted the NextGenU.org model of free eLearning to the Kenyan context and assessed training impacts (Frank et al., 2016; Galway et al., 2014; Rossa-Roccor et al., 2017). Further details on eDATA-K program of research, its methods, associated theory of change, successful trainees acquisition of substance use services related competencies, and positive impact on stigma are available in other publications (Clair et al., in press, 2016a, 2016b, 2016c, 2019, 2022; Hsiang-Te Tsuei et al., 2017).

The research questions addressed by this article are the following:

  1. 1.

    In patients with moderate to high alcohol-related risk levels, attending typical primary care clinics, what is the impact on alcohol consumption of receiving the ASSIST feedback by lay health workers compared to those receiving that same feedback by lay health workers in addition to a BI delivered by clinicians, with clinicians and lay health workers trained via blended-eLearning, as observed in two pragmatic randomized control trials (one in public and one in private facilities) in an LMIC context?

  2. 2.

    What is the experience of patients and health workers involved in those randomized control trials (RCTs)?

Methods

Per best practices in studying complex interventions, eDATA-K used a mixed-methods developmental evaluation, including parallel qualitative data collection alongside the RCTs, to strengthen the understanding of the innovation, its implementation, and impact (Patton, 2011). The use of multiple methods triangulates the findings, therefore enhancing explanatory capacity and validity (Cathain et al., 2013; Clark et al., 2008; Kaner et al., 1999; O'Cathain et al., 2014; Spillman, 2014; Tarrow, 2004).

Ethical approvals were obtained from the University of British Columbia and the Kenya Medical Research Institute ethics boards; RCTs were registered with ClinicalTrials.gov (NCT02388243).

Setting and Context

eDATA-K involved public and private facilities in separate RCTs, as prior studies on mental health interventions show different results between public and private settings (Patel et al., 2010). Public facilities were from two similar counties selected for their comparability and population characteristics, each including about 1 million Kenyans living in typical suburban and rural conditions. Private facilities were in Nairobi or an adjacent county. Selection criteria for the facilities were providing primary care, willingness, and ability to engage in research (electricity, staffing, sufficient patient attendance) and have a minimum of one staff who completed the screening course (lay health worker in public facilities, support staff in private facilities) and one clinician (nurse, physician or clinical officer) who completed the primary care course. The NextGenU.org courses were completed in the prior phase of eDATA-K. Eight of eleven public and two of four private facilities that participated in the training phase had enough trained health workers to participate in the RCT.

The courses included modules on the ASSIST screening tool and associated BI, communication, confidentiality, and stigma (Clair et al., 2019; Tsuei et al., 2017). ASSIST-linked BIs components considered important in generating change in substance use have been summarized with the FRAMES acronym (Humeniuk et al., 2010a):

  • Feedback (personally relevant)

  • Responsibility (helping people realize their need for change and that they have choices related to their substance use and behaviors)

  • Advice (provision of clear advice to reduce harm from continued use, delivered in a non-judgmental manner)

  • Menu of option (suggesting a range of strategies)

  • Empathy

  • Self-efficacy (eliciting strengths and capacity to change)

RCT Design and Procedures

The design and procedures are based on the seminal study by Fleming et al. (1997) and illustrated in supplementary Figure S.1. Anyone 18 or older presenting to a participating facility seeking outpatient consultation, when a screener was available, was offered the general health screening consisting of the RAPA for physical activity level (Topolski et al., 2006), weight and height for nutritional status, alongside the ASSIST. The inclusion of other health issues than alcohol use masked the focus on substance use decreasing recruitment bias and stigma. Those screened received verbal feedback, with short written advice and directed to the research staff if their ASSIST score fell in the included range. Eligible participants were randomized to receive no more intervention than the simple feedback already received (Fb group) or feedback plus a BI (Fb + BI group). Study staff, independent from intervention delivery, provided the relevant instructions based on allocation from a computer-generated randomization list concealed until allocation. After randomization, only those in the Fb + BI group were instructed to show/discuss their ASSIST score with their clinician. Clinicians were instructed to provide BIs only to those presenting their scores and to offer one or two brief follow-up visits. Screeners and clinicians were blinded to the randomized allocation. Study staff were blinded at follow-up by carrying follow-up on participants from a different clinic than those they randomized. Baseline and follow-up questionnaires included demographic surveys and gold-standard, timeline follow-back methods for assessing self-reported alcohol consumption (Hjorthøj et al., 2012; Sobell & Sobell, 1992). Participants presenting for follow-up data collection were offered soap as compensation for their time.

