Bangladesh, an LMIC in South Asia with a total population of 171.5 million, has undergone major demographic and epidemiologic transitions in recent decades, and NCDs now account for approximately 70% of all deaths nationally [40]. The prevalence of hypertension and diabetes among Bangladeshi adults is estimated to be 27.4% and 9.8%, respectively [41,42]. The awareness, treatment, and control rates among people living with these conditions remain alarmingly low [41,42]. Furthermore, tobacco use, insufficient fruit or vegetable intake, and overweight are highly prevalent [43]. Over the past decade, the Government of Bangladesh has tried to implement national multisectoral actions to strengthen NCD care, most notably through the establishment of dedicated NCD care delivery points (known as “NCD Corners”) in subdistrict hospitals (“Upazila health complex” in local language) since 2011. The initiative’s planned decentralization to village-level primary care facilities (community clinics [CCs]) has not materialized [41]. Consequently, access to public NCD services remains severely limited in rural Bangladesh [41]. The Simple app has been progressively implemented across primary NCD care facilities in Bangladesh since 2020. As of current reporting, more than 180 facilities use the platform to manage over 350,000 hypertension and diabetes patients. The app’s integration with the national health management information system enables real-time performance monitoring and feedback for health facilities.

This hybrid effectiveness-implementation trial simultaneously evaluates intervention effectiveness and implementation strategies [44], using a 3-arm, quasi-experimental design. The aims of this trial are (1) to evaluate the effectiveness of a multicomponent, decentralized care model for hypertension and diabetes management within the public primary care systems, compared to both usual care and a standalone digital health intervention (Simple app); (2) to examine implementation processes, including explanatory factors influencing intervention effectiveness and barriers to and facilitators of delivery and sustainability; and (3) to undertake an economic evaluation. We hypothesize that compared with usual care, the multicomponent decentralized primary care—supported by the Simple app—will improve all steps along the hypertension and diabetes care continuum. Conversely, we hypothesize that the mobile health intervention alone (Simple app) may improve BP and glycemic control compared with usual care but will have a limited impact on rates of screening, diagnosis, and treatment; multicomponent integrated care will lead to a higher treatment success rate compared to the mobile health intervention alone.

The study is being conducted across 3 subdistricts in the Dinajpur district, Rangpur division, northern Bangladesh (Figure 1 and Table 1). The multicomponent intervention is being implemented in the Chirirbandar subdistrict, while the digital health–only intervention is being implemented in the Parbatipur subdistrict. The Biral subdistrict serves as the reference site. Primary outcomes include BP and glycemic control rates, assessed by population-based repeated cross-sectional surveys with independent samples, supplemented by facility-based prospective cohort data. Additionally, a mixed methods process evaluation is being conducted to capture the quantity, fidelity, adaptations, reach, and context of the interventions. The study duration is 36 months, including an intervention period of 24 months.

This study was approved by the Institutional Review Board of James P Grant School of Public Health, BRAC University (reference number: IRB-16 November-23‐041). Written informed consent was obtained from all participants. Identifying information associated with the study participants will be kept confidential through unique identifying numbers on all paper forms and computer-based files to protect privacy and ensure confidentiality of data being collected. Participants received no compensation for participation.

In the subdistrict (Parbatipur) with digital health–only intervention, initial training on the national protocols for hypertension and diabetes management and the Simple app in the NCD corner was conducted in July 2024. Biannual refresher training will be done throughout the implementation period. A project assistant is hired to help with the Simple app data entry and to follow up with overdue patients. CCs and CHWs are essentially not involved in care provision. Patient pathways remain the same as in usual care.

The usual care subdistrict (Biral) receives biannual training on national protocols for hypertension and diabetes management. Existing usual care is being provided by the subdistrict NCD corner, including screening, treatment initiation, drug refill, and routine follow-up. CCs and CHWs are not involved in NCD care provision. The local government agreed to delay the rollout of the Simple app in the comparison subdistricts until the end of the study to avoid contamination.

