Feasibility of achieving the 2025 WHO global tuberculosis targets in South Africa, China, and India: a combined analysis of 11 mathematical models

Summary Background The post-2015 End TB Strategy proposes targets of 50% reduction in tuberculosis incidence and 75% reduction in mortality from tuberculosis by 2025. We aimed to assess whether these targets are feasible in three high-burden countries with contrasting epidemiology and previous programmatic achievements. Methods 11 independently developed mathematical models of tuberculosis transmission projected the epidemiological impact of currently available tuberculosis interventions for prevention, diagnosis, and treatment in China, India, and South Africa. Models were calibrated with data on tuberculosis incidence and mortality in 2012. Representatives from national tuberculosis programmes and the advocacy community provided distinct country-specific intervention scenarios, which included screening for symptoms, active case finding, and preventive therapy. Findings Aggressive scale-up of any single intervention scenario could not achieve the post-2015 End TB Strategy targets in any country. However, the models projected that, in the South Africa national tuberculosis programme scenario, a combination of continuous isoniazid preventive therapy for individuals on antiretroviral therapy, expanded facility-based screening for symptoms of tuberculosis at health centres, and improved tuberculosis care could achieve a 55% reduction in incidence (range 31–62%) and a 72% reduction in mortality (range 64–82%) compared with 2015 levels. For India, and particularly for China, full scale-up of all interventions in tuberculosis-programme performance fell short of the 2025 targets, despite preventing a cumulative 3·4 million cases. The advocacy scenarios illustrated the high impact of detecting and treating latent tuberculosis. Interpretation Major reductions in tuberculosis burden seem possible with current interventions. However, additional interventions, adapted to country-specific tuberculosis epidemiology and health systems, are needed to reach the post-2015 End TB Strategy targets at country level. Funding Bill and Melinda Gates Foundation


Introduction
In May, 2014, the World Health Assembly approved the post-2015 End TB Strategy, setting "ambitious but feasible" targets for reducing the global burden of tuberculosis by 2035. 1,2 The strategy is aiming for a 50% reduction in global tuberculosis incidence and a 75% reduction in global tuberculosis mortality by 2025, and 90% and 95% reductions in these outcomes, respectively, by 2035. 2 Policy makers must identify what interventions, and at which level of scale-up, will be needed to meet these targets at country level.
The End TB targets are deliberately ambitious, and any single intervention (defi ned here as a group of activities leading to an improvement in a specifi c area of tuberculosis control-eg, treatment outcomes) is unlikely to achieve these goals. 3 Instead, national tuberculosis programmes will need improvements across the tuberculosis care pathway, together with preventive measures.
The End TB Strategy describes two phases of future eff orts to control tuberculosis. 2 In phase 1, the focus of this Article, progression towards the 2025 milestones will largely depend on optimising the use of existing tools, enabled by investments in universal health coverage and social protection. 4 Post-2025 in phase 2, novel tools (diagnostics, drugs, and vaccines) are expected to enable further acceleration of tuberculosis decline towards the 2035 targets. 1 For both phases of the End TB Strategy, policy makers require guidance about which interventions and technologies to use-questions that are unlikely to be answered by empirical studies, given the diffi culty of testing all possible approaches at high scale before policy decisions are made.
Mathematical modelling is a powerful tool to support policy discussions, because several hypothetical intervention strategies can be compared in a systematic framework to project future trends. [5][6][7] Multimodel exercises for HIV 8,9 have illustrated how diff erences in model design can infl uence results, yet this structural variation is not apparent in a single-model analysis. By comparing answers to the same question using diff erent models, we can identify fi ndings robust enough to account for betweenmodel variation, and as such contribute to the evidence needed to commit the resources for national-level policy initiatives. The points where model projections diverge can signal important knowledge gaps to be addressed by future research. Paired with information on use of resources, projections of health impact also can be used to estimate the cost-eff ectiveness of competing policy options and to design optimum policy portfolios, which is beyond the scope of this paper.
To provide results relevant to individual countries, modelled analyses should be consistent with existing evidence on local tuberculosis epidemiology (eg, incidence, mortality, prevalence, and multidrug-resistant [MDR] tuberculosis) and tuberculosis control activities (eg, treatment success, and linkage to care), and tailor intervention scenarios to local needs and capabilities. Because the required information is not always available in a systematic way in the public domain, involvement of local experts is key.
In this Article, we describe epidemiological projections from 11 independently developed dynamic transmission models of tuberculosis, exploring the feasibility of the 2025 End TB Strategy targets in China, India, and South Africa. These analyses explore a range of policy scenarios, incorporating perspectives from national tuberculosis programmes and advocacy communities. China, India, and South Africa account for approximately 40% of the global tuberculosis burden 10 and are ideal to explore the feasibility of the targets in a country context because of their distinct combinations of epidemiological characteristics, health systems, and levels of tuberculosis prevention and care pathway activities. The likelihood of achieving the global targets will depend, to a large extent, on progress in these high-burden countries.

