pmc10169900346113Lancet Public HealthLancet Public HealthThe Lancet. Public health2468-2667390951331158788710.1016/S2468-2667(24)00151-8HHSPA2026495ArticleRisk factors underlying racial and ethnic disparities in tuberculosis diagnosis and treatment outcomes, 2011–19: a multiple mediation analysis of national surveillance dataReganMathildaBarhamTerrikaLiYunfeiSwartwoodNicole AAsayGarrett R BeelerCohenTedHorsburghC RobertJrKhanAwalMarksSuzanne MMylesRanell LSalomonJoshua ASelfJulie LWinstonCarla AMenziesNicolas ADepartment of Global Health and Population, Harvard T H Chan School of Public Health, Boston, MA, USA (M Regan PhD, Y Li ScD, N A Swartwood MSPH, N A Menzies PhD); Office of Health Equity (T Barham PhD, R L Myles PhD) and Division of Tuberculosis Elimination (G R Beeler Asay PhD, A Khan PhD, S M Marks MPH, J L Self PhD, C A Winston PhD), National Center for HIV, Viral Hepatitis, STD, and Tuberculosis Prevention, US Centers for Disease Control and Prevention, Atlanta, GA, USA; Yale School of Public Health, New Haven, CT, USA (Prof T Cohen DPH); Department of Epidemiology, Department of Biostatistics, and Department of Global Health, School of Public Health, Boston University, Boston, MA, USA (Prof C R Horsburgh Jr MD); Department of Medicine, School of Medicine, Boston University, Boston, MA, USA (Prof C R Horsburgh Jr); Department of Health Policy, Stanford University, Stanford, CA, USA (Prof J A Salomon PhD)

Contributors

Conception and design: MR, TB, GRBA, AK, SMM, JLS, CAW, and NAM. Analysis: MR. Interpretation: all authors. Accessed and verified underlying data: MR, YL, and NAM. First draft of manuscript: MR. Review and editing of manuscript, and decision to submit for publication: all authors. All authors had access to presented and output data.

Correspondence to: Dr Mathilda Regan, Department of Global Health and Population, Harvard T H Chan School of Public Health, Boston, MA 02115, USA, mathildaregan@hsph.harvard.edu309202482024018202598e564e572https://creativecommons.org/licenses/by-nc-nd/4.0/This is an Open Access article under the CC BY-NC 4.0 license.SummaryBackground

Despite an overall decline in tuberculosis incidence and mortality in the USA in the past two decades, racial and ethnic disparities in tuberculosis outcomes persist. We aimed to examine the extent to which inequalities in health and neighbourhood-level social vulnerability mediate these disparities.

Methods

We extracted data from the US National Tuberculosis Surveillance System on individuals with tuberculosis during 2011–19. Individuals with multidrug-resistant tuberculosis or missing data on race and ethnicity were excluded. We examined potential disparities in tuberculosis outcomes among US-born and non-US-born individuals and conducted a mediation analysis for groups with a higher risk of treatment incompletion (a summary outcome comprising diagnosis after death, treatment discontinuation, or death during treatment). We used sequential multiple mediation to evaluate eight potential mediators: three comorbid conditions (HIV, end-stage renal disease, and diabetes), homelessness, and four census tract-level measures (poverty, unemployment, insurance coverage, and racialised economic segregation [measured by Index of Concentration at the ExtremesRace–Income]). We estimated the marginal contribution of each mediator using Shapley values.

Findings

During 2011–19, 27 788 US-born individuals and 57 225 non-US-born individuals were diagnosed with active tuberculosis, of whom 27 605 and 56 253 individuals, respectively, met eligibility criteria for our analyses. We did not observe evidence of disparities in tuberculosis outcomes for non-US-born individuals by race and ethnicity. Therefore, subsequent analyses were restricted to US-born individuals. Relative to White individuals, Black and Hispanic individuals had a higher risk of not completing tuberculosis treatment (adjusted relative risk 1·27, 95% CI 1·19–1·35; 1·22, 1·11–1·33, respectively). In multiple mediator analysis, the eight measured mediators explained 67% of the disparity for Black individuals and 65% for Hispanic individuals. The biggest contributors to these disparities for Black individuals and Hispanic individuals were concomitant end-stage renal disease, concomitant HIV, census tract-level racialised economic segregation, and census tract-level poverty.

Interpretation

Our findings underscore the need for initiatives to reduce disparities in tuberculosis outcomes among US-born individuals, particularly in highly racially and economically polarised neighbourhoods. Mitigating the structural and environmental factors that lead to disparities in the prevalence of comorbidities and their case management should be a priority.

Introduction

Racial and ethnic disparities in tuberculosis outcomes in the USA have been well documented.1–3 An analysis of national tuberculosis surveillance data from 2003 to 2018 found that US-born Black, Hispanic, and Native American people with tuberculosis disease had higher risks of mortality and treatment discontinuation, relative to White people.3 Similarly, studies have observed that tuberculosis incidence rate ratios have remained elevated or increased for all racial and ethnic groups relative to White people over the past two decades.1,2 These disparities have persisted despite the overall steady decline in tuberculosis incidence and mortality in the USA during this period.4 Less is known about the mechanisms that underlie these disparities.

