We estimated four study populations for this analysis: (1) non-US–born Asian persons, (2) US-born Asian persons, (3) non-US–born Hispanic persons, and (4) US-born Hispanic persons, all with diagnosed diabetes. We estimated US resident Asian and Hispanic persons aged ≥15 years with diagnosed diabetes, stratified by race, ethnicity, age, and whether the person is US-born or non-US–born. We first calculated population estimates for each age-, nativity-, ethnicity-, and race-stratum using data from the American Community Survey (ACS) 2022 5-year public-use microdata files ⁹. We categorized persons as non-Hispanic Asian or Hispanic (of any racial identity), following US Census Bureau conventions ¹⁰; referred to as Asian or Hispanic persons, respectively. We then estimated the prevalence of diagnosed diabetes each age-, nativity-, ethnicity-, and race-stratum using data from the 2020–2022 National Health Interview Survey ¹¹. Persons were identified as having diagnosed diabetes if they reported having ever been told by a healthcare professional that they have diabetes. Those reporting gestational diabetes without reporting diabetes outside of pregnancy were not included in the study populations.

To estimate Mtb infection prevalence for persons with diagnosed diabetes in each age-, nativity-, ethnicity-, and race-stratum, we used data on interferon gamma release assay (IGRA) positivity from the 2011–2012 National Health and Nutrition Examination Survey (NHANES) ¹², which represent the most recent nationally-representative Mtb infection prevalence data. Restricting the estimation to include IGRA-positive persons with only diagnosed diabetes resulted in small cell sizes. To increase precision of Mtb infection prevalence estimates, we included IGRA-positive persons with either diagnosed and undiagnosed diabetes, consistent with determination guidance from the American Diabetes Association ¹³. We estimated Mtb infection prevalence by adjusting the prevalence of IGRA positive test results (from NHANES) using published estimates of IGRA sensitivity and specificity ¹⁴, following a published approach ⁵. Details on survey variables used are provided in the Supplement.

We estimated the outcomes of a one-time Mtb infection testing and treatment intervention offered in 2024 to each of the four study populations (non-US–born Asian persons, US-born Asian persons, non-US–born Hispanic persons, and US-born Hispanic persons) with diagnosed diabetes. We assumed that Mtb infection testing would be conducted with IGRA, and TB disease would be excluded from treatment for Mtb infection with a two-view chest x-ray. IGRA test sensitivity was assumed to be 78% (95% uncertainty interval: 65%–91%) for US-born persons and 79% (70%–90%) for non-US–born persons ¹⁴. IGRA test specificity was assumed to be 98% (96%–99%) for US-born persons and 99% (96%–100%) for non-US–born persons ¹⁴.

Individuals receiving a positive IGRA test and negative chest x-ray result would be offered treatment to prevent TB disease. We assumed treated persons would be prescribed one of three self-administered regimens: 25% would receive a regimen of weekly isoniazid (900 mg) and rifapentine (900 mg) for 3 months (3HP), 7% would receive daily isoniazid (300 mg) and rifampin (600 mg) for 3 months (3HR), and 68% would receive daily rifampin (600 mg) for 4 months (4R), with these percentages based on recent prescribing patterns ¹⁵. We assumed isoniazid-containing regimens (3HP and 3HR) had a 0.3% risk of hepatoxicity ¹⁶. The treatment acceptance rate was assumed to be 77% (70%–85%) ¹⁷ and treatment completion rate was 76% (57%–95%) ¹⁵. The probability of cure given treatment completion was assumed to be 93% (90%–96%) ¹⁸. We assumed persons with Mtb infection and diabetes had a rate ratio for TB progression of 1.60 (1.47–1.73) compared to persons without diabetes ⁵. Similarly, we assumed that all persons with diabetes in the study populations experienced a rate ratio of 1.93 (1.84–2.03) for all-cause mortality, as compared to the US general population ¹⁹. We assumed that persons developing TB disease in the future would be identified and initiated in TB disease treatment based on current performance of TB case detection and treatment outcomes ¹.

This intervention scenario was compared to a ‘no intervention’ scenario that assumed any future Mtb infection- and TB disease-related health services would eventually be provided with the same coverage and effectiveness as in the intervention scenario, but without the one-time Mtb infection testing and treatment intervention offered to persons in the study populations. Persons who were not offered the intervention or did not accept treatment in 2024 were eligible to receive treatment in future years, through routine TB health services. These persons had an age-stratified probability of progression to TB disease and/or TB death prior to their future diagnosis and treatment.

