Policy Implications of Mathematical Modeling of Latent Tuberculosis Infection Testing and Treatment Strategies to Accelerate Tuberculosis Elimination

Tuberculosis (TB) disease is the leading cause of death globally from a single infectious organism. ¹ However, TB is both curable and preventable. In the United States during the past 2 decades, a national coordinated multi-agency policy response implemented in 1992, along with other influences (eg, new federal and state funding), led to a decrease in the number of TB cases reported in the United States, from 26 673 in 1992 to 9105 in 2017, a 65.9% decline. ^2,3 The 1992 national policy response was launched as a result of multidrug-resistant TB outbreaks that occurred during 1985-1992. That response included support for improved TB diagnostics, infection control, monitoring of TB treatment, and investigation of persons who had recent contact with persons who had infectious TB. ⁴ Mathematical modeling of TB during 1995-2014 in the United States estimated that approximately 145 000 to 319 000 TB cases were averted, yielding societal benefits (in 2014 US dollars) of $3.1 billion to $14.5 billion. ⁴

Recently, however, the US decline in the number of TB cases and in the incidence of TB has slowed substantially. During 2014-2017, the annual percentage decrease in the number of TB cases and in the TB incidence rate was <2% annually compared with 4%-5% during the previous decade. ³ An estimated 13% of US TB cases (and only 8% of US TB cases among non–US-born persons) is attributed to recent transmission, with nearly all the remaining TB cases assumed to be due to reactivation of latent TB infection (LTBI). ³ To address this slowing in TB decline, the Centers for Disease Control and Prevention (CDC) recognized the need to improve prevention of LTBI reactivation in its recommendations to expand LTBI testing with interferon gamma release assays (IGRAs) and use of short LTBI treatment regimens among persons at greatest risk of TB. ^{5 -9}

Because guidelines for LTBI testing and treatment in the United States can be difficult to interpret, we summarize the current guidelines and then present results of recent mathematical modeling of the impact of LTBI testing and treatment in furthering US TB elimination. We identify populations that were modeled for LTBI testing and treatment that resulted in the greatest future TB reduction and those that were the most efficient to test and treat. We also summarize results on the most cost-effective test and treatment to use in specific populations. Consistent findings across models are highlighted. Finally, we discuss challenges to translating modeling into policies and to implementing the modeled scenarios, as well as possible future modeling.

Current Guidelines

Current guidelines for LTBI diagnosis and treatment are complex and potentially confusing, leaving room for interpretation on who should be tested and on how best to test or treat. ^{6 -9} The guidelines can be difficult to implement for multiple reasons. Most clinicians and patients are unfamiliar with LTBI and TB, their diagnosis and treatment regimens, and their risk assessment procedures. We briefly summarize the guidelines to show the uncertainty and multiple decision points for which mathematical modeling can help in decision making about LTBI testing and treatment.

Clinical practice guidelines from the American Thoracic Society (ATS), the Infectious Diseases Society of America (IDSA), and CDC recommend that clinicians “weigh the likelihood of infection, the likelihood of progression to TB if infected, and the benefit of therapy” in determining whether to test populations for LTBI. ⁷ CDC discourages LTBI testing unless there are plans to rule out TB disease and to ensure completion of LTBI treatment among those who are diagnosed with LTBI. ¹⁰ Populations mentioned in the diagnostic guidelines as having an increased risk of infection with Mycobacterium tuberculosis include: those who have had recent close contact with infectious TB, mycobacteriology laboratory personnel, immigrants from countries with a high burden of TB (ie, where TB incidence is >20 per 100 000 persons), and residents and employees of high-risk congregate settings (eg, homeless shelters, correctional institutions). ⁷ Populations listed as having a high risk of progressing to TB if infected include: children aged <5 years, persons with HIV infection, persons diagnosed with silicosis (a lung disease characterized by scarring of the lungs and breathing passages that results from inhalation of silica particles), persons taking immunosuppressive medications, or persons who have an abnormal chest radiograph consistent with previous TB disease. ⁷ Populations identified as having an intermediate progression risk include persons with a clinical predisposition to TB progression if infected (eg, persons diagnosed with diabetes, persons diagnosed with chronic renal failure, or intravenous drug users). ⁷ The benefit of preventing TB is especially great for persons with HIV, who have high rates of TB-related mortality and among whom TB transmission is increased because of immunosuppression and association of HIV positivity with other factors correlated with recent TB transmission (eg, substance use, homelessness). ^11,12

The US Preventive Services Task Force (USPSTF), an entity independent of CDC, focused its recommendations on TB and LTBI testing of asymptomatic adults who were born in or previously lived in countries with increased TB prevalence or who live in or have lived in high-risk congregate settings. The USPSTF did not make recommendations for TB testing of populations in which it is considered the standard of care (eg, persons receiving immunosuppressive medications or persons living with HIV infection or silicosis) or of populations who are part of public health surveillance (ie, persons having contact with infectious TB patients) or employees tested by occupational health staff members concerned with workplace safety. ⁶

