Protocol for a randomised controlled trial to evaluate the effectiveness of improving tuberculosis patients’ treatment adherence via electronic monitors and an app versus usual care in Tibet

We will use a multicentre, parallel-group, individually randomised controlled trial with one intervention and one control arm, using a 1:1 allocation ratio, to evaluate whether the intervention is superior to the control treatment. The study design is informed by the Medical Research Council framework [18] on complex interventions and implementation science principles [19] with an embedded theory-based process evaluation [20] to examine operational questions regarding acceptability, cultural appropriateness, and burden of technology use. In addition, we will conduct an incremental cost-effectiveness analysis to inform future scale-up. We also plan to do a follow-up study to compare the intervention’s impact on TB relapse rates after patients finish their treatment.

We will implement the study in two districts of Shigatse Prefecture in China’s Tibetan Autonomous Region: one urban area called Samzhubze, and one rural area called Sa’gya. Shigatse Prefecture is located west of Lhasa with an average altitude of 4000 m above sea level. Population density in Shigatse is very low (4/km²). The prefecture covers an area of 182,000 km² of harsh and rugged terrain with paved roads connecting county centres. Over 90% of its 800,000 residents are ethnic Tibetans. The 2014 TB survey revealed that the prevalence rate of pulmonary TB was 758/100,000 (0.76%), almost twice China’s national average [21, 22].

The two project sites currently face challenges to provide standard DOT to ensure TB treatment adherence. TB care is only available from public health providers including: (1) TB dispensaries located in Centre for Disease Control (CDC) facilities for clinical care and general coordination, (2) township hospital (primary care facilities) physicians for home visits and supervision of village physicians, and (3) village physicians, often as treatment supervisors. According to China’s National Tuberculosis Programme (NTP) guidelines, new pulmonary TB patients should be treated in the community and should visit the district TB dispensary every month to refill their medicines. However, the serious shortage of health staff, plus the long travel distances and severe weather conditions make regular home visits often impossible. In practice, patients typically receive self-administered therapy (SAT) with limited calls/visits from health staff, and loss to follow-up rates have been very high. Most patients visit the TB dispensary once every 2 months, if not longer. Across Shigatse in 2016, only 72% (769/1073) of new pulmonary TB cases completed treatment. Among patients who did not complete treatment 83% (252/304) were lost to follow-up. Treatment completion rates vary greatly by district. In our two study districts a total of 269 patients were registered in 2016, but only 38% (102) completed treatment, while 60% (162) were lost to follow-up and 1% (3) died. Many patients, mostly poor farmers, were lost to follow-up during the intensive phase as they received little education and support from health workers due to a lack of communication.

In each district, patients with presumed TB and newly confirmed pulmonary TB will be referred to the TB dispensaries by health workers from other public and private health facilities. Individuals can also present themselves if suspecting having TB. All patients will be diagnosed in the TB dispensary according to national and international guidelines of TB care [23]. Eligible TB patients are those: (1) aged ≥ 15 years, (2) who are starting on standard 6-month short-course chemotherapy on an outpatient basis, (3) are free from any communication impairment (mental, visual, auditory or speech), and (4) do not have any family members within the same household who have already been enrolled into the trial. We will recruit patients into the study via TB physicians in the TB dispensaries who will screen patients for eligibility, and explain the study purpose and obtain informed consent. The TB physicians also act as TB coordinators responsible for reporting data to the NTP. Based on existing routine TB data across the two districts we expect to recruit 300 new pulmonary TB patients in 15 months. We will maximise patient enrolment through fully informed consultations provided by TB physicians.