We estimated intervention outcomes using a published transmission-dynamic model of TB in the United States ²⁰. This deterministic, compartmental model simulates TB epidemiology and the long-term consequences of Mtb infection treatment interventions and was calibrated to reproduce trends in the US population and TB epidemiology through 2023. This model is available on GitHub as a package for the R programming language (MITUS) ²¹. The model stratifies the US population by age, US-born status, Mtb infection status, and a set of generalized risk strata representing factors associated with elevated TB progression rates and mortality. We used these strata to represent the Mtb infection progression risk and mortality associated with diabetes. Figure S1 presents a schematic state diagram of the model, which includes model compartments and transitions. Further details on model assumptions and calibration can be found in the Supplement.

We constructed cohorts within the model representing persons in the estimated study populations: Asian and Hispanic persons aged ≥15 years with diagnosed diabetes stratified by age and nativity. Each population was parameterized with their unique age and nativity distribution and their respective Mtb infection prevalence, rates of progression to TB disease, and background mortality rates. Model parameters describing the Mtb infection testing and treatment cascade (test sensitivity and specificity, distribution of treatment regimens, treatment effectiveness, etc.) did not differ between baseline and intervention scenarios.

We used the mathematical model to estimate TB cases averted, TB deaths averted, life years gained, and quality-adjusted life years gained (QALYs) for the intervention scenario compared to no intervention. TB cases included incident TB among the study populations as well as secondary cases among contacts, as simulated by the transmission model. TB deaths included all deaths occurring among persons with TB disease (including untreated persons and those receiving treatment). Life years were calculated as the number of TB deaths in each age group multiplied by life expectancy for that age group. Age-group-specific life expectancies are given in the Table S1 ²². QALYs were calculated to include years lost due to TB mortality and the reductions in quality of life associated with Mtb infection treatment, untreated TB disease, and TB treatment ²³. The disutility weight associated with potential Mtb infection treatment hepatotoxicity associated with isoniazid was 0.25 (0.19–0.31) ²⁴, applied for 14 days ²⁵.

Untreated TB disease and TB treatment during the first month had a disutility weight of 0.25 (0.34–0.17); this value was based on standard gamble utility estimates for persons receiving TB treatment compared to control, at treatment initiation ²⁶. After one month of treatment, disutility weights decreased to 0.11 (0.21–0.02) through completion; this value was based on standard gamble utility estimates for persons receiving TB treatment compared to control, averaged over months 2–6 after diagnosis ²⁶. All parameters used in the QALY calculation, including additional parameters, such as duration of TB disease, treatment, and adverse events, can be found in Table S2. Calculations for each health outcome are described in the Supplement. To capture intervention consequences over the lifetime of the study cohort, all health outcomes were evaluated between the years 2024–2100.

We estimated costs under three perspectives: a TB health services perspective, a health services perspective, and a societal perspective. The TB health services perspective included costs associated with the diagnosis and treatment of Mtb infection and TB disease. The health services perspective included TB health services costs as well as future non-TB health services costs. The societal perspective included all health services costs as well as the productivity losses associated with TB healthcare visits, morbidity, and mortality, adjusted for non-healthcare spending ²⁷. We present results calculated using the TB health service perspective as the main analysis and report results for the additional perspectives in the Supplement.

Per-patient Mtb infection treatment costs were based on published estimates and varied by regimen ²⁸. Per-person costs of treatment with 3HP totaled $557, treatment with 3HR totaled $1,054, and treatment with 4R totaled $1,276. Details of the cost components for these regimens can be found in Table S3. The per-patient cost of treating TB disease was assumed to be $25,285, based on cost analyses conducted by the Centers for Disease Control and Prevention ^29,30. For all treatment costs, we assumed an uncertainty range of ± 25%. Annual total healthcare expenditures by age and diabetes status were based on analyses of Medical Expenditure Panel Survey data ³¹. Annual total non-healthcare spending was calculated from the US Bureau of Labor Statistics’ Consumer Expenditure Surveys ³². Productivity losses were calculated using age-group-stratified US life-expectancies (Table S1),²² the additional mortality risk associated with diabetes,³³ estimated productivity lost to diagnosis and treatment²⁵, and productivity estimates (annual and lifetime) based on an analysis of US market and non-market productivity (Table S4, S5) ³⁴. All costs are reported in 2022 US dollars. This study is reported as per the Consolidated Health Economic Evaluation Reporting Standards 2022 (CHEERS 2022) ³⁵. The CHEERS checklist (Table S9) and detailed cost inputs are provided in the Supplement.

We estimated incremental cost-effectiveness ratios (ICERs)—calculated as the incremental difference in mean costs divided by the incremental difference in mean QALYs—to evaluate the cost-effectiveness of the Mtb infection testing and treatment intervention for the four study population strata, as compared to the no intervention scenario. Future costs and health benefits were discounted at 3% per year ³⁶.

