Abstract
Background Ambitious population-based screening programs for latent and active tuberculosis (TB) were implemented in the Republic of the Marshall Islands (RMI) in 2017 and 2018.
Methods We used a transmission dynamic model of TB informed by local data to capture the historical dynamics of the RMI epidemic. We then used the model to project the future epidemic trajectory following the active screening interventions, as well as considering a counterfactual scenario with no intervention. We also simulated future scenarios including periodic interventions similar to those previously implemented, to assess their ability to reach the End TB Strategy targets and TB pre-elimination in RMI.
Findings The screening activities conducted in 2017 and 2018 were estimated to have reduced TB incidence and mortality by more than one third in 2020, and are predicted to achieve the End TB Strategy milestone of 50% incidence reduction by 2025 compared to 2015. Screening interventions had a considerably greater impact when latent TB screening and treatment was included, compared to active case finding alone. Such combined programs implemented at the national level could achieve TB pre-elimination around 2035 if repeated every two years, and around 2045 if repeated every five years.
Interpretation Our model suggests that it would be possible to achieve TB pre-elimination by 2035 in RMI through periodic repetition of the same interventions as those already implemented in the country. It also highlights the importance of including latent infection testing in active screening activities.
Funding Australian National Health and Medical Research Council and World Health Organization.
Introduction
Every year, around ten million people develop tuberculosis (TB) disease globally.1 The causative agent of TB (Mycobacterium tuberculosis, M.tb) was first identified in 1882 – and yet more than 130 years later, TB is estimated to be the world’s leading cause of death from a single infectious agent.1 The availability of effective vaccines and treatments has not been sufficient to eliminate TB. Consequently, disease elimination is not expected in the coming decades without dramatic changes to current approaches to public health management. Improving case detection is a promising way to progress the global TB response because millions of diseased individuals are currently undetected.1 We now know that ACF activities can identify significant numbers of individuals with active disease who may not have been detected otherwise or may only have been detected after substantial delays. Recent large-scale randomised trials of active case finding (ACF) interventions in high burden settings have demonstrated the efficacy of these approaches and generated great hope for dramatically improved control by reducing the pool of infectious individuals.2,3 Further in the past, large-scale ACF programs achieved dramatic and sustained reductions in TB burden in developed countries, such as Australia, after the Second World War.4
While classic ACF approaches focus on active disease only, screening and treatment of latent TB infection (LTBI) can also be included in these ambitious programs in order to prevent future TB reactivation in infected individuals. Population-wide use of preventive treatment (PT) is not a new intervention. Large-scale cluster-randomised trials of isoniazid PT were conducted in the late 1950s in Alaska, Greenland, and Tunisia, all demonstrating significant reductions of TB incidence.5–7 However, a more recent trial conducted in a South African mining population with high prevalence of HIV was not successful at reducing TB burden.8 This demonstrates that the effectiveness of mass PT programs depends on the characteristics of the population and epidemic in which the intervention is implemented,9 and that it is difficult to anticipate the effects of future intervention by extrapolating observations made in other settings.
Ambitious population-level screening programs were recently completed in the Republic of the Marshall Islands (RMI), an archipelago nation in the western Pacific. RMI is a high TB burden country with an estimated TB incidence approaching 500 cases-per-100,000-persons-per-year in 2019 according to the World Health Organization (WHO).10 The screening activities were conducted in the two most populous islands of RMI – Ebeye and Majuro – together representing nearly three quarters of the national population. In 2017, a first ACF intervention aimed to screen the entire adult population (those aged 15 years or older) of Ebeye Island for active TB disease and treat those with suspected disease. The following year, an even more comprehensive intervention was conducted on Majuro Atoll to screen individuals of all ages for both latent and active TB and provided appropriate treatment for both these conditions. These programs have enabled identification and treatment of a large number of individuals with latent and active TB. It is now critical to estimate the long-term effects of these interventions and to identify effective follow-up approaches that would sustain significant reductions of TB incidence in RMI.
Mathematical modelling has the ability to predict the long-term effects of control interventions, as well as to guide future TB control by considering different hypothetical scenarios of intervention. In this modelling study, we incorporated local data collected during the interventions into a transmission dynamic model of TB in order to assess whether the End TB Strategy targets and pre-elimination could be achieved in RMI under the current strategy as well as under various future intervention scenarios.
Methods
Overall approach
We used a deterministic compartmental model to simulate M.tb transmission in RMI using a similar approach to previously published studies.11–14 After calibrating the model to local data using Bayesian techniques, we projected the long-term effect of the large-scale LTBI and TB screening activities undertaken in RMI in 2017 and 2018. The projected epidemic trajectories were compared to a counterfactual scenario corresponding to the predicted epidemiological trajectory in the absence of these interventions. We then projected repeated periodic interventions similar to those previously implemented and covering the entire country every two, five or ten years, to assess prospects for reaching the End TB Strategy milestones and targets, and the pre-elimination incidence threshold (see Table 1).
