Background

The World Health Organization (WHO) estimated that 1.12 million children (< 15 years) developed tuberculosis (TB) in 2018, representing 11% of the overall TB case load. WHO also estimated 205,000 child TB deaths that year [1], making TB a top ten cause of death in children under five years worldwide [2]. Recent modelling suggests that almost all children dying from TB (96%) are untreated and that 80% are aged below five years, with treatment not started likely due to not being diagnosed with TB [2]. Indeed, only 512,000 paediatric TB cases were notified to WHO in 2018, representing a treatment coverage of approximately 46% [1]. Childhood TB therefore remains undiagnosed and underreported, mostly due to the challenges in confirming its diagnosis. This in turn is largely because of the paucibacillary nature of disease, and the difficulty in obtaining respiratory samples from young children [3].

Pneumonia is the leading cause of death in children under the age of five years worldwide. There were an estimated 120 million pneumonia episodes in children younger than 5 years in 2011, including 14 million severe episodes [4], of which 1.3 million led to death. In 2014, WHO revised its definitions for pneumonia and severe pneumonia, the latter only justifying referral from primary health centres to higher level facilities for inpatient treatment [5]. However, there are no global estimates of mortality specifically attributable to severe pneumonia as currently defined by WHO. Studies in children hospitalized with severe pneumonia conducted in East and East-central Africa showed an inpatient mortality ranging from 8.7 to 32% [6,7,8]. Children yet remain at high risk for death even after discharge, notably during the first 3 months, with severe acute malnutrition an important factor of poor outcome [6].

In high TB burden settings, there is growing evidence that TB is common in children with pneumonia [9]. Although TB is a chronic disease in adults, recent data shows that the duration of respiratory symptoms before admission can be acute in children with severe pneumonia associated with TB [10]. A systematic review showed that up to 23% of children admitted to hospital with an initial diagnosis of pneumonia were later diagnosed with TB [10, 11]. This is particularly true in the African and South-East Asian WHO regions, which accounted for 30 and 40% of all paediatric TB cases in 2019 respectively [1]. In these regions, the case fatality rate for childhood pneumonia associated with TB is high, ranging from 4 to 21% [10], with younger age, malnutrition and HIV infection increasing the risk of death [12, 13]. However, the diagnosis of TB in children with severe pneumonia remains low. Indeed, the current WHO standard of care (SOC) for young children with pneumonia considers a diagnosis of TB only if the child has a history of prolonged symptoms or fails to respond to antibiotics [14]. Therefore, many children with TB-associated severe pneumonia are currently missed or diagnosed too late, which is likely to affect their outcome.

In 2013, WHO updated its policy to include Xpert MTB/RIF (Cepheid, USA) as the initial test for the diagnosis of TB in children, based on a meta-analysis showing a pooled sensitivity and specificity of Xpert MTB/RIF performed on gastric lavages of 66% (95% confidence interval 51–81) and 98% (95% confidence interval 96–99), respectively, when compared with culture [15, 16]. Although data on the performance of Xpert MTB/RIF in children with pneumonia are limited, in Bangladesh, sensitivity on gastric aspirates or sputum samples compared to culture in this group was equivalent to that reported in other studies [17]. The next-generation of Xpert MTB/RIF assay, Xpert MTB/RIF Ultra (Ultra), has a lower detection threshold (similar to culture), which is expected to improve the diagnosis of children with paucibacillary TB [18, 19].

Previous studies in Africa and Asia have shown that alternative specimen collection methods such as nasopharyngeal aspirates (NPA) and stool samples are easier to implement, and are better tolerated in young and sick children [20,21,22,23,24]. These methods do not require a child to fast, as mandatory for gastric aspirates, and are more suitable than induced sputum in children with severe respiratory deficits [25]. Recent studies have shown similar sensitivity of Xpert MTB/RIF on the combination of one stool and one NPA as compared to two induced sputum or two gastric aspirates [20, 26].

We therefore propose a diagnostic trial to assess the impact on mortality of adding the systematic early detection of TB using Ultra performed on NPA and stool samples to the WHO SOC for children with severe pneumonia, followed by immediate initiation of anti-TB treatment in children testing positive on any of the samples. Our hypothesis is that in high TB burden countries, testing young children with severe pneumonia for TB and starting those who test positive on anti-TB treatment on the day of presentation, could reduce all-cause mortality through reduction of mortality attributed to TB.