We conceptualise race and ethnicity as social constructs representing inter-related historical and contemporary discriminatory processes and structures that adversely affect health, including inequitable living and working conditions, socioeconomic deprivation, poor access to health care, under-representation in public health surveillance and outreach programmes, psychosocial stress, and interpersonal racism and discrimination.5 Although many of these pathways could potentially explain disparities in tuberculosis outcomes, few studies have empirically assessed their role and relative importance. A previous study combining tuberculosis surveillance data and US census data found that socioeconomic status explains approximately half of racial and ethnic disparities in tuberculosis incidence.6 Comorbidities associated with a greater risk of tuberculosis mortality, including HIV,7 diabetes,8 and end-stage renal disease,9 disproportionately affect some racial and ethnic minority populations.10–12 Although increased access to antiretroviral therapy (ART) has lowered the risk of mortality for individuals with tuberculosis–HIV co-infection,7 racial and ethnic disparities persist in access to HIV care and ART retention.10

A growing body of social epidemiology research has used causal mediation methods to assess the extent to which more proximal risk factors mediate the association between race and ethnicity and health outcomes.13 However, to the best of our knowledge, no study has applied this approach to tuberculosis outcomes in the USA. In this analysis, we used a sequential multiple mediation approach to examine several potential mediators simultaneously.14 With use of national tuberculosis surveillance data on cases among US-born and non-US-born individuals (representing approximately 27% and 73% of cases reported in the USA, respectively) we aimed to: (1) assess whether risk factors arising from discriminatory processes and structures (exposure to neighbourhood-level social deprivation, homelessness, and comorbidities) contribute to disparities for groups with higher risk of adverse tuberculosis outcomes (diagnosis after death, treatment discontinuation, and death during treatment); and (2) quantify the strength of these associations.

MethodsStudy population

We extracted data on individual-level outcomes and risk factors for tuberculosis cases reported to the National Tuberculosis Surveillance System (NTSS)15 among both non-US-born individuals and individuals from the 50 US states and District of Columbia (including individuals born abroad of a US citizen or born in a US-affiliated jurisdiction; hereafter referred to as US-born) during 2011–19. We excluded individuals diagnosed with multidrug-resistant tuberculosis (1·08%, n=919), because the criteria for successful treatment completion differ for these individuals (eg, clinical guidelines in effect during the period of the data recommended a treatment duration of up to 24 months, so we would need to exclude 2019 data to ensure that we captured the full treatment duration for these individuals).

This analysis of routinely collected public health data was reviewed by the US Centers for Disease Control and Prevention (CDC), was deemed not research, and was conducted consistent with CDC policy and applicable federal law (45 CFR part 46.102[l][2]) which includes no requirement of informed consent.

Analytic approach

We examined potential disparities in tuberculosis outcomes among US-born and non-US-born people. In cases where disparities were present, we used a sequential mediation approach14 to estimate the extent to which risk factors mediate the association between race and ethnicity (the exposure) and tuberculosis outcomes. This approach allows us to examine multiple mediators simultaneously, while accommodating cases in which the mediators affect one another. Mediation analysis yields the indirect effect (the effect through measured mediators), the direct effect (the effect through unmeasured mediators), the total effect or overall association between exposure and outcome, and the percent mediated. Going forward we do not use the terms direct effect and indirect effect as is conventional for mediation analyses, because the term direct effect could be taken to imply that race or ethnicity directly affects tuberculosis outcomes, yet there is no biological basis for this interpretation. Instead, differences not captured by a given set of mediators are assumed to result from unmodelled mediators.

Exposure

The exposure variable of interest was race and ethnicity (representing the social construction of race and ethnicity and how that construction negatively affects the lives of traditionally marginalised communities). We extracted self-reported data on race and ethnicity and categorised individuals according to US census single-race classification conventions as: Non-Hispanic Black, Non-Hispanic White, Hispanic, Non-Hispanic Asian, Non-Hispanic American Indian or Alaska Native, Non-Hispanic Native Hawaiian or Other Pacific Islander or Non-Hispanic Other (Multiracial). Hereafter, these classifications are referred to as Black, White, Hispanic, Asian, American Indian or Alaska Native, Native Hawaiian or Other Pacific Islander, and Multiracial. Individuals who identified as Hispanic and Multiracial were classified as Hispanic. We excluded individuals with missing data on race and ethnicity from the analysis (0·27%, n=226).

Mediators

Potential mediators were selected based on a literature review of tuberculosis risk factors in the USA.6,16,17 We extracted data on four individual-level potential mediators from the NTSS: laboratory HIV test result, diagnosis of diabetes (type 1 or 2) at time of tuberculosis diagnosis or before, end-stage renal disease or chronic renal failure at time of tuberculosis diagnosis or before, and whether the patient experienced homelessness at any point in the year preceding tuberculosis diagnosis.

We extracted data on four census tract-level measures of social vulnerability from the American Community Survey: percentage of individuals below 150% poverty line, percentage of civilian (age ≥16 years) individuals experiencing unemployment, percentage of the total civilian non-institutionalised population without health insurance, and the Index of Concentration at the ExtremesRace–Income (ICERace–Income), a racialised measure of economic segregation.18 The measure ranges from −1 (least privileged) to 1 (most privileged) and compares the proportion of a census tract that is both high income and White to the proportion that is low-income and Black (including Hispanic). We used crosswalk files from the US Department of Housing and Urban Development19 to link zip codes in the NTSS dataset with census tracts in the American Community Survey data. Because zip codes can contain multiple census tracts, we used data on population weighting from the Department of Housing and Urban Development to calculate a weighted average of census tract data, giving more populous tracts a greater weight.

Outcomes

We analysed three tuberculosis outcomes (diagnosis after death, treatment discontinuation, and death during treatment), as well as one summary outcome representing failure to complete treatment (diagnosis after death, treatment discontinuation, or death during treatment). Diagnosis after death included all causes of death and all sites of tuberculosis disease. Treatment discontinuation includes patients who refused treatment or were lost to follow-up. Death during treatment included any deaths that occurred between initiation of treatment and treatment completion or discontinuation. For the outcomes of treatment discontinuation and death during treatment, we excluded individuals who stopped treatment due to adverse drug reactions (0·30%, n=242) or unspecified other reasons (0·02%, n=17).

We adjusted for patient sex at birth, age category (<1, 1–4, 5–14, 15–24, 25–34, 35–44, 45–54, 55–64, 65–74, 75–84, and ≥85 years), calendar year, and location, based on US Census Bureau regional division (Pacific, Mountain, West North Central, West South Central, East North Central, East South Central, South Atlantic, Middle Atlantic, and New England).