The ATS/IDSA/CDC guidelines recommend the use of IGRAs rather than tuberculin skin tests (TSTs) to diagnose LTBI in persons aged >5 years who are likely to be infected with Mycobacterium tuberculosis, have a low or intermediate risk of disease progression, have sufficient reason for LTBI testing, and either have a history of Bacillus Calmette–Guerin vaccination or are unlikely to return to have their TST read. ⁷ The USPSTF also stated that IGRAs may be preferable to use for testing persons who received a Bacillus Calmette–Guerin vaccination or for those who are unlikely to return for TST reading. ⁶ For persons who are likely to be infected and have a high risk of disease progression, the ATS/IDSA/CDC guidelines cite insufficient data to recommend a preference of IGRA over TST and allow consideration of a second test if an initial test is negative, with infection defined as positivity on either test. ⁷ Although stating that persons at low risk for TB should not be tested, the ATS/IDSA/CDC guidelines suggest a second diagnostic test (either IGRA or TST) if such testing occurs and the initial test is positive, with infection diagnosed only if both initial and secondary tests are positive. ⁷

CDC preferred and strongly recommended a regimen for LTBI treatment for persons aged >2 years, including for persons with HIV on antiretroviral therapy with acceptable drug interactions, which includes 3 months of once-weekly isoniazid-rifapentine for 12 weeks (3HP) by directly observed therapy (DOT) or self-administered therapy (SAT). ⁹ A preferred, strongly recommended regimen for HIV-negative persons is 4 months of daily rifampin (4R). ⁹ Another preferred but conditionally recommended regimen for adults and children of all ages, including for persons with HIV on antiretroviral therapy with acceptable drug interactions, is 3 months of daily isoniazid plus rifampin (3HR). ⁹ Alternatively recommended regimens (with higher toxicity and lower completion rates than shorter rifamycin-based regimens) include 6 months of daily isoniazid (6H), which is strongly recommended for HIV-negative persons, and 9 months of daily isoniazid (9H), which is conditionally recommended for both HIV-positive and HIV-negative persons. ⁹ The regimens have been determined through clinical trials to be similarly efficacious in preventing TB; however, greater treatment completion rates were found using the shorter regimens than the longer regimens. ^{13 -15} Consequently, CDC recommends using shorter regimens whenever possible. ⁹

Modeling of TB Incidence and LTBI Prevalence in the United States

Mathematical modeling of TB incidence; LTBI prevalence; interventions to diagnose, treat, or prevent TB disease or LTBI; intervention costs; and cost effectiveness can inform decision making and best use of resources. In 2014, CDC’s National Center for HIV/AIDS, Viral Hepatitis, STD, and TB Prevention Epidemiologic and Economic Modeling Agreement (NEEMA) funded 3 universities to develop models relevant to the National Center for HIV/AIDS, Viral Hepatitis, STD, and TB Prevention’s public health programs in the United States ¹⁶ : (1) the Emory University Coalition for Applied Modeling for Prevention (working with Johns Hopkins Bloomberg School of Public Health at Johns Hopkins University [JHU]), (2) the Harvard University Prevention Policy Modeling Lab (PPML), and (3) the University of California at San Francisco’s Consortium for the Assessment of Prevention Economics (CAPE). NEEMA-supported modeling has produced estimates of future TB incidence and LTBI prevalence for the US population, 4 US states that account for more than half of all TB cases in the United States, and populations already identified as being at high risk for TB. We summarize key results of NEEMA TB modeling activities that have been published or publicly presented and that have programmatic implications for which populations to test, which test to use, and which treatments are most cost effective. ^{17 -25} A consensus of NEEMA modeling to date concurs with the aforementioned guidelines and suggests that, for greatest effectiveness and cost effectiveness, targeted testing policies should focus on LTBI testing and treatment among non–US-born persons and persons with HIV, respectively. Testing with an IGRA appears to be most cost effective. Treatment with 3HP by either SAT or DOT is more cost effective than treatment with isoniazid alone in these populations. One-time testing and treatment of populations at risk of TB can have a substantial impact in the near term that sustains lower TB rates into the future. We present study results detailing these findings.

Whether to Expand Testing and Treatment for LTBI

Concordant with previous studies, ²⁶ PPML researchers projected that, among policy scenarios modeled, the greatest progress toward TB elimination (defined as <1 case per million persons per year) ²⁷ could be achieved by intensifying LTBI testing and treatment. ¹⁷ Specifically, modeling indicates that LTBI testing and treatment of immigrants before arrival in the United States might reduce the TB rate by 4.0 cases per million US residents by 2050, and that doubling LTBI testing and treatment of all US residents at high risk of TB disease might reduce the TB rate by 3.2 cases per million US residents by 2050 ¹⁷ compared with projections of not expanding testing and treatment.

