Using Machine Learning Techniques and National Tuberculosis Surveillance Data to Predict Excess Growth in Genotyped Tuberculosis Clusters.

The early identification of clusters of persons with tuberculosis (TB) that will grow to become outbreaks creates an opportunity for intervention in preventing future TB cases. We used surveillance data (2009-2018) from the United States, statistically derived definitions of unexpected growth, and machine-learning techniques to predict which clusters of genotype-matched TB cases are most likely to continue accumulating cases above expected growth within a 1-year follow-up period. We developed a model to predict which clusters are likely to grow on a training and testing data set that was generalizable to a validation data set. Our model showed that characteristics of clusters were more important than the social, demographic, and clinical characteristics of the patients in those clusters. For instance, the time between cases before unexpected growth was identified as the most important of our predictors. A faster accumulation of cases increased the probability of excess growth being predicted during the follow-up period. We have demonstrated that combining the characteristics of clusters and cases with machine learning can add to existing tools to help prioritize which clusters may benefit most from public health interventions. For example, consideration of an entire cluster, not only an individual patient, may assist in interrupting ongoing transmission.

US Tuberculosis Rates among Persons Born Outside the United States Compared with Rates in Their Countries of Birth, 2012-20161.

The US Centers for Disease Control and Prevention recommends screening populations at increased risk for tuberculosis (TB), including persons born in countries with high TB rates. This approach assumes that TB risk for expatriates living in the United States is representative of TB risk in their countries of birth. We compared US TB rates by country of birth with corresponding country rates by calculating incidence rate ratios (IRRs) (World Health Organization rate/US rate). The median IRR was 5.4. The median IRR was 0.5 for persons who received a TB diagnosis <1 year after US entry, 4.9 at 1 to <10 years, and 10.0 at >10 years. Our analysis suggests that World Health Organization TB rates are not representative of TB risk among expatriates in the United States and that TB testing prioritization in the United States might better be based on US rates by country of birth and years in the United States.

Robustness of NHANES Estimates of the US Prevalence of a Positive Tuberculin Skin Test.

BACKGROUND: A single 2-year National Health and Nutrition Examination Survey (NHANES) cycle is designed to provide accurate and stable estimates of conditions with prevalence of at least 10%. Recent NHANES-based estimates of a tuberculin skin test (TST) ≥10 mm in the noninstitutionalized US civilian population are at most 6.3%. METHODS: NHANES included a TST in 1971-1972, 1999-2000, and 2011-2012. We examined the robustness of NHANES-based estimates of the US population prevalence of a skin test ≥10 mm with a bias analysis that considered the influence of non-US birth distributions and within-household skin test results, reclassified borderline-positive results, and adjusted for TST item nonresponse. RESULTS: The weighted non-US birth distribution among NHANES participants was similar to that in the overall US population; further adjustment was unnecessary. We found no evidence of bias due to sampling multiple participants per household. Prevalence estimates changed 0.3% with reclassification of borderline-positive TST results and 0.2%-0.3% with adjustment for item nonresponse. CONCLUSIONS: For estimating the national prevalence of a TST ≥10 mm during these three survey cycles, a conventional NHANES analysis using the standard participant weights and masked design parameters that are provided in the public-use datasets appears robust. See video abstract at, http://links.lww.com/EDE/B636.

Factoring Prior Treatment into Tuberculosis Infection Prevalence Estimates, United States, 2011-2012.

To refine estimates of how many persons in the United States are candidates for treatment of latent tuberculosis, we removed from analysis persons who self-reported prior treatment on the National Health and Nutrition Examination Survey 2011-2012. We estimate that 12.6 million persons could benefit from treatment to prevent active tuberculosis.

Simple Estimates for Local Prevalence of Latent Tuberculosis Infection, United States, 2011-2015.

We used tuberculosis genotyping results to derive estimates of prevalence of latent tuberculosis infection in the United States. We estimated <1% prevalence in 1,981 US counties, 1%-<3% in 785 counties, and >3% in 377 counties. This method for estimating prevalence could be applied in any jurisdiction with an established tuberculosis surveillance system.

Statistical Method to Detect Tuberculosis Outbreaks among Endemic Clusters in a Low-Incidence Setting.

We previously reported use of genotype surveillance data to predict outbreaks among incident tuberculosis clusters. We propose a method to detect possible outbreaks among endemic tuberculosis clusters. We detected 15 possible outbreaks, of which 10 had epidemiologic data or whole-genome sequencing results. Eight outbreaks were corroborated.

Age-Period-Cohort Analyses of Tuberculosis Incidence Rates by Nativity, United States, 1996-2016.