Following recruitment, patients will be randomised to either the intervention or the control arm in a near 1:1 ratio, using a computer-generated randomised permuted block design. The design will use random block lengths of 2, 4 or 6, and include an initial unbalanced block, with a maximum of one additional patient in either arm, to further hide allocation sequences, which means that the ultimate sample size and allocation ratio will vary very slightly from what is planned dependent on the unbalanced block. The allocation will be stratified by district, with one district having approximately 160 patients and another having approximately 140 patients, with a near equal allocation ratio in each. JPH will generate the allocation sequence, and will have no further role in the randomisation or allocation process. The allocation sequence will be used by the study team to print individually numbered allocation cards, which will be placed in sequentially numbered, sealed, opaque envelopes before being delivered to study sites. The TB physicians will allocate each recruited patient after their enrolment by opening the correct envelope in the sequence, and will provide the e-monitor setup according to the patient’s allocation. Research team members will educate the TB physicians on the importance of following the correct sequence, and local CDC staff will make monthly visits to the TB dispensaries to carry out spot checks and ensure the randomisation procedures are being correctly followed. It will not be possible to blind providers or patients due to the nature of the intervention. However, we will take measures to ensure a blinded outcome evaluation using: (1) the ‘PROBE’ design of blinded outcome evaluation [24], and (2) blinding those analysing the data to treatment status.

All patients will be treated according to the standard WHO DOTS programme and the China NTP guidelines using isoniazid, rifampin, ethambutol and pyrazinamide for 2 months (3 months for sputum-smear-positive patients whose sputum smears have not converted to negative at the end of 2 months), followed by isoniazid and rifampin for 4 months, under daily fix-dose combination (FDC) for the entire treatment course [23]. According to the NTP guidelines, patients will be given the choice of SAT, or treatment with a supervisor, mostly village physicians but sometime a township hospital health worker. The health workers who act as treatment supervisors will receive RMB 60 (US$10) from the government when completing 6-month support to a TB patient. The township hospital health workers also have the role of visiting patients’ homes, and supervising village physicians. In the two districts patients should visit the TB dispensary at least every 2 months to meet with their TB physician and refill their FDC medications using the e-monitor box, which will collect data on the number of medications unused and refilled each time they are opened. When patients meet with the TB physician they will also count any leftover medicines. E-monitors will be returned to district TB dispensaries at the end of treatment and reused when possible.

Figure 1 shows the framework of intervention strategies. The e-monitors in the intervention arm have two main functions beyond storing the TB medication. First, they will remind patients to take their medicine on time using human voice recordings. Second, they will transmit patients’ adherence history to a Cloud-based server, linking with computers and a smartphone app. At recruitment the TB physician will demonstrate how to use the e-monitor box and WeChat app. The treatment supervisor will visit the patient’s home within the first week of treatment, and solve any remaining problems. During recruitment or at the first home visit by the treatment supervisor, a family member will be chosen to act as a treatment supporter. The family member must live in the same house as the patient, care about the patient, and be literate in using the WeChat app. Then, patients or their family members (if the patients are not able to use a smartphone or WeChat) will be invited to setup the WeChat app on their smartphones, and connect with their TB physician and treatment supervisor. When appointed, the family member will be trained by the TB physician and the treatment supervisor about their responsibilities for providing psychological support to patients and facilitating patients use of the e-monitor and the WeChat app. TB physicians and treatment supervisors will add the patient or the family member into their WeChat. Patients, treatment supporters, supervisors and physicians will be able to have direct but distinct communications. In addition, patients will receive audio/video-based health education messages through WeChat sent by the treatment supervisor and TB physician.

Using the WeChat app and also their computers connected with e-monitors through the Cloud, health workers will be able to monitor patient adherence history and outpatient visits. Patients who report consistently taking their medications will continue SAT. TB treatment supervisors or TB physicians will identify patients at high risk of being lost to follow-up based on whether the patient: (1) expresses serious concerns about maintaining their adherence, (2) skips three consecutive doses or (3) refuses to start or continue treatment. Treatment supervisors will be trained that if their patient is considered a high risk of becoming lost to follow-up they should explore with the patient their reasons for non-compliance and what help may be needed. The treatment supervisor or TB physician will then try to initiate VOT while providing any help needed. The treatment supervisor will document the start and end time of VOT. Patients who accept the invitation to begin VOT will be given instructions either in the TB dispensary or at home during the visit of their treatment supervisor. Using the WeChat app VOT can be done with the TB supervisor via a live video conversation or via recorded video/pictures showing them taking their medicines. VOT will be provided until three consecutive doses are completed on time, then the patient will switch back to SAT using their e-monitors. Patients who do not agree to VOT will be called daily by treatment supervisors until three consecutive doses are completed. We anticipate that less than 20% of TB patients will need VOT, and mostly during the initial phase or at the end of the treatment.