The primary outcome is the proportion of treated patients achieving or maintaining disease control according to national standards and WHO PEN (WHO Package of Essential Noncommunicable Diseases Interventions) protocols, with hypertension control defined as systolic/diastolic BP <140/90 mm Hg and glycemic control as fasting capillary glucose < 7.0 mmol/L or random capillary glucose <11.1 mmol/L. Effectiveness is being assessed primarily using data from repeated community-based surveys. Secondary outcomes of this study include hypertension and diabetes care cascade (ie, percentage of patients ever screened, percentage aware of their condition, and percentage on treatment), changes in health behaviors (eg, smoking cessation among hypertension and diabetes patients, and meeting recommended weekly physical activity level). To complement assessment of population-level changes using repeated community-based surveys, facility-level data will be extracted to evaluate changes in care quality at primary care facilities.

Specific aim 2 is evaluated by using the RE-AIM (reach, effectiveness, adoption, implementation, and maintenance) framework (Table 2) with patient assessments and stakeholder interviews, informed by previous studies [25,55]. Implementation fidelity and process evaluation are guided by the MRC guidelines on process evaluation of complex interventions [56] and the WHO’s NCD Facility-Based Monitoring Guidance [57]. The program theory is depicted in Figure 3. Data sources include training reports, prescription practice captured by the Simple app, facility records, and patient registries. The Simple app captures essential data related to patient background, prescriptions and titration, dates of follow-up visits, and longitudinal BP and BG records. Baseline qualitative data were collected from patients, health care providers, CHWs, and public health managers to inform the evaluation of barriers and enablers of implementing the interventions in the primary care system for improved NCD outcomes. Another round of qualitative data collection will be conducted upon completion of the interventions.

We anticipated some minor practical refinements and adaptations of the intervention components during the implementation process. Any refinements and the rationale for these are being documented for transparent reporting, guided by the program theory and updated MRC guidance.

For specific aim 3, the primary cost-effectiveness measure is the incremental cost per 1-percentage-point increase in the proportion of participants achieving control status over the 24-month intervention period. Detailed costs by input categories and steps in intervention components will be estimated. Both health system costs and patient costs will be compared between the intervention and usual care arms.

Three community-based surveys are undertaken at months 0, 12, and 24. For each evaluation survey, an independent random sample of adults aged 40 years and above who are community residents in 3 study subdistricts is randomly selected through a multistage cluster sampling approach. We opted for a quasi-experimental design with repeated cross-sectional surveys, instead of a traditional randomized trial, to capture changes in the entire care continuum at the population level and to generate real-world evidence on effectiveness and implementation strategies for greater external validity [58,59]. A buffer zone between the 2 intervention subdistricts (ie, Chirirbandar and Parbatipur) was created to mitigate potential spillover effects between contiguous communities. Given the available resources and time constraints, we randomly select a total of 15 villages from each subdistrict. The villages are divided into segments of approximately 250 households without disrupting the boundaries of the villages. One segment per village is then randomly selected. Subsequently, a list of adults aged 40 years of age within each village of the selected segment is obtained. From each segment (cluster), an equal number of women and men ≥40 years of age are randomly selected by using systematic random sampling with the condition that no more than 1 adult male and 1 adult female is included from the same household.

The planned sample size was determined to be 6750 participants for each community survey, with an equal number of clusters per subdistrict (15 clusters per subdistrict, or 45 clusters total) and similar cluster sizes (150 participants: 75 females and 75 males per cluster). Assuming a prevalence of diagnosed hypertension of 40%, and 15% for diabetes among adults aged 40 years and above, based on South Asia Biobank [60] data (unpublished findings), we expected to collect information from 2700 individuals with hypertension and 1000 individuals with diabetes. Assuming a conservative intraclass correlation coefficient of 0.02 [25,61], and a 2-sided type I error rate of 5%, the survey has 80% power to detect a difference as small as 7 percentage points between the intervention and reference groups for BP control, and a 10 percentage point difference for glycemic control. Power calculations were performed using Stata 17 (Stata Inc.) power analysis for a clustered 2-sample proportions test [62]. The planned sample size for binary outcomes ensures sufficient power for continuous specifications of the outcomes [25,61,63].

Baseline community surveys used interviewer-administered questionnaires and physical examinations. The baseline questionnaire, an adapted version of the WHO STEPS (World Health Organization’s STEPwise approach to NCD risk factor surveillance) questionnaire, covers sociodemographic characteristics, comorbidities, medication use, and health behaviors. The questionnaire was translated into Bengali and piloted for clarity and digitized using Kobo Toolbox for data collection. BP was measured by research staff using Omron HEM-7120. Capillary BG after at least 8 hours of fasting was measured using Accu-Chek Instant (Roche Diabetes Care).