Tuberculosis model
The base model consisted of six compartments that each represent a particular clinical status with regards to M.tb infection or disease (Figure 1). LTBI was modelled using two sequential compartments (E and L) in order to capture the declining risk of disease progression over time from infection.15,16 Individuals progressing to active TB (I) were stratified based on their form of disease: smear-positive pulmonary TB (PTB), smear-negative PTB and extrapulmonary TB. Diseased individuals may either recover spontaneously, receive treatment (T) or die. Recovered individuals were modelled as a separate compartment (R), such that differential susceptibility to future infection could be considered, as compared to the original risk of infection associated with the “susceptible” population. Reinfection was also permitted for latently infected individuals and was modelled through a flow returning from the late latent compartment to the early latent compartment, with LTBI assumed to confer partial protection against reinfection (Table 2).
The simulated population was then stratified by age according to five categories: 0-4, 5-14, 15-34, 35-49 and 50+ years old. Finally, the model was further stratified at the population level to explicitly simulate three different locations: Majuro Atoll, Ebeye Island and all other islands of RMI. We incorporated heterogeneous mixing by age and location. The impact of type 2 diabetes was accounted for by adjusting the age-specific rates of progression from latent to active disease based on age-specific diabetes prevalence estimates. Age-specific diabetes prevalence was assumed to increase progressively between 1950 and 2020 and remained constant in the future in our base case analysis. Both increasing and decreasing future diabetes prevalence were explored in sensitivity analyses. The effect of BCG vaccination was modelled by adjusting the age-specific susceptibility to infection and this adjustment varied with time to reflect the progressive scale-up of BCG coverage.
We used programmatic data to inform the simulated detection and treatment processes and previously published estimates to inform the natural history of TB,17 as well as the rates of progression from latent to active TB.16 Model parameters were fitted to local data on population size, TB prevalence, LTBI prevalence and TB notifications using an adaptive Metropolis algorithm. This calibration approach allowed consideration of the substantial uncertainty around the most critical model parameters (Table 2). In particular, various profiles of time-variant passive TB screening were considered to cover a wide range of assumptions regarding the future performance of passive case finding. The code used to implement the model is publicly available on Github (https://github.com/monash-emu/AuTuMN/tree/master/apps/tuberculosis/regions/marshall_islands).18
Modelled interventions
When LTBI screening and treatment were implemented, latently infected individuals were assumed to transition to the recovered compartment on completion of treatment. The transition rate associated with the preventive treatment was defined as the product of a time-variant LTBI screening rate, the sensitivity of the LTBI test and the efficacy of preventive treatment. ACF was modelled through an additional flow from the undetected disease compartment to the treatment compartment. The rate governing this transition was defined by a time-variant ACF screening rate, which was multiplied by the sensitivity of the diagnostic test used during the ACF intervention.
The time-variant functions used to characterise both the LTBI screening rate and the ACF screening rate were defined such that the modelled proportions of screened individuals at the end of the interventions were consistent with those measured in the field. In Ebeye, 85% of individuals aged 15 years old and over were screened for active TB, while in Majuro 81% of the entire population were screened for both LTBI and active TB. In order to capture the future effect of increased TB awareness due to the large-scale community interventions, we increased the rate of passive screening of active TB from 2018 - the end point of the completed intervention, considering relative improvements in the screening rate ranging from zero to 50%.
In addition to simulating the previously implemented interventions, we modelled scenarios in which the community-wide interventions would be repeated periodically at the national level, starting from 2021. We considered interventions repeated every two, five or ten years in separate scenarios and we assumed a similar proportion of the population screened to that of the intervention conducted in Majuro in 2018. Finally, we also estimated the effect of intensive contact tracing and preventive treatment provision in the contacts of individuals with active TB.
Role of the funding sources
The funders of the study had no role in the collection and analysis of data. TI and KR are employed by the WHO, one of the funders of this study, but contributed as technical experts. They both contributed to developing the concept, interpreting the data, reviewed the manuscript and agreed to publish it.
Results
Figure 2 presents the base case model fits to local data and posterior parameter estimates are shown in the Supplement (Table S3).