TB physicians, township public health staff and village physicians will receive training on a revised operational NTP guideline that incorporates using the e-monitor box, WeChat, and VOT based on our previous experience [3], as well as how to communicate using the tools.

The control arm will practice usual care except that patients will use deactivated e-monitors without automated voice reminders to only collect treatment adherence data. The data will be encrypted on the Cloud and will thereby not be available to health workers during the study. At each visit to the TB dispensary, patients will refill their medications in their e-monitor boxes every 2 months. Treatment supervisors are advised to visit patients at least once a week according to the NTP guideline, but this will be at their own discretion. We will not invite patients to connect with their TB physician or treatment supervisors through the WeChat app. Treatment supervisors will contact patients through traditional means, such as physical visits or phone calls. We will not select family treatment supporters.

The primary outcome will be the rate of poor adherence measured per month across the 6/7 months of the standard WHO DOTS programme for new pulmonary TB patients. We will calculate the outcome from monthly level data for each patient indicating the number of doses missed per month, with poor adherence in a given month defined as the patient having missed ≥ 20% of doses in that month (equivalent to missing ≥ 6 out of 30 doses in a given month). This threshold has been commonly used in other disease areas [25] and in a similar trial as a relevant indicator of treatment adherence [16]. Our secondary outcomes are: (1) the patient-level percentage of total doses missed over the 6/7 months of treatment (calculated for each patient based on the total number of doses missed out of the total possible number of doses), and (2) a patient-level binary indicator of overall poor adherence (defined as ≥10% of total doses missed, which is the NTP definition of non-adherence). Our remaining secondary outcomes will be based on World Health Organisation (WHO) standard definitions of TB treatment outcomes [26]: (3) treatment completion rate (‘a TB patient who completed treatment without evidence of failure but with no record to show that sputum smear or culture results in the last month of treatment and on at least one previous occasion were negative, either because tests were not done or because results are unavailable’), (4) loss to follow-up rate (‘a TB patient who did not start treatment or whose treatment was interrupted for two consecutive months or more’), (5) poor treatment outcome rate (defined as death, treatment failure (‘a TB patient whose sputum smear or culture is positive at month 5 or later during treatment’) or patient loss to follow-up), and (6) sputum conversion rate at the end of the second month [27].

As there are limited analytical sample size approaches available for longitudinal outcomes we used a simulation approach. We based our subsequent power calculation on our primary outcome of monthly poor adherence, and assumed a 40% monthly poor-adherence rate for the control arm [16]. For our simulations we created six binary poor-adherence outcomes (i.e. one per treatment month) for patients, varying the total number of patients between simulations to explore power, but always assuming a 1:1 treatment arm allocation ratio. The outcomes were randomly generated from a Bernoulli distribution with a fixed probability of success (i.e. probability of monthly poor adherence), which was always set to 0.4 for the control arm, but was varied between simulations for the intervention arm. In addition, in each simulation all patients’ outcomes were generated with a fixed level of correlation between their monthly outcomes, based on an exchangeable correlation structure. These outcome data were generated using the R package simstudy [28], based on the Emrich and Piedmonte algorithm [29]. We explored the effect on power of assuming an AR1 correlation structure instead, but as our simulations showed that this resulted in less conservative estimates of power, and because we had no similar data to explore the robustness of this assumption and no strong reason to assume an AR1 structure, we assumed an exchangeable structure instead. After generating the simulated data we then used the R package geepack [30] to fit a generalised estimating equation (GEE) to the data with binomial errors and an identity link, which allowed estimation of the absolute difference in treatment arm outcome proportions (see the ‘Statistical analysis’ section for more details) [31]. The GEE included a covariate for treatment arm and a categorical variable for month. We then extracted the Wald-based two-sided p value associated with the coefficient for the treatment effect estimate. Finally, for each set of parameter assumptions we repeated this process 1000 times and calculated the resulting power based on the proportion of p values ≤ 0.05. We assumed no missing data due to patients who become lost to follow-up still being able to contribute 100% of their outcomes in the study. Assuming an expected monthly poor-adherence rate of 40% in the control arm and a moderate within-patient correlation between monthly poor-adherence outcomes of 0.5, we require 300 patients to detect an absolute reduction of 12 percentage points in the intervention arm outcome (i.e. an intervention-arm monthly poor-adherence rate of 28%) with 81.5% power.

We will assess the feasibility of implementing the intervention and running our trial processes through a pilot study that will precede the main trial. Specifically, during the preparation period we will invite four new pulmonary TB patients (two in each district) to be randomised into either the intervention or control arm as pilot cases, allowing us to test the randomisation and recruitment processes. We will then follow them up for 2 weeks to assess the feasibility and acceptability of: (1) using the e-monitor boxes for medication storage, (2) using the Cloud-based server for monitoring adherence history, and (3) using the WeChat app for patient-health worker communication. We will also pilot test the feasibility of using VOT in the two patients who are allocated to the intervention. We will revise the implementation plan based on our experiences from the pilot.

Theoretical framework: we will employ an adapted version of the Unified Theory of Acceptance and Use of Technology (UTAUT) model [32, 33] to guide our process evaluation to assess issues that influence patients’ behaviour change in relation to both the technology used, such as acceptance, feasibility, appropriateness, adaptability, as well as adherence behaviours (Fig. 2). The UTAUT has been widely used to understand information technology adoption in general. The adapted UTAUT model has four constructs through which to explore technology-oriented factors that shed light on user experiences interacting with a given technology. These factors are: (1) performance expectancy, (2) effort expectancy, (3) social influence, and (4) facilitating conditions. Performance expectancy includes the perceived usefulness and personal outcome expectations associated with technology use. Effort expectancy is the perceived ease of use and complexity of the technology. Social influence includes subjective norms and technology use within a user’s social context. Facilitating conditions include perceived behavioural control and wider contextual circumstances that support the use of technology. Together these factors provide a comprehensive understanding of the patient’s experience with the technology.

We also included adherence relevant aspects of the WHO adherence framework into our theoretical framework [34]. Condition-related factors are the illness-related demands faced by the patient which ultimately impact the patients’ risk perception, treatment beliefs and the priority that they place on adherence. While therapy-related factors include side effects, treatment duration, treatment failures and experience of side effects [34]. These indicators may either motivate or dissuade patients from adhering to treatment plans. Though our intervention may satisfy technology-oriented needs of a patient, their treatment plan and their lived experience may pose unforeseen challenges to their behavioural intention and change. All the above factors will be considered with contextual and person-oriented factors including gender, age, socioeconomic factors, experience and voluntariness of technology use.

Our process evaluation is designed to explore factors highlighted by the modified UTAUT model from the perspective of patients, treatment supervisors, and physicians. We will record the e-monitor’s defects, connectivity, and the operation within existing health and logistics conditions through programme logbooks. We will also record each supervision trip. We will report the proportion of patients having a smartphone, able to enrol in, and their technology use. We will measure the accuracy of patient adherence history collected by e-monitors against anti-TB drug refills/ counting in TB dispensaries at each time of patient refills. District TB physicians will record the persistence of using e-monitors over 6/7 months, the proportion of patients who need VOT, start VOT, and the persistence of VOT (over 3 days until satisfactory adherence).