OBJECTIVES: To assess changes in US tuberculosis (TB) incidence rates by age, period, and cohort effects, stratified according to race/ethnicity and nativity. METHODS: We used US National Tuberculosis Surveillance System data for 1996 to 2016 to estimate trends through age-period-cohort models. RESULTS: Controlling for cohort and period effects indicated that the highest rates of TB incidence occurred among those 0 to 5 and 20 to 30 years of age. The incidence decreased by age for successive birth cohorts. There were greater estimated annual percentage decreases among US-born individuals (-7.3%; 95% confidence interval [CI] = -7.5, -7.1) than among non-US-born individuals (-4.3%; 95% CI = -4.5, -4.1). US-born individuals older than 25 years exhibited the largest decreases, a pattern that was not reflected among non-US-born adults. In the case of race/ethnicity, the greatest decreases by nativity were among US-born Blacks (-9.3%; 95% CI = -9.6, -9.1) and non-US-born Hispanics (-5.7%; 95% CI = -6.0, -5.5). CONCLUSIONS: TB has been decreasing among all ages, races and ethnicities, and consecutive cohorts, although these decreases are less pronounced among non-US-born individuals.

Tuberculosis Among Foreign-Born Persons Diagnosed ≥10 Years After Arrival in the United States, 2010-2015.

The majority of tuberculosis (TB) cases in the United States are attributable to reactivation of latent TB infection (LTBI) (1). LTBI refers to the condition when a person is infected with Mycobacterium tuberculosis without signs and symptoms, or radiographic or bacteriologic evidence of TB disease. CDC and the U.S. Preventive Services Task Force (USPSTF) recommend screening populations at increased risk for LTBI, including persons who have lived in congregate settings at high risk and persons who were born in, or are former residents of countries with TB incidence ≥20 cases per 100,000 population (2-4). In 2015, foreign-born persons constituted 66.2% of U.S. TB cases (5). During the past 30 years, screening of persons from countries with high TB rates has focused on overseas screening for immigrants and refugees, and domestic screening for persons who have newly arrived in the United States (6,7). However, since 2007, an increasing number and proportion of foreign-born patients receiving a diagnosis of TB first arrived in the United States ≥10 years before the development and diagnosis of TB disease. To better understand how this group of patients differs from persons who developed TB disease and received a diagnosis <10 years after U.S. arrival, CDC analyzed data for all reported TB cases in the United States since 1993 in the National TB Surveillance System (NTSS). After adjusting for age and other characteristics, foreign-born persons who arrived in the United States ≥10 years before diagnosis were more likely to be residents of a long-term care facility or to have immunocompromising conditions other than human immunodeficiency virus (HIV) infection. These findings support using the existing CDC and USPSTF recommendations for TB screening of persons born in countries with high TB rates regardless of time since arrival in the United States (2,3).

Characteristics of tuberculosis cases that started outbreaks in the United States, 2002-2011.

A review of 26 tuberculosis outbreaks in the United States (2002-2011) showed that initial source case-patients had long infectious periods (median 10 months) and were characterized by substance abuse, incarceration, and homelessness. Improved timeliness of diagnosis and thorough contact investigations for such cases may reduce the risk for outbreaks.

A field-validated approach using surveillance and genotyping data to estimate tuberculosis attributable to recent transmission in the United States.

Tuberculosis genotyping data are frequently used to estimate the proportion of tuberculosis cases in a population that are attributable to recent transmission (RT). Multiple factors influence genotype-based estimates of RT and limit the comparison of estimates over time and across geographic units. Additionally, methods used for these estimates have not been validated against field-based epidemiologic assessments of RT. Here we describe a novel genotype-based approach to estimation of RT based on the identification of plausible-source cases, which facilitates systematic comparisons over time and across geographic areas. We compared this and other genotype-based RT estimation approaches with the gold standard of field-based assessment of RT based on epidemiologic investigation in Arkansas, Maryland, and Massachusetts during 1996-2000. We calculated the sensitivity and specificity of each approach for epidemiologic evidence of RT and calculated the accuracy of each approach across a range of hypothetical RT prevalence rates plausible for the United States. The sensitivity, specificity, and accuracy of genotype-based RT estimates varied by approach. At an RT prevalence of 10%, accuracy ranged from 88.5% for state-based clustering to 94.4% with our novel approach. Our novel, field-validated approach allows for systematic assessments over time and across public health jurisdictions of varying geographic size, with an established level of accuracy.

Controlling the last known cluster of Ebola virus disease - Liberia, January-February 2015.