Methods

Aim

The primary objective of the TB-Speed Pneumonia trial is to evaluate the impact on all-cause mortality at 12 weeks post inclusion of adding systematic early detection of TB with Ultra, performed on one NPA and one stool sample in young children with severe pneumonia, followed by immediate anti-TB treatment initiation in children with a positive Ultra result, in high TB incidence countries, as compared to the WHO SOC alone. The study only includes children with community-acquired pneumonia, excluding children already hospitalized who have developed a nosocomial pneumonia.

Secondary objectives will assess the impact of the systematic TB detection on TB case detection, time to TB treatment initiation, inpatient mortality, duration of initial hospitalization and hospital readmission rate, and will assess the feasibility and acceptability of the intervention.

Trial design

TB-Speed Pneumonia is an international, cluster-randomised trial with a stepped wedge design. Stepped wedge trials are randomised controlled trial in which clusters successively switch from control to intervention, in an order randomly assigned, until all clusters are eventually exposed to the intervention (Table 1). In our study, all hospitals (clusters) start by implementing the WHO SOC for severe pneumonia (control arm) and are randomly allocated a time at which they will transition to implementing the intervention. Depending on the time at which they are enrolled in the study, children pertain either - and exclusively – to the control arm, or the TB-Speed intervention arm.

Table 1 Stepped wedge implementation of the intervention in participating hospitals
Full size table

Study settings

The impact of this innovative approach may vary with TB incidence as well as geographical and seasonal variability that can affect the prevalence and aetiology of pneumonia in young children. To provide a better basis for the generalisability of results, the trial takes place in six countries. These include high and very high tuberculosis incidence countries with different epidemiological and environmental backgrounds, in Sub-Saharan Africa (Cameroon, Cote d’Ivoire, Mozambique, Uganda, and Zambia) and South East Asia (Cambodia) (Table 2). The trial is implemented in 15 national or regional reference hospitals with previous research experience.

Table 2 Implementing sites
Full size table

Randomisation

Randomisation is stratified by the estimated country TB incidence rate, classified as either high (100 to < 300/100,000 patients-years; Cameroon, Côte d’Ivoire and Uganda) or very high (≥ 300/100,000 patients-years; Cambodia, Mozambique and Zambia) [9] (Table 1). Within these strata, a computer-generated random sequence will determine the order in which hospitals move from control to intervention. The statistician of the international coordination team, based at University of Bordeaux, will prepare the randomization sequence before the start of the trial.

The time when a new cluster receives the intervention is called a step. In the TB-Speed Pneumonia study, the period between two successive steps is 5 weeks (Table 1). The international coordination team and country research teams are blinded to the randomisation order. The international coordination team however is informed 10 weeks in advance (i.e. two periods) of the next site to switch, while study sites are notified 5 weeks prior to their crossover date to initiate program planning.

Study population

Any child younger than five years presenting with signs and symptoms of presumptive severe pneumonia to outpatient, emergency units, intensive care unit, or paediatric departments of the selected hospitals is screened for eligibility for the trial as soon as possible. Eligibility to participate includes the following criteria: (1) aged 2 to 59 months, (2) newly hospitalized for WHO-defined severe pneumonia (Table 3), (3) informed consent signed by parent/guardian. Ongoing TB treatment or history of intake of anti-TB drugs in the last 6 months is the only exclusion criterion.

Table 3 WHO criteria for severe pneumonia
Full size table

Trial intervention strategy

The WHO standard of care for children with severe pneumonia

All children admitted in the hospital and presenting with WHO-defined severe pneumonia are immediately managed as part of routine care per the WHO SOC for children with severe pneumonia (Table 4).

Table 4 The WHO standard of care for young children with severe pneumonia
Full size table

As recommended by the WHO for TB assessment in the context of the SOC, if the child presents with persistent cough and fever for more than two weeks, and signs of pneumonia after adequate antibiotic treatment, s/he is evaluated for TB using routine procedures. Xpert or Ultra, depending on local availability, can be used in children with clinical suspicion of TB (chronic symptoms, failure to respond to antibiotic treatment or TB exposure) according to the clinician’s judgement. This can be done using the standard sample collection methods usually implemented at the inpatient ward.

The TB-Speed intervention

The intervention consists of the WHO SOC for children with severe pneumonia plus the trial intervention consisting of systematic, rapid detection of TB on the day of hospital admission using the Ultra assay performed on 1 NPA and 1 stool sample. Ultra testing on NPA is performed immediately, either at the hospital laboratory with the standard GeneXpert device, or undertaken in the ward or in a side-laboratory next to the ward using a one-module GeneXpert device (G1 Edge®, Cepheid). The sample flow has been organised in order to reduce turnaround time for results to 3 h. Ultra testing on stool, which requires prior processing, is performed at the hospital laboratory. Drugs are available at the inpatient level to enable immediate initiation of TB treatment, as soon as a positive Ultra result is released.