We used Monte Carlo simulation to propagate parameter uncertainty throughout the analysis (probabilistic sensitivity analysis).³⁷ To account for uncertainty in calibrated parameters, we simulated results for 1,000 parameter sets generated by the Bayesian approach used to calibrate the mathematical model ³⁸. For parameters not included in the calibration (e.g., cost inputs, utility weights), we sampled from probability distributions representing the uncertainty in each of these inputs. For parameters that had a published uncertainty range, we use these values to parameterize the distribution; otherwise, we assume an uncertainty range of ± 25%. We calculated point estimates for each outcome as the mean of these values and 95% uncertainty intervals as the 2.5^th and 97.5^th percentiles.

We conducted one-way sensitivity analyses to understand the robustness of the results to variation in individual parameters. These parameters and their ranges are presented in Table S6. Analyses were conducted in R (version 4.2.1) using the survey (version 4.0), MASS, (version 7.3.60), and Rcpp (version 1.0.10) packages ^39–42. All codes used in this analysis are publically available at https://github.com/PPML/tttDiabetes.

We estimated that there were 292,900 (95% uncertainty interval: 144,900–644,200) US-born Asian, 1,296,600 (1,006,300–1,682,400) non-US–born Asian, 2,204,500 (1,759,900–2,762,900) US-born Hispanic, and 2,164,100 (1,766,000–2,664,900) non-US–born Hispanic persons aged ≥15 years with diagnosed diabetes in the United States in 2024 (Table 1). Figure S2 shows age-stratified estimates of diagnosed diabetes prevalence in each population. Figure S3 shows the age distribution of each population. The estimated percentage of the population with diagnosed diabetes that was aged 55 years or older ranged from 54% among non-US–born Asian persons to 71% among US-born Asian persons.

The non-US-born Asian population with diagnosed diabetes had the highest Mtb infection prevalence at 31.6% (11.9% –56.7%) and estimated prevalence was lowest among the US-born Hispanic persons at 4.7% (0.5%–12.0%) (Table 1). Figure S4 shows the estimated Mtb infection prevalence for all populations, stratified by age group, origin of birth, and diabetes status. Estimated Mtb infection prevalence was higher among non-US–born persons compared with US-born persons, across all age groups and diabetes statuses.

Table 2 shows the health outcomes of the Mtb infection testing and treatment intervention in each population over the period 2024–2100. Per 100,000 persons tested, the number of TB cases averted ranged from an estimated 239 (225–254) among non-US–born Asian persons to 8 (7–8) among US-born Hispanic persons. The number of TB deaths averted per 100,000 persons tested ranged from an estimated 53.7 (51.4–56.1) among non-US–born Asian persons to 1.3 (1.2–1.4) among US-born Hispanic persons. QALYs gained per 100,000 persons tested followed the same pattern and were highest among non-US–born Asian persons (584 [546–628]) and lowest among US-born Hispanic persons (17 [15–19]).

Table 3 presents the incremental cost of the intervention in each population from the TB health service perspective, per 100,000 persons tested for Mtb infection. These incremental costs of the intervention were higher among non-US–born persons—$22.88 ($13.97–$31.80) million among non-US–born Asian persons and $20.31 ($12.47–$28.31) million among non-US–born Hispanic persons— reflecting the higher Mtb infection prevalence among these populations compared with US-born populations. Incremental cost estimates from the health services and societal perspective are shown in Table S7, and Figure S5 compares incremental costs across all perspectives. Discounted incremental TB health services, all health services, and societal costs associated with the intervention are presented in Table S8.

The incremental cost per QALY gained varied across the four study populations, with interventions among the two non-US–born populations having greater QALYs gained and lower incremental costs than those among US-born populations (Tables 2 and 3). Figure 1 depicts the calculated ICERs for each population. The ICERs calculated for the non-US–born Asian population and the Hispanic population with diagnosed diabetes were $66,664 ($41,456–$93,625) and $68,749 ($43,136–$97,044) per additional QALY gained, respectively. ICERs for US-born Asian and Hispanic populations with diagnosed diabetes were substantially higher: $856,671 ($533,506–$1,234,032) and $1,081,646 ($673,142–$1,551,264) per additional QALY gained, respectively. Figure S6 presents incremental costs and health benefits on the cost-effectiveness plane, estimated from the societal perspective. These ICERs ranged from $76,035 ($49,265–$107,005) among the non-US–born Hispanic population to $1,102,605 ($695,169–$1,576,240) among the US–born Hispanic population.

Figure 2 presents the results of one-way sensitivity analyses for key parameters, stratified by each intervention population with diagnosed diabetes. Across all populations, the ICER was most sensitive to uncertainty in Mtb infection prevalence, with higher Mtb infection prevalence estimates associated with lower ICER values (i.e., more cost-effective). The percentage change in ICERs ranged from a 13% decrease to a 49% increase as Mtb infection prevalence among non-US–born varied by +/−25%. ICER percentage change was higher when Mtb infection prevalence was varied +/−25% among US-born persons, ranging from a 50% reduction to a 746% increase.