Cost-effectiveness and resource implications of aggressive action on tuberculosis in China, India, and South Africa: a combined analysis of nine models.

BACKGROUND
The post-2015 End TB Strategy sets global targets of reducing tuberculosis incidence by 50% and mortality by 75% by 2025. We aimed to assess resource requirements and cost-effectiveness of strategies to achieve these targets in China, India, and South Africa.


METHODS
We examined intervention scenarios developed in consultation with country stakeholders, which scaled up existing interventions to high but feasible coverage by 2025. Nine independent modelling groups collaborated to estimate policy outcomes, and we estimated the cost of each scenario by synthesising service use estimates, empirical cost data, and expert opinion on implementation strategies. We estimated health effects (ie, disability-adjusted life-years averted) and resource implications for 2016-35, including patient-incurred costs. To assess resource requirements and cost-effectiveness, we compared scenarios with a base case representing continued current practice.


FINDINGS
Incremental tuberculosis service costs differed by scenario and country, and in some cases they more than doubled existing funding needs. In general, expansion of tuberculosis services substantially reduced patient-incurred costs and, in India and China, produced net cost savings for most interventions under a societal perspective. In all three countries, expansion of access to care produced substantial health gains. Compared with current practice and conventional cost-effectiveness thresholds, most intervention approaches seemed highly cost-effective.


INTERPRETATION
Expansion of tuberculosis services seems cost-effective for high-burden countries and could generate substantial health and economic benefits for patients, although substantial new funding would be required. Further work to determine the optimal intervention mix for each country is necessary.


FUNDING
Bill & Melinda Gates Foundation.


Introduction
The World Health Assembly's post-2015 End TB Strategy formalises goals for aggressive action against tuber culosis, including reductions in global incidence by 50% and mortality by 75% by 2025. 1 To meet these targets, major advances are needed in high-burden countries. The TB Modelling and Analysis Consortium conducted a multimodel evaluation to assess the goals' feasibility, 2 fi nding that aggressive but feasible scale-up of existing approaches could achieve the reductions described by the global targets in South Africa but not in India or China.
If targets can be met, understanding whether doing so represents the best use of funding or is even aff ordable is crucial. Conversely, if targets cannot be met, expansion of tuberculosis services is not without value. Although the End TB Strategy provides an important consensus to invigorate the fi ght against tuberculosis and attract funding, Ministries of Health also need to consider local priorities and programmatic constraints. In this context, an understanding of the resources required for scale-up and a comparison of the performance of competing intervention approaches are crucial.
In this analysis, we aimed to describe the costs and health outcomes of aggressive intervention against tuberculosis, and to assess cost-eff ectiveness, fi nancial implications, and patient economic burden of these interventions. Although previous studies [3][4][5] have assessed the cost-eff ectiveness of various interventions in highburden settings, few have compared multiple interventions simultaneously and assessed aff ordability. Quantifi cation of the eff ect of these interventions on patient-incurred costs is also important, in view of the high disease burden in low-resource settings 6,7 and the growing policy interest in catastrophic health-care spending, 7,8 a concern addressed explicitly in the End TB Strategy. 1

Overview
In collaboration with national tuberculosis programme representatives for each country, we defi ned scenarios for scaling-up existing interventions to high yet feasible coverage by 2025. We projected long-term outcomes using multiple indepen dently developed models of tuberculosis epidemiology and health services, and estimated costs by synthesising model outputs with empirical cost data and expert opinion on implementation approaches. We estimated health eff ects (disability-adjusted life-years [DALYs] averted) and resource implications for the 20 year period (2016-35), and calculated costs from multiple perspectives. For each country we compared intervention scenarios with a base case in which present intervention coverage is maintained.

Countries
We undertook this study for China, India, and South Africa because of their substantial tuberculosis burden and their contrasting HIV and tuberculosis epidemiology and organisation of tuberculosis services. In China, tuberculosis incidence is 68 cases per 100 000 population and mortality is three per 100 000 population 8 (equivalent to 3% of global tuberculosis mortality), and the country has achieved progressive reductions in tuberculosis burden in the past three decades. 9 In India, tuberculosis incidence is 167 per 100 000 population and mortality is 19 per 100 000 population (accounting for 17% of global tuberculosis mortality), and a large private sector provides roughly half of all tuberculosis care. 10 In South Africa, tuberculosis incidence is 834 per 100 000 population and mortality is 178 per 100 000 population (equivalent to 6% of global tuberculosis mortality), with both incidence and case fatality driven by the HIV epidemic.