We will conduct semi-structured in-depth interviews to understand acceptability, cultural appropriateness, and burden of technology use. We will recruit and interview 24 patients (18 from the intervention arm and 6 from the control arm), 12 family treatment supporters, at least 12 treatment supervisors, six township treatment supervisors and four TB physicians. Among patients in the intervention arm, we aim to explore differences between patients who demonstrate adherence and those who do not. We also aim to explore differences in technology uptake, adoption and sustained use between patients and identify whether there are different stages of uptake and adoption. Questions will explore smartphone literacy, experience and usability of the new technology, overall satisfaction, and any concerns (e.g. understanding, learning experience, time consumption, communication quality, confidentiality, data-usage costs). Due to culture norms, we will pay specific attention to women regarding confidentiality/privacy issues in communications using WeChat and the VOT.

We will conduct an incremental cost-effectiveness study that will collect costs of the implementation package including staff, equipment, and supplies – ‘incremental’ to those required in usual (control arm) care. Overall incremental cost-effectiveness ratios for the service intervention and health outcomes, based on our primary and secondary endpoint of treatment completion rate, will be computed where possible. Simple unit costs of implementing the intervention including cost per district and per patient will be computed. The total per patient cost will be calculated as the sum of three elements: (1) the cost of clinic consultation estimated with physician’s time input and unit salary; (2) the cost of patient follow-up measured with frequency of home visit/using WeChat, average time input for the two approaches and their unit salaries, plus travel cost where applicable; and (3) medication cost including inpatient and outpatient health service charge, and indirect cost (travel, caregiver cost, etc.) when a patient is hospitalised. The unit implementation cost will be calculated as the sum of: (1) cost of staff time of both trainers and trainees; and (2) preparation of intervention materials including e-monitors and printings. Using published data, as available, we will estimate the benefits to: human health including potentially reduced multiple-drug resistance TB or TB death, and productivity gains.

We plan to follow up all patients using their national ID numbers which will be recorded in the NTP reporting system as a routine practice in Tibet. Follow-up will start from when a participant completes their treatment with any treatment outcomes including loss to follow-up. We will track any relapse cases reported in Tibet using the national TB reporting systems within 12 months of their treatment completion. Some patients may travel to other provinces to work where national ID number reporting is not mandatory for TB cases. To mitigate this gap, we will conduct a short survey regarding their TB status and the place of treatment (if any) by the end of the follow-up study through their treatment supervisors. We will also collect all mortalities using the vital registration system in China where the national ID has to be input. The primary endpoint of the follow-up study is TB relapse and the secondary endpoint is mortality. We will evaluate whether these outcomes differ between the intervention and control arms using the same methods of analysis as outlined below for the main trial indicators. However, where patient follow-up data cannot be collected we will conduct complete-case analyses by excluding those patients with missing data. This study will be reported separately.

The trial protocol has obtained ethical approval from the Office of Research Ethics at the University of Toronto (Ref: 36569) and the Ethics Review Committee of the Tibet Centre for Disease Control and Prevention (Ref: 006).

In addition to the requirement by the China NTP, all health staff in this study will have to sign an agreement to keep all patient information confidential and only use it for the purpose of TB care. All patient information stored in the WeChat app at the provider’s end will be deleted immediately once the patient has completed treatment or been lost to follow-up. Patient information, including name, age, sex, address, education, profession, diagnosis, and treatment outcomes, will be routinely collected by the NTP. We will also collect routinely recorded treatment outcomes data from the two districts, including conversion at the second month, loss to follow-up, death, and sputum smear results at the fifth and sixth months. We will collect all these data during monthly monitoring visits in the two districts. We will use double entry and check a random subset of the data to ensure data quality. We will use these data to calculate our secondary indicators. The interventions are unlikely to increase potential risks to patients. However, any adverse events of the interventions will be reported to the research coordinators in the Shigatse CDC by TB physicians within 24 h and will be recorded in work logs.

