A decision-analytic model, including a short-term decision tree model followed by a 20-year Markov model, was constructed to simulate the long-term TB-related outcomes of a hypothetical cohort of adult HIV-infected individuals undergoing TBI screening from the perspective of healthcare provider in the US (Fig. 1a,b).
Decision-analytic modelling is a well-accepted tool to provide a computational framework for evaluation of the cost and clinical outcomes of health interventions over time, using evidence-based probabilities of clinical events, utility and cost inputs from multiple sources21,22. Model-based analysis examines the result robustness over the model input uncertainties using sensitivity analysis, providing information on thresholds of influential parameters on the cost-effectiveness of the health intervention23. The cost-effectiveness findings generated by decision-analytic modelling facilitate policy and decision makers on informed decision of resources allocation to generate optimal benefit on population health24.
All hypothetical individuals (both US-born and foreign-born) first entered a decision tree model to evaluate the immediate outcomes of TBI testing and subsequent treatment (Fig. 1a). A total of twelve testing strategies were examined: No testing strategy, three single-test strategies (TST; T-SPOT.TB; QFT-Plus), four sequential strategies to confirm negative tests (TST-negative followed by T-SPOT.TB; TST-negative followed by QFT-Plus; T-SPOT.TB-negative followed by TST and QFT-Plus-negative followed by TST) and four sequential strategies to confirm positive tests (TST-positive followed by T-SPOT.TB, TST-positive followed by QFT-Plus, T-SPOT.TB-positive followed by TST and QFT-Plus-positive followed by TST).
In all testing strategies, a hypothetical individual might or might not have TBI, and this individual might be either not tested, or tested positive or negative for TBI. For no testing strategy, no TBI testing was carried out and no TBI treatment was provided. For the three single-test strategies, a test-positive result indicated for treatment of TBI and a test-negative result required no sequential testing or treatment. In the four confirm-negative sequential strategies, an initial positive testing indicated for treatment. An initial negative testing required a sequential testing (as specified in the strategy), and treatment was provided to those with a positive result. In the four confirm-positive sequential strategies, a sequential testing was conducted for those with initial positive testing, and treatment was indicated for those who were tested positive in both sequential tests. After the testings, the individual (with or without TBI) might or might receive treatment of TBI (3-month once-weekly rifapentine 900 mg plus isoniazid 900 mg (3HP)6,14) and further entered the Markov model (at the health state of “TBI” or “no TBI”) for a maximum of 20 years with yearly cycle.
The Markov model resembled the disease progression of TBI, including five health states: “no TBI”, “TBI”, “TB disease”, “TB recovered”, and “death” (Fig. 1b). The hypothetical individuals proceeded through health states in each model cycle according to transition probabilities. All individuals might die from all causes in every yearly cycle. Of those who survive in each yearly cycle, the individuals with no TBI might acquire TBI. Patients in the “TBI” state might develop TB disease. TB disease patients were managed according to the CDC guidelines for adults with HIV14, and the patients might or might not achieve treatment success. Patients proceeded to the “TB recovered” state if the treatment success was achieved, and the recovered patients might experience TB disease recurrence in the following yearly cycles. Those with TB disease who failed treatment would die or receive palliative care until death.
All model inputs are shown in Table 1. A search for clinical probabilities was performed in the Medline covering the period 2000–2022 and public data of the CDC and World Health Organization (WHO) to estimate input parameters for the model using keywords “Mycobacterium infection”; “latent tuberculosis infection”; “active tuberculosis”; “tuberculosis skin test”; “interferon-gamma release assay”; “T-SPOT.TB”; “QFT-Plus”; and “human immunodeficiency virus infection”. The inclusion criteria for clinical studies were: (1) Reports written in English; (2) adult HIV-infected patients and (3) TBI test results and/or treatment outcomes were reported. Systematic review and meta-analysis were preferred sources for the model inputs. A study was included if the data relevant to the model inputs were available. If multiple sources were obtained for model input, the weighted average was used as the base-case value, and the high or low values formed as the range for sensitivity analysis.
The proportion of foreign-born (18.9%) and US-born (81.1%) individuals among HIV-infected patients were approximated from the findings of a cross-sectional study on epidemiology data from the National HIV Surveillance System for people with HIV infection (n = 328,317 patients) during 2010–201725. A prospective clinical study (n = 10,740) on TBI diagnostics reported the TBI prevalence among subgroups based on age, foreign birth outside the USA, and HIV infection. The prevalence of TBI was reported to be 31.7% in foreign-born and 4.20% in US-born HIV-infected patients10. The study also reported the sensitivities and specificities of TBI testing strategies both in foreign-born and US-born HIV-infected patients, and were adopted for the model inputs of sensitivities and specificities of TST, T-SPOT.TB and QFT-Plus10.
The treatment of TBI (3HP once-weekly) completion rate was approximated from the results of the directly observed therapy for TBI in an open-label, phase 4 randomized clinical trial (n = 1002 patients) in the outpatient tuberculosis clinics in the US, Spain, Hong Kong, and South Africa. Of those patients enrolled in the US, the treatment completion rate was 85.40%26. A prior US health economic analysis of TBI screening for at-risk populations had adopted a 90% TB risk reduction from the 9-month isoniazid preventive treatment27, and a randomized noninferiority trial (n = 7731) found 3HP therapy to be as effective as the 9-months isoniazid therapy in subjects at high-risk for TB28. The present model, therefore, adopted 90% as the TB risk reduction from the 3HP treatment for TBI. The outcomes of TBI patients who initiated but failed to complete the TBI treatment were assumed to be the same as those without treatment.
The yearly probability of progression from TBI to TB disease in HIV-infected patients (1.82%) was approximated from the results of an epidemiology study of reactivation TB in the US, using data on TB cases (n = 39,920) reported to the CDC during 2006–200829. The WHO reported the treatment success rate of TB disease among HIV-infected patients to be 70.00% in the US, and the mortality rate among those who failed anti-TB therapy was estimated to be 59.65%30. The yearly TB recurrence probability among HIV-infected patients (4.09%) was estimated from the findings of two prospective clinical trials conducted by the Tuberculosis Trials Consortium in the US and Canada (n = 1244 culture-positive TB patients) on the outcomes of anti-TB treatment31. Age-specific all-cause mortality rates were retrieved from the US Life table reported by WHO32.
Health utility inputs
The effectiveness of testing strategies was evaluated in terms of the quality-adjusted life-year (QALY), estimated using the utility and duration of time spent in each of the health states33. The health utility weights of TBI were not significantly different from those of the healthy general population34,35. Hence, the utility model input of TBI adopted the age-specific health utility (0.92 for age 18–65 years), previously generated by the -related quality of life study using the US national health measures and surveys36. The base-case age of the HIV-infected individuals for foreign-born (46.4 years) and US-born (49.6 years) were retrieved from the prospective clinical study evaluating TBI diagnostics in the US10. The utility value of TB disease (0.69) was approximated from a cross-sectional survey on the impact of TB on health utility37. The utility of TB disease treatment success (0.88) was retrieved from the findings of health-related quality of life studies in TB patients37,38. The QALYs accumulated over the model time horizon were discounted to the year 2022, with an annual discounting rate of 3%.
The cost analysis was conducted from the perspective of the US healthcare provider. Direct medical costs included in the model were costs of diagnostic tests, treatment of TBI, TB disease treatment, palliative care, and TB-related mortality. The cost of TST was approximated from the Medicare-allowable physician fee schedule39. Cost of T-SPOT.TB and QFT-Plus were adopted from Centres for Medicare and Medicaid Services clinical laboratory fee schedule40. The direct treatment cost per case of TBI was obtained from published reports by CDC41. Similarly, according to a CDC report, the per-patient cost of treating TB was USD21,955, and the cost of TB-related mortality was USD37,49942. The direct cost of palliative care per admission was retrieved from a meta-analysis study of the health economics of palliative care for hospitalized adults with serious illness43. All costs in the present analysis were adjusted to the year 2022, according to the Consumer Price Index in the US44. The future incurred costs were discounted to the year 2022 by 3% annually.
Cost-effectiveness, sensitivity and scenario analyses
All analyses were performed using TreeAge Pro 2022 (TreeAge Software Inc, Williamstown, MA, USA) and Excel 365 (Microsoft Corporation, Redmond, WA, USA). The primary outcomes of the model were direct medical cost, QALY, and incremental cost per QALY gained (ICER). The ICER of a strategy was calculated if the strategy gained additional QALYs at a higher cost than the next less costly strategy: ICER = ∆Cost/∆QALYs.
A strategy was dominated when it gained either (1) lower QALYs at a higher cost or (2) lower QALYs with higher ICER than another strategy. The dominated strategies were removed from further cost-effectiveness analysis. A strategy was accepted as the preferred cost-effective treatment if it gained (1) higher QALYs at a lower cost, or (2) higher QALYs at a higher cost, and the ICER was less than the willingness-to-pay (WTP) threshold22,45.
All model inputs were examined by the one-way sensitivity analysis over the ranges specified in Table 1. The one-way sensitivity analysis was used to describe the association between each input variable and the primary outcomes (cost and QALY) for examination of the robustness of base-case findings. The cost and QALY were further recalculated 10,000 times in the probabilistic sensitivity analysis using Monte Carlo simulations, by randomly drawing each model input value from the parameter-specific distribution, to evaluate the impact of uncertainty in all variables simultaneously. The details of distributions used in the probabilistic sensitivity analysis are provided in Supplementary Table S1. The mean incremental cost and QALY gained along with the 2.5th and 97.5th percentiles were computed to estimate the uncertainty interval of the simulation. The probabilities of each strategy to be accepted as cost-effective were assessed in the acceptability curves over a broad range of WTP threshold from 0 to 200,000 USD/QALY46,47,48.
In the present study, all individuals were tested for TBI once at the entry of the model. Individuals who are at high risk for ongoing or repeated exposure to TB disease, such as incarceration, traveling in a high-TB incidence area, homelessness, and living in a congregate setting, are recommended to test for TBