Diabetes mellitus remains one of the leading causes of mortality worldwide and represents a significant public health challenge. In 2024, an estimated 589 million adults (11.1% of the global population aged 20-79 years) were living with diabetes, and this figure is projected to rise to 853 million (13%) by 2050 [1]. Notably, around 95% of this increase is expected to occur in low- and middle-income countries [1]. The global burden is primarily driven by type 2 diabetes (T2D), which accounts for more than 90% of diagnosed cases and is associated with adverse cardiovascular, metabolic, maternal, and neonatal outcomes [1].

In Argentina, the prevalence of diabetes mellitus increased from 5.4% in 2021 to 14% in 2024, with estimates suggesting it may reach 15.4% by 2050 [1]. Combined with high levels of obesity (25.3%) and physical inactivity (44.2%), this trend highlights the urgent need to enhance prevention, early detection, and chronic disease management across the national health system [1,2]. The prevalence of gestational diabetes is estimated at 9.8% [3], further underscoring the national public health concern.

Considering that over 40% of the Argentine population relies exclusively on the public health care system [4], several initiatives have sought to strengthen T2D and gestational diabetes management at the primary care level. Strategies such as health care team training, patient education, scheduled follow-ups, and reminder phone calls have shown improvements in metabolic control and service use, particularly when both health care providers and patients were involved [5,6]. Other multicomponent programs have combined team training with educational outreach visits and digital health strategies, improving access to laboratory tests and ophthalmologic care, blood pressure control, and follow-up visits at 12 months [7]. Despite these efforts, sustaining effective strategies over time remains challenging, especially in resource-constrained primary care settings [8-10].

Mobile phones are now widely used to deliver health information remotely, often through text messaging [11]. This modality provides people with accessible health information and practical guidance to support daily self-management of their health. Given its relatively low cost, potential scalability, and the fact that text messaging does not require advanced technological skills, mobile communication offers a promising way for public health systems seeking to improve care for people living with chronic diseases [12]. These digital tools provide practical and cost-effective alternatives to conventional health promotion strategies, and studies evaluating these interventions generally report high feasibility and acceptability [13].

Argentina has high mobile phone and internet connectivity: 92% of the population uses a mobile phone and 86% uses the internet (Q4 2021), placing the country among the most digitally connected in Latin America [14,15]. WhatsApp is also widely used; 86% of internet users report weekly use in 2021 [16]. This high level of digital connectivity stands in contrast to Argentina’s persistent social inequalities, where nearly 40% of the population currently lives below the poverty line [17]. As a result, mobile technology is deeply integrated into daily life, particularly through WhatsApp, the most widely used and culturally accepted messaging platform in the country, offering a feasible channel for delivering health interventions.

Between 2021 and 2024, a Diabetes Prevention and Care Program was implemented within Argentina’s public health system, targeting low-income areas of 3 locations in Argentina: Salta, San Juan, and Tandil (Province of Buenos Aires). The program included training for health care teams, as well as screening and care strategies for 3 target populations: people with diabetes, those at high risk of developing the disease, and pregnant women.

A key component of the program was a WhatsApp-based intervention that delivered educational messages aimed at enhancing knowledge, promoting healthy lifestyle changes, supporting medication adherence, and functioning as motivational prompts to enhance self-management behaviors [18].

Using WhatsApp via business APIs (application programming interfaces) allows us to pair a familiar channel with delivery and read status indicators that enable auditable reach, and to implement an explicit, channel-specific opt-in to document and reconfirm consent. This supports a minimally intrusive, scalable approach to public health messaging within the public primary-care system. This study aimed to describe the implementation and process evaluation of this WhatsApp-based intervention within Argentina’s public health system, using the Carroll Implementation Fidelity Framework [19], focusing on the challenges encountered, strategies used, and lessons learned across 3 target populations.

This study was conducted between 2021 and 2024 in 3 urban settings of Argentina: Salta (capital city of Salta Province, northwest region), San Juan and Gran San Juan (covering 4 of the 5 health zones of San Juan Province, Cuyo region), and Tandil (a midsized city in Buenos Aires Province).

A total of 40 primary care centers (PCCs) were included. Each PCC is responsible for a geographically defined catchment area, composed mostly of low-income neighborhoods where residents rely primarily on public health services. Eligible PCCs had to be located in socioeconomically disadvantaged areas, report at least 800 monthly outpatient visits, use community health workers (CHWs), and provide free medications through the national and provincial essential medicines supply program. PCCs were selected in coordination with local health authorities to ensure comparability with other centers in each city.

The population included adults residing in the catchment areas of the selected PCCs. Eligibility criteria included adults (aged 18 years or older) with T2D, people at moderate to high risk of developing T2D, and pregnant women. Participants were recruited through two complementary approaches: (1) opportunistic recruitment at PCCs, when individuals attended for routine visits or other health reasons; and (2) community-based recruitment, conducted by trained health workers through door-to-door household visits within the PCC catchment areas and community outreach activities (eg, neighborhood health fairs and community centers). CHWs used standardized screening forms, and received prior training on recruitment protocols, and field activities were supervised by PCC coordinators to ensure consistency across sites.

Trained health care workers administered the validated Finnish Diabetes Risk Score (FINDRISC) questionnaire which was used to screen for people at moderate or high risk of developing T2D [20]. Data were collected using an electronic form developed in REDCap (Research Electronic Data Capture; Vanderbilt University) [21,22]. CHWs and nurses, previously trained to use the system, gathered sociodemographic and clinical information, including age, gender, cardiovascular risk factors, diabetes history, lifestyle behaviors (eg, fruit and vegetable intake and physical activity), and current treatments. Anthropometric measurements, such as height, weight, and abdominal circumference, were also collected to calculate diabetes risk using the FINDRISC tool integrated into the system.

Participants with moderate or high risk of T2D, or with diabetes, or pregnant, were enrolled in this study.

The following components were required to operate the messaging system: (1) data repository: REDCap was used to store participant information, message templates, and the algorithms for message delivery; (2) Twilio and WhatsApp integration: a verified Meta Business account was required to configure and authorize the WhatsApp Business API via Twilio; (3) cell phone lines: dedicated local numbers were acquired through Twilio to enable automated message delivery via WhatsApp; (4) preapproved message templates: WhatsApp mandated that all message templates be reviewed and approved before deployment; and (5) API-based communication: the control system communicated with Twilio via a PHP-based API, while Twilio interacted with the control system via webhooks.

As illustrated in Figure 1, each message was generated in REDCap through an external module that applied Meta-approved templates. Templates included a common header, a personalized greeting with the participant’s first name, and the educational content scheduled for that week. Each message was tagged with both a participant ID and a message ID.

Delivery and read receipts were retrieved from Twilio logs and matched to participants using phone number and message text. These logs were consolidated weekly through R (R Project for Statistical Computing; R Foundation) scripts to cover the full study period.

Contact details were stored in a separate REDCap project from health data, and linkage was made only via a study-specific participant ID. Phone numbers were retained solely to enable message delivery.

Access to Twilio logs was restricted to account administrators, who downloaded them to a secure shared folder for statistical analysis. Twilio retained logs of up to 10,000 lines (≈3-6 months of activity), which were then consolidated. Data transfers were encrypted in transit (https). Data were stored on cloud servers without encryption at rest. No additional procedures were implemented to minimize exposure of phone numbers, which were necessary for message delivery.

Participants retained the right to withdraw at any time, either by replying with an opt-out request or by directly contacting this study’s team using the information provided in the consent form.

The evaluation was based on Carroll Implementation Fidelity Framework [19], which assesses both the implementation process and the extent to which participants received the intervention as intended (Table 1).

This framework has been widely applied in previous studies [27-29] to evaluate the consistency and quality of intervention delivery. It focuses on key domains such as adherence, quality of delivery, participant responsiveness, and contextual factors.

In our process evaluation, using data collected at various stages of the intervention, adherence was examined by determining whether the planned activities were carried out as intended. This included verifying that all key messages, including opt-in (ie, initial confirmation of consent to receive weekly messages), welcome, weekly, and final messages, were successfully generated and delivered to participants. Additionally, we analyzed the coverage of the intervention across the 3 participant subgroups: people with T2D, those at moderate or high-risk of developing diabetes, and pregnant participants.

Participant responsiveness was measured by calculating the opt-in acceptance rate. We also tracked the proportion of messages that were read by participants relative to those delivered. Participants’ withdrawal from the intervention was also documented.

Contextual factors that may have impacted implementation were also considered, including technical issues and challenges reported by participants.

Descriptive statistics were used to summarize sociodemographic and clinical characteristics overall and by target group, and to describe implementation fidelity indicators. Group comparisons assessed differences in access (valid cellphone number) and acceptability (opt in). As a sensitivity analysis, multivariable logistic regression models including sex, age, educational attainment, and comorbidity category were fitted for each outcome; odds ratios with 95% CIs are reported in Table S1 in Multimedia Appendix 2 and Table S1 in Multimedia Appendix 3.

This study was reviewed and approved by 3 institutional ethics committees in Argentina: the Institutional Review Board for Research Studies (Consejo Institucional de Revisión de Estudios de Investigación) of the Hospital Privado de Comunidad in Mar del Plata, Province of Buenos Aires (approved date: July 7, 2022); the Research Ethics Committee of the Ministry of Public Health of San Juan (May 12, 2021); and the Provincial Commission for Biomedical Research of the Ministry of Public Health of Salta (March 8, 2021).

Participation in this study was voluntary, and written informed consent was obtained from all participants during the screening process. Consent covered both study data collection and the receipt of digital messages. Participants confirmed their enrollment by explicitly opting in to the messaging system. They retained the right to withdraw at any time, either by replying directly to the system or by contacting this study’s team using the information provided in the consent form. Participants did not receive financial or material compensation for their participation in this study, and no identifiable images of individual participants are included in this paper or its supplementary materials.

Confidentiality was strictly maintained, and all procedures complied with Argentina’s Personal Data Protection Law (Ley 25.326). The use of WhatsApp Business API required attention to privacy and data protection. Messages were sent using standardized templates and contained no personal health information. Participant data were stored in REDCap with password-protected access. While messages were transmitted over encrypted https connections, data were stored on secure cloud servers.

A total of 11,029 participants were enrolled in this study, of whom 9983 (90.5%) had a valid mobile phone number registered in the system. Among these, 32.8% (3276/9983) had a diagnosis of T2D, 53.3% (5320/9983) were identified as being at moderate or high risk based on the FINDRISC questionnaire, and 13.9% (1387/9983) were pregnant women (Figure 2).

Table 2 summarizes the main sociodemographic and clinical characteristics of participants enrolled in the Diabetes Prevention and Care Program.

Most participants were women (n=7725, 77.4%), with a mean age of 44.5 (SD 15.6) years. Participants with a valid cellphone number tended to be slightly younger compared to those who could not be reached. Age distribution also varied across target groups, from a mean of 51.9 (SD 13.6) years among people with diabetes to 26.4 (SD 5.9) years among pregnant participants.

More than half of all participants (n=5332, 53.4%) had completed or partially completed secondary education. Among those with a valid cellphone number, a higher proportion had completed secondary education, whereas among participants without a valid phone number, the majority had only completed primary education. In target groups, the proportion with secondary education was higher among the pregnant women group (1008/1387, 72.7%). In contrast, people with diabetes showed lower educational attainment, with 46.3% (1516/3276) having completed only primary education or less.

Comorbidities were common, affecting 76% (7584/9983) of participants overall. Those with a valid cellphone number more frequently reported two or more conditions.

The most commonly reported delivery issues included phone numbers not enabled for WhatsApp, users who had not accepted WhatsApp’s terms of service, outdated or unsupported versions of the WhatsApp app, and improperly formatted phone numbers.

Beyond individual delivery issues, the system faced broader contextual challenges. Some technical difficulties were encountered during the delivery of educational text messages via WhatsApp using the Twilio interface. At some point of this study, during periods of high system load, the server’s processing capacity was insufficient to manage the volume of data exchange until the server was upgraded. This occasionally resulted in delays in the registration of message delivery status, particularly when response signals were received before outgoing messages could be properly logged. Additionally, for pregnant women, some opt-in messages could not be generated due to missing or incorrectly coded gestational age data.

Additional limitations included suboptimal code performance and intermittent communication failures between the messaging control system and the REDCap platform. The underlying code was subsequently optimized to handle the increased message volume as more participants were enrolled.

To ensure accurate message tracking, Twilio logs were manually retrieved every week and matched with REDCap data using participant phone numbers.

This study examined the implementation of a WhatsApp-based intervention for people with T2D, moderate or high risk of developing this condition, and pregnant women, in the context of Argentina’s public health system, focusing on adherence, engagement, delivery quality, and contextual barriers. The intervention followed a structured delivery process, from opt-in messages to weekly health education content and final closing messages.