Statistical analysis

We used descriptive statistics to examine the distribution of demographic characteristics across race and ethnicity and place of birth (US-born or non-US-born). We evaluated the correlation between our social vulnerability variables using Pearson correlation coefficients. We used multivariable log-binomial models to assess the relative risk (RR) of each tuberculosis outcome for each racial and ethnic group compared with White people, stratifying by place of birth. If we did not observe evidence of a disparity (RR <1) for a racial and ethnic group or place of birth, we did not include them in subsequent analyses. We fit multivariable log-binomial models to estimate the association between race and ethnicity and potential mediators, using White people as the reference group.

We used the R package CMAverse to estimate the total effect, mediated effect, unmediated effect, and percent mediated, while assessing all potential mediators simultaneously.20 We estimated the percent mediated as follows: RRunmediated × (RRmediated – 1)/(total effect – 1). The percent mediated could exceed 100% if there were unmeasured mediators that had a protective effect on the outcome. We calculated mediation effects using the g-formula approach because it is doubly robust.13 Standard errors were derived from 100 bootstrap replications. Further details on the mediation approach are provided in the appendix (p 1).

To examine the marginal effects of individual mediators, we examined them sequentially. Because the relative contribution of each mediator in sequential mediation analysis depends on the order in which they are entered into the model, we used Shapley values21 to estimate the marginal percent mediated by each mediator averaged across all possible sequences of model entry. We did not report the percent mediated (or Shapley values) in instances in which there was not a significant mediated effect for the association between race and ethnicity and the outcome.

We imputed missing covariate and potential mediator data using the R package Amelia22 and Rubin’s rule.23 Missing data from covariates and other potential mediators were less than 5% apart from the ICERace–Income variable (8%). Details on imputation and the estimation of Shapley values are included in the appendix (p 1). All statistical analyses were carried out in R (version 4.3.0).

To assess the robustness of our results, we carried out several sensitivity analyses. First, we divided our dataset into two subgroups by time period (2011–15 and 2016–19) to assess whether the relative contribution of HIV changed between these two periods. Second, we re-ran our mediation models using a complete-case dataset. For the HIV variable, we re-coded all cases that were not confirmed positive or negative as negative. Third, we re-ran the mediation analyses, stratifying by sex.

Role of the funding source

Employees of the funder participated as coauthors on the study, contributing to study design, data analysis, data interpretation, manuscript preparation, and the decision to submit for publication. The funder had no other role in these processes.

Results

During 2011–19, 27 788 tuberculosis cases were reported to the NTSS among US-born individuals and 57 225 among non-US-born individuals. After excluding cases with missing data on race and ethnicity (n=226) and cases of multidrug-resistant tuberculosis (n=919), 27 605 US-born cases and 56 253 non-US-born cases remained in our analysis. Demographic characteristics of the analytic sample are shown in the appendix (pp 2–3). The majority of the study population was male among both US-born (n=17 788, 64%) and non-US-born (n=33 310, n=59%) individuals. The greatest proportion of US-born individuals was located in the South Atlantic region of the USA (n=6571, 24%), whereas the greatest proportion of non-US-born individuals was located in the Pacific region (n=18 125, 32%). We observed positive correlations between the four social vulnerability variables (census tract-level poverty, unemployment, uninsurance, and racialised economic segregation (appendix p 5).

Associations between race and ethnicity and adverse tuberculosis outcomes are presented in table 1 (for US-born individuals) and the appendix (p 6; for non-US-born individuals). Relative to White individuals, we observed higher risks of diagnosis after death among Black individuals (adjusted RR [aRR] 1·24, 95% CI 1·07–1·44) and Hispanic individuals (1·46, 1·19–1·78). Black individuals had a higher risk of treatment discontinuation (1·28, 1·06–1·54), whereas Asian individuals had a lower risk (0·43, 0·21–0·77). We observed higher risks of death during treatment among Black individuals (1·30, 1·20–1·41), Hispanic individuals (1·21, 1·06–1·37), and American Indian or Alaska Native individuals (1·34, 1·09–1·63). Overall, the risk of not completing treatment (ie, the summary outcome) was higher among Black individuals (1·27, 1·19–1·35), Hispanic individuals (1·22, 1·11–1·33), and American Indian or Alaska Native individuals (1·19, 1·01–1·39; table 1). We did not observe evidence of disparities (RR <1) in tuberculosis outcomes for non-US-born individuals and therefore restricted our subsequent analyses to US-born individuals. Relative to US-born and non-US-born White individuals combined, we observed a significantly lower risk of treatment incompletion for non-US-born Black (0·83, 0·75–0·91), Hispanic (0·91, 0·85–0·97), and Asian (0·77, 0·73–0·82) individuals (appendix p 6).

The posited relationships between variables in the mediation analysis are shown in the appendix (p 4). Associations between race and ethnicity (the exposure) and potential mediators are presented in table 2. Relative to White individuals, we observed that Black, Hispanic, and American Indian or Alaska Native individuals with active tuberculosis had significantly higher risks of living in the most disadvantaged quintiles of census tracts for poverty, unemployment, lack of health insurance, and racialised economic segregation. Black individuals were more likely to have a positive laboratory HIV test (aRR 1·19, 95% CI 1·12–1·26) and to have experienced homelessness in the past year (1·32, 1·23–1·43). Relative to White individuals, all other racial and ethnic groups had a higher risk of diabetes and end-stage renal disease.

Results from the mediation analysis are shown in table 3 (for Black and Hispanic individuals) and the appendix (p 7; for American Indian and Alaska Native, Asian, Native Hawaiian and Other Pacific Islander, and Multiracial individuals). For risk of diagnosis after death, the disparity between Black and White individuals was significantly mediated by the eight measured mediators (comorbid HIV, end-stage renal disease, and diabetes; homelessness; and the most disadvantaged quintiles of poverty, unemployment, lack of insurance, and racialised economic segregation, assessed simultaneously; joint percent mediated 70%, p=0·0017). The joint effect through measured mediators had an aRR of 1·17 (95% CI 1·06–1·27) and the effect through unmeasured mediators had an aRR of 1·08 (0·89–1·26). The biggest marginal contributions were HIV infection (38%), end-stage renal disease (28%), and racialised economic segregation (15%). We did not observe a significant effect through measured mediators or unmeasured mediators among Hispanic individuals.