RCT Participants’ Recruitment

Consenting participants at moderate or high risk from alcohol use (a score ≥ 11) and ≥ 18 years old were eligible unless they met exclusion criteria. Exclusion criteria were pregnancy, having attended an alcohol treatment program in the previous year, or reporting symptoms of suicide or severe neurological/psychiatric impairment (e.g., psychosis).

Statistical Analyses

Statistical analyses were performed through SPSS v23 and Stata v14.0, with group allocation concealed until the end result. All analyses used a 2-tailed alpha level of 0.05. Per SPSS, each RCT arm needed 146 participants to achieve 80% power using the expected change in alcohol consumption from prior studies (Kaner et al., 2007). A wealth index using principal component analysis was built from the socio-economic status (SES) survey, per best practices for LMICs (Chuma & Molyneux, 2009; Howe et al., 2012; Vyas & Kumaranayake, 2006). Independent samples t tests assessed unadjusted differences in alcohol consumption between groups at every time point, whereas paired t tests were performed to assess changes in each study group between baseline and follow-ups.

Intention-to-treat analyses were performed based on generalized estimating equation (GEE) models of the alcohol quantity consumed over the last 7 days, in grams, adjusting for SES, with appropriate imputation, taking clustering into account (Zeger et al., 1988). To investigate the effect of loss to follow-up, complete cases and imputed analysis were compared in a sensitivity analysis (Jakobsen et al., 2017).

Qualitative Methods and Analyses

Qualitative data were analyzed through a phenomenological lens (Dowling, 2007; Norlyk & Harder, 2010; Sandelowski, 2000a, 2000b). Data was collected via field observation, focus group discussions (FGDs), and key informant interviews (KIIs). FGDs were used to encourage dialogue and joint meaning-making (Krueger, 1994; Morgan & Krueger, 1993; Powell & Single, 1996; Rabiee, 2004; Wilkinson, 1998). FGD guides included questions about their general study experience and the perceived impact of the ASSIST screenings with or without BI. (Betts et al., 1996; Krueger, 1994; Morgan & Krueger, 1993; Powell & Single, 1996; Rabiee, 2004; Sandelowski, 2000a, 2000b; Wilkinson, 1998). If selected participants could not participate in an FGD, KII was conducted.

A stratified random sampling framework was used to recruit diverse RCT participants who provided signed consent. FGDs and KIIs were audio-recorded, with co-/moderators taking written notes on body language and major immediate emerging themes and summarizing their understanding of participants’ main points. All participants received a small monetary compensation.

Kenyan moderators conducted the FGDs and KIIs in Kiswahili, other local languages, or English, according to participants’ preferences. Audio recordings were transcribed, translated, and imported into NVivo, a qualitative data management software. Latent-content analysis using a constant comparison method was used to analyze the data (Boeije, 2002).

Field observations conducted through regular site visits, and in course-related events, included recorded discussions with research staff and records of meetings and other significant interactions by qualitative research staff with participants and other staff.

Canadian and Kenyan researchers read all transcripts and independently recorded common emerging words, phrases, and themes. Similar themes were formulated into a preliminary coding framework with in-depth discussion of opposing/unique themes or interpretations. Cycles of discussion, reflection, and data immersion generated new themes until data saturation was reached and discrepancies resolved in group consultations. As insight developed, researchers jointly re-categorized data in a reductionist process to extrapolate meaning (Boeije, 2002; Dowling, 2007; Norlyk & Harder, 2010; Sandelowski, 2000a, 2000b) which was further informed by outreach to the larger research team, a subset of health workers, and key collaborators, including other research staff RAs who had numerous discussions (while collecting qualitative and quantitative data) with patients on their experience of the screening/BI.

Results

RCT Participants

RCT patient recruitment and loss to follow-up are presented in Fig. 1. Of the 683 patients in public facilities and 516 in private facilities, respectively, 91% and 97% were male (mean age = 38.1 and 36.7 years). There were few SES differences between groups (Table 1). There were differences in wealth quintile distribution in public facilities, while age and marital status differed in private facilities. Baseline alcohol consumption was similar in Fb + BI and Fb groups in public facilities; in private facilities, it was slightly higher in the Fb + BI group.