Research in context
Evidence before this study The post-2015 Global TB Strategy envisions and is aiming for a 50% reduction in tuberculosis incidence and a 75% reduction in tuberculosis mortality by 2025, using existing or near-existing tools. Given that this period starts in 2016, there is an urgent need to inform policy discussions on how these targets can be reached on a country level. Modelling can be a powerful tool to address this need by projecting the potential eff ect of a combination of diff erent interventions. Additionally, by using multiple models to address the same question, it can be used to identify fi ndings robust to between-model variation, increasing confi dence in the conclusions. We reviewed existing modelling studies that assessed the individual and combined impact of a range of existing interventions, and other multimodelling exercises in the fi eld of tuberculosis. We built on a systematic review by the TB Modelling and Analysis Consortium who gathered all tuberculosis modelling papers and extended the review to June, 2015. PubMed was searched using the following search query: (tuberculosis OR TB) AND ((mathem* AND (model OR models)) OR (mathem* modell*) OR (mathem* modeling) OR (modeling OR modelling) OR "Population Dynamics" [MeSH Terms] OR "Population Dynamics" OR "System Dynamics" OR "Computer Simulation" OR "Computer Simulation" [MeSH Terms])". We also did specifi c searches in mathematical modelling journals, and searched private libraries, and references of existing modelling reviews. We only included English language papers and we used no date restrictions.
Ours is the fi rst study to compare multiple tuberculosis models to answer a public health question. Single models have usually evaluated a single intervention, making it diffi cult to understand the full potential, including potential synergy, or non-synergy, of a combination of interventions implemented simultaneously. Other multimodel exercises have been published, most notably in the fi eld of HIV, which focused on questions around antiretroviral therapy scale-up to inform UNAIDS policy.

Added value of this study
This study highlights the uncertainty in the natural history of tuberculosis that drives between-model diff erences, while still identifying relatively consistent fi ndings of public health importance. It explores how a range of existing interventions across the tuberculosis care pathway, scaled up to country-specifi c levels, can take China, India, and South Africa towards the 2025 global tuberculosis targets. Our results show that expansion of existing interventions should enable South Africa to reach the 2025 targets, while for India and China additional context-specifi c activities are likely to be needed.

Implications of all the available evidence
Although major reductions in tuberculosis burden seem possible with current tools and 2025 targets might be met in South Africa, additional interventions, adapted to the country-specifi c tuberculosis epidemiology and health systems, are likely to be needed to reach the post-2015 Global TB Targets in other key countries such as China and India. These might include interventions such as tackling the latent tuberculosis infection reservoir in elderly people in China and undernutrition in India. This decision making can be informed by rigorous data analysis and the logical framework that mathematical models provide.

Participating models
After a global call from the TB Modelling and Analysis Consortium for expressions of interest, 11 modelling groups contributed results for at least one of the countries (China, India, or South Africa). These models varied in their frameworks, population stratifi cations, and approaches used to model disease and intervention mechanisms. An overview of participating models and references is in table 1.

Data and country context for baseline
To calibrate the models and provide a baseline scenario, modellers were provided calibration targets refl ecting tuberculosis burden (as incidence and mortality) in 2012 11,12 and tuberculosis control activities. Because half the participating models only included adult groups (aged ≥15 years), calibration targets and results also focused on adults (appendix). For China, the calibration targets included 2000 and 2010 tuberculosis prevalence values as estimated by national surveys. 13 For South Africa, models were calibrated to refl ect an estimated 2-5% decline in the annual incidence of tuberculosis in 2012 and projected scale-up of antiretroviral therapy (ART) coverage to 77% of HIV-positive adults by 2025 10,14 (see appendix section 1 for additional details of the calibration process, including sources).