As one of the three West African countries highly affected by the 2014-2015 Ebola virus disease (Ebola) epidemic, Liberia reported approximately 10,000 cases. The Ebola epidemic in Liberia was marked by intense urban transmission, multiple community outbreaks with source cases occurring in patients coming from the urban areas, and outbreaks in health care facilities (HCFs). This report, based on data from routine case investigations and contact tracing, describes efforts to stop the last known chain of Ebola transmission in Liberia. The index patient became ill on December 29, 2014, and the last of 21 associated cases was in a patient admitted into an Ebola treatment unit (ETU) on February 18, 2015. The chain of transmission was stopped because of early detection of new cases; identification, monitoring, and support of contacts in acceptable settings; effective triage within the health care system; and rapid isolation of symptomatic contacts. In addition, a "sector" approach, which divided Montserrado County into geographic units, facilitated the ability of response teams to rapidly respond to community needs. In the final stages of the outbreak, intensive coordination among partners and engagement of community leaders were needed to stop transmission in densely populated Montserrado County. A companion report describes the efforts to enhance infection prevention and control efforts in HCFs. After February 19, no additional clusters of Ebola cases have been detected in Liberia. On May 9, the World Health Organization declared the end of the Ebola outbreak in Liberia.

Initiation of a ring approach to infection prevention and control at non-Ebola health care facilities - Liberia, January-February 2015.

From mid-January to mid-February 2015, all confirmed Ebola virus disease (Ebola) cases that occurred in Liberia were epidemiologically linked to a single index patient from the St. Paul Bridge area of Montserrado County. Of the 22 confirmed patients in this cluster, eight (36%) sought and received care from at least one of 10 non-Ebola health care facilities (HCFs), including clinics and hospitals in Montserrado and Margibi counties, before admission to an Ebola treatment unit. After recognition that three patients in this emerging cluster had received care from a non-Ebola treatment unit, and in response to the risk for Ebola transmission in non-Ebola treatment unit health care settings, a focused infection prevention and control (IPC) rapid response effort for the immediate area was developed to target facilities at increased risk for exposure to a person with Ebola (Ring IPC). The Ring IPC approach, which provided rapid, intensive, and short-term IPC support to HCFs in areas of active Ebola transmission, was an addition to Liberia's proposed longer term national IPC strategy, which focused on providing a comprehensive package of IPC training and support to all HCFs in the country. This report describes possible health care worker exposures to the cluster's eight patients who sought care from an HCF and implementation of the Ring IPC approach. On May 9, 2015, the World Health Organization (WHO) declared the end of the Ebola outbreak in Liberia.

Estimated rate of reactivation of latent tuberculosis infection in the United States, overall and by population subgroup.

We estimated the rate of reactivation tuberculosis (TB) in the United States, overall and by population subgroup, using data on TB cases and Mycobacterium tuberculosis isolate genotyping reported to the Centers for Disease Control and Prevention during 2006-2008. The rate of reactivation TB was defined as the number of non-genotypically clustered TB cases divided by the number of person-years at risk for reactivation due to prevalent latent TB infection (LTBI). LTBI was ascertained from tuberculin skin tests given during the 1999-2000 National Health and Nutrition Examination Survey. Clustering of TB cases was determined using TB genotyping data collected by the Centers for Disease Control and Prevention and analyzed via spatial scan statistic. Of the 39,920 TB cases reported during 2006-2008, 79.7% were attributed to reactivation. The overall rate of reactivation TB among persons with LTBI was estimated as 0.084 (95% confidence interval (CI): 0.083, 0.085) cases per 100 person-years. Rates among persons with and without human immunodeficiency virus coinfection were 1.82 (95% CI: 1.74, 1.89) and 0.073 (95% CI: 0.070, 0.075) cases per 100 person-years, respectively. The rate of reactivation TB among persons with LTBI was higher among foreign-born persons (0.098 cases/100 person-years; 95% CI: 0.096, 0.10) than among persons born in the United States (0.082 cases/100 person-years; 95% CI: 0.080, 0.083). Differences in rates of TB reactivation across subgroups support current recommendations for targeted testing and treatment of LTBI.

Trends in tuberculosis - United States, 2013.

In 2013, a total of 9,588 new tuberculosis (TB) cases were reported in the United States, with an incidence rate of 3.0 cases per 100,000 population, a decrease of 4.2% from 2012. This report summarizes provisional TB surveillance data reported to CDC in 2013. Although case counts and incidence rates continue to decline, certain populations are disproportionately affected. The TB incidence rate among foreign-born persons in 2013 was approximately 13 times greater than the incidence rate among U.S.-born persons, and the proportion of TB cases occurring in foreign-born persons continues to increase, reaching 64.6% in 2013. Racial/ethnic disparities in TB incidence persist, with TB rates among non-Hispanic Asians almost 26 times greater than among non-Hispanic whites. Four states (California, Texas, New York, and Florida), home to approximately one third of the U.S. population, accounted for approximately half the TB cases reported in 2013. The proportion of TB cases occurring in these four states increased from 49.9% in 2012 to 51.3% in 2013. Continued progress toward TB elimination in the United States will require focused TB control efforts among populations and in geographic areas with disproportionate burdens of TB.