Intervention scenarios
With input from national tuberculosis programme representatives for each country, we defi ned scenarios describing scale-up of specifi c interventions to high coverage by 2025 (table 1), making use of currently available intervention options and considering local policy preferences and capacity. Although intervention areas were defi ned in advance, country experts deter mined whether additional scale-up of an intervention was appropriate for their country, and the anticipated scale and pace of coverage improvements. Scenario descriptions included the activities required to produce coverage changes: for example, in South Africa, improving access

Research in context
Evidence before this study The World Health Assembly's post-2015 End TB Strategy proposes aggressive action to reduce tuberculosis incidence and mortality worldwide. Major reductions in high-burden countries will be essential for achieving these targets. Before this study, little quantitative evidence existed on the feasibility of eff orts to reach these targets in high-burden settings and on the cost of implementing aggressive service expansion. We reviewed published work (English-language articles identifi ed on PubMed, supplemented by the authors' familiarity of the relevant literature) on the cost-eff ectiveness of tuberculosis interventions relevant to China, India, and South Africa. Most studies addressed only one intervention area, and few investigated issues of aff ordability. Although the economic burden of tuberculosis on individuals and households is known to be high, few studies have estimated patient-incurred costs as part of cost-eff ectiveness analysis. None of the studies reviewed were designed to address the issues raised by the global End TB Strategy targets. Collaborative analysis with multiple models to assess intervention cost-eff ectiveness has previously been done for HIV policy changes but not for tuberculosis policy.

Added value of this study
In this study, we found that substantial improvements in the reach and quality of tuberculosis care might be cost-eff ective according to conventional criteria, despite requiring substantially increased funding compared with current practice. By estimating patient-incurred costs at the same time as health service costs, we were able to understand the relative magnitude of these eff ects-for some interventions, patient cost savings are larger than the additional costs borne by health services, leading to net cost savings under a societal perspective. By comparing multiple intervention approaches, our fi ndings reveal the relative effi ciency of each approach in the generation of health benefi ts. From these comparisons, eff orts to improve access to care seemed the most benefi cial and cost-eff ective in each setting. Our fi ndings also show substantial variation in results predicted by diff erent models, pointing to important uncertainties in the evidence base for predicting long-term costs and health outcomes.

Implications of all the available evidence
Combined with evidence from previous cost-eff ectiveness studies, the results of our analysis lend support to eff orts to scale up tuberculosis services, motivated by the End TB targets. Our fi ndings also reveal wide diff erences in the eff ect and effi ciency of diff erent approaches, implying that countries will need to carefully consider the approaches taken to service expansion. The variation in results also shows a clear need for further empirical research to strengthen the evidence base used for tuberculosis policy modelling, and thereby improve the reliability of future analyses.
to tuberculosis diagnosis could be achieved through outreach to underserved areas and symptom screening for individuals attending primary care. Specifi cation of activities helped to defi ne the feasible extent of scale-up, and allowed the costs of each scenario to be estimated (see appendix p 3 for a description of the disease course, treatment, and fi nal outcomes).
We assessed interventions separately and also considered a scenario representing the combination of all interventions for each country. A base case scenario represented continuation of current practice, with service coverage held at existing levels. Services not specifi cally addressed in a scenario were held at current coverage levels, with two exceptions: for South Africa, we assumed that antiretroviral therapy (ART) coverage would increase to 77% by 2025; and for India, we assumed all retreatment tuberculosis cases would receive drug susceptibility testing by 2019. Both policies have high-level commitment in their respective countries.

Modelling approach
We projected policy outcomes using multiple independently developed Mycobacterium tuberculosis transmission models. 3,[11][12][13][14][15][16][17][18][19] Models had to represent the major mechanisms determining tuberculosis outcomes in each setting, be consistent with existing evidence on epidemiology and service provision, and simulate outcomes needed to estimate summary health and economic eff ects. Of 11 models in the modelling exercise, nine fulfi lled these requirements and contributed inputs for the economic analysis ( Reduced initial default and improved retention and cure rates for both drug-sensitive and drug-resistant cases Combination All of the above All of the above All of the above MDR=multidrug-resistant. IPT=isoniazid preventive therapy. ART=antiretroviral therapy. *Represents summary outcomes across public and private sectors. †Active case fi nding in specifi c risk groups was not considered because of modelling limitations.