Figure 3 presents the future projected epidemic trajectories with and without implementation of the active screening interventions. At the national level, we estimated that the interventions had reduced TB incidence by around 37% in 2020 from 447 (95% credible interval 353-548) to 281 (95% CrI 221-363) per-100,000-persons-per-year. TB mortality was predicted to have decreased by around 42% from 127 (95% CrI 108-160) to 74 (95% CrI 61-100) per-100,000-persons-per-year in 2020 due to the screening interventions. The community-wide interventions had a considerably greater effect on the epidemic in Majuro than in Ebeye. In Majuro, the interventions achieved an estimated 55% reduction in the local TB incidence and mortality in 2020, compared to the counterfactual no intervention scenario, with effects sustained until at least 2050 with no further intervention. In Ebeye, the estimated incidence reduction induced by the screening activities reached 43% in 2050, compared to the counterfactual no intervention scenario. If current programmatic conditions were continued with no future ACF interventions and assuming enhanced TB awareness following the interventions, we predicted that RMI could achieve the 2025 End TB incidence milestone (Figure 4). However, the country would fall short of reaching any TB mortality targets as well as the 2035 TB incidence target.
The results of our analysis considering different assumptions for the future trend of diabetes prevalence are shown in Supplementary Figure S2. This analysis included the interventions previously conducted in Majuro and Ebeye and assumed that TB control would remain similar to the current programmatic situation until 2050. If diabetes prevalence increased by 20% by 2050, the predicted TB incidence in 2050 would be 140 (95% CrI 73-245) per-100,000-persons-per-year. In contrast, a decline of 20% in diabetes prevalence was estimated to be associated with a further decrease in TB incidence to 92 (95% CrI 51-151) per-100,000-persons-per-year in 2050.
The results of the projections considering repeated interventions at the country level every two, five or ten years are presented in Figure 4. If LTBI screening was not included in the ACF programs, we predicted that the 2035 End TB Strategy targets would not be reached, regardless of the intervention frequency considered. Combining ACF with LTBI screening and treatment would have considerably greater impacts on future TB burden. We estimated that all the End TB Strategy targets would be reached, whether this intervention was repeated every two, five or ten years. Similar results to those presented in Figure 4 are shown on a log-scale in Supplementary Figure 3, also displaying the pre-elimination threshold defined as a TB incidence rate of 1 case per-100,000-persons-per-year.19 This showed that pre-elimination could be achieved around 2035 with community-wide screening of latent and active TB repeated every two years at the country level, and around 2045 with the same intervention implemented every 5 years. The ten-year cycle was not predicted to reach pre-elimination by 2050.
Our analysis considering the use of preventive treatment provided through intensive contact tracing showed that such an intervention would yield a comparable overall effect compared to using community-wide ACF at the country level every 5 years. This intervention would not achieve any of the targets set by the End TB strategy for 2035.
Discussion
In 2017 and 2018, unprecedented TB and LTBI screening activities were conducted in RMI. These ambitious community-led interventions were supported by exceptional efforts produced by local and external stakeholders and supported by volunteers. Success has already been demonstrated through the large number of detected individuals with latent or active TB who completed treatment. Our modelling projections now suggest that these efforts will have considerable effects on the long-term trajectory of the local TB epidemics, and we estimate that periodic use of interventions such as those implemented in RMI could achieve pre-elimination goals over the coming decades.
ACF is a high-intensity, high-resource effort. Going beyond TB case-finding by adding treatment of LTBI greatly increases the cost and the time to screen a population, but our model demonstrates a considerably higher impact when both active and latent TB are addressed together. This suggests that considerable resources are likely to be required to yield the dramatic impacts on TB burden we projected. However, our analysis also shows that impacts could be sustained in the long term and that pre-elimination could be achieved through such interventions. This means that substantial returns on investment are anticipated as many TB cases and deaths will be averted over several decades. In addition to the human and societal benefits, these direct health effects also translate into economic savings because of the known catastrophic impact that TB has on the finances of individuals, families and countries.20 Our results therefore highlight the importance of adopting a long-term vision when planning TB control and when funding the TB response.
Our analysis has important implications for future TB control in RMI. First, we suggest that while it is critical to address the issue of latent infection, classic approaches relying on tracing the contacts of diagnosed TB cases would be unable to achieve dramatic reductions in burden. An explanation for this is that such strategies would not capture infections that result from old transmission events, whereas these remote infections may still contribute to the future burden of active TB through late reactivation. This suggests that population-wide screening of LTBI may be a necessary component of any program aiming to achieve TB pre-elimination in RMI. The critical importance of reducing the size of the latent infection pool had already been demonstrated by previous modelling works, which suggested that even if transmission totally ceased from 2015 onwards, this would be insufficient to achieve the End TB Strategy targets at the global level.21 In the present analysis, we demonstrate that these goals could be achieved by using large-scale screening interventions that are realistic, practical and utilise only existing technologies, given that such interventions have already been implemented in the field. Our study also reinforces the importance of addressing risk factors and comorbidities in addition to TB itself. In particular, we found diabetes to be a key determinant of the long-term trends of the TB epidemic in the RMI context, consistent with findings from other Pacific Island settings.14 This calls for multifactorial approaches that combine TB-specific measures with interventions addressing TB risk factors, whether these tools are medical or social. Finally, our model suggests a clear pathway to reaching TB pre-elimination in RMI. Indeed, we estimated that TB incidence could be brought below one new case-per-100,000-persons-per-year by 2035 if the ambitious intervention conducted in Majuro in 2018 could be repeated every two years from 2021 at the national level.