Which Populations to Test and Treat

JHU modeled TB incidence separately among non–US-born persons, persons with HIV, persons with diagnosed diabetes, and persons experiencing incarceration or homelessness in the 4 states that account for more than half of all TB cases in the United States (California, Florida, New York, and Texas). ¹⁸ Although the effectiveness and cost effectiveness of targeting prevention efforts to specific populations differ by state, ^19,20 JHU estimated that targeting LTBI testing and treatment to half of the non–US-born population in these 4 states could lower national TB incidence by 25% to 31% during the next 10 years. ²¹ LTBI testing and treatment delivered to less numerous populations with higher-than-average TB risk (eg, persons with HIV) was more efficient than LTBI testing and treatment of non–US-born persons but did not substantially reduce TB incidence in the United States. ²¹ The most-to-least cost-effective populations to test and treat were: persons with HIV, non–US-born persons, persons experiencing homelessness, incarcerated persons, and persons with diabetes. ²⁰ JHU concluded that only by providing interventions to non–US-born persons could substantial reductions in TB incidence be made. ²¹

CAPE researchers modeled testing of various populations, types of tests, and treatment regimens in California, the state with the highest number of TB cases in the United States. ²² Researchers found that testing of non–US-born persons with medical risk factors for TB progression (HIV infection, tumor-necrosis-factor-alpha inhibitors use, receipt of solid organ transplant) was most cost effective, followed by testing of US-born persons with HIV, non–US-born persons with other medical risk factors (diabetes, end-stage renal disease, smoking), US-born persons with other medical risk factors, and non–US-born persons without medical risk factors. ²² CAPE researchers also estimated that TB pre-elimination (<10 TB cases per million persons per year) in California could be achieved by 2065 through a 10-fold increase in annual testing and treatment (using 6H) of non–US-born persons and of all persons with medical risk factors for TB progression. ²³ A one-time doubling of current testing and treatment of non–US-born persons using IGRA and 3HP/DOT would be cost effective at <$100 000 per quality-adjusted life year gained. ²³ A 23% increase in LTBI testing and treatment of all USPSTF-recommended populations at high risk ⁶ in California could prevent approximately 40% of new TB cases in the first decade post-implementation, with targeted testing and treatment of non–US-born persons having the greatest impact. ²⁴

Which Test and Treatment to Use in Specific Populations

CAPE modeling found that among LTBI tests, IGRA was most cost effective when compared with TST. Among treatment regimens, 3HP/DOT treatment was more cost effective than 4R, 6H, or 9H. ²²

In modeling testing algorithms of non–US-born persons living in the United States with or without HIV infection or diabetes, PPML estimated that using an IGRA and treatment with 3HP/SAT was cost effective compared with no testing or treatment. ²⁵ However, among non–US-born US residents with HIV, the most cost-effective testing algorithm was dual testing with IGRA and TST, with positivity determined by either test, because of the high risk of TB progression and increased sensitivity using 2 tests. ²⁵

NEEMA Collaboration With State TB Controllers

In developing modeling topics, models, and Internet web tools, NEEMA researchers collaborated with willing state and local TB staff members. The aforementioned modeling studies incorporated data from collaborating states, some of which are already using NEEMA modeling results and/or NEEMA-developed web tools. For example, in January 2016, the California TB Control Branch of the California Department of Public Health (CDPH), in collaboration with the Curry International TB Center and the California TB Controllers Association, launched a risk assessment tool for California health care providers to promote LTBI testing and treatment for persons with birth, travel, or residence in a country with an elevated TB rate (prioritizing persons with medical risk factors for TB progression), current or planned immunosuppression, or close contact with persons with infectious TB. ²⁸ Furthermore, to spur health care provider use of 3HP, CDPH used CAPE modeling results ²² to support inclusion of rifapentine in a large health system formulary. ²⁹ In addition, when PPML published its cost-effectiveness modeling results, CDPH affirmed the findings in an associated editorial. ³⁰ Also using CDPH data, projected estimates from the 3 transmission models (by CAPE, JHU, and PPML), which were independently developed and structured differently to model different areas, were directly compared during the historical and projected period to assess model estimates under various assumptions. ³¹ The comparison found that one-time LTBI testing and treatment among non–US-born persons produced sustained reductions in projected TB incidence using each of the 3 models, with reactivation of LTBI being a more important contributor to future TB transmission than to recent TB transmission. ³¹ Model result differences were due to uncertain model parameters, differing values for which were assumed by the 3 modeling groups. ³¹