Intervention scenarios
We defi ned a framework of enhancements of tuberculosisprogramme activities using existing tools, which were grouped into intervention scenarios (fi gure 1). Through detailed discussions with representatives from national tuberculosis programmes and the global advocacy community (Stop TB Partnership), we defi ned two distinct levels of scale-up within those intervention scenarios. Table 2 summarises national tuberculosis programme and advocacy scenario sets for each country; see appendix seciton 2 for members and affi liations for each group. The fi rst intervention scenario looked at improvements in access to high-quality care, defi ned as individuals with tuberculosis disease having access to the best level of care available in their local context. Tuberculosis care in China and India is provided through many health-care providers. 15 Since the quality of care is known to diff er strongly between providers, 16 Access to high quality care (#1)

Figure 1: TB Care and Prevention framework
The patient care pathway from disease to completion of treatment (blue boxes and arrows). Areas aff ected for enhancing current tuberculosis programme activities (ie, intervention scenarios) are shown in grey boxes and arrows, with the number (#x) to link them to activities in table 2 and the appendix section 3.   received the higher quality care as provided in the Centers for Disease Control (China) or the public sector (India). For South Africa, models explored a new policy of expanded screening of health-centre visitors for symptoms of tuberculosis disease. A second set of intervention scenarios modelled improvements in the tuberculosis care pathway, which included replacing sputum smears with a molecular diagnostic test such as GeneXpert as the fi rst laboratory test, increasing linkage to care for individuals diagnosed with tuberculosis disease, and improved treatment outcomes for those linked to care. We also estimated the eff ect of active case fi nding for tuberculosis disease in the general population, implemented as simple screening of a proportion of the population for disease (table 2, appendix), either on its own, or as screening for active disease along with preventive therapy for individuals with latent tuberculosis infection. For South Africa, an additional intervention scenario estimated the impact of providing continuous isoniazid preventive therapy, with screening for active disease before initiation for individuals receiving ART. A combination intervention scenario estimated the overall impact of all interventions run simultaneously. Only models that contributed results to all individual interventions reported the combined intervention.

Country
Under the advocacy scenarios, models estimated the potential eff ect on tuberculosis incidence and mortality if countries were able to screen and treat 30% (India and China) or 50% (South Africa) of the general population (as a proxy for identifying a similar proportion of the burden through active-case fi nding in high-risk groups) for tuberculosis disease and latent tuberculosis infection twice a year (table 2). With the exception of the active case fi nding and preventive therapy for the general population, each intervention scenario was operationalised as specifi c programmatic activities tailored to each country context (table 2). Modellers were asked to refl ect these activities as closely as possible using their respective model structure and parameterisation (see appendix section 3 for guidance provided and the implementation of the intervention scenarios for each model).  Economic development and related investments in universal health coverage, components of Pillar 2 in the End TB Strategy 1 were considered critical enablers of other intervention scenarios-eg, for access to high-quality care and treatment success (table 2, appendix section 3). Thus, we did not model separately potential benefi ts for interventions such as achievement of universal health coverage, cash-transfer programmes, or the preventive eff ect of poverty-reduction eff orts on tuberculosis outcomes.
Models reported on tuberculosis incidence, mortality, and prevalence for the period 1990-2025. Because half the models captured adults only (table 1), our main outcomes were the change in adult (aged ≥15 years) incidence and mortality between 2015 and 2025 in the baseline scenarios. Additionally, we recorded the incremental impact of individual intervention scenarios and the overall impact of the baseline plus the combination intervention scenarios, and the cumulative cases and deaths averted. Additional outcomes reported  (see table 2) as best as possible within their model framework, and provide an implementation narrative (see appendix section 3). Inevitably, simplifi cation will have occurred to fi t the intervention within the model structure. For example, in South Africa, the method of implementing the intervention scenario of isoniazid preventive therapy for HIV positive individuals receiving antiretroviral therapy will depend on whether the model had a separate compartment for isoniazid preventive therapy to track the number of individuals who were screened (as part of annual re-screening for tuberculosis) and have separate tuberculosis progression rates. See appendix section 3 for guidance and specifi c implementation.
by modellers included MDR tuberculosis prevalence in new or retreatment cases, latent tuberculosis infection prevalence, and the proportion of disease after recent infection. These outcomes were used to understand the diff erences between models and ensure internal model consistency. After fi tting to the calibration targets, modelling groups were provided with guidance on how to implement the intervention scenarios (appendix section 3) in view of the diff erences in model structures.
Additionally, we established minimum requirements to model structure for contribution to each scenario (appendix section 3).

