Most individuals infected with Mycobacterium tuberculosis develop a subclinical latent tuberculosis infection (LTBI), which can progress to active disease many years later (Barnett et al., 1971). In high-income, low tuberculosis (TB) burden countries in Western Europe, North America, and Australasia, cases of ‘reactivation TB’ are now the primary driver of TB incidence. Delayed progression is particularly relevant for individuals migrating from high- to low-disease prevalence settings, as high infectious exposure prior to migration can cause elevated incidence of clinical disease after arrival. This is increasingly the case for TB in higher-income countries of destination. Among 15 high-income countries in Western Europe, the percentage of TB cases occurring among non-native-born populations in 2018 ranged from 51 % in Ireland to 90 % in Sweden (European Centre for Disease Prevention and Control, 2019; International Monetary Fund, 2019). In New Zealand and Australia this percentage was 78 % (Institute of Environmental Science and Research Ltd, 2015) and 86 % (Toms et al., 2017), respectively, for 2014. In the United States, the corresponding percentage has risen from 47 % to 70 % between 2000 and 2018 (U.S. Centers for Disease Control and Prevention, 2019) and epidemiological projections suggest continued increases (Menzies et al., 2018a). Evidence from transmission linkage studies suggests that the majority of new U.S. TB cases arise from reactivation of established LTBI, rather than recent infection (Yuen et al., 2016). Recent evidence also suggests that an increasing fraction of U.S. TB cases among non-U.S.-born persons arise among older individuals and those who have been in the United States for 10 years or more (Armstrong et al., 2019; Tsang et al., 2017).

Given their major contribution to TB incidence, non-native-born individuals represent a priority group for TB prevention and enhanced case surveillance. However, in many high-income countries, non-native-born individuals represent a substantial fraction of the population and targeting all such individuals for these services would require covering a large number of people. For example, non-U.S.-born individuals living in the United States in 2017 represented 13.7 % of the population, or 44.5 million individuals (U.S. Census Bureau, 2018). More precise approaches for targeting services could allow more efficient use of scarce prevention resources and reduce potential harms for those expected to receive minimal benefit from screening and treatment to prevent TB. In this study we report the distribution of TB incidence in the non-U.S.-born population between 2000 and 2016, in order to describe trends within and across migrating cohorts and identify high-risk groups that might benefit most from prevention and enhanced surveillance. We used case notification data from the U.S. National TB Surveillance System (NTSS) (U.S. Centers for Disease Control and Prevention, 2016), and population estimates derived from the American Community Survey and 2000 Decennial Census (Ruggles et al., 2015). These data represent 122, 118 non-U.S.-born TB cases, arising during more than 620 million person-years of exposure. This cohort is substantially larger than analogous cohorts from other similar settings (Aldridge et al., 2016; McBryde and Denholm, 2012; Chemtob et al., 2015; Menzies et al., 2018b) and provides a high-resolution description of differences in TB risk within the non-U.S.-born population.

The two regression models yielded smoothed TB incidence trends among the non-U.S.-born resident population, enabling stratification by age, entry year, and length of time in the United States, and, in the case of the second model, by birth country.

In recent decades the epidemiology of TB in the United States has shifted towards reactivation of latent infection as the main driver of new cases of disease. The majority of new TB cases now arise among the non-U.S.-born population, reflecting elevated exposure to infection before U.S.-arrival. As TB becomes increasingly concentrated in the non-U.S.-born population, it is important to distinguish differences in risk within this population, to better understand the current state of the TB epidemic in the United States.

For this study we analyzed data on 122,118 non-U.S.-born TB cases arising over 620 million person-years of exposure between 2000 and 2016. This large dataset allowed us to document major differences in TB risk within the non-U.S.-born population as a function of age and entry year, years since entry, and birth country. The interaction of these factors produces incidence risks for individual population strata that differ by several orders of magnitude and suggests clear opportunities for TB prevention programs. These granular insights would not have been possible to discern directly from the raw data of case counts; a previous analysis which pooled case data over several years to address sparse counts had much more limited ability to resolve the epidemiology of TB among the non-U.S.-born (Cain et al., 2008).

Our approach allowed us to estimate the component of TB risk attributable to the year in which an individual moved to the United States. We found that earlier entry cohorts experienced a higher risk of incident TB between 2000–2016, controlling for other factors associated with TB risk. For each of these arrival cohorts non-U.S.-born rates were found to be highest in the first few years since entry and increase with age at the time of arrival. While this is a well-known feature of TB epidemiology in the non-U.S.-born, our analytic approach allowed us to project these trends over the lifetime of an individual arrival cohort, demonstrating incidence risk to continue declining up to 40 years after arrival (though at a slower rate than observed shortly after entry). Within a given entry cohort, incidence rates were found to drop quickly with years since entry for those entering at a younger age, a finding consistent with the fact that infants and young children with TB would have developed disease as a result of recent infection. By contrast, older entrants retain a higher level of risk over time since entry, reflective of higher rates of reactivation from infection in the past. Rates were projected to rise again in later years among some who entered in 2016 at older ages and this is consistent with the upturn observed among older entrants in cohorts arriving in 1990 and 2000.