There is robust evidence that text messaging interventions can support chronic disease management, such as in T2D [30,31]. However, limited evidence exists on how these programs are implemented in routine primary care settings, particularly in low-resource environments [32]. Our findings contribute to addressing this gap by documenting the novel use of WhatsApp for digital health messaging in the real-world public health sector in Argentina, providing insights on feasibility, potential scalability, and engagement in resource-constrained settings.

While most messages were successfully delivered, a small proportion were affected by technical issues, such as invalid cellphone numbers (approximately 1046/11,029, 10%) or intermittent connectivity problems. Similar challenges have been reported in other mobile health interventions for chronic disease [27,33]. Steinman et al [33] reported that 40% of participants did not receive messages due to frequent changes in personal mobile numbers driven by cost-saving strategies, while Van Olmen et al [27] found that nearly one-third of project phones became unusable because of subscription cancellations, breakage, or loss, despite providing SIM cards and devices. Both studies also described infrastructural limitations, such as weak network coverage and recurrent software errors [33].

In comparison, the proportion of unreachable participants in our study was lower, suggesting that integrating WhatsApp within existing care structures, with validated contact information and proactive follow-up, may help mitigate some of these barriers. Nonetheless, the percentage of nonreached people highlights possible inequities: those who could not be reached tended to be older, more often male, and with lower educational levels compared to those successfully contacted. Although the differences were modest, these patterns reflect potential digital exclusion linked to socioeconomic and demographic disparities. These trends are consistent with prior evidence showing that older age is negatively associated with adoption of mobile health technologies, while higher educational attainment increases access and use [34]. These patterns could be related to the lower digital literacy, physical or cognitive limitations, and greater concerns about privacy for older adults, whereas people with lower education are more likely to experience restricted access to smartphones, unstable connectivity, or difficulties interpreting health messages. Taking into account the inequities in phone ownership, access, and use [35], future interventions should consider tailored recruitment strategies to ensure that no population is systematically excluded due to digital access limitations, such as providing guidance for phone access and registration, using complementary communication channels for those without WhatsApp, and improving the process of validating contact information.

Another critical component of implementation is participant engagement and acceptability. In this study, we assessed these aspects by examining the read rate of the initial opt-in message, the proportion of participants who agreed to receive weekly messages, the weekly message read rates during the intervention, and dropout rates over time. Engagement levels were high: approximately two-thirds of participants accepted the intervention, over 90% (7778/8538) read the opt-in message, weekly read rates averaged 82.2% (SD 29.8), and only 6.1% of participants dropped out. Our findings are consistent with previous research [24,31,36], showing that text messaging interventions are generally well-received by both participants and health care providers. The engagement observed in this study may be related to the intervention’s carefully designed structure, including its integration into routine primary care practice and the proactive efforts of health care teams to validate contact information.

These findings support the potential of WhatsApp as an effective platform to reach and engage broad populations and sustain participation, particularly when integrated into existing care structures. Compared to SMS, WhatsApp might offer advantages such as greater user familiarity, support for multimedia content, and faster message visibility, which could enhance engagement in low-resource settings.

Participant engagement levels differed across the 3 target groups. People with diabetes showed the highest acceptance and lowest dropout rates, likely reflecting a greater perceived need for ongoing support. Those at risk of diabetes had the lowest acceptance, while pregnant women showed higher acceptance of the messages than the at-risk group, but had a higher dropout rate. In this sense, although there is evidence that text messaging interventions can be effective during pregnancy, particularly for promoting healthy behaviors [37-40], the higher dropout rates observed in this group may reflect the competing demands pregnant women often face, such as medical appointments, childcare responsibilities and work, as well as potential preferences for more personalized messaging which may limit their ability to remain engaged with supplementary programs [41,42].

In our study, people identified as being at risk of developing diabetes were the least likely to accept the intervention. This group was, on average, younger than those with diabetes, which may have contributed to a lower perceived awareness, diminished need for support, and reduced motivation to engage with preventive interventions. These findings highlighted the importance of designing targeted strategies that enhance risk awareness and encourage participation in early-stage prevention efforts. Other studies have reported comparable trends, highlighting that engagement levels significantly impact long-term outcomes, particularly in prediabetes self-management programs [43,44].