For risk of treatment discontinuation, the disparity between Black and White individuals was significantly mediated by measured mediators (aRR 1·11, 95% CI 1·00–1·23; joint percent mediated 50%, p=0·14). The effect through unmeasured mediators had an aRR of 1·13 (0·90–1·36). The biggest marginal contributors to the disparity were HIV infection (26%), homelessness (18%), census tract-level unemployment (16%) and census tract-level poverty (11%). We did not observe a disparity in treatment discontinuation between Hispanic and White individuals.

For risk of death during treatment, the disparity between Black and White individuals was significantly mediated by the measured mediators (aRR 1·23, 95% CI 1·16–1·30; joint percent mediated 74%, p=0·0002). The biggest marginal contributors for Black individuals were HIV infection (17%), end-stage renal disease (17%), poverty (12%), and unemployment (11%). The effect through unmeasured mediators had an aRR of 1·08 (0·96–1·21). The disparity between Hispanic and White individuals for death during treatment was also significantly mediated by measured mediators (1·26, 1·15–1·37; joint percent mediated 122%, p=0·0002). Of note, a percent mediated greater than 100% means that there were unmeasured mediators with an inverse association with the outcome. The effect through unmeasured mediators had an aRR of 0·96 (0·80–1·11). The biggest marginal contributors to disparities for Hispanic individuals were end-stage renal disease (27%), diabetes (24%), racialised economic segregation (28%), HIV (18%), and lack of insurance (14%). We observed a statistically significant effect through unmeasured mediators (1·56, 1·18–1·92) for American Indian and Alaska Native individuals but not through measured mediators (0·88, 0·75–1·00).

For the summary outcome, risk of incomplete treatment, the disparity between Black and White individuals was significantly mediated by the measured mediators (joint percent mediated 67%, p<0·0001). The effect through measured mediators had an aRR of 1·20 (95% CI 1·15–1·24), and the effect through unmeasured mediators had an aRR of 1·10 (1·01–1·20). The biggest marginal contributors were HIV infection (20%), end-stage renal disease (15%), and census tract-level poverty (12%). The joint percent mediated through measured mediators was 65% for Hispanic individuals (p<0·0001), and the biggest marginal contributors were end-stage renal disease (18%) and HIV (17%). The effect through measured mediators had an aRR of 1·15 (1·07–1·22) and the effect through unmeasured mediators had an aRR of 1·08 (0·95–1·21).

The mediation analysis by time period is shown in the appendix (pp 8–14). We did not observe differences in the relative contribution of HIV during the period 2011–14 compared with the period 2015–19 (we were not able to compare the two time periods for all outcomes because the percent mediated was not significant in some cases). The sex-stratified mediation analysis is shown in the appendix (pp 15–18). For incomplete treatment, we observed a higher joint percent mediated for Black men (83%, p=0·0031) compared with Black women (53%, p<0·0001). HIV was a greater contributor to the disparity for men (26%) compared with women (14%), whereas diabetes was a slightly higher contributor for women (7% vs 4% for men). Sensitivity analyses comparing a complete case analysis to the main models using imputed data are shown in the appendix (pp 19–25). The two approaches yielded similar results with the exception of the Shapley values for diagnosis after death; for Black individuals and Hispanic individuals, we observed a greater contribution due to HIV in the imputed analysis compared with the complete case analysis.

Discussion

This is the first analysis to use a sequential multiple mediation analysis to estimate the marginal contribution of risk factors underlying racial and ethnic disparities in tuberculosis outcomes in the USA. We restricted our mediation analysis to US-born individuals, because we did not observe disparities in tuberculosis outcomes among non-US-born individuals. Our findings suggest that exposure to neighbourhood-level socioeconomic disadvantage, economic and racial polarisation, and the presence of comorbidities (HIV, diabetes, and end-stage renal disease) have important roles in explaining disparities in tuberculosis outcomes for US-born Black and Hispanic individuals.

A positive HIV diagnosis and comorbid end-stage renal disease were two of the largest contributors to disparities in treatment completion for Black and Hispanic individuals. Although this is the first study to evaluate comorbidities as potential mediators of the association between race and ethnicity and tuberculosis outcomes, our findings are consistent with previous descriptive research. Black and Hispanic individuals in the USA are disproportionately affected by HIV, and Black individuals with tuberculosis are disproportionately likely to have comorbid HIV.10,24 HIV co-infection is associated with an increased risk of post-mortem tuberculosis diagnosis25 and death during treatment.11 Black individuals in the USA are two to four times more likely to develop kidney failure than White individuals but are less likely to receive a transplant.26 Individuals with tuberculosis and end-stage renal disease have a 2–3-times higher risk of mortality relative to the general tuberculosis population.27 Diagnosis of tuberculosis among individuals with HIV or end-stage renal disease is more likely to be delayed due to atypical clinical presentation.10,27

Our finding that diabetes was one of the largest contributors to the disparity in death during treatment for Hispanic individuals is consistent with earlier research that observed a higher prevalence of diabetes among Hispanic individuals with tuberculosis,28 and a higher risk of death for individuals with comorbid tuberculosis and diabetes.8,28 Individuals with comorbid tuberculosis and diabetes are more likely to have more severe case presentation at diagnosis and to respond more slowly to tuberculosis treatment.29 National guidelines recommend baseline diabetes screening for individuals with tuberculosis who are at greater risk for diabetes, including Hispanic individuals,30 and there are global recommendations for bidirectional screening (ie, also screening for tuberculosis among patients with diabetes).31 Previous research also suggests that targeted latent tuberculosis testing and treatment for individuals with diabetes could be beneficial, because diabetes increases the risk of progression to active tuberculosis.28