Fig. 1
figure 1

RCT participant selection and loss to follow-up

Full size image
Table 1 Baseline characteristics by intervention groups, public and private facilities
Full size table

Changes in Alcohol Consumption over Time

There were statistically significant decreases in alcohol consumption, the primary outcomes of interest, at almost all time points in both public and private facilities (Fig. 2 and Table 2). The average decrease of alcohol consumed by public facility participants at 6 months was 341.6 and 309.6 g, respectively, in the Fb + BI group and Fb group. In private facilities, the reductions were 314.1 and 236.4 g. There were no statistically significant differences in the amount consumed between Fb and Fb + BI at follow-up time points (Table 3).

Fig. 2
figure 2

Changes in alcohol consumption in the last week over 6 months

Full size image
Table 2 Decrease in alcohol consumption from baseline to 6 months post-intervention in public facilities and private facilities (paired t test)
Full size table
Table 3 Difference in alcohol consumption between treatment groups for each time point (t test)
Full size table

Similarly, the imputed GEE results, adjusted for SES (Table 4), still showed that alcohol consumption reduced significantly over time in public and private facilities, with similar decreases in the Fb + BI and Fb groups in both facility types. These results persisted in sensitivity analysis with raw data (Supplementary Table S.1).

Table 4 Socioeconomic status and changes in alcohol consumption in public and private facilities (GEE)
Full size table

Statistically significant covariates included gender, wealth, household size, and age (Table 4). Male participants decreased less in both types of facilities. In public facilities, participants on the poorest quintile decreased more than the wealthiest, while in private facilities, participants of the lowest three quintiles reduced less than the wealthiest quintile. The larger the household size, the larger the decrease in alcohol consumption in public facilities. An increase in age was associated with a decrease in private facilities.

Participating Facilities and Health Workers

The FGDs and KIIs with patients included 86 participants, and those with health worker had 31, for a total of 117 (Table 5). Participating public facilities included two outpatient departments and seven community clinics involving 24 clinicians (70% female) and 24 screeners (58% female), with 13 (54%) of each group took part in FGDs or KIIs. Participating private facilities included one private clinic and one private hospital outpatient department involving 5 clinicians (25% female) and 9 screeners (56% female), of those 3 clinicians (60%) and 2 screeners (22%) participated in the FGDs or KIIs.

Table 5 FGD and KII participants
Full size table

Qualitative Findings

The iterative process of thematic and latent content analysis, with constant comparison, of the data from the FGDs and KIIs plus field observations, confirmed the unexpected finding of a similar and meaningful decrease in alcohol use in both groups and revealed several likely thematic etiologic mechanisms (Tables 6–9): (1) meaningful impact on alcohol consumption, (2) fundamental shift in knowledge, attitude, and practice by health worker, (3) contributing factors to screeners delivering the FRAMES components of BIs, and (4) other research design elements supporting change.

Table 6 Theme 1: Report of meaningful decrease in alcohol consumption in: examples from both RCT groups
Full size table
Table 7 Theme 2: Fundamental shift in knowledge, attitude, and practice by HCW
Full size table
Table 8 Theme 3: Contributing factors to screeners delivering the FRAMES components of BIs
Full size table
Table 9 Theme 4: Other research design element supporting change
Full size table

Theme 1. Meaningful Impact on Alcohol Use and Other Domains Reported by All Groups

RCT patient participants in both groups reported reduced alcohol consumption with many improvements in mental and physical health and in other dimensions of their lives, which they attributed to their participation in the study, per Table 6 and these quotes:

Like before, I would say that I cannot watch a game without drinking, but right now, I can watch without any alcohol. Initially, it is like I cannot see drinks in a bar and just sit but nowadays, I can just sit next to them without touching. I am also too busy with my life which was so messy, I do my general cleaning over the weekend, I can visit my family and parents and I can get to work at the right time. (FGD men, Fb, private facilities)

Since I was married, I had never had a child with my husband, that is why am saying it is very bad [alcohol]. We have stayed for long without me getting pregnant. I have been drinking a lot. After I joined this study, I stopped completely taking alcohol and I became pregnant immediately. This child I am holding here is as a result of this study… (FGD women, Fb+BI, county 1

Corroborating these patients’ reports, screeners and clinicians noticed similar impacts (Table 6).