Cost estimation
Models produced standardised outputs describing service use under each scenario, and outputs to estimate the economic burden of tuberculosis on patients and households. We also did a systematic review 20 to collate and synthesise unit costs for each country. Empirical tuberculosis cost data are limited; 20 where local values were unavailable we adapted estimates from other settings using local prices, holding country consultations to ensure face validity. We developed country-specifi c cost models to combine service use estimates with unit costs. We divided cost categories into diagnosis (ie, active or passive screening, tuberculosis diagnosis, and drug resistance testing), fi rst-line treatment for active disease, multidrugresistant (MDR) tuberculosis treatment, treatment of latent infection for HIV-positive individuals (ie, isoniazid preventive therapy; South Africa only), and programme overheads (ie, high-level overheads supporting service delivery). We estimated direct intervention costs and changes in the costs of core services (eg, passive detection and treatment) indirectly aff ected by policy change.
To estimate programme overheads, we examined past programme expenditures and consulted with WHO and country experts. Patient-incurred costs were calculated as the sum of direct medical costs (ie, fees paid to providers to receive care [eg, consultation fees] and to purchase drugs), direct non-medical costs (ie, expenses incurred to receive care, such as travel costs, excluding fees paid to providers), and indirect productivity costs (ie, income loss due to tuberculosis symptoms or treatment, and opportunity cost of productive activity forgone due to untreated active disease or time taken to receive treatment). Several intervention strategies used re imbursements or incentives paid for attending care to off set patient-incurred costs.
Additional details on the costing approach and input values are shown in the appendix pp 17-31. We report economic costs from several perspectives: a tuberculosis service perspective, representing costs borne by national tuberculosis programmes and associated service providers; a health service perspective, summing tuberculosis service costs (for South Africa, costs or cost savings associated with ART were also taken into account); and a patient perspective, including costs or cost savings realised by individuals with active tuberculosis or receiving tuberculosis care; as well as a societal perspective combining patient and health service costs. We did sensitivity analyses for unit costs and programme overheads.

Summary health outcomes
We measured health benefi ts using DALYs. Models produced standardised outputs for calculating DALYs averted compared with the base case. Outputs included deaths by age and year, and the yearly population distribution across tuberculosis-related and HIV-related health states (appendix pp 8-10). Disability weights (appendix p 11) were derived from a multicountry valuation study, 21 and remaining life expectancy (appendix p 12) from country-specifi c life tables 22 (values not truncated at the analytic horizon).

Scenario comparisons
In all scenarios, the assumptions were that interventions would reach peak coverage before 2025 and that they would be extended to 2035 maintaining 2025 coverage levels. This 20 year evaluation period balanced confl icting concerns: that longer-term projections would be increasingly unreliable, but that a short evaluation period  would exclude important policy consequences, as health benefi ts are lagged relative to implementation costs. We summed costs and DALYs over the 20 year period (2016-35). For cost-eff ectiveness analyses, we discounted these outcomes at 3% per year. Costs represent 2014 US$ (equal to 6·1 Chinese Yuan, 60·9 Indian Rupees, and 10·8 South African Rand). Model results are presented individually and averaged across models. Costeff ectiveness ratios were calculated as mean incremental cost divided by mean incremental health benefi ts for each scenario, as compared with base case. To describe aff ordability, we compared annual undiscounted costs for each scenario, averaged across models with equal weights. As we estimated economic costs, annual results are a smoothed version of actual fi nancial needs.

DPC and PD are employees of the Bill & Melinda Gates
Foundation, which funded the study. These authors were acting as subject matter experts rather than agency representatives, and did not have veto power over any study decision. The corresponding author had full access to all data and fi nal responsibility for the decision to submit for publication.

