By maximizing the EM program’s implementation, the overall objective is to improve the treatment adherence and health outcomes of TB patients. The study’s specific goals include:

The optimized EM program’s implementation and evaluation under the theory of implementation science are synthesized in the study’s theoretical model (Figure 1). To improve TB patients’ treatment adherence and health outcomes in China, the EM program will be implemented. The Consolidated Framework for Implementation Research (CFIR) framework will be used to identify the facilitators and barriers to this (23). The Expert Recommendations for Implementing Change (ERIC) protocol will then be used to determine the best implementation strategies for the EM program in the context of China’s remote regions (24). The implementation strategies will be assessed in terms of implementation outcome, service outcome, and patient outcome using the RE-AIM framework (25). The implementation and efficacy of the optimized EM program will be assessed using these indicators, which include reach, effectiveness, adoption, implementation, and maintenance, based on the theoretical model. Figure 1.

This economic evaluation adopts a societal perspective incorporating three key cost components: (1) Direct medical costs including EMM procurement and maintenance (with cloud service fees), healthcare personnel training differentiated by urban/rural rates, and outpatient consultations adjusted according to Inner Mongolia’s tiered pricing system; (2) Patient-incurred costs comprising travel expenses calculated as round-trip distance × local transit fare, pastoralists’ fuel costs for private vehicles, and lost herding income quantified via sheep unit equivalents/day with seasonal grazing adjustments; (3) Societal benefits such as averted MDR-TB costs projected over 10 years using WHO-CHOICE unit costs (including indirect costs from transmission interruption), productivity gains measured through the human capital approach applied to employment data, and Quality-Adjusted Life Years derived from EQ-5D-5L surveys at baseline/6/12 months using Chinese health status preference weights. Analytical approaches will report incremental cost-effectiveness ratios for both adherence improvement (ICER) and quality-adjusted life years (ICUR). Sensitivity analyses include discount rate variations (base case 3%, range 0–5%), time horizon scenarios (5/10/20-year projections), probabilistic sensitivity analysis with 10,000 Monte Carlo iterations, and geographic cluster subgroup analyses (Chifeng/Hohhot/Baotou clusters), with all costs standardized to 2025 RMB using China’s CPI health component.

The human capital approach was selected over the friction cost method for productivity loss estimation based on the following considerations: (1) Methodological Appropriateness‌: For chronic conditions like TB requiring long-term therapy (standard 6-month regimen), the human capital approach more comprehensively captures the prolonged productivity losses. This is particularly critical for pastoral patients, since treatment interruption may cause cumulative income impacts spanning several years due to the unique demands of nomadic livelihoods. (2) Research Objective Alignment: As this study aims to assess the comprehensive impact of optimized interventions on patients’ adherence and quality of life, the human capital method better reflects the long-term economic benefits arising from improved treatment compliance. (3) Regional Specificity: In Inner Mongolia’s pastoral regions, TB patients’ mobile lifestyles imply that treatment interruptions can result in sustained negative livelihood impacts over extended periods, which the human capital approach can effectively quantify. (4) Data Feasibility: Friction time, vacancy period, and other key parameters required for the friction cost method are lacking for Inner Mongolia, while employment and wage data essential for the human capital approach are more accessible and verifiable through local statistical sources. (5) International Standards Compatibility: The WHO recommends the human capital approach for economic evaluations in TB, facilitating international comparison of findings and policy implications. (6) Policy Relevance: From a public health decision-making perspective, estimates derived from the human capital approach more accurately reflect the true economic burden of TB on individuals and society in the Inner Mongolian context.

Consequently,‌ to calculate productivity losses ‌using this approach‌, we will source employment data and wage rates for Chifeng, Hohhot, and Baotou from the 2025 data ‌published‌ by the local Bureau of Statistics and the ‌Human Resources and Social Security‌ departments.

We will enhance the study’s rigor by systematically tracking participants using their national ID numbers within Inner Mongolia’s NTP reporting system. Excluded participants (κ ≤ 0.6) will receive separate bias analysis based on baseline characteristics. Starting post-treatment, this process will capture various outcomes, including potential relapses and even cases of loss to follow-up. We’ll monitor relapse cases through the national TB reporting systems for 24 months after treatment completion. Recognizing the possibility of participants relocating to provinces where national ID reporting for TB cases is not mandatory, we’ll address this gap by conducting a concise survey via treatment supervisors at the study’s conclusion. This survey will gather TB status and treatment location information. Mortality data, requiring national ID inputs, will be collected from China’s vital registration system. The primary focus lies on TB relapse, with mortality as a secondary aspect. Analytical methods mirroring the main trial indicators will be used to compare outcomes between intervention and control groups. In cases of missing follow-up data, a complete-case analysis will be performed and documented separately.