Using statistical methods and genotyping to detect tuberculosis outbreaks.

BACKGROUND: Early identification of outbreaks remains a key component in continuing to reduce the burden of infectious disease in the United States. Previous studies have applied statistical methods to detect unexpected cases of disease in space or time. The objectives of our study were to assess the ability and timeliness of three spatio-temporal methods to detect known outbreaks of tuberculosis. METHODS: We used routinely available molecular and surveillance data to retrospectively assess the effectiveness of three statistical methods in detecting tuberculosis outbreaks: county-based log-likelihood ratio, cumulative sums, and a spatial scan statistic. RESULTS: Our methods identified 8 of the 9 outbreaks, and 6 outbreaks would have been identified 1-52 months (median=10 months) before local public health authorities identified them. Assuming no delays in data availability, 46 (59.7%) of the 77 patients in the 9 outbreaks were identified after our statistical methods would have detected the outbreak but before local public health authorities became aware of the problem. CONCLUSIONS: Statistical methods, when applied retrospectively to routinely collected tuberculosis data, can successfully detect known outbreaks, potentially months before local public health authorities become aware of the problem. The three methods showed similar results; no single method was clearly superior to the other two. Further study to elucidate the performance of these methods in detecting tuberculosis outbreaks will be done in a prospective analysis.

Using genotyping and geospatial scanning to estimate recent mycobacterium tuberculosis transmission, United States.

To determine the proportion of reported tuberculosis (TB) cases due to recent transmission in the United States, we conducted a cross-sectional study to examine culture-positive TB cases with complete genotype results (spoligotyping and 12-locus mycobacterial interspersed repetitive unit-variable-number tandem repeat typing) reported during January 2005-December 2009. Recently transmitted cases were defined as cases with matching results reported within statistically significant geospatial zones (identified by a spatial span statistic within a sliding 3-year window). Approximately 1 in 4 TB cases reported in the United States may be attributed to recent transmission. Groups at greatest risk for recent transmission appear to be men, persons born in the United States, members of a minority race or ethnic group, persons who abuse substances, and the homeless. Understanding transmission dynamics and establishing strategies for rapidly detecting recent transmission among these populations are essential for TB elimination in the United States.

Unexpected decline in tuberculosis cases coincident with economic recession - United States, 2009.

BACKGROUND: Since 1953, through the cooperation of state and local health departments, the U.S. Centers for Disease Control and Prevention (CDC) has collected information on incident cases of tuberculosis (TB) disease in the United States. In 2009, TB case rates declined -11.4%, compared to an average annual -3.8% decline since 2000. The unexpectedly large decline raised concerns that TB cases may have gone unreported. To address the unexpected decline, we examined trends from multiple sources on TB treatment initiation, medication sales, and laboratory and genotyping data on culture-positive TB. METHODS: We analyzed 142,174 incident TB cases reported to the U. S. National Tuberculosis Surveillance System (NTSS) during January 1, 2000-December 31, 2009; TB control program data from 59 public health reporting areas; self-reported data from 50 CDC-funded public health laboratories; monthly electronic prescription claims for new TB therapy prescriptions; and complete genotyping results available for NTSS cases. Accounting for prior trends using regression and time-series analyses, we calculated the deviation between observed and expected TB cases in 2009 according to patient and clinical characteristics, and assessed at what point in time the deviation occurred. RESULTS: The overall deviation in TB cases in 2009 was -7.9%, with -994 fewer cases reported than expected (P < .001). We ruled out evidence of surveillance underreporting since declines were seen in states that used new software for case reporting in 2009 as well as states that did not, and we found no cases unreported to CDC in our examination of over 5400 individual line-listed reports in 11 areas. TB cases decreased substantially among both foreign-born and U.S.-born persons. The unexpected decline began in late 2008 or early 2009, and may have begun to reverse in late 2009. The decline was greater in terms of case counts among foreign-born than U.S.-born persons; among the foreign-born, the declines were greatest in terms of percentage deviation from expected among persons who had been in the United States less than 2 years. Among U.S.-born persons, the declines in percentage deviation from expected were greatest among homeless persons and substance users. Independent information systems (NTSS, TB prescription claims, and public health laboratories) reported similar patterns of declines. Genotyping data did not suggest sudden decreases in recent transmission. CONCLUSIONS: Our assessments show that the decline in reported TB was not an artifact of changes in surveillance methods; rather, similar declines were found through multiple data sources. While the steady decline of TB cases before 2009 suggests ongoing improvement in TB control, we were not able to identify any substantial change in TB control activities or TB transmission that would account for the abrupt decline in 2009. It is possible that other multiple causes coincident with economic recession in the United States, including decreased immigration and delayed access to medical care, could be related to TB declines. Our findings underscore important needs in addressing health disparities as we move towards TB elimination in the United States.