Theme 2: Fundamental Shift in Knowledge, Attitude, and Practice by Health Worker

This theme, divided into several subthemes, provides evidence of the changes and contributing factors to those changes, as presented below and in Table 7.

Subtheme 2.A. New Confidential, Empathic Services with Decreased Stigma, Attracting Patients

Prior to eDATA-K, neither public nor private facilities’ roles included addressing risky levels of alcohol use, according to field observations and discussions with the county and facility leadership, as well as health workers themselves at the beginning of the project. Private insurance policies specifically excluded medical care coverage for those who abused alcohol. Health workers fundamentally shifted their approach and attitudes toward people presenting with risky levels of alcohol use, from denying or delaying services to welcoming them and providing empathic assessment and interventions for the first time in these communities. Our results support that health workers developed self-efficacy and offered new evidence-based, non-judgmental, empathic, and confidential alcohol and substance use services, as well as providing better general care to those patients.

…you know that is something that is disturbing you, but you can't stop. So, if you find somebody that is willing to help you and is telling you that he is helping you and that the information is confidential, OK, you will feel uneasy at first, but if you see that it is helping you, then you become relaxed. (FGD men, Fb+BI, county 1)

I have gained knowledge and that knowledge is power… For example, my father used to take a lot of alcohol, and at one point, he happened to divorce my mother. So, I used to despise the people who take alcohol because a lot happened to me. When they come, I never attend to them... They could wait all day until another person comes in the afternoon shift to attend to them… After being mentored at Africa Mental, I discovered that these are people we can still take care of. At first, maybe it's because I didn't know how to approach them, I didn't know what to tell them, etc. Right now, I've learnt a lot and I know how to handle them better. (FGD clinicians and screeners, county 2)

Accordingly, patients in both groups reported exposure, for the first time, to new knowledge, empathy, and confidential feedback from health workers, which they appreciated and found transformative. As such, the facilities started attracting patients seeking these new services in particular:

I can say that there is a lot of change in the sense that like for instance, you find that me working in the facility, you find patients or clients coming in even when them they are not [from] around… I have seen it, so you will find that maybe some have been screened, others have heard it from their fellows who had been screened, you know, in the community. So, you find them bringing themselves, you know, to the facility. They want to know more. (FGD screeners, county 1)

Subtheme 2.B Provision of Other ASSIST-Linked BI FRAMES Components

Our finding supports the widespread delivery of all the FRAMES components of effective BI by clinicians and screeners. Provision of E, empathy, is documented in subtheme 2.A. Provision of the feedback, responsibility, and advice components is illustrated in these quotes:

If you realize someone is taking that, it is good to advise them on the consequences. Then you let them make the decision. (FGD screeners, county 1)

So, when I came here, I was just so happy, we talked and if there was anything to be shared, we did and they gave me advice. They told me the disadvantages of drugs and I saw there was not any advantage of taking alcohol. So, I can say they were ok. We talked as sisters and brothers. (KII woman, Fb, county 1 #2)

Participants in all groups gave examples of various strategies used to decrease their alcohol consumption, corresponding to the “Menu of option” (M). Strategies most frequently mentioned were finding activities to occupy oneself instead of drinking and changing social networks. Other strategies included avoiding triggering locations, using money differently, getting support from loved ones or spirituality, committing to change, refusing offers to drink, reducing slowly to quit, changing the timing of the start or end of the drinking, reviewing the information provided, and reminding oneself of the negative consequence of drinking. Also, in all groups, RCT participants expressed developing self-efficacy:

Like when I came for counselling, I think it was the second time the nurse that was attending to me, we were talking about things like self-esteem, how to stop hating yourself. She even helped me on how to be positive in life, so I started seeing myself as worthy and that I can also be somebody in life. (FGD women, Fb+BI, county 1)

Participants in the Fb + BI group elaborated more on their strategies than the Fb participants. Some clinicians, but not screeners, reported consciously building self-efficacy in their patients.