Role of the funding source
The funder of the study had no role in study design, data collection, data analysis, data interpretation, or writing of the report. The corresponding author had full access to all the data in the study and had fi nal responsibility for the decision to submit for publication.

Results
The modelling took place between March and December, 2014. Of 11 participating models, six provided projections for China, fi ve for India, and eight for South Africa. Three modelling groups (Harvard, AuTuMN and TIME) contributed results for all three countries and one model (IDM) modelled the China and South Africa epidemics. Of 11 models, nine provided results for all interventions (table 1). Figure 2 shows baseline calibration and projections for incidence and mortality. In China, where models were calibrated to 2000 and 2010 prevalence targets (appendix sections 1 and 4), historical and projected trends were similar between models. For India, the historical uncertainty was propagated in the projections, although most models predicted a declining incidence trend consistent with recent WHO projections. 10 In South Africa, models were calibrated to an epidemiological burden and trend in 2012, but diverged over time as the projected baseline change in incidence between 2015 and 2025 ranged from 0 to 25%. For China and India, baseline changes in incidence were 11-27% and 0-19%, respectively, suggesting that with tuberculosis-programme activity, the tuberculosis burden was most likely to continue its decline.
In the national tuberculosis programme scenarios, the incremental contribution of individual intervention scenarios (compared with the baseline) diff ered strongly between countries. In China, the additional impact of individual interventions on tuberculosis incidence and mortality was small for all scenarios (fi gure 3, appendix section 4). By contrast, for India, improving access to high-quality care substantially reduced tuberculosis incidence beyond baseline trends by a median of 20% (range 5-41%). In the same setting, activities solely aimed at further improving care for patients already accessing high-quality care made little diff erence to baseline trends. In South Africa, although we noted some between-model variation, most intervention scenarios showed substantial impact, with prevention (#6: continuous isoniazid preventive therapy for individuals receiving ART), case fi nding (#1: screening at primary-health clinics), and improvements in linkage to care and treatment success (#3) reducing tuberculosis incidence by a further median of 16% (range 8-51%), 20% (7-35%), and 8% (0-25%), respectively.
In South Africa, the 2025 End TB Strategy targets seem feasible, because the model projections showed that a combination of prevention, case fi nding, and improvements in care (fi gure 4) reduced incidence and mortality with a median of 55% (range 31-62%) and 72% (65-82%), respectively, and averted a cumulative 1·2 million (0·7 million-1·8 million) cases of tuberculosis and 298 000 (193 000-453 000) deaths from tuberculosis between 2015 and 2025. In China, median cases and achievement in tuberculosis control, but our results show that the targets and the tools enlisted to achieve them will need adapting to provide countries with a path that is both ambitious but also feasible. Achievement of the targets at a global level will be challenging, because more modest contributions from one country will need to be compensated by other countries going beyond these already highly ambitious targets. Additionally, for countries where the standard package of interventions is likely to be insuffi cient, new strategies need to be developed that tackle country-specifi c drivers, such as the ageing population of individuals with tuberculosis in China 21 and high levels of undernutrition in India. 22,23 As they stand, the advocacy scenarios would involve community-based tuberculosis screening for substantial parts of the population, twice a year. With current tools, this raises substantial issues around feasibility, resources, and evidence for impact. 24 Also, we did not quantify potential negative eff ects, including false-positive treatment, and regimen side-eff ects. However, the advocacy scenarios illustrate a key point of addressing the latent tuberculosis infection reservoir in the population, particularly for settings like China where current high levels of tuberculosis programme performance (and resulting relatively minor contribution of transmission to tuberculosis incidence 25 ) means substantial gains can be made in this area. Tools are needed to reduce the volume (and associated costs) of screening for active disease and treatment of latent tuberculosis infection, such as a postexposure vaccine, or a screening test that detects individuals with latent tuberculosis infection who are likely to progress in the next 5-10 years, 26 all of which are part of Pillar 3 of the End TB strategy. 1 We collated the best available data from published reports and country experts, but emphasise that improved information about tuberculosis epidemiology and the current tuberculosis care pathways is still needed. For example, substantial uncertainty exists about tuberculosis incidence and mortality in India, prevalence of latent tuberculosis infection, and treatment volume provided in public and private sectors, which can aff ect model projections. 10,15 The epidemiological eff ect of activities often relied on expert opinion in the absence of reliable data, which adds uncertainty to model projections. Further evidence about effi cacy of interventions activities is needed to inform policy discussions at a global and national level.
In this project we aimed to examine the impact of major policy options in the tuberculosis response. As such, some specifi c interventions that might be considered for specifi c settings were not included such as addressing undernutrition in India and age-specifi c screening for latent tuberculosis infection in China, which have been explored in other models. [21][22][23] Also, we did not include scenarios that focused on active case fi nding in high-risk groups, such as miners or people living in informal settlements. To deaths averted by the combination intervention scenarios were 312 000 (range 42 000-764 000) cases of tuberculosis and 65 000 (10 000-93 000) deaths from tuberculosis. In India the corresponding fi gures were 3·1 million (1·2 million-5·8 million) cases of tuberculosis and 1·1 million (0·8 million-2·1 million) deaths from tuberculosis. Despite these projected substantial health gains, the proportional reductions between 2015 and 2025 were estimated to fall short of the post-2015 End TB Strategy targets for China and India (fi gure 4).
The impact of annual screening for active disease (#4 active-case fi nding in the general population) was small when compared with the eff ect of treating latent tuberculosis infection, particularly for China where #5 (active-case fi nding followed by treatment of latent tuberculosis infection) achieved a median of 64% (range 52-81%) reduction in tuberculosis incidence (fi gure 3), and the impact of the combined intervention exceeded the 2025 global targets.