Estimated Latent Tuberculosis Infection Prevalence and Tuberculosis Reactivation Rates Among Non-U.S.-Born Residents in the United States, from the 2011-2012 National Health and Nutrition Examination Survey.

Increased testing and treatment of latent tuberculosis infection (LTBI) among US-residents who were born (or lived) in countries with high rates of TB can hasten progress toward TB elimination. We calculated LTBI prevalence using QuantiFERON®-TB Gold In-Tube results from the 2011 to 2012 National Health and Nutrition Examination Survey (NHANES). LTBI prevalence was highest for persons born in India (31.7%, 95% confidence interval [21.2, 44.5]). Non-Hispanic white persons had the lowest LTBI prevalence (6.3% [1.9, 18.9]). TB reactivation rate, defined as the number of TB cases not associated with recent transmission per 100 person-years of life with LTBI, was highest for persons born in Vietnam [0.183 (0.117, 0.303)]. Reactivation rates were lower among persons who had resided in the United States for ≥ 10 years than among those who had resided for < 10 years. Results among high risk populations can guide LTBI targeted testing and treatment among non-U.S.-born residents.

Birth cohort effect on latent tuberculosis infection prevalence, United States.

BACKGROUND: Latent tuberculosis infection (LTBI) prevalence in the United States decreased approximately 60% in the three decades between the 1971-1972 and 1999-2000 National Health and Nutrition Examination Survey (NHANES) surveys. We examined the effects of birth cohort on LTBI prevalence over time. METHODS: Using weighted data analysis software to account for NHANES survey design, we calculated the difference in LTBI prevalence between 1971-1972 and 1999-2000 for birth cohorts corresponding to 5-year intervals (1912-1916, 1917-1921,1922-1926, 1927-1931, 1932-1936, 1937-1941, 1942-1946). RESULTS: LTBI prevalence was significantly lower in 1999-2000 compared to 1971-1972 for cohorts born in 1926 or earlier (19% versus 5%), but not for cohorts born 1927-1946 (9% versus 7%). Adjustment for cohort restriction and foreign-birth did not qualitatively change the results. CONCLUSIONS: Although older age groups have higher rates of TB infection than younger groups, nationally representative U.S. survey data suggest that observed LTBI prevalence in older people represents an underestimate of infection, because of the birth cohort effect and waning immunologic reactivity.

Increasing proportions of advanced pulmonary tuberculosis reported in the United States: are delays in diagnosis on the rise?

RATIONALE: Delays in the diagnosis of tuberculosis (TB) can result in progression to advanced disease. Patients with pulmonary TB and advanced disease are more likely to transmit disease and fail treatment. OBJECTIVES: To examine clinical, epidemiological, and geographic factors associated with advanced pulmonary TB to further understanding of delayed diagnosis and transmission. METHODS: Pulmonary tuberculosis cases in persons older than 15 years of age reported to the U.S. National Tuberculosis Surveillance System with advanced disease (cavitation on chest radiograph and acid-fast bacilli smear-positive sputum result) were compared with those without advanced disease using trend and binomial regression analysis. MEASUREMENTS AND MAIN RESULTS: There were 35,584 cases of advanced pulmonary tuberculosis (APT) and 125,077 cases of non-APT reported from 1993 through 2006. Proportions of pulmonary TB cases with APT increased from 18.5% in 1993 to 26.1% in 2006, and the increase in the proportion of APT was most notable for national TB rates below 6.6 per 100,000. At the county level, the association between APT and low TB incidence has grown incrementally since 2000. The proportion of APT increased greatest among whites (65.4%), the employed (63.3%), and the U.S. born (59.2%). The prevalence of APT was 44% greater among persons with multidrug-resistant TB compared with those without it. CONCLUSIONS: This study highlights the need for TB diagnosis at early stages of the disease to minimize APT and decrease the risk of transmission. Additional efforts should concentrate on reducing time to treatment initiation in low-incidence areas and among groups traditionally seen as being at low risk for TB disease.