Theme 3: Contributing Factors to Screeners Delivering the FRAMES Components of BIs

Subtheme 3.A. The ASSIST Tool Design

Field observation revealed that a major contributing factor to the effective delivery of the Feedback element is that it was integrated into the ASSIST questionnaire and feedback form by design. The ASSIST-linked BI manual states p.11:

…feedback is the provision of personally relevant information which is pertinent to the client, and is delivered by the health worker in an objective way. ‘Much of the feedback given in an ASSIST-linked BI can be delivered by reading directly from the ASSIST feedback report card.’ (Humeniuk et al., 2010a)

Questions embedded in the ASSIST and information in the report card are designed to increase insight into one’s level of use and associated consequences, classifying respondents’ score in terms of low, moderate, or high-risk levels for each substance category with specific examples of health consequences. As the ASSIST questionnaire was administered by the screeners, including the provision of the feedback with the report card, it contributed to Fb group participants receiving this key element of BIs.

The Responsibility component, described as “Communicating with clients in terms such as … ‘How concerned are you by your score?’ enables the client to retain personal control over their behaviour and its consequences and the direction of the intervention” p.12 (Humeniuk et al., 2010b). As that question is asked in the feedback form, again by design, the screeners included this other FRAMES component in their patients’ interactions.

Subtheme 3.B Opportunities for Screeners to Learn and Provide all the FRAMES Components

Field observations revealed that screeners had the opportunity to develop BI skills above and beyond administering the ASSIST screening. As part of the course, clinicians and screeners observed a demonstration of effective and ineffective screening and BI. Furthermore, many screening course trainees were exposed to the BI modules of the primary care course:

Also, the screening course trainees and the primary care management trainees frequently completed the training together, advancing through the module as a group in the different facilities. This exposed the trainees of the screening course to some of the primary care management modules. (FGD clinicians, county 1)

This was due to access to one computer in most facilities, with screeners and clinicians being available at similar times to complete the training (before/after work or during breaks). Additionally, in each clinic, a copy of recommended BI steps was posted to support clinicians, and a printed copy of the self-help manual was available. With screeners motivated to help the patients, they reported accessing those.

Subtheme 3.C. Health Workers Personal Motivation to Learn and Help Those Affected by Alcohol Use

Trainees who completed either course were very motivated to do so because they desire to help others in their family and community. As such, they wanted to provide as much support as they could to those with whom they interacted. Screeners did not see their role in eDATA-K as research staff administering a tool to assess study eligibility but as concerned health workers wanting to help address an issue in their community. They derived personal satisfaction from the improvement they saw — encouraging them to continue without being asked to do so (Table 8):

Maybe I like learning new things, adding knowledge on what I have and also like he has said, our family many of them are drunkards, and I really wanted to help them. I have already started talking to some, and they have started dropping, not that they have dropped completely but at least. (FGD screeners, county 1)

Let’s say when we are here in the hospital, they believe in us according to the relationship we have outside the hospital. When in the hospital, we interact with many people and happen to socialize with them. That is an added advantage, especially to the screeners because we don‘t come from far, so those who meet with during the screening are our dear people. (FGD screeners, county 2)

Field observation also confirmed the motivation of screeners in private facilities. However, in private facilities, the screeners were not lay health workers from the communities where the patients were located and did not have as many opportunities to interact with and support participants after the initial screening.

Theme 4. Other Research Design Elements Potentially Supporting Change (Table 9)

Subtheme 4.A. Repeated Assessments by Research Staff

Repeated assessments by research staff were seen by some RCT participants as educational opportunities and a potential source of support. The baseline and follow-up assessments included questions about alcohol use and related consequences such as sexual activities risks, theft, violence, housing, ability to work, depression, suicidality, quality of life, and self-stigma. Participants, especially in county 2, and some in private facilities, reported these questionnaires, and the support from the research staff helped them realize the impact of their substance use, decrease their use, and celebrate their progress, generating self-efficacy.

Subtheme 4.B. Soap Seen as an Incentive or Reward for Change and Staying Engaged

RCT participants were given soap as a form of compensation for the time spent answering the research surveys administered by the research staff. Some RCT participants, especially in public facilities, attributed meanings to the provision of soap. They saw it as an encouragement to be “clean” and to take care of themselves and their families. Some mentioned perceiving the soap as an incentive to change, to stay engaged in the change process, or as a reward for decreasing their alcohol consumption. However, in the FGD Fb + BI private facilities, the soap was criticized, with participants wishing to receive something more valuable, like baseball caps or T-shirts, explaining they did not need soap. In public facilities, some participants would have been preferred food or money for transport to the healthcare facility.