We found that homelessness was one of the biggest contributors to the disparity in treatment discontinuation between Black and White individuals. Our findings are consistent with a previous analysis of tuberculosis surveillance data that reported that individuals experiencing homelessness have more than twice the odds of treatment discontinuation.32 Tuberculosis treatment guidelines have stressed the importance of providing appropriate housing for individuals experiencing homelessness who are undergoing tuberculosis treatment,33,34 as well as linkage with additional services and support, including access to treatment for mental health disorders, substance use disorders, and HIV, which could disproportionately affect individuals experiencing homelessness.33 Surveillance of individuals experiencing homelessness for focused testing and treatment is challenging and might require innovative approaches, such as the incorporation of multiple locations per individual in spatial mapping.35

All of the assessed census tract-level measures of deprivation contributed to disparities in tuberculosis outcomes to varying extents. Although these measures do not necessarily reflect individual circumstances, they represent local access to a variety of resources that affect health, including quality health care, education, housing, green spaces, and infrastructure.36 Furthermore, prolonged exposure to social disadvantage (at both the individual and neighbourhood level) has been shown to lead to chronic psychosocial stress and impaired immune function.8 Although the marginal contribution of census tract-level poverty was smaller than that of comorbid health conditions, it consistently contributed to disparities in tuberculosis outcomes for Black and Hispanic individuals. Although previous analyses have found an association between geographical measures of poverty and tuberculosis incidence,37 few studies have examined area-level deprivation as a risk factor for other tuberculosis outcomes. Our findings suggest that not only neighbourhood-level poverty, but also residence in polarised neighbourhoods (as assessed by the ICERace–Income measure of racialised economic segregation) contributes to the disparities in tuberculosis outcomes for Black and Hispanic individuals. A growing body of literature has shown associations between the ICERace–Income measure and adverse health outcomes and has identified hospital quality and comorbidities as potential points of intervention.38 Health facilities in more polarised neighbourhoods are more likely to have quality gaps and patients are more likely to experience discrimination and implicit bias from providers.38

This study has several limitations. First, we have not identified all pathways underlying the disparities, as indicated by the significant effect through unmeasured mediators for incomplete treatment for Black individuals. Due to the unavailability of data, we could not evaluate individual-level socioeconomic status, substance misuse, access to care, recent incarceration, or quality of health services as potential mediators. We did not examine current incarceration as a potential mediator because previous research has shown an inverse association with death during treatment.29 Second, we are unable to untangle the relationship between the mediators, including temporality and directionality. For example, comorbidities might be the result of neighbourhood-level social deprivation. Third, our census tract-level measures of social deprivation are linked to individuals based on zip code at time of diagnosis, but we were unable to assess how long an individual lived in a particular census tract and to what extent they were exposed to neighbourhood-level deprivation. Additionally, because the individual-level zip code for individuals who were incarcerated at time of diagnosis (4% of the study population) represents the facility’s zip code rather than their home address, there is some systematic misclassification of zip codes. This could also be the case for individuals who reported experiencing homelessness in the past year (10% of the study population) because the zip code could reflect the location where they sought care rather than their neighbourhood. As a result, the neighbourhood-level deprivation variables for these individuals might not be accurate. Fourth, due to missing data on HIV status we used imputed data in our main models. However, our results were consistent with sensitivity analyses using complete case data, with the exception of our findings for diagnosis after death (which might have been underpowered due to the small number of cases). The Shapley values for HIV for diagnosis after death should be interpreted with caution. However, because we recoded unknown HIV status as negative, our complete case estimates could underestimate the effect due to HIV if some of these unknown cases are positive. Fifth, due to the small number of American Indian and Alaska Native individuals in our dataset, the study could have been underpowered to evaluate mediators for this population. Similarly, sensitivity analyses of results that examined two time periods and sex stratification might be underpowered. Sixth, we were unable to look at clustering among contacts, because these data are not reportable to the CDC. Finally, although our mediation model is doubly robust, requiring only correct specification of the exposure or outcome model, there is likely to be unmeasured confounding of both models. As a result, our mediation effects should not necessarily be interpreted as causal.

Reducing racial and ethnic disparities in tuberculosis should be a priority for public health interventions and policy. Although further research is needed to confirm these findings and examine additional potential mediators, our findings suggest that many of the measured pathways we observed are modifiable. Geographically targeted interventions to increase resources to health facilities and improve quality of health services are needed in the most polarised neighbourhoods. Engagement with community members in underserved areas is crucial to identify the barriers they face along the tuberculosis continuum of care, particularly regarding timely diagnosis and treatment adherence. Although we did not examine potential mediators for reduced risk of adverse tuberculosis outcomes, such as we observed for most non-US-born racial and ethnic populations, this merits a separate investigation and could help identify facilitators of treatment completion. Although our sex-stratified analysis should be interpreted with caution, our findings suggest that further research is needed to identify additional mediators of disparities in tuberculosis outcomes among women. Our results also suggest that interventions to improve coordination of care and quality of care for individuals with comorbid HIV, diabetes, and end-stage renal disease could substantially reduce racial and ethnic disparities in the likelihood of successful tuberculosis treatment completion. Improved access to health insurance, pre-exposure prophylaxis, and antiretroviral therapy is needed to improve treatment outcomes for individuals with comorbid HIV and tuberculosis and to prevent new tuberculosis and HIV diagnoses.14,34 Targeted tuberculosis education and testing among individuals with HIV and end-stage renal disease, as well as the use of nucleic acid amplification testing for rapid diagnosis, could improve tuberculosis case detection and reduce delays in diagnosis.11 Improved coordination of care between providers who are treating tuberculosis and comorbid conditions is also needed, particularly if tuberculosis is treated in the public sector and other conditions in the private sector.28,31 Our findings also underscore the importance of preventing these comorbidities in the first place. Mitigating the structural and environmental factors that lead to disparities in the prevalence of comorbidities and their case management is an important component of addressing tuberculosis disparities. Interventions that address social needs through coordination of services should continue to be supported and may need to be expanded.39,40

Supplementary MaterialAcknowledgments

We thank Donna McCree, Andrew Hill, Cindy Imai, and Jiaying Stephanie Su for their contributions. This project was funded by the US CDC National Center for HIV, Viral Hepatitis, STD, and Tuberculosis Prevention Epidemiologic and Economic Modeling Agreement (grant number 5NU38PS004651). The findings and conclusions in this report are those of the authors and do not necessarily represent the views of the CDC.