Results
The models for China and South Africa had fairly consistent results in terms of incremental costs (fi gure 1). In China, introduction of Xpert MTB/RIF seemed more expensive than expansion of access to care and improvement of treatment quality, because of high diagnostics costs and increased volume for MDR tuberculosis treatment. In South Africa, expansion of access to care had higher costs than other singleintervention scenarios, because of the high costs of expanding screening in primary care. Results for India showed clear diff erences between models. In particular, the results for expansion of access to care diff ered, with some models predicting cost savings over the 20 year period. This diversity of results points to the uncertain consequences of private sector intervention-central to several scenarios modelled for India-and diff erent assumptions about the eff ect of shifting patients from low-quality care to high-quality care.
For incremental patient-incurred costs, most intervention scenarios showed cost savings compared with the base case (fi gure 2). These cost savings resulted from reduced disease burden and from the inclusion of social protection or incentives paid for attending care, or both, in many intervention scenarios. By contrast, incremental patient-incurred costs were positive for the scenario of introducing Xpert in China, because of improved diagnosis of MDR tuberculosis and high costs that patients incur at present to receive second-line treatment. For India and South Africa, expansion of access to care generated the greatest patient cost savings among the single-intervention scenarios. Figure 3 presents incremental costs, health benefi ts (ie, DALYs averted), and cost-eff ectiveness ratios from both health service and societal perspectives for each intervention scenario compared with the base case (see appendix pp 13-15 for results of individual models). Results for South Africa included any costs or cost savings from changes in ART service volume. Exclusion of these costs reduced cost-eff ectiveness ratios by 10-20% (appendix p 16). In China, Xpert introduction averted fewer DALYs at higher cost than expansion of access to care and improvement in treatment, the results for which largely overlap. This fi nding might be related to the relatively minor increase in sensitivity of the Xpert algorithm over current practice and the low success rates for MDR tuberculosis treatment at present, both of which were features of the modelled scenarios. In India, expansion of access to care seemed the most attractive among the single-intervention strategies. On average, the scenario of improving treatment averted more DALYs at lower cost than Xpert introduction or active case fi nding, although results for this strategy vary widely. Active case fi nding in the general population generated minimal health benefi ts and comparatively high costs per DALY averted. In South Africa, tuberculosis services are predominantly publicly provided, and thus patient-incurred costs are lower relative to health service costs. Consequently, cost-eff ectiveness ratios diff ered little between health service and societal perspectives. Cost-eff ectiveness ratios were similar across scenarios, although the magnitude of eff ects was substantially greater for expansion of access to care than for other single-intervention scenarios. Figure 4 presents average annual tuberculosis service costs for each scenario, showing the budgetary implications of aggressive scale-up. In all three countries, the combination scenario required substantially increased funding. In China and South Africa, resource requirements peaked at about three times existing tuberculosis service costs. In China and India, cost increases were projected to decline over time, whereas for South Africa high spending levels were expected to persist, which is attributable to high ongoing costs of expanding access to care.
Major sources of uncertainty relating to costs that aff ected our results included, for China, additional programme investments needed to support scale-up, costs of Xpert introduction, and costs of providing second-line treatment. For India, the cost of implementing public-private partnerships to expand access was the major source of uncertainty, followed by uncertainty around programme costs and Xpert introduction costs. For South Africa, the costs of implementing mobile health services to increase access was the major source of uncertainty, followed by the costs of implementing symptom screening in primary care. Full results of the sensitivity analysis are given in appendix pp 31-33.