Discussion

In summary, this is one of the largest mixed-method studies of the effectiveness of blended-eLearning on health worker’s ability to support patients in decreasing harm from alcohol use, filling an important gap in the literature. Two RCTs (one in public and one in private facilities, total n = 1,205) and an extensive parallel qualitative study (field observations and 18 FGDs or KIIs with 117 participants) indicate that LMIC trainees (lay health workers, support staff or primary care clinicians) in public and private facilities, in outpatient departments and community clinics, in rural and urban areas, can successfully change their practice, decrease their stigma, and help participants decrease their alcohol consumption with significant improvements in patients’ lives. Other studies on blended-eLearning or eLearning rarely assess patient outcomes, and even less in such a comprehensive manner, and were mainly carried out on health sciences students, not practicing health workers, with only a few studies specific to LMICs (George et al., 2014; Hew & Cheung, 2014; Lewis et al., 2014; Liu et al., 2016; Maloney et al., 2012, 2015; Marrinan et al., 2015; Sandars, 2010; Shorbaji et al., 2015; Sinclair et al., 2015; Walsh et al., 2010).

A significant finding, supported by qualitative and quantitative results, is that motivated lay health workers and support staff, who completed a short blended-eLearning course, delivered an intervention with equivalent alcohol-related patient outcomes to clinician-led (physicians, nurses, or medical officers) additional BI. Other studies have demonstrated the effectiveness of lay health workers’ interventions to decrease alcohol use risk or improve other mental disorders (van Ginneken et al., 2021), but none by training health workers through free, blended-eLearning or eLearning, to our knowledge.

There is also a paucity of qualitative studies on lay health workers’ contributions to effective interventions (Glenton et al., 2011), which is another gap that is partially addressed by our study. The confounding factor from lay health workers going beyond the provision of the ASSIST results, by offering additional support in the community, limits the purity of assessment of the added BI’s effect by clinicians to the ASSIST screening and feedback. Support staff in private facilities could not provide that extra support in the community. Despite that, the alcohol consumption decrease was similar for those who received the Fb or the Fb + BI in private facilities, suggesting that extra community support is not a necessary component of effective intervention, above the delivery of the FRAMES components with decreased stigma.

The profound change in attitude and practice in the health workers and their entire clinic attracted new patients just for that service. This finding is key, as stigma and a lack of emphasis on the need to provide services to those with risky substance use contribute to the dismal access to services in both HIC and LMICs (Corrigan & Watson, 2002; Corrigan et al., 2004; Room, 2005; van Boekel et al., 2013). Factors promoting change from educational interventions also found influential in other studies, including personal motivation, support by the leadership, an opportunity to learn elements of BI, and being part of a group changing its attitude globally (Embleton et al., 2013; Morris & Chen, 2009; Sengwana & Puoane, 2004).

The decrease by more than 235 g/week at 6 months in all groups is of unprecedented magnitude (14 g being one alcohol serving). A Cochrane review found a mean intervention decrease of − 21.56 g/week (95% CI − 31.56, − 11.55) at 6 months and 20 g/week (95% CI − 28.36, − 11.8) at 1 year (Kaner et al., 2018). A few studies have reported a reduction of around 100 g (Kaner et al., 2018). Naturally, studies with higher baseline consumption may observe a larger decrease (Kaner et al., 2018). In our RCTs, the baseline consumption was high, averaging 326 to 423 g/week. That level was comparable to other studies with only men, who constitute more than 90% of our sample (Kaner et al., 2018).

Social desirability bias, which could have increased by the end of the study from all the attention paid to substance use issues and regression toward the mean, could have inflated the decrease in consumption (Dawson-Saunders & Trapp, 1994). Furthermore, some may find the measurement of reduced alcohol consumption in grams as an insufficient outcome to demonstrate a meaningful reduction in alcohol-related adverse consequences, believing that only sobriety matters as an outcome. However, findings from the field observation, FGDs and KIIs, with patients and health worker, in all settings and groups support the plausibility of a sizeable quantitative decrease in alcohol use resulting in a meaningful impact on patients’ lives.