Implementation procedure

Aggregated data on severe pneumonia management were collected in all hospitals two months prior to the start of the study to document routine practices for the SOC for severe pneumonia in children (number of hospitalizations, inpatient mortality, antibiotic use, access to oxygen therapy and other supportive care). Adherence to the WHO SOC for severe pneumonia was reinforced by initial study training and is monitored throughout the study implementation. The study provided equipment for oxygen therapy (oxygen concentrators) where needed, and pulse oximeters. It is expected that the provision of equipment for oxygen therapy and training on pneumonia case management in children will reduce the variation in characteristics and practices between sites.

Study visits and assessments

After written informed consent is obtained from parent(s)/guardian(s) by the study clinician or study nurse, the baseline visit includes a complete clinical evaluation, a digitalized chest X-ray (CXR), HIV and malaria testing, and a complete blood count. For children in the intervention arm, initial bacteriological specimen collection is done as soon as possible and within 24 h of hospital admission, including one NPA collected by the nurse on the day of admission, and one stool sample collected as soon as the child is able to produce stool. Additionally, blood samples and leftovers from NPA and stool are collected for future biomarkers studies in children for whom parent(s)/guardian(s) give their consent for biobanking. Data about routine care as well as additional study strategies are collected. All children are followed up for a total duration of 12 weeks, with four systematic protocol visits planned at day 3 (and/or at hospital discharge), 2 weeks after discharge (as recommended by the WHO SOC), and 12 weeks after recruitment. Each follow-up visit comprises a clinical evaluation, and collection of medical history since the last visit, an evaluation of adherence to TB treatment if initiated, and TB drug prescription and dispensation to cover the time until the next visit (Table 5). Parent(s)/guardian(s) are invited to bring their child back to the hospital in case of new symptoms for an unscheduled (extra) visit. During the final visit at 12 weeks, a second digital CXR is performed, as well as an assessment of TB disease evolution and TB treatment outcome in those diagnosed with TB (improvement, treatment failure, death, or lost to follow-up).

Table 5 Study assessments and specimen collection
Full size table

Tolerability and acceptability of the intervention

Assessment of the tolerability and acceptability of NPA and stool specimen collection procedures will be undertaken in a subset of children. Tolerability is defined by the child’s perceived level of discomfort/distress/pain as assessed by the child him/herself, the parents/guardians and the nurses using the Wong-Baker Face scale, the Visual Analog Scale, and the FLACC (Face Legs Activity Cry Consolability) behavioural scale, respectively. Nurses’ and parents’ acceptability regarding the whole sampling and testing strategy will be evaluated using both quantitative (self-reported questionnaire) and qualitative methods (semi-structured interviews).

Safety assessment

Occurrence of Adverse Events (AEs) is monitored at each visit by study nurses and clinicians for children receiving the TB-Speed intervention. Expected AEs occurring from NPA collection include, by decreasing order of frequency: cough, nausea, local trauma/nose bleeding, sneezing, vomiting, and in rare cases dyspnoea/low oxygen saturations and bradycardia < 60/bpm [33]. No AEs are expected from stool sample collection. Since this is a diagnostic trial without investigational medicinal product, and very low expected risk of AEs linked to the intervention, there is no systematic notification of severe AEs (SAEs) to the sponsor with the exception of: 1) death; 2) grade 4 clinical AEs (excluding asymptomatic biological grade 4 AEs); or 3) SAEs related to NPA collection. Grading will be done using the 2017 Division of AIDS Table for Grading the Severity of Adult and Pediatric Adverse Events [27].

Endpoints

Primary study endpoint and measure

The primary endpoint is all-cause mortality 12 weeks after recruitment. Mortality due to severe pneumonia (related or not to TB) is expected to occur early. We expect that a 12-week period is long enough to assess impact on mortality of TB treatment empirically started in children with poor clinical progress during the first weeks of follow-up. If the children are not brought to the Week 12 visit, parents will be contacted, and home visits will be organized in order to collect the vital status of the child.

Secondary study endpoints and measures

Several secondary endpoints will further compare the intervention and control arms and document the feasibility and acceptability of the intervention (Table 6). Of note, some endpoints are collected all along the trial for process monitoring and assessment of the feasibility of the TB-Speed intervention. Among others, they include the proportion of children with NPA and stool samples collected as per protocol and the time to sample collection and Xpert result after collection.

