This work was conducted in Malawi, a country in Southeastern Africa with 19 million people. Malawi has a 10.6% adult HIV prevalence and 1.1 million people living with HIV [13]. Malawi has a mature HIV program that is approaching the 95-95-95 targets [13]. Malawi has long had a passive index-based program, with the Department of HIV/AIDS promoting routine distribution of “family referral slips.” In early 2018, through a PEPFAR-led demonstration project, implementing partners began introducing ACT in dozens of facilities, and in early 2019, voluntary assisted partner notification, Malawi’s ACT approach, was adopted as part of Malawi’s national HIV testing policy.

Malawi has a dire human resource shortage with less than 2 physicians, 0.02 psychiatrists, and 0.01 psychologists per 100,000 people, some of the lowest global rates [14,15]. To address its formidable HIV burden with limited human resources, Malawi has task shifted many HIV-related activities, especially counseling tasks, to lay cadres. In 2015, to respond to gaps in HIV testing, Malawi introduced a cadre of HIV diagnostic assistants to improve coverage of early infant diagnosis, viral load testing, and rapid HIV antibody testing [16]. HIV diagnostic assistants are lay persons with secondary education and 4 weeks of preservice training. In its first year, nearly 1200 HIV diagnostic assistants were deployed to 450 facilities, resulting in improved diagnostic indicators [16]. Community health workers hired by local nongovernmental organizations are another lay cadre responsible for a range of HIV-related tasks, including community tracing. HIV diagnostic assistants and community health workers conduct most ACT implementation [17]. HIV diagnostic assistants typically diagnose HIV-positive indexes and support contact elicitation and selection of ACT options, and community health workers typically conduct tracing.

Throughout Malawi’s national program, nongovernmental organizations implementing partners play a critical role in hiring and supervising these lay cadres. Tingathe Program, one of the largest PEPFAR-implementing partners in Malawi, and the implementation lead on this research, was initiated by Baylor College of Medicine Children’s Foundation-Malawi in partnership with the Malawi Ministry of Health. Tingathe takes a family-focused approach to the HIV epidemic by supporting the provision of high-quality, comprehensive HIV services. Tingathe has over 10 years of experience implementing HIV testing, care, and treatment programs in Malawi and currently provides support to Malawi’s HIV care and treatment program with more than 1000 staff in 95 Malawi Ministry of Health facilities.

As Malawi’s National HIV testing program evolved from a client-led passive referral model to a provider-supported ACT model, we sought to understand the barriers and facilitators to ACT implementation. This process was guided by the Consolidated Framework for Implementation Research (CFIR) [18]. The CFIR is a determinants framework comprised of five domains (intervention characteristics, individual characteristics, outer and inner organizational settings, and processes) and 39 constructs within these domains that interact with one another [18].

We were also interested in digitizing some or all of these implementation strategies. To facilitate meaningful and sustainable digital programs, technological, human, and cost features must be taken into consideration. We used human-centered design thinking to inform potential adaptations of our implementation strategies into a digital platform. Human-centered design thinking is an approach that integrates the possibilities of (1) technology, (2) the preferences of people, and (3) the requirements for business viability [19] in the process of product develpoment.

In October 2019, we conducted a series of in-person focus group discussions, each 2 hours long, on (1) barriers and facilitators to ACT implementation and (2) the potential role of digital modalities to enhance the program. Tingathe staff were considered key informants and were purposively selected if they held ACT training as well as supervisory and clinical mentorship roles within the organization (N=10). Thus, they had routine contact with hundreds of ACT implementers at dozens of facilities. We also conducted one 2-hour-long, semistructured in-depth interview with a key informant who was not available for the focus group discussions.

All participants provided informed consent to participate. The discussion was facilitated using a semistructured guide. Data were audiorecorded, summarized, and transcribed. Data were reduced into themes organized within the constructs of the CFIR and human-centered design.

The study was reviewed and approved by the National Health Science Research Committee (20/01/2467), the local ethics committee in Malawi, and the Baylor College of Medicine Institutional Review Board (H-47655). Informed consent was obtained from all participants.

Of the 11 key informants, 3 (27%) were district-level supervisors, 4 (36%) were facility-level supervisors, 2 (18%) were health care workers who provided ACT services, and 2 (18%) were data collection supervisors; 5 (45%) were female, and 6 (55%) were male. All were employed by Tingathe program.