Vast differences exist between related to patients and service providers in LMICs vs. HICs in terms of availability and level of training, access to primary care, general population level of awareness of substance use impacts, cultural norms, and social contexts (Awenva et al., 2010; Babor et al., 2013; O'Donnell et al., 2013). Such differences might have contributed to the unexpectedly much-larger decrease, considering the novelty of the intervention, newer expression of empathy, and a significant reduction in stigma, as well as tremendous new knowledge and skills transfer to patients, as observed in our qualitative findings.

Previous research showed that empathy is a strong mediator of the therapeutic effect (Moyers & Miller, 2013). The courses explicitly covered stigma and empathy, both in the written material, exercises, in-person demonstrations, and experiential learning activities, with some meant to trigger reflection on stigma, as per best practices to decrease stigma (Clair et al., 2019, in press; Livingston et al., 2012).

Furthermore, the first three FRAMES components (feedback, responsibility, and advice) are key to effective BI (Humeniuk et al., 2010a), and there was ample evidence that these three components were delivered to participants in all groups and settings. The design of the ASSIST screening, with its insightful questions, and the feedback form, which covers some elements of BI, seems to contribute to the delivery of those three components of FRAMES, per our qualitative findings. Other studies using the ASSIST screening tool have observed significant and similar decreases in both groups (Assanangkornchai et al., 2015), which supports that the ASSIST feedback tool can be effective. The repeated use of the ASSIST and other research surveys, at 3 and 6 months, might have generated further decrease after the initial intervention. Therefore, our studies provide some insight into improving methods for further trials. Further research could confirm if the delivery of the ASSIST results via its feedback form is an effective intervention. Also, since the soap seemed to motivate some participants, as a reward or as an encouraging gesture, it could be interesting to study such incentives as part of contingency management interventions for substance use in LMIC populations having difficulties accessing even such a basic necessity of life.

Another limitation is the small number of screeners in the private facilities with the FGD, including both screeners and clinicians and the lack of identifiers of who was speaking in the audio recording. However, field observations align with the interpretation of the FGDs and KIIs.

This study has several strengths, including that it was conducted with health workers in a variety of clinical settings where Kenyans usually seek care, addressing an important gap in studying intervention for alcohol use in LMICs. It is a large mixed-method study with rich inferential qualitative findings aiding the interpretation of the RCTs’ results, offering an opportunity for triangulating findings and contextualizing conditions that supported those results (Fortin et al., 2021; Pope et al., 2007; Spillman, 2014; Todd, et al., 2004). The loss to follow-up was small in public (7%) and larger (30%) in private facilities. Using intention-to-treat analysis with sensitivity analysis with or without imputation showed similar outcomes, congruent with a lack of a significant attrition bias. The 6-month follow-up period is also adequate, as per the Cochrane review, with showed minimal changes from 6 to 12 months (Kaner et al., 2018). Other strengths include that the randomization was carried out with high fidelity, with few participants in the Fb group receiving a formal BI by a clinician and appropriate concealment of group allocation during quantitative analysis. Furthermore, our study included qualitative researchers of multiple backgrounds, and the interpretation of the findings was corroborated with participating health workers and the research staff who carried out the field observations. In conclusion, and consistent with other studies (van Ginneken et al., 2021), our findings found that motivated lay health workers and support staff can learn and deliver BI as effectively as primary care clinicians when using tools such as the ASSIST report card and covering most FRAMES components. Furthermore, significant changes in health workers’ knowledge, skills, and stigmatizing vs. empathic attitudes led to significant reductions in alcohol overuse and meaningful patient life improvements. The acquisition of such transformative competencies through blended-eLearning courses was feasible even in an LMIC, regardless of facilities being private or public, outpatient department or community clinics, and in rural or urban areas. Educators can use the eDATA-K blended-eLearning NextGenU.org competency-based courses created through assembling freely available high quality online learning content with mentored and peer activities to teach effective screening and brief intervention. These findings suggest a major scale-up of human resources to address the significant burdens from risky alcohol use is feasible through blended-eLearning at multiple provider levels, including lay health workers, using the eDATA-K courses. In addition, the feasibility has been highlighted in the recent COVID-19 pandemic during which several universities have used these courses as they were forced to switch to online learning, which have been used by several universities when they needed to switch to online learning during the COVID-19 pandemic. The model could also potentially address other health priorities by creating other continuing education courses or health education curricula in collaboration with NextGenU.org.