Discussion
The post-2015 End TB Strategy aims to reinvigorate action on tuberculosis control and achieve substantial and rapid reductions in incidence and mortality. In this study, we assessed the costs and cost-eff ectiveness of aggressive expansion of tuberculosis services with existing technology and interventions. Compared with current practice, all intervention scenarios in India and South Africa-and all but the scenario of Xpert introduction in China-had a cost per DALY averted that fell below the country's gross domestic product (GDP) per person, even before patient cost savings were considered. GDP per person is a conventional threshold for identifying highly cost-eff ective inter ventions, 23,24 yet many potential public health interventions meet this criterion. 25 Recent work has highlighted that this threshold might not adequately refl ect opportunity costs in many settings. 26 However, in view of the very low costs per DALY averted for many interventions and the substantial reductions in economic burden for patients, the results suggest that some form of expansion of tuberculosis services is likely to be cost-eff ective for each country, and this fi nding is robust because it is supported by the results of every participating model. Involvement of experts from each country was crucial for the development of realistic policy scenarios. The process of developing these scenarios revealed the importance of local epidemiology and care patterns in determining the relevant interventions for a particular setting, such that important interventions for one country (eg, private sector intervention in India and isoniazid preventive therapy for HIV-positive individuals in South Africa) were thought irrelevant or to have minimal benefi t for the other countries. Therefore, the activities suggested for each scenario diff ered between    followed by expansion of access to care, although with a higher cost per DALY averted than other interventions in South Africa. In India and South Africa, expanding access to care had a substantially greater eff ect than improving treatment, whereas in China this diff erence was only marginal, refl ecting China's past success in improving case detection. For both India and China, Xpert introduction had a smaller health eff ect and higher costs than improving treatment or expanding access, despite Xpert producing major improvements in the detection of MDR tuberculosis. This fi nding is related to low treatment quality for MDR tuberculosis in these two countries at present, characterised by poor retention and cure rates, and by high treatment costs. If the major consequence of Xpert introduction is to expand access to MDR tuberculosis treatment (as assumed in this study), then cost-eff ectiveness ratios are likely to remain high while the treatment costs of MDR tuberculosis remain high. Despite our broadly positive cost-eff ectiveness fi ndings, our results raise concerns regarding aff ordability. In all three countries, the annual tuberculosis service costs of the combination intervention were more than double the base case levels and remained elevated throughout the 20 year projection (fi gure 4). For India and South Africa, the average annual increase in spending required in the fi rst 5 years of the combination scenario (compared with the base case) represented 3-4% of present government health sector funding, compared with only 0·15% in China. Only in India was the eff ect on tuberculosis burden suffi cient to return the budget to the same level as the base case after 20 years. Although human resource needs and other health system constraints were not modelled explicitly, they might also present major challenges to scale-up.
Although aff ordability and health system constraints remain a challenge, a key fi nding of this study is that, across all three countries, more aggressive tuberculosis policy would substantially reduce the economic burden of tuberculosis on patients and their families. These results refl ect reductions in direct costs and reductions in income loss from early identifi cation and eff ective treatment and prevention. For India and China, several interventions directly targeted the costs borne by patients, using such reimbursements and monetary incentives to the patient to improve patient adherence to eff ective diagnosis and treatment. Factoring the eff ects on patientincurred costs into the cost-eff ectiveness ratio will substantially strengthen the investment case for expanding tuberculosis services, with both expansion of access to care and improvement in treatment producing net savings in societal costs in the evaluation period.
Despite this study's strengths and breadth, care should be taken in drawing conclusions about the costeff ectiveness of any one intervention approach. Although we investigated a range of interventions, because of the complexity of using multiple models we considered only a restricted set of options. In reality, each intervention could be implemented at diff erent scale and in various combinations with one another, and other interventions might also be considered. By comparing our set of intervention scenarios with the base case, we provide an initial scoping of the eff ects that might be achieved with each approach and highlight broad priority areas. However, ideally a full set of mutually exclusive strategies would be compared simultaneously, to identify the optimal set of services for a given budget or cost-eff ectiveness threshold. A consequence of our more limited approach is that even though the combination scenario seemed cost-eff ective compared with current practice, it is possible that another combination-potentially involving a subset of interventions or components of the intervention scenarios, more or less aggressive scale-up, or interventions not be considered in this analysis-might be optimal.
To our knowledge, we are the fi rst to use multiple independently developed models to assess the costeff ectiveness and aff ordability of tuberculosis inter vention options. An important benefi t of this approach was the opportunity to compare model results and in so doing understand the variation in predictions across models. Research in HIV has highlighted the variation in results possible when modelling complex disease and health system dynamics, 25,27 and by using multiple models we were able to identify major uncertainties that would remain undetected by single-model analyses. We found substantial variation in the net health and economic consequences predicted for several inter vention scenarios, and in the rankings of interventions implied by these results. Since we applied a standard cost model, this variation is primarily due to uncertainty in the processes of tuberculosis epidemiology and interventions in high-burden settingsrealised as diff erences in model structure and parameterisation-as well as uncertainty about how specifi c programme actions will aff ect tuberculosis epidemiology and outcomes. 28 For costs, we had few empirical data for several interventions, and for the programme costs of supporting scaled-up service provision and addressing health system constraints and increasing use. Our results are sensitive to these uncertainties. Because of the complexity of using multiple models, we were unable to systematically investigate uncertainty in individual epidemiological parameters, yet these uncertainties are also likely to be consequential. These uncertainties are related to the nature of the scenarios we examined, which were designed to extend the reach and quality of services far above current levels. Although the use of multiple models provides some indication of the uncertainty, these projections will not include unanticipated (and therefore unmodelled) factors that will limit the health eff ect of interventions or increase the costs. The major coverage expansions described by the intervention scenarios are unprecedented for tuberculosis control, and the possibility of unanticipated challenges might be higher than those for more conventional policy options.
In this study, we examined the cost-eff ectiveness of a set of tuberculosis interventions using multiple models, bringing together a community of country experts, modellers, and economists. In view of the limitations described previously, this study is a crucial fi rst step in supporting resource allocation to and within tuberculosis control programmes. We found that a wide range of context-sensitive interventions are likely to be costeff ective and alleviate fi nancial burden, but at substantial cost. Further work is needed to inform tuberculosis policy. In the future, policy decisions will ideally involve country-led planning processes-exemplifi ed by South Africa's investment case analysis in 2016-which can more fully examine the range of candidate policies and attendant implementation challenges, validate modelling assumptions, and evaluate budget needs against options for increasing funding.