Facility-based data collection is being conducted via the Simple app by on-site staff. Demographic and NCD history, medication use, BP/BG measurements, and treatment dosages are being recorded. In usual care subdistricts, the same data are being collected through patient registry reviews. Additional facility-level data, such as medication availability, device functionality, and treatment success rates, are being collected for process evaluation. An acceptability and utility survey of the Simple app for NCD management will be administered to health care providers in intervention subdistricts at months 12 and 24. All providers involved in hypertension or diabetes care at NCD corners and CCs will be invited to participate.

All health care providers in primary care facilities who are directly involved in NCD care were approached for qualitative data collection at baseline. This included 15 doctors and nurses from 3 subdistrict NCD corners and 15 CHCPs from CCs. Additionally, 6 public health officials (2 from each subdistrict) were approached for an in-depth interview (IDI). Furthermore, 9 focus group discussions (FGDs) with patients living with hypertension, diabetes, or both conditions were conducted. For each FGD, 8‐12 patients were included. Participants were purposively selected to maximize the diversity of the sample on sociodemographic characteristics (eg, age, sex, religion, socioeconomic status, and geographic distance to health facilities) and NCD history (new and experienced). Qualitative data collection will be repeated by the end of the study with the same group of public health officials and NCD care providers. In cases of staff turnover, the ones who are in position at the time of data collection will be approached. Patient FGDs will be repeated with the same groups of participants. Moreover, all 15 CHWs will be approached for an FGD session. FGD/IDI guides were developed based on the WHO’s 7-domain framework of health care delivery [64], chronic care model [10], and prespecified program theory. Interviews will be audio-recorded and are expected to each take 30‐45 minutes to complete. Qualitative data collection will be performed at baseline and endline surveys.

The qualitative data collected from health administrators, clinicians, and CHWs in intervention subdistricts provide qualitative insights on barriers and facilitators to NCD care delivery, experiences with the Simple app, study participation, workflow restructuring, and teamwork with clinics and CHWs. Data collection at endline allows participants to reflect on changes in these aspects over time. Patient FGDs explore barriers and facilitators to care accessibility and perceptions of care quality, including specific intervention components like drug refills, CHW home visits, and CHCP follow-ups.

Hypertension- and diabetes-related treatment costs were collected for economic evaluation, including direct and indirect costs (eg, transportation, food, childcare, and lost work time) at baseline. Health service delivery costs, including staff, transportation, laboratory, training, utilities, and overheads, are being assessed using microcosting to track time and resources spent on activities [65]. Intervention costs account for the setting, target population, resources, and consumables such as medical supplies and overheads.

To protect privacy and ensure confidentiality of data being collected, identifying information associated with the study participants is being kept confidential through unique identifying numbers on all paper forms and computer-based files. This file linking names and study numbers is password-protected, only accessible to authorized study personnel. All electronic data are encrypted, password protected, and stored in secure computer networks. All study personnel were trained to follow standard protocols.

Serious adverse events, including death, hospitalization, and other conditions that result in persistent or significant disability, were handled by medical professionals at the subdistrict NCD corner as per standard protocol. Health care providers at CC were required to send patients back to the subdistrict NCD corner for evaluation when persistent poor control of BP and BG is observed among patients managed at CCs. Adverse events were recorded by study personnel.

Baseline characteristics of participants, including sociodemographic characteristics, medical history, anthropometrics, and lifestyle factors, will be compared between the 2 intervention and 1 control subdistricts using 1-way ANOVA or Rao-Scott χ² tests. Analyses will be stratified by hypertension and diabetes status. A difference-in-difference estimate for hypertension and diabetes will be implemented with multivariate logistic regression analyses, which take the following form:

The β₃ coefficient captures the difference in change over time. Clustered standard errors at the village segment level will be specified to allow for intragroup correlation, relaxing the usual requirement that the observations within a cluster are independent [66]. Several individual- and area-level covariates (β₄) will be included in the analytic models to mitigate confounding by potential different sample composition. These covariates will include age, sex, education, wealth, tobacco use, physical activity, self-reported fruit and vegetable intake, self-reported history of chronic disease (eg, CVD, chronic respiratory disease, cancer, and chronic kidney disease), and rural or urban status. Given that the evaluation follows a multistage process whereby village segment clusters are randomly sampled from each subdistrict, and individuals are sampled randomly from the village segment clusters, a clustering adjustment for standard error is necessary to avoid inflating the precision of the estimated intervention effect [67,68]. Marginal effects will be calculated to illustrate how the predicted probability (ie, the proportion of patients with controlled conditions) changes over time among the three groups. Potential heterogeneity of effects will be explored in subgroup analyses, defined by age, sex, religion, and socioeconomic status. Several sensitivity analyses will be performed to check the robustness of the findings to critical model assumptions and missing data. While the “parallel” assumption should be tested with repeated measures prior to the intervention, this was not feasible given the budget and other limitations. Instead, we are testing the assumption with a negative control in treatment or outcome. Second, we will test the robustness of the results to different modeling approaches to handle geographical clustering of participants and over time. Third, missing data patterns and potential mechanisms (ie, whether missing at random can be assumed) will be assessed. If missing data only affect a small proportion (ie, <5%) of the sample, listwise deletion will be done; otherwise, missing data will be handled with multiple imputation with chained equations if missing at random assumption is reasonable.

The facility-based cohort data collected longitudinally will capture patients’ treatment trajectory over the study period. Changes in the proportion of patients with a controlled condition will be analyzed with generalized estimating equation Poisson regression with robust variance [69,70]. Continuous changes in systolic BP and diastolic BP from baseline will be assessed with linear mixed effects models.

For qualitative data, dual Bengali-English language speakers on the study team will transcribe and translate the audio-recorded IDIs and FGDs into English. The English language version of the transcripts will then be coded and analyzed thematically using NVivo, a qualitative data software program developed by Lumivero. Qualitative data will allow the identification of specific barriers to and facilitators of interventions by exploring the experiences of patients and clinicians engaged in different models of NCD care delivery. Qualitative data analysis will be guided by prespecified program theory, the RE-AIM framework, and standard grounded theory to identify themes, build and apply codebooks, and describe thematic characteristics, patterns, and relationships [71,72].

For the economic evaluation, we will compare the costs and effects of intervention with the usual care group, both from financial and economic perspectives [73]. Financial cost per unit outcome will be calculated by dividing the total cost by the quantifiable unit of outcome (screening, treatment, retention in care, and control) for each of the three groups. Incremental cost-effectiveness ratios (ICERs) will be computed to compare the additional costs and effects of each intervention with the usual care. A nonparametric bootstrap with 1000 replications will be used to estimate 95% CIs around the point estimate of ICERs. All local currency (Bangladeshi Taka) costs will be converted to US dollars using the prevailing exchange rates and will be adjusted for inflation rates, discounted at 3%, and expressed in the 2025 US dollar present value terms. Sensitivity analysis will explore the robustness of the results to alternative probability distributions, time horizon, and uncertainty of key variables.

Descriptive statistics were reported for baseline characteristics of the sample stratified by subdistricts. Percentage estimates for key primary and secondary outcomes were age-standardized to the 2023 Bangladeshi population using age groups 40‐49, 50‐59, and ≥60 years, derived from the United Nations’ Population Division of the Department of Economic and Social Affairs World Population Prospects. We used a direct standardization approach using the STDIZE and STDWEIGHT Stata commands. The preliminary analyses were performed using Stata 18.0 (StataCorp Inc).

To ensure that the strategies developed are feasible, contextually appropriate, and sustainable in real-world settings, stakeholder groups (including government agencies, health care providers, and community representatives) have been actively engaged to enable cocreation. At the study design phase, input from relevant government agencies, major nongovernmental organizations, and community representatives was sought. During implementation, monthly meetings with subdistrict-level public health officials and CHWs were convened to facilitate bidirectional exchange of knowledge. As a member of the Bangladesh national committee for the revision of the national hypertension and diabetes protocol, the site principal investigator has shared findings with national policymakers. A stakeholder advisory group has been convened, with its first annual meeting planned for the end of 2025.