Table 6 Secondary endpoints
Full size table

Data analysis

We will perform final data analysis to answer the objectives at the end of the study after data review and database closure. At the request of the trial Independent Data Monitoring Committee (IDMC), an interim analysis on safety and feasibility data is planned at the point when all children enrolled during the first 9 months of the study have been followed for 12 weeks; there are no stopping rules planned and no formal statistical test will be used. We will conduct a descriptive analysis of the endpoints and other measured variables and characteristics (Table 6). Quantitative variables will be summarized using means, standard deviation, median and interquartile range. Categorical variable will be analysed using frequencies and proportions.

The analysis of the primary endpoint will be performed following an intention-to-treat principle. Patterns of missing values will be analysed but no imputation will be done. As the literature regarding statistical analysis of stepped wedge designs is constantly growing, we will follow the recommendations of Hemming and colleagues [28], using the Hussey and Hughes model for the primary analysis [29]. Thus, we will used generalised linear mixed models with logit link and binary distribution. We will model individual binary response (vital status) with condition (control vs intervention) and time-period as fixed effects and sites as random effects to adjust for clustering of children. Sensitivity analyses will be performed to control for 1) important prognostic factors such as age, malnutrition, HIV infection, severity criteria measured at recruitment, 2) seasonality which could be different in different countries and, 3) different model specifications to better adjust secular trends and cluster heterogeneity [28]. These model specifications are developed and detailed in a statistical analysis plan that will be validated by the IDMC before performing the analysis. All results will be reported as estimates of effect (odds ratios for binary variables) and corresponding 95% confidence intervals. We will conclude that the intervention is better than the control if the mortality odds ratio is statistically lower than 1 (p < 0.05).

All analyses will be performed using R (version 3.6.0 or higher) and findings will be reported as per the CONSORT extension for reporting of stepped-wedge cluster-randomized trials [30].

Power and sample size

Sample size calculations were performed based on several assumptions. (1) The expected proportion of TB cases in children with pneumonia would be 15 and 24%, in high and very high incidence settings, respectively, with a proportion of 33% of culture confirmed cases among TB cases overall (2). In the context of our intervention, Ultra could detect the majority of confirmed TB cases, versus a hypothesized TB detection rate of 25% in routine conditions. (3) The intervention would raise awareness about TB in site clinicians and would lead to an increase in detection rate from 15 to 50% between the control and the intervention arm. (4) We estimate that in the control arm the overall mortality will reach 15%, in line with mortality associated to severe pneumonia in previous studies, and that the intervention would therefore lead to a 30% reduction in the overall mortality rate (10.5%).

According to these assumptions, the sample size for an individual randomized trial would be 1730 children. For financial and logistical reasons, we retained an ICC value of 0.005, corresponding to a design effect of 2.16 [31]. Using the estimated mortality of 15% in the control arm, an expected reduction in mortality in the experimental arm of 30%, an alpha of 0.05, a power of 80%, an ICC of 0.005 and 1% of incomplete data, the corresponding sample sizes would be 3780 children. This sample size randomised across the two strategies, in 15 hospitals over 16 periods, resulted in a mean of 15.8 children enrolled per hospital per time period (or 252 children per hospital for the entire study), with no restriction if included numbers exceed this target.

Data management and confidentiality

Patient data will be recorded into an electronic case report form by study nurses through single data entry on tablets using the REDCap application. All electronic data will be kept on a password-protected, secured server hosted at the University of Bordeaux, accessible only to researchers involved in this trial. Each trial participant will be assigned a unique study identification code.

Oversight

The trial is coordinated primarily by the international Central Coordination Unit at University of Bordeaux, France and is overseen by the Trial Steering Committee. A Scientific Advisory Board (SAB) provides advice on the relevance and validity of the project design and implementation, monitors progress and ensures scientific and ethical integrity of the project. An IDMC acts as a consultative board for the SAB and the sponsor. It has access to overall safety and efficacy data, as well as to any information justifying continuation or discontinuation of the trial. However, the IDMC will not apply stopping rules and interim analyses as per standard clinical trials since the stepped wedge design does not allow for it.

Sub-studies

Cost-effectiveness study

We hypothesise that benefits in terms of survival and increased Disability-Adjusted Life Years (DALYs) over a lifetime horizon will justify extra costs incurred by systematic Ultra testing in children with severe pneumonia. A mathematical model will be developed to project health economic outcomes, including TB cases and mortality in children with severe pneumonia. Cost-effectiveness analysis will be from the health payer perspective and only direct healthcare costs will be included. Budget impact analysis will be conducted to evaluate the expected costs of implementing the TB-Speed approach on healthcare budget at 2- and 5-year horizons in the countries participating in the project.

Biomarkers studies

Baseline samples, including NPA and stool leftovers, whole blood and plasma samples collected at inclusion, will be frozen and stored at the country clinical trial unit laboratory. The trial provides a unique opportunity to investigate a number of TB biomarkers, which could discriminate active disease from latent TB infection as well as TB from non-TB pneumonia using transcriptomic approaches. It also permits the study of the molecular epidemiology of Mycobacterium tuberculosis, and further characterisation of the proteomic, metabolic and immunologic profiles of children presenting with signs of severe pneumonia, with or without TB [32,33,34,35,36]. Biological samples will be retained for 10 years after study completion, unless there are objections expressed by parent(s)/guardian(s).

Trial status

Recruitment to the trial started in March 2019. Recruitment will continue until September 2020, with the last visit of the last participant in December 2020. The current protocol is Version 2.0 dated November 22nd, 2019.

Discussion

TB-Speed Pneumonia is a pragmatic diagnostic trial with a stepped-wedge cluster-randomised design, and a stratified randomisation. The stepped wedge cluster-randomised design enables evaluation of the study hypothesis using a pragmatic and operational approach, i.e. for an intervention which remains to be tested in a real-world setting. To our knowledge, this is the first time a stepped wedge trial has been implemented at the international level [37,38,39].

Strengths and challenges of the stepped-wedge design

The stepped wedge design is particularly relevant where it is predicted that the intervention will ultimately do more good than harm, as is the case for our intervention, but where there is uncertainty as to its effectiveness and its safety in a specific setting and population [40]. The choice of the stepped wedge design was also based on our hypothesis that the intervention would raise TB awareness among clinicians and may lead to more empirical TB treatment initiated in the intervention arm as compared to the control arm, thus benefitting all children hospitalized, and impacting beyond the research settings. Since clusters are expected to remain until the study ends, they will all eventually implement the intervention. The stepped-wedge design is therefore particularly adapted to capture such effects. From a logistical point of view, a phased roll-out of the intervention is easier to implement in the context of an international multicentre study. Randomisation per hospital is also advantageous due to the difficulty at health facility level of randomising children individually to one of the two strategies, additionally minimising contamination between the two arms.

A major limitation of stepped wedge designs is that blinding of the study team to the intervention is not possible. However, since the primary trial endpoint (mortality at 12 weeks) is not subjective, there is no risk of ascertainment bias. As the intervention relates to the management of children upon hospital admission, outcomes are estimated only from individuals with no prior exposure to the control, thus avoiding carryover (residual) effect [41]. Moreover, in order to guarantee that children benefit from the same quality of care across study sites, adherence to WHO SOC for severe pneumonia will be monitored throughout the study implementation.

Challenges to this design include the need for repeated training activities and increasing workload for coordination teams as more clusters start the intervention. Although all hospitals will finally implement the intervention, some of them will remain in the control arm for a long time, requiring continuous engagement to avoid drop-out and demotivation. In this study, the number of participants is not capped, which can be challenging for resource planning. Since all study sites are expected to start on the same day, as per the stepped wedge design, any delay in opening a site impacts all others.

In this multi-country study with different epidemiological and environmental backgrounds, one can expect to face variability in the number of recruitments across clusters and time-periods (seasonal effect). Although the impact of unequal cluster size on statistical precision has been previously investigated, further research is needed to apply previous results to binary outcomes and to consider size variation across clusters and time-periods [42].

Conclusion

TB-Speed Pneumonia will test an innovative intervention to diagnose TB in children with severe pneumonia, in a pragmatic cluster randomised stepped-wedge trial. In addition to testing the main hypothesis that molecular detection and early treatment will reduce TB mortality in children, the strength of such pragmatic research is that it helps to provide some evidence regarding the feasibility of the intervention as part of routine care. A systematic review, commissioned by WHO in 2019, provided additional evidence on the use of Xpert MTB/RIF and Ultra as initial diagnostic tests for TB in children, specifically in NPA and stool specimens. The sensitivity of Xpert MTB/RIF (as compared to liquid culture of a respiratory specimen) was of 46 and 61% on NPA and stool specimens respectively, as compared to 65% on sputum and 73% on gastric specimens, and the sensitivity of Ultra was of 46 and 73% on NPA and sputum, respectively. The specificity for all samples was above 98% (97% for Ultra) [43].

Should this intervention be successful, safe and well tolerated, it could be systematically implemented in order to increase TB diagnosis in children with severe pneumonia and reduce childhood mortality due to TB.