Economic data collection will be enhanced through patient cost diaries recording monthly travel frequency/costs (verified against public transport tariffs) and income loss days (cross-checked with local employment records). Herding income impacts will be assessed using baseline livestock holdings from household surveys and seasonal price fluctuations from regional commodity markets.

The Xi’an Jiaotong University Health Science Center’s Ethics Committee granted ethics approval (approval number: 2020–119).

Inner Mongolia CDC will regularly gather patient data, such as name, age, gender, education, occupation, diagnosis, and treatment results. The results of sputum smears performed at the fifth and sixth months, as well as information on conversion at the second month, loss to follow-up, and death, will also be recorded as routinely tracked treatment outcomes (8). The treatment adherence data, collected by the electronic monitors and downloaded from the cloud-based database, will be normalized as percentages of prescribed doses to enable standardized comparisons across patients with varying prescription regimens. To ensure that all missed doses are accurately recorded, we will compare the adherence history obtained from e-monitoring devices with pill counts manually collected at the conclusion of each outpatient visit. Cases with κ ≤ 0.6 (indicating device-pill count discrepancies) will trigger qualitative interviews to identify root causes (e.g., device tampering, operational errors, non-adherence), and will be analyzed thematically via NVivo 12.0 per Landis and Koch criteria. Adherence data quality will be further ensured through triple validation‌: (1) Monthly unannounced pill counts (10% participants stratified proportionally across Chifeng, Hohhot, and Baotou populations) with a validity threshold of κ > 0.8 (SPSS 28.0); (2) Anti-tampering protocols: Randomized audits + 20 × digital microscopy for blister integrity; (3) Tiered corrections: Device recalibration if 0.6 < κ ≤ 0.8; data exclusion followed by root-cause analysis if κ ≤ 0.6. While cases with κ ≤ 0.6 will undergo root-cause analysis and qualitative interviews to identify potential biases, these cases will remain in the primary Intention-to-treat (ITT) analysis with appropriate statistical handling. Finally, data on implementation costs, such as the number of trainees, the trainers’ labor and travel costs, and the price of the EMM boxes and training materials, will be gathered using the researchers’ work logs. We will implement double entry and examine a representative sample to ensure data entry accuracy.

While stratified random sampling improves validity, two methodological considerations emerge: (1) Potential underrepresentation of mobile populations, adjusted via entropy-weighting with approved anonymized mobility data; (2) Monthly audits could miss short-term adherence variations, mitigated by randomized timing and blister pack integrity checks. The κ thresholds follow Landis and Koch but require local context interpretation.

ITT analysis will be performed for the primary analysis, including all randomized participants regardless of adherence data quality or device-pill count discrepancies.‌ For cases with κ ≤ 0.6 adherence discrepancies, a three-tier multiple imputation strategy will be implemented to maintain the ITT principle.

The Enhanced Multiple Imputation Procedure for κ ≤ 0.6 Cases includes: (1) Level 1 (Baseline and Treatment Patterns) - initial imputation based on observed baseline characteristics (age, gender, education) and treatment assignment status (as fixed covariate); (2) Level 2 (Qualitative Insights) – contextualized imputation incorporating key barriers identified in qualitative interviews (e.g., device tampering, operational errors); and (3) Level 3 (Device and Location Factors) - population-specific imputation considering device types and residence classification (urban/rural). The multiple imputation procedure will use Multiple Imputation by Chained Equations with 20 imputed datasets, using baseline characteristics, treatment assignment status, time variables, and observed adherence patterns. The following variables will undergo imputation: age, gender, education, urban/rural residence type, and monthly adherence rates. Final effect estimates will be obtained by combining results across all imputed datasets using Rubin’s rules.

The Sensitivity Analysis Framework comprises: (1) Primary analysis - ITT analysis including all randomized participants with multiple imputation for missing data; (2) Sensitivity analysis 1 – Complete-case analysis including only cases with κ > 0.6 to assess whether the primary ITT results are consistent in the subset of high-quality data (κ > 0.6), providing a baseline for comparison with the imputed results for cases with κ ≤ 0.6‌; (3) Sensitivity analysis 2 – Household-pairs analysis – This analysis examines adherence trajectories using shared random intercept models, specifically for cohabitating patient pairs (non-immediate family members) with device-recorded versus pill-count adherence agreement κ > 0.8; (4) Sensitivity analysis 3 – Bayesian analysis using Markov chain Monte Carlo methods with appropriate prior specifications to evaluate the influence of prior distribution choices on posterior effect estimates‌; and (5) Sensitivity analysis 4 - Bootstrapped confidence intervals (1,000 iterations) for cases with κ ≤ 0.6 adherence discrepancies to validate the robustness of the multiple imputation approach under different statistical paradigms.