Themes arose within the five following domains of CFIR, which guided our implementation package development process: (1) intervention characteristics, (2) individual characteristics, (3) outer organizational settings, (4) inner organizational settings, and (5) processes.

With respect to intervention characteristics, complexity of ACT implementation was identified as a formidable barrier. Complexities included sensitivities around discussing sexual behavior with indexes, multiple potential contacts for each index, multiple tracing options for each contact, challenges with obtaining correct locator information for sexual contacts, and concerns around index safety.

This level of complexity, along with the minimal preservice training of lay health workers delivering ACT [16], exposed a set of individual characteristics, which were low competence around basic training principles and poor self-efficacy among both health workers and supervisors. For example, Tingathe staff reported that health workers frequently were uncomfortable taking sexual histories, did not understand which partners needed to be tested, and did not know how to help indexes decide which tracing methods to select. Supervisors suggested that modeling these counseling behaviors would enhance skills.

In the outer setting, diverse patient needs of both indexes and contacts were not being met, as counseling was conducted generically in a non–client-centered manner and without sensitivity to potential stigma and intimate partner violence or abandonment.

Networks and communication were identified as challenges in the inner setting: health workers in many sections of a single health facility (eg, antenatal care, pediatrics, and antiretroviral therapy) interacted with potential indexes and contacts, but responsibilities were not clearly delineated, and coordination between health workers was minimal.

To address these challenges, several processes were proposed: training that consisted of engaging strategies (role modeling) and planning strategies (practice and feedback) to address perceived intervention complexity, low health worker competence and self-efficacy, as well as patient needs. To address network and communication challenges, group problem-solving approaches that incorporated reflection and evaluation were suggested.

Synthesizing these findings, we arrived at two sets of implementation strategies, which are as follows: (1) enhanced health worker training to improve ACT counseling sensitivity and competence; and (2) group problem-solving to facilitate ACT coordination (Figure 1). This set of strategies aligns with a seminal review showing the combination of training and group problem-solving to frequently improve health worker practices in low- and middle-income countries (LMIC) [20]. Our specific training and problem-solving approaches were guided by theory and evidence, which will be explained here in detail.

Themes were also categorized within human-centered design thinking domains, which guided the adaptation of these implementation strategies into a blended learning delivery platform (Figure 2). The results were organized around considerations of (1) technology, (2) desirability, and (3) the requirements for business viability [19].

Applied to rural Malawian health facilities, important technological challenges were discussed, including limited internet connectivity, limited computer literacy, and nonuniversal personal mobile devices, a theme observed in comparable studies in the region [38,39]. These considerations led us to deliberate on content that does not require continuous internet connectivity. They also suggested the importance of user-friendly applications, as has been observed in similar settings [40].

To address desirability for Malawian health workers, participants proposed content that was primarily audio- and video-based to ensure limited reading for this low literacy audience. Studies in similar low-resource environments also proposed similar multimedia elements to address challenges with computer literacy [41]. Participants suggested content be created with Malawian cultural and linguistic considerations in mind. Smaller segments were proposed to enable learning to occur in brief increments when workload permits.

With respect to business aspects, we selected tablets, which are less costly than laptops. We opted for simple visuals and basic editing to limit the budget. For digital learning packages to be cost-effective, it is important to minimize up front development costs, maximize the number of users, or, preferably, do both simultaneously.

We synthesized findings from the formative work to digitize the package. Based on the WHO’s recommendation that digital modalities for training health workers in LMICs complement, rather than replace, face-to-face learning, we opted for a blended learning approach [42,43]. “Blended learning” combines the best features of digital learning (ie, quality, consistency, and convenience) with the best features of face-to-face learning (ie, interactivity and group engagement) [44,45]. Blended learning is typically more effective than either electronic or face-to-face learning alone for acquiring new knowledge and skills [25,46], and specifically for improving health worker counseling and communication [47-49].

The resulting blended learning implementation package translated our training plus group problem-solving approaches for the possibility of decentralized delivery. We included both individual asynchronous learning sessions and synchronous interactive small-group sessions. The individual learning sessions contain descriptions of ACT skills, video vignettes modeling these skills, and embedded questions assessing comprehension. The small-group learning session contains tablet-guided practice role plays, individual feedback, and facilitated group discussions. It was designed for facility-level conduct guided by ACT data. We developed a model where all health workers involved in ACT would meet to examine their facility’s performance, identify gaps in performance, and propose actionable solutions. Each meeting would review progress toward proposed solutions, as well as their impact on ACT indicators.