Declaration of interests

We declare no competing interests.

See Online for appendix

Data sharing

The National Tuberculosis Surveillance System operates under an Assurance of Confidentiality issued by the Centers for Disease Control and Prevention (CDC) under Sections 306 and 308(d) of the Public Health Service Act (42 USC 242k and 242m[d]). Data are reported voluntarily to CDC by state and local health departments on a case report form called the Report of Verified Case of Tuberculosis (OMB number 0920-0026). The Assurance of Confidentiality prevents disclosure of any information that could be used to directly or indirectly identify patients. For more information, see the CDC/Agency for Toxic Substances and Disease Policy on Releasing and Sharing Data at http://www.cdc.gov/maso/Policy/ReleasingData.pdf. A limited dataset is available at http://wonder.cdc.gov/TB-v2013.html. Researchers may apply to analyse additional data at CDC headquarters by contacting TBInfo@cdc.gov.

ReferencesKhanA, MarksS, KatzD, Changes in tuberculosis disparities at a time of decreasing tuberculosis incidence in the United States, 1994–2016. Am J Public Health 2018; 108: S321–26.30383425 KimS, CohenT, HorsburghCRJr, Trends, mechanisms, and racial/ethnic differences of tuberculosis incidence in the US-born population aged 50 years or older in the United States. Clin Infect Dis 2022; 74: 1594–603.34323959 ReganM, LiY, SwartwoodNA, Racial and ethnic disparities in diagnosis and treatment outcomes among US-born people diagnosed with tuberculosis, 2003–19: an analysis of national surveillance data. Lancet Public Health 2024; 9: e47–56.38176842 Centers for Disease Control and Prevention. Reported tuberculosis in the United States, 2021. 2022. https://www.cdc.gov/tb/statistics/reports/2021/default.htm (accessed March 16, 2024).BaileyZD, KriegerN, AgénorM, GravesJ, LinosN, BassettMT. Structural racism and health inequities in the USA: evidence and interventions. Lancet 2017; 389: 1453–63.28402827 CantwellMF, McKennaMT, McCrayE, OnoratoIM. Tuberculosis and race/ethnicity in the United States: impact of socioeconomic status. Am J Respir Crit Care Med 1998; 157: 1016–20.9563713 SchmitKM, ShahN, KammererS, Bamrah MorrisS, MarksSM. Tuberculosis transmission or mortality among persons living with HIV, USA, 2011–2016. J Racial Ethn Health Disparities 2020;7: 865–73.32060748 BakerMA, HarriesAD, JeonCY, The impact of diabetes on tuberculosis treatment outcomes: a systematic review. BMC Med 2011; 9: 81.21722362 RomanowskiK, ClarkEG, LevinA, CookVJ, JohnstonJC. Tuberculosis and chronic kidney disease: an emerging global syndemic. Kidney Int 2016; 90: 34–40.27178832 SullivanPS, Satcher JohnsonA, PembletonES, Epidemiology of HIV in the USA: epidemic burden, inequities, contexts, and responses. Lancet 2021; 397: 1095–106.33617774 HassanS, GujralUP, QuarellsRC, Disparities in diabetes prevalence and management by race and ethnicity in the USA: defining a path forward. Lancet Diabetes Endocrinol 2023; 11: 509–24.37356445 AssariS. Racial disparities in chronic kidney diseases in the United States; a pressing public health challenge with social, behavioral and medical causes. J Nephropharmacol 2015; 5: 4–6.28197489 NaimiAI, SchnitzerME, MoodieEE, BodnarLM. Mediation analysis for health disparities research. Am J Epidemiol 2016; 184: 315–24.27489089 VanderWeeleTJ, VansteelandtS. Mediation analysis with multiple mediators. Epidemiol Methods 2014; 2: 95–115.25580377 Centers for Disease Control and Prevention. Tuberculosis (TB) (Mycobacterium tuberculosis) 2009 case definition. 2021. https://ndc.services.cdc.gov/case-definitions/tuberculosis-2009 (accessed July 14, 2024).WaittCJ, SquireSB. A systematic review of risk factors for death in adults during and after tuberculosis treatment. Int J Tuberc Lung Dis 2011; 15: 871–85.21496360 LeeDC, LiangH, ShiL. The convergence of racial and income disparities in health insurance coverage in the United States. Int J Equity Health 2021; 20: 96.33827600 KriegerN, WatermanPD, SpasojevicJ, LiW, MaduroG, Van WyeG. Public Health monitoring of privilege and deprivation using the Index of Concentration at the Extremes (ICE). Am J Public Health 2016; 106: 256–63.26691119 US Department of Housing and Urban Development. HUD USPS Zip code crosswalk files. 2021. https://www.huduser.gov/portal/datasets/usps_crosswalk.html#data. (accessed Feb 2, 2023).ShiB, ChoiratC, CoullBA, VanderWeeleTJ, ValeriL. CMAverse: a suite of functions for reproducible causal mediation analyses. Epidemiology 2021; 32: e20–22.34028370 ShapleyLS. A value for n-person games. Ann Math Stud 1953; 28: 307–12.HonakerJ, KingG, BlackwellM, AmeliaII. A program for missing data. J Stat Softw 2011; 45: 1–47.MarshallA, AltmanDG, HolderRL, RoystonP. Combining estimates of interest in prognostic modelling studies after multiple imputation: current practice and guidelines. BMC Med Res Methodol 2009; 9: 57.19638200 MarksSM, KatzDJ, DavidowAL, PagaoaMA, TeeterLD, GravissEA. The impact of HIV infection on TB disparities among US-born Black and White tuberculosis patients in the United States. J Public Health Manag Pract 2020; 26: E5–12.32732731 Centers for Disease Control and Prevention. Mortality among patients with tuberculosis and associations with HIV status—United States, 1993–2008. MMWR Morb Mortal Wkly Rep 2010; 59: 1509–13.21102404 BoulwareLE, PurnellTS, MohottigeD. Systemic kidney transplant inequities for Black individuals: examining the contribution of racialized kidney function estimating equations. JAMA Netw Open 2021; 4: e2034630.33443577 OkadaRC, BarryPM, SkarbinskiJ, ChitnisAS. Epidemiology, detection, and management of tuberculosis among end-stage renal disease patients. Infect Control Hosp Epidemiol 2018; 39: 1367–74.30231948 ArmstrongLR, KammererJS, HaddadMB. Diabetes mellitus among adults with tuberculosis in the USA, 2010–2017. BMJ Open Diabetes Res Care 2020; 8: e001275.PascopellaL, BarryPM, FloodJ, DeRiemerK. Death with tuberculosis in California, 1994–2008. Open Forum Infect Dis 2014; 1: ofu090.25734158 NahidP, DormanSE, AlipanahN, Official American Thoracic Society/Centers for Disease Control and Prevention/Infectious Diseases Society of America Clinical Practice guidelines: treatment of drug-susceptible tuberculosis. Clin Infect Dis 2016; 63: e147–95.27516382 HarriesAD, KumarAM, SatyanarayanaS, Communicable and non-communicable diseases: connections, synergies and benefits of integrating care. Public Health Action 2015; 5: 156–57.26393110 BamrahS, Yelk WoodruffRS, PowellK, GhoshS, KammererJS, HaddadMB. Tuberculosis among the homeless, United States, 1994–2010. Int J Tuberc Lung Dis 2013; 17: 1414–19.24125444 MarksSM, SelfJL, VenkatappaT, Diagnosis, treatment, and prevention of tuberculosis among people experiencing homelessness in the United States: current recommendations. Public Health Rep 2023; 138: 896–907.36703605 Centers for Disease Control and Prevention. Prevention and control of tuberculosis among homeless persons. Recommendations of the Advisory Council for the Elimination of Tuberculosis. MMWR Recomm Rep 1992; 41: 13–23.1314323 WorrellMC, KramerM, YaminA, RaySM, GoswamiND. Use of activity space in a tuberculosis outbreak: bringing homeless persons into spatial analyses. Open Forum Infect Dis 2017; 4: ofw280.28480272 KriegerN. Theories for social epidemiology in the 21st century: an ecosocial perspective. Int J Epidemiol 2001; 30: 668–77.11511581 BakhshY, ReadheadA, FloodJ, BarryP. Association of area-based socioeconomic measures with tuberculosis incidence in California. J Immigr Minor Health 2023; 25: 643–52.36445646 Larrabee SonderlundA, CharifsonM, SchoenthalerA, CarsonT, WilliamsNJ. Racialized economic segregation and health outcomes: a systematic review of studies that use the Index of Concentration at the Extremes for race, income, and their interaction. PLoS One 2022; 17: e0262962.35089963 National Academies of Sciences, Engineering, and Medicine; Health and Medicine Division; Board on Health Care Services; Committee on Integrating Social Needs Care into the Delivery of Health Care to Improve the Nation’s Health. Integrating social care into the delivery of health care: moving upstream to improve the nation’s health. Washington, DC: National Academies Press, 2019.HackerK, HouryD. Social needs and social determinants: the role of the Centers for Disease Control and Prevention and public health. Public Health Rep 2022; 137: 1049–52.36367214