Discussion
Using multiple, independently developed tuberculosis transmission models, we explored the feasibility of achieving the post-2015 End TB Strategy targets in three high-burden countries, each with diff erent epidemiology and existing levels of tuberculosis control. By projecting the impact of combinations of existing tools in China, India, and South Africa, we showed the importance of country context in assessing whether and how these global targets might be achieved at a country level. For South Africa, the 2025 milestones of 50% reduction in incidence and 75% reduction in mortality appear feasible with existing tools, whereas for India and China, these targets appear unfeasible.
Contrasting results between countries refl ect the diff erences between epidemiological context and responses of local tuberculosis epidemics. Whereas in China, two decades of steady improvement in reach and quality of basic tuberculosis services have led to nearly a two-thirds reduction in tuberculosis prevalence between 1990 and 2010, 13 South Africa's tuberculosis programme was overwhelmed by the eff ects of HIV, 18 and is only just turning a corner. 10,19 The room for further improvement with current tools diff ers widely between countries, which leaves high-performing tuberculosis programmes, such as the one in China, with the question of how to achieve additional reductions. India faces specifi c challenges around private providers of tuberculosis care, who are common throughout the southeast Asia region, and our results show that improving the quality of tuberculosis care in the private sector is essential. This process is underway, partly through an expansion of government subsidies to pay for individuals to access eff ective tuberculosis diagnosis and treatment through the private sector. 20 The post-2015 End TB Strategy targets are laudable in their ambition, and describe what would be a great adequately capture these dynamics, and the impact of targeted interventions in these populations in a reasonable way, we believe that epidemiological models require specifi c model structure, and credible data on size and tuberculosis burden in each population, as well as a reasonable estimate of mixing within and between the general and high-risk populations. When more data become available, these choices can be revisited.
We did not report parametric uncertainty, or the relationship between model structure and predicted outcomes. These issues represent areas for future research. What is clear is that, as models aim to capture greater complexity, the structural and parametric uncertainty that can be expressed increases. One illustration is the more pronounced divergence in baseline model projections for South Africa where HIV is a key driver of the tuberculosis burden. Models that capture the interaction between tuberculosis and HIV and the eff ect of ART create additional opportunities for model diff erences and resulting divergence of baseline projections. However, since there is no one true model structure, multimodel exercises such as ours are important to identify fi ndings robust to the structural uncertainty, as we have here.
Our study provides unique insights on the feasibility of these global epidemiological 2025 targets at the country level, and illustrates the challenges ahead. In further work, these epidemiological projections have been linked to costs 39 to explore cost-eff ectiveness, aff ordability, and poverty alleviation. Such information is vital as policy makers and the global tuberculosis community assess the health gains and economic costs that would come with scaling up existing tools to meet the fi rst targets of the End TB Strategy.

Declaration of interests
We declare no competing interests.