For this study we will allocate a unique participant number for each patient in our dataset to mask patient identifiable information such as ID, names, and addresses. Research staff will only have access to the participant number and non-identifying information such as treatment outcomes. All the patients’ national ID numbers will be recorded in the TB reporting system as routinely required, which will be used to track any relapse cases 12 months after the patient has completed their treatment.

We will measure the adherence indicators based on data collected by the e-monitor in both intervention and control arms, which will be downloaded from the Cloud-based database. To ensure that all missed doses are accurately recorded we will cross-check the adherence history recorded from e-monitors with pill counts manually recorded at the end of each outpatient visit, and we will record as the true value whichever is the larger number of missed doses from the two data sources. The TB physician will also conduct a short questionnaire survey asking: (1) if patients have had access to and/or used the WeChat and e-monitors, (2) how much time patients have spent using the e-monitors and WeChat, and (3) any medical costs. We will also conduct a questionnaire for physicians and treatment supervisors at the end of the trial period to measure their time cost in supporting patient treatment. We will collect implementation costs using researchers’ work logs, including the number of trainees, trainers’ labour and travel costs, and the cost of training materials and the e-monitor boxes. We will also record and audio tape the interviews with patients and health staff for the process evaluation. Information from the interviews will only be accessed by research staff, and anonymised during analysis.

Our primary analysis population will be the ‘intention-to-treat’ population, which will include all randomised patients analysed according to their original treatment allocation, irrespective of their subsequent adherence or how they were actually treated. The Consolidated Standards of Reporting Trials (CONSORT) guidelines for reporting parallel-group randomised trials recommend presenting absolute measures of effect for binary outcomes due to their increased relevance for practice and policy, as well as relative measures of effect. Generalised estimating equations (GEEs) have been shown to produce robust estimates of the absolute difference in treatment effect for binary outcomes from multi-level trial data [31]. Therefore, we will analyse our monthly binary primary outcome of poor adherence using a GEE, with binomial errors and an identity link, to estimate the absolute difference in the proportion of poor adherence across all treatment months between the intervention and the control arms (intervention minus control). We will assume an exchangeable correlation structure for the repeated outcomes within individuals, and estimate parameter standard errors as ‘Huber-White’ or ‘robust’ standard errors. In this analysis we will also adjust for effects (covariates) of month (as a categorical variable), stratum, and a range of sociodemographic characteristics: age (categorical: 15–49, 50 + years), sex, marriage status (married, never married/not currently married), and employment status (employed: farmer, employed: other, unemployed/retired). We will base our principle inference about the effectiveness of the intervention on this main analysis. We will also repeat this analysis but excluding all covariates other than month and stratum as a sensitivity analysis to explore how robust the treatment effect is when not adjusting for those potentially imbalanced covariates. To explore how the treatment effect differs between treatment months we will also repeat our main analysis but also include an interaction between treatment group and month (as a categorical variable) in the model, while also only including stratum as an additional covariate because our parameter to sample size ratio will be constrained. Because the approach of using an identity link in a GEE with binomial errors is not yet widely used, and to facilitate meta-analyses of odds ratios (which are more commonly estimated) we will also repeat the main analysis but using a GEE with binomial errors and a logit link to estimate the odds ratio for the odds of poor adherence in the intervention arm compared to the control arm (based on taking the exponential of the treatment effect coefficient). We do not expect any missing data because we are only collecting basic covariate data at baseline, and if patients are lost to follow-up at any point then they will still contribute 100% of their primary outcome data because their adherence status would be validly recorded as non-adhering from the point at which they were lost to follow-up. However, if their primary outcome data are missing for an unexpected reason (e.g. loss of data) then we will ensure that the ‘intention-to-treat’ analysis population is maintained in all analyses by imputing the missing outcome data assuming the worst outcome of poor adherence for those months where data are missing.