Covariate-adjusted analyses‌ will be used for all outcomes, with individual consideration of participant clustering in community health centers and township hospitals. Treatment adherence will be analyzed as a continuous variable (percentage) while maintaining the predefined categorization threshold of ≥ 80% for high adherence versus < 80% for non-adherence. Intervention-control group differences will be compared using t-tests and χ 2 tests (20). Intervention effects will be examined using generalized estimating equation (GEE) models incorporating spatial autocorrelation terms to quantify contamination effects, with sensitivity analyses comparing: (1) Household-clustered GEE models using household IDs as clustering variables with exchangeable correlation structure, adopting an intraclass correlation coefficient threshold < 0.1 based on Inner Mongolia’s low population density (20.19 persons/km²) and κ weak consistency standards (0–0.2); and (2) Bootstrapped confidence intervals (1,000 iterations) for cases with κ ≤ 0.6 adherence discrepancies (8).

Cases with κ ≤ 0.6 will be evaluated via bootstrapped confidence intervals (1,000 iterations) and qualitative contextualization. These cases remain included in the primary ITT analysis. For continuous outcomes, mean differences and standardized mean differences will be used to assess intervention effectiveness. For binary outcomes, odds ratios and risk differences will be calculated. Additionally, a mixed-effects model that allows health facility-level estimates to be random effects and accounts for facility- and time-level clustering will be used to compare pre- and post-intervention proportions for reach and adoption (20). Implementation and maintenance will be evaluated using a thematic analysis methodology, to identify emerging themes. The data will be translated, transcribed, verified‌, cleaned, and then analyzed using NVivo 12.

Professor Min Su from Inner Mongolia University and Professor Xiaolin Wei from the University of Toronto will serve as co-guarantors of the trial. The trial dataset will be completely accessible to each of them. To protect the security and privacy of the patients involved, as well as to ensure that all data are collected following ethical standards and only used for research, a data management committee led by outside members will be established. The steering committee for the externally led trial will be made up of representatives from China CDC, Inner Mongolia CDC, and other significant trial team members. It will gather every year. Significant protocol changes will be discussed during the meetings (8).

The WHO recommends using an electronic monitor, which is recognized as an evidence-based intervention (9). This has received much use in primary healthcare facilities and is regarded as the tool with the best chance of increasing TB treatment adherence (47). Health policymakers and researchers must urgently determine how to maximize the positive impact of the electronic monitor on TB patients’ treatment adherence and negative outcomes. In addition to developing strategic options for the scalability and generalizability of the optimized interventions in remote areas of China and other LMICs, this study will be the first to develop and implement interventions that improve treatment adherence and health outcomes for TB patients. The study has several main contributions. Firstly, a CFIR framework will be used to pinpoint the difficulties and opportunities associated with putting electronic monitors in place, as well as potential areas for intervention alterations and improvements. To optimize the EM program in the context of China’s remote regions, we will secondly use the ERIC protocol to identify appropriate implementation strategies. Thirdly, a 12-month pragmatic, parallel, cluster-randomized trial will be used to evaluate the implementation strategies of the optimized EM program using the RE-AIM framework. An electronic monitor implementation strategy will be developed as a result of this study, which is crucial for those who are scaling up activities in remote areas of China and other LMICs.

However, this study confronts five challenges. First, economic analyses may systematically underestimate three components: (1) informal caregiving burdens; (2) TB transmission externalities; and (3) intergenerational educational impacts. Second, despite employing gold-standard cluster-randomized controlled trials (20) with 20-km contamination buffers, residual risks persist through two primary pathways: (1) offline medication-sharing networks; (2) transient mobility events below the 20-km detection threshold. ‌Notably, low population density (20.19/km² in Inner Mongolia) may mitigate contamination risks from such mobility events. Third, confounding factors, bias, and temporal trends could limit the validity of the findings. Fourth, while AHP-optimized strategy prioritization enhances rigor, the weight allocation might reflect evaluators’ professional biases. Furthermore, although many TB patients in Inner Mongolia lack literacy, they can still use smartphones and WeChat (48), because (1) contacts in WeChat can be identified by icons and (2) WeChat enables smartphones to send audio and video recordings rather than just text. A family member with strong literacy will also be used to teach the TB patient how to use WeChat.

The trial was registered at Current Controlled Trials: July 29, 2023 (ISRCTN15169616). We started to recruit patients from January 11, 2023. At the original submission of the protocol, we have completed our pilot studies.

This study will be done over a 28-month period, with a 3-month preparation and pilot phase, a 15-month participate recruitment phase, with a 6/7-momth treatment phase for all patients, and a 3-month data analysis and write-up stage (8). A 24-month follow-up study of all participates will be conducted after they end treatment under the optimized EM program to look at the treatment outcomes, and implementation strategies between the treatment arms. Results of the follow-up study will be reported in a separate paper from the trial results. See Table 2 for details of the trial timeline. Table 2.