Association between race and ethnicity and tuberculosis outcomes among US-born individuals with tuberculosis, 2011–19

Diagnosis after deathTreatment discontinuationDeath during treatmentIncomplete treatment*
n/N (%)aRR (95% CI)†n/N (%)aRR (95% CI)†n/N (%)aRR (95% CI)†n/N (%)aRR (95% CI)†
White (n=8815)342/8810 (3·9%)1 (ref)188/8357 (2·2%)1 (ref)891/8357 (10·7%)1 (ref)1421/8699 (16·3%)1 (ref)
Black (n=10 362)366/10 356 (3·5%)1·24 (1·07–1·44)278/9923 (2·8%)1·28 (1·06–1·54)958/9923 (9·7%)1·30 (1·20–1·41)1602/10 289 (15·6%)1·27 (1·19–1·35)
Hispanic (n=5766)145/5765 (2·5%)1·46 (1·19–1·78)151/5572 (2·7%)1·04 (0·83–1·31)293/5572 (5·3%)1·21 (1·06–1·37)589/5717 (10·3%)1·22 (1·11–1·33)
Asian (n=1182)11/1182 (0·9%)0·78 (0·40–1·35)10/1164 (0·9%)0·43 (0·21–0·77)44/1164 (3·8%)1·22 (0·91–1·59)65/1175 (5·5%)0·93 (0·73–1·16)
American Indian or Alaskan Native (n=1026)30/1025 (2·9%)1·04 (0·69–1·50)22/990 (2·2%)0·92 (0·57–1·41)97/990 (9·8%)1·34 (1·09–1·63)149/1020 (14·6%)1·19 (1·01–1·39)
Native Hawaiian or Other Pacific Islander (n=267)4/267 (1·5%)1·19 (0·37–2·74)3/256 (1·2%)0·56 (0·14–1·46)12/256 (4·7%)1·41 (0·87–2·25)19/260 (7·3%)1·14 (0·73–1·67)
Multiracial (n=187)4/187 (2·1%)0·97 (0·30–2·20)5/182 (2·7%)1·19 (0·43–2·55)14/182 (7·7%)1·31 (0·76–2·01)23/186 (12·4%)1·20 (0·80–1·67)

Denominators for each outcome do not correspond to the totals for each racial and ethnic group because some individuals had missing data or were excluded (as specified in the Methods section). aRR=adjusted relative risk.

Summary outcome combining diagnosis after death, treatment discontinuation, and death during treatment.

Adjusted for sex, year, age group, and regional division.