For our secondary outcome of the patient-level percentage of total doses missed over 6 months of treatment, assuming the outcome is normally distributed we will analyse it using a linear regression model including the same covariates, other than month, as used in our primary outcome main analysis. If the data are substantially non-normally distributed we will apply a suitable transformation to the data to achieve normality. If any dose data are missing we will ensure that the ‘intention-to-treat’ analysis population is maintained by imputing the missing outcome data assuming the worst outcome, i.e. that the dose(s) were missed. For all remaining secondary outcomes, which are all binary, we will use a ‘logistic regression average-risk-difference’ analysis approach to obtain absolute measures of treatment effect. In these analyses we will again adjust for the same covariates, other than month, as used in the main primary outcome analysis. This approach has been shown to provide robust estimates of absolute treatment effects for RCTs with binary outcomes [35]. As with our primary outcome we will also conduct covariate-sensitivity analyses for all secondary outcomes where we will repeat the above analyses but without adjusting for any covariates other than stratum. For our secondary outcome of overall poor adherence if any dose data are missing we will follow the same approach as for our analysis of the patient-level percentage of total doses missed outcome. All patients should be assigned a WHO standard TB treatment outcomes. However, if any patients are missing these outcomes then we will assume the ‘worst’ outcome (e.g. for the treatment completed outcome we would assume they did not complete treatment).

Lastly, we will also conduct a small number of subgroup analyses to explore whether treatment effects vary between the subgroups. For our primary outcome we will use the same main analysis approach planned above (of using a binomial-identity GEE), but we will add an interaction between treatment group and the relevant subgroup identifier into the analysis to estimate the difference in treatment effect between the subgroups. We will conduct subgroup analyses for our primary outcome for the subgroups (1) sex, (2) age, and (3) travel time to the relevant district TB dispensary, and we will compare between men and women, less than the median age and greater or equal than the median age, and less than the median travel time and greater or equal than the median travel time. We will also conduct subgroup analyses for our secondary outcome of treatment completion for the same range of subgroups. To do this we will use a logistic regression model that includes covariates for treatment, subgroup, the interaction between treatment and subgroup, and the same range of covariates as noted in our main analysis except month.

Both quantitative and qualitative data will be analysed. Quantitative data will be summarised, described and analysed using appropriate statistical methods, including multiple linear regression for continuous outcomes and multiple logistic regression for binary outcomes. Qualitative data analysis will use a thematic approach to discover emergent themes. Notes will be reviewed after each interview to identify emergent topics and allow for exploration in subsequent interviews. Data will be translated and transcribed, then analysed using NVivo 10.

Prof Xiaolin Wei from the University of Toronto and Dr. Jun Hu from Shigatse CDC will be the co-guarantors of the trial who have full access to the trial dataset. An external-member-led data management committee (DMC) will be established to: (1) safeguard the safety and privacy of patients involved, and to (2) ensure that all data are collected according to agreed ethical guidelines, properly stored and only used for research purposes. We will also form an external-member-led trial steering committee (TSC), consisting of members from China NTP, Tibetan TB programme, key members of the trial team. Annual meetings will be held for both the DMC and TSC. Important protocol modifications will be discussed during the meetings.

The trial was registered at Current Controlled Trials: ISRCTN52132803 on 9 November 2018 (http://​www.​isrctn.​com/​ISRCTN52132803). We started to recruit patients from 26 November 2018. At the original submission of the protocol, we have completed our pilot studies and have recruited five patients.

The study will be done over a 28-month period, with a 3-month preparation and pilot phase, a 15-month patient recruitment phase, with a 6/7-month treatment phase for all patients, and a 3-month data analysis and write-up stage. We will disseminate trial results through research articles and policy briefs. We also plan to follow-up all participants 12 months after they end treatment using the TB reporting system to look at relapse and mortality rates between the treatment arms. Results of the follow-up study will be reported in a separate paper from the trial results. See Fig. 3 for details of the trial timeline.