PPML worked with the Massachusetts Department of Public Health to develop a Massachusetts TB projection model called Tabby2; PPML also produced Tabby2 models for 10 other states using state-specific historical data. ¹⁶

JHU collaborated with CDPH, the Texas Department of State Health Services, the New York State Department of Health, and the Florida Department of Health to identify priority populations and interventions to model. JHU published and presented its findings, including those on cost effectiveness of interventions by population and state. ^20,21

NEEMA modeling benefited greatly from collaboration with state and local TB staff members. In fact, its success is a direct result of such collaboration. Some collaboration was one-time, to obtain a list of priority modeling topics or data inputs. Other collaboration was long-term and often intense, including participation in monthly NEEMA progress telephone calls, in-depth review of results or draft journal articles, and testing of web tools. The level of collaboration by TB staff members was determined by state or local programs, some of which had more time or expertise to devote to such efforts. It is hoped that all state or local TB programs will benefit from NEEMA models developed for each area, which are planned in the coming year or two.

Future Modeling Through NEEMA

NEEMA-supported activities have provided modeling evidence to predict how national and state policies of enhanced LTBI testing and treatment could accelerate TB elimination in the United States. NEEMA also developed web-based estimation tools ¹⁶ that national, state, and local health care providers, public health programs, researchers, and policy evaluators can use to project TB incidence and LTBI prevalence in populations at high risk for TB, from which they can estimate how much staffing and other resources will be needed to prevent TB. Additional work in progress at the time of this article includes modeling the impact on the US TB burden of improvements in TB control in countries where most non–US-born persons with TB were born.

Although mathematical models can help identify the populations, diagnostics, and treatments that have the greatest potential impact and the TB trends that might result if they were applied, models and modeling results are limited by uncertainty in key inputs, such as the size of risk populations, sensitivity and specificity of diagnostic tests, treatment completion and effectiveness, risk of TB progression, and costs of interventions and disease. There is also uncertainty about LTBI prevalence, risk of progression, and lifespan of specific populations. NEEMA modeling in the next 5 years will attempt to better define the uncertainties of model inputs and to improve modeling accuracy. Moreover, modeling is based on assumptions that may not be true in all situations. Persons who are born in the United States or who have never resided in countries with a high incidence of TB but who fall into other risk categories (eg, live or work in high-risk congregate settings, some health care workers) likely benefit from LTBI testing and treatment, which could reduce TB morbidity and expensive outbreak and contact investigations. However, the full impact of these activities is not included in current NEEMA modeling. Future research is needed to collect or improve these key model inputs and to reduce their variability by population and setting to strengthen current models.

Although models can be improved, perhaps the greatest challenge remains in translating modeling into policies and in implementing the policies to achieve the projected impact. Translating modeling into policies will require persuasive communication of modeling results to state and local policy makers in language that they can understand and that highlights the feasibility of achieving results to benefit their jurisdictions or the TB elimination effort. Policy makers will then need to commit resources and legislators will need to allocate the resources to implement the modeled scenarios. Implementation challenges are potentially even greater. For example, to achieve a projected 40% decline in TB incidence in 10 years, massive efforts would be needed to hire and train additional staff members; purchase diagnostic tests and medications using a budget adequate to conduct recruitment, testing, and treatment; coordinate public and private sector involvement; and track success. Budget estimates should include costs from the beginning of targeted testing and treatment efforts, which are rarely available through published literature. To improve implementation chances, modeling could be conducted using NEEMA-developed tools that use historical data from the implementing area (eg, state or local TB programs) along with user-defined inputs, such as populations for testing, population sizes and characteristics, unit costs, and other values specific to the area. Implementing area-specific modeling scenarios could be facilitated by identifying private or public partners willing to commit resources to help further TB elimination in the area. Future NEEMA modeling at state and local levels will attempt to examine issues related to implementation.

NEEMA-supported activities will continue to build on data that have already been collected and analyzed, model structures and tools already developed, and tackle additional topics to model new diagnostics and interventions to accelerate the decline in US TB case counts and incidence. State and local TB controllers can use models already developed on the basis of local epidemiology to identify the most effective and cost-effective interventions and plan for TB elimination in their areas. Earlier efforts at modeling TB elimination were undertaken in a neighborhood in Texas. ³² Although implementing TB elimination activities state by state or city by city might seem daunting, according to Philosopher Lao Tzu: “The journey of a thousand miles begins with one step.” ³³ Current modeling can be extended to project TB outcomes for all populations at risk, which is necessary to identify trends in TB disparities and to examine the impact of interventions that might reduce or eliminate health inequities. These findings could inform state- and locality-specific resource allocation and policy revisions to bring TB elimination closer to reality in their jurisdictions and in the United States.