Association between race and ethnicity and potential mediators among US-born individuals with tuberculosis, 2011–19

Percentage in poverty*Percentage unemployed*Percentage without insurance*Percentage least privileged (ICERace–Income)†HIVDiabetesEnd-stage renal diseaseHomelessness
White1 (ref)1 (ref)1 (ref)1 (ref)1 (ref)1 (ref)1 (ref)1 (ref)
Black3·35 (3·14–3·57)2·83 (2·72–3·04)1·85 (1·72–1·99)6·39 (5·86–6·98)1·19 (1·12–1·26)1·39 (1·30–1·50)2·49 (2·05–3·03)1·32 (1·23–1·43)
Hispanic2·97 (2·76–3·20)2·07 (1·93–2·23)2·77 (2·58–2·98)1·86 (1·63–2·11)1·06 (0·99–1·14)2·49 (2·29–2·70)2·72 (2·13–3·48)0·85 (0·76–0·96)
Native American or Alaskan Native4·58 (4·18–5·01)2·60 (2·37–2·85)2·75 (2·49–3·04)1·14 (0·77–1·61)0·87 (0·74–1·02)1·59 (1·36–1·84)2·29 (1·51–3·38)0·93 (0·78–1·09)

Values are adjusted RR (95% CI); RR was adjusted for year, age group, and regional division. Definitions of the potential mediators are provided in the Methods section. ICERace–Income=Index of Concentration at the ExtremesRace–Income. RR=relative risk.

Highest quintile.

Lowest quintile.

Mediation analysis for Black and Hispanic individuals compared with White individuals among US-born individuals with tuberculosis, 2011–19

Diagnosis after deathTreatment discontinuationDeath during treatmentIncomplete treatment*
Black individualsHispanic individualsBlack individualsHispanic individualsBlack individualsHispanic individualsBlack individualsHispanic individuals
Total effect, aRR (95% CI)†1·25 (1·07–1·43)1·40 (1·16–1·63)1·26 (1·04–1·47)1·15 (0·85–1·43)1·33 (1·22–1·44)1·22 (1·05–1·36)1·32 (1·22–1·42)1·24 (1·11–1·37)
Effect through unmeasured mediators, aRR (95% CI)†1·08 (0·89–1·26)1·25 (0·99 1·52)1·13 (0·90–1·36)1·24 (0·88–1·59)1·08 (0·96–1·21)0·96 (0·80–1·11)1·10 (1·01–1·20)1·08 (0·95–1·21)
Effect through measured mediators, aRR (95% CI)†1·17 (1·06–1·27)1·11 (0·97–1·26)1·11 (1·00–1·23)0·92 (0·86–1·43)1·23 (1·16–1·30)1·26 (1·15–1·37)1·20 (1·15–1·24)1·15 (1·07–1·22)
Overall percent mediated, % (p value)70% (p=0·0017)37% (p=0·13)50% (p=0·14)NA74% (p=0·0002)122% (p=0·0002)67% (p<0·0001)65% (p<0·0001)
Marginal percent mediated by individual mediators, %
 Poverty7%6%11%NA12%9%12%9%
 Unemployment−16%3%16%NA11%0%7%6%
 No health insurance1%−2%−5%NA1%14%0%0%
 Racialised economic segregation (ICERace–Income)15%2%−9%NA6%28%4%9%
 HIV38%14%26%NA17%18%20%17%
 End-stage renal disease28%18%−3%NA17%27%15%18%
 Diabetes3%−5%−4%NA7%24%5%3%
 Homelessness−5%0%18%NA0%−2%2%−1%

Percent mediated >100% means there are unmeasured mediators that have a protective effect on the outcome. aRR=adjusted relative risk. NA=not applicable. ICERace–Income=Index of Concentration at the ExtremesRace–Income.

Summary outcome combining diagnosis after death, treatment discontinuation, and death during treatment.

Adjusted for sex, year, categorical age, and regional division.

Research in contextEvidence before this study

We searched PubMed from inception up to Jan 29, 2024, to identify previous studies on the topic of racial and ethnic disparities in tuberculosis outcomes, using the keywords “United States” AND “tuberculosis” AND (“race” OR “ethnicity” OR “disparities”) AND (“mortality” OR “diagnosis” OR “treatment”) AND (“risk factors” OR “HIV” OR “structural racism” OR “poverty” OR “diabetes” OR “renal disease”). No restrictions were applied to the language or type of publication. This search returned 213 results, which we reviewed for relevance. We identified nine studies that evaluated risk factors for adverse outcomes along the tuberculosis care continuum in the USA. Several studies restricted their analysis to a single subgroup (such as American Indian and Alaska Native individuals or incarcerated individuals) or a single risk factor (such as HIV or diabetes). Most studies adjusted for race and ethnicity alongside other risk factors in models looking at associations with tuberculosis outcomes. Some studies looked at the interaction between race and ethnicity and other risk factors. HIV was associated with a higher risk of tuberculosis mortality and failure to complete treatment. Diabetes was associated with a higher risk of hospitalisation, particularly for Hispanic individuals.

Added value of this study

This study conceptualises race and ethnicity as a social construct representing historical and contemporary discriminatory processes that lead to disparities in exposure to tuberculosis risk factors (comorbidities, poverty, unemployment, lack of health insurance, and racialised economic segregation). We therefore used a sequential multiple mediator approach to examine the marginal contribution of various risk factors to racial and ethnic disparities along the care continuum. Our findings suggest that socioeconomic disadvantage, economic and racial polarisation, and the presence of comorbidities (HIV, diabetes, and renal disease) are the biggest contributors to disparities in tuberculosis treatment completion for Black and Hispanic individuals, relative to White individuals.

Implications of all the available evidence

Public health actions are needed to mitigate structural and environmental factors that lead to racial and ethnic disparities in the prevalence of tuberculosis comorbidities and their case management. Further research is needed to clarify the ways in which economic and racial segregation contribute to racial and ethnic disparities in tuberculosis outcomes.