Comparison of the two regression models confirmed the importance of accounting for birth country to understand differences in TB incidence rates between successive cohorts. By doing so, our model incorporates the changing demographics over time of the originating countries, where more immigrants came from Western Europe in the mid-20th century as opposed to greater proportions of arrivals from Latin America and Asia in the latter part of the 20th century through to the present day. A steady decrease of IRR relative to incidence in 2016 has mostly prevailed since 1950 for individual countries of birth and failing to control for birth country was found to mask this phenomenon.

Per capita GDP of the birth country was found to be a strong predictor of incidence rates in the United States, with the top five countries by IRR falling in the bottom third of countries with respect to per capita GDP. While the top five countries in terms of annual cases (Mexico, Philippines, India, Vietnam, China) currently receive greater attention in the United States, there remain smaller populations within the United States whose TB incidence rates are considerably higher (U.S. Centers for Disease Control and Prevention, 2019; Tsang et al., 2020). If these populations are relatively localized in the United States, these groups might benefit from additional TB prevention and control resources. It is noteworthy that Mexico and China are in the middle tertile of per capita GDP and have rates below the overall non-U.S.-born rate (IRR < 1) whereas the Philippines, India, and Vietnam are all in the lower per capita GDP tier and have rates above the average. The model also identifies birth countries for which more recent entry cohorts have higher TB incidence, controlling for age at entry and years since entry. In particular, more recent entrants from Myanmar (MMR), Liberia (LBR), and Kenya (KEN) are predicted to face greater TB risks than earlier entrants from these countries.

Our analytic approach has limitations associated with estimation of population denominators which have previously been discussed (Menzies, 2019). While NTSS data are largely complete and consistent (Yelk Woodruff et al., 2015) notwithstanding small levels of missingness noted here, data sources informing the numerator and denominator are subject to biases, for example in NTSS with recall peaks for initial or mid-years of decades and some data discontinuities in first year since entry reporting. We attempted to correct for these expected sources of bias, but our models would not capture unexpected disruptive events affecting denominators (for example, refugee inflows in certain years) or numerators (such as the unexpected drop in TB cases in 2008–09 (Winston et al., 2011)).

Model results for a specific calendar year describe only those individuals in historic entry cohorts who have survived to that year. Results should not be interpreted as referring to the complete cohort arriving in any past year. Estimates could thus be biased if there were, for example, higher survival rates among those from higher income or lower TB prevalence countries, in which instance we would be underestimating rates among historic entry cohorts.

Country-specific IRRs defined as the WHO-estimated TB incidence rate divided by the non-U.S.-born rate for that country were recently published elsewhere (Tsang et al., 2020). These IRRs should not be confused with ours which are relative to the overall non-U.S.-born rate in a given year; comparators are explicitly stated throughout the manuscript whenever IRRs are stated. An IRR > 1 here indicates elevated rates for a specific country whereas in (Tsang et al., 2020) it indicates a lower rate for the country in question (when compared to the WHO rate).

In this work we adopted an analytic approach that attempted to separate systematic trends in the data from the sampling uncertainty associated with small population strata and observed reporting biases. This led us to specify flexible functions to describe the relationship between continuous predictors and incidence rates, and random effects to regularize the estimates for individual countries of birth. Our approach required analytic decisions that might have led to both over- or underfitting of the data. As an example of the former, it is possible that the sharp increase in incidence rates 10 years after entry for individuals aged 75 years old in the 2000 entry cohort (Fig. 2, top left) results from over-fitting to random variation in the empirical data. For the latter, our decision to include different countries of birth in the same model may have hidden patterns specific to individual countries. More closely-fitted models could be developed in future work, with the usual tradeoffs in interpretability and accuracy (Caruana et al., 2015). Indeed, development of models specifically for individual countries of birth or sub-national geographies within the United States could be useful for identifying population groups most at risk of TB, or revealing epidemiological patterns not addressed in this analysis.

These results provide a nuanced description of TB epidemiology for a population group currently contributing over two-thirds of new TB cases in the United States. These findings support the development of TB services that could accelerate progress towards TB elimination in the United States and may also be relevant for understanding trends in other high-income countries.