The strengths of our study include the fact that our calibration approach was able to capture key disease indicators accurately while incorporating uncertainty around the most fundamental parameters. Furthermore, the model was directly informed by the most relevant data possible, since these were directly measured in the field, including through the interventions themselves. This includes data on the prevalence of active and latent TB which are critical to replicate the local TB burden accurately, along with TB notification data that ensured that the historical trends in case detection were captured appropriately. Finally, simulations were conducted using state-of-the-art computing techniques that are publicly available and have been extensively tested and documented,18 such that our model could be easily reused to assist TB control in RMI as well as in other settings.
Limitations of our study include the significant uncertainties that remain in TB epidemiology. In particular, model projections could be refined with improved knowledge about the effect of preventive therapy on the risk of future reinfection, although our model considers a broad range of assumptions regarding this parameter. Similarly, even if recent studies have produced estimates for the rates of progression from latent to active TB,16,22 the rate of late reactivation remains poorly characterised due to the limited data available.23 Refining estimates of this parameter would undoubtedly increase the accuracy of predictions related to intervention involving preventive treatment. Second, while our model demonstrates the effects of two different strategies for ACF in a high-incidence area, it may not capture all impacts of coordinated TB screening efforts. More nuanced impacts to a local or national TB program could include implementation of improvements to TB diagnosis, scaling up efforts to identify close contacts, and increasing resources for LTBI identification and management among contacts and other high-risk groups. These programmatic changes are more difficult to quantify in a way that could be incorporated into a model, and could also lead to sustainable reductions in TB incidence. Third, the model allows interactions between individuals of different islands, although empiric data were not available to quantify this social mixing process. Therefore, the extent of the diffusion of the interventions effects between the different locations relies on our assumption that 95% of interpersonal contacts occur between individuals of the same geographical stratum. Finally, we stress that the projections presented in this study are intended to be primarily relevant to the RMI context and, given the variability of TB epidemics in different settings, drawing quantitative conclusions for other settings would require dedicated analyses.
Repeating screening of LTBI within the same population may present some practical issues that are not considered in this modelling study. It can be difficult to interpret positive tests in individuals who were previously treated since LTBI tests frequently remain positive after treatment. Thus, some individuals may be treated unnecessarily if all positive tests were interpreted as evidence of current infection. Conversely, reinfection occurs, and some infections may be missed if positive tests were systematically interpreted as false positives in previously treated individuals, in which case testing would become redundant for this population. These issues suggest that repeating active screening programs every five years or every ten years may be more realistic than using two-year cycles. They also highlight the importance of developing highly accurate infection screening tools that could distinguish between previous and current infections or to identify biomarkers that could predict short-term progression risk.
In conclusion, our analysis suggests that while the End TB Strategy targets may not all be reached in RMI under the current programmatic situation, it would be possible to achieve pre-elimination in the next ten-to-twenty years through periodic repetition of the same interventions as those already implemented in the country, using a similar screening rate at the national level. If this is to occur, it is a rare example of the capability of existing tools to achieve END TB targets.
Data Availability
The data and code used to implement the study are available on a Github repository.
https://github.com/monash-emu/AuTuMN/tree/master/apps/tuberculosis/regions/marshall_islands
Contributions
RR, BMW, TI and JMT designed the study. AL, JNJ, TJ and MKL contributed to the collection and processing of the data used to inform the model. RR, BMW, ESM and JMT designed the mathematical model. RR and JMT implemented the code. RR conducted the experiments. All authors contributed to the interpretation of the findings. RR wrote the first version of the manuscript. All authors contributed to the final version of the manuscript.
Declaration of interests
The authors declare that there is no conflict of interest.
Acknowledgments
This work was funded by the Australian National Health and Medical Research Council (Project Grant APP1144570 and Fellowship Grant APP1142638) and the Western Pacific Regional Office of the World Health Organization. We gratefully thank Dr Richard Brostrom for his detailed description of the interventions, for providing access to data and for his advice on how to capture the local TB epidemiology accurately. We also thank the nurses, doctors and all persons involved in the implementation of the screening activities and who contributed to collecting the data used to inform this modelling study. In particular, we thank the late Dr. Kennar Briand, Herokko Neamon, Risa Bukbuk and Mareta Hauma for their contribution to this work.