Sociodemographic factors also influenced acceptability in our study, with men, older participants, and those with lower educational attainment consistently less likely to opt in across all target populations. These patterns underscore that even within targeted populations, digital interventions may reach participants unevenly, potentially reinforcing existing sociodemographic inequities. An analysis of barriers and facilitators of mobile health adoption among people with low socioeconomic position supports these trends [45], identifying multiple factors such as digital literacy, trust, and need for in-person support, that are critical for successful adoption and use. Similarly, a systematic review of mobile health use [34] found that women were more likely to adopt mobile health services than men, indicating that lower engagement among men may reflect differences in health-seeking behavior and health awareness. In light of these findings, tailored strategies that consider gender, age, and educational level, such as proactive outreach, simplified content, and personalized relevant messaging, may enhance relevance and comprehension and could help maximize engagement across all subgroups [35].

Another key aspect in our study was the resolution of technical barriers, particularly those related to the high volume of data processed. One of the main issues encountered was system overload due to the high rate of outgoing messages, which exceeded the initial server’s capacity. This led to delays in updating message delivery status, complicating the integration with the REDCap platform and obstructing real-time monitoring. To address these challenges, substantial improvements were implemented to both the server infrastructure and the system’s code. The server was upgraded with expanded memory and transitioned to a more robust disk configuration, increasing its capacity to handle large data volumes. In parallel, code optimizations, particularly in the communication protocols with REDCap, streamlined data processing and improved log matching. These enhancements strengthened the reliability of message delivery and tracking, reduced the need for manual interventions, and improved overall system efficiency.

Additionally, using WhatsApp via the Business API, a familiar and widely accessible platform in Argentina, minimized technical barriers. The API enabled automated logging of delivery and read statuses, and captured explicit, channel-specific opt-ins and opt-outs, providing auditable, high-fidelity process data. Together, these technical improvements contributed to the intervention’s feasibility.

This real-world implementation study demonstrated that WhatsApp-based text messaging interventions were successfully integrated into public health services; however, several limitations should be considered when interpreting these results.

First, while messages were tailored by participant subgroups (people with T2D, at-risk of diabetes, or pregnant participants), they were not customized at the individual level. This lack of individual tailoring may have limited the intervention’s acceptability and adherence, as more personalized content or follow-up could have increased relevance and engagement [11]. However, we implemented a local validation of the message’s content at the regional level and included data privacy warnings in some messages, thereby enhancing the contextual relevance and participant trust.

Second, the intervention included general reminder messages encouraging participants to attend medical appointments, but lacked an individualized reminder system. Evidence suggests that individualized reminders, with specific dates and times, can significantly improve attendance in primary care settings [31,46]. Future integration with electronic health records could enable automated, personalized notifications. Third, another limitation of this study is the potential for selection and participation bias. Participants without a valid cellphone number who could not be reached differed modestly in terms of age and educational level compared to those who could be reached. Participation in the intervention also depended on willingness to opt in, which may have led to the overrepresentation of individuals who are more digitally connected, engaged with the health system, or have higher health literacy. As a result, engagement estimates may not be fully generalizable. Future implementations should consider strategies to enhance digital inclusion and monitor differential participation from the outset. Additionally, randomized controlled trials could provide complementary high-quality evidence on intervention effectiveness.

Fourth, despite WhatsApp’s widespread use in Argentina, engagement with messages varied across participants. This suggests that high platform penetration alone does not guarantee acceptance or behavioral change. Future adaptations should consider factors such as message content, perceived relevance, and user trust to optimize engagement across different subgroups. Finally, the engagement metrics reported reflect process outcomes rather than behavioral or clinical impact. Changes in self-care practices or glycemic control are beyond the scope of this evaluation and will be analyzed separately alongside other components of the Diabetes Prevention and Care Program.

This study evaluated the implementation of a WhatsApp-based text messaging intervention for diabetes prevention and care in 3 locations within Argentina’s public primary health care system. The intervention successfully reached key target groups (people with diabetes, individuals at risk, and pregnant women), demonstrating the feasibility of integrating digital messaging into routine care. However, engagement varied across participants, and several technical and contextual factors likely influenced message adherence and acceptance.

Integrating digital tools into primary care and adapting them to local contexts can strengthen service delivery, but future efforts should focus on improving interoperability with electronic health records, personalizing content by cohort and comorbidity profile, and ensuring equitable access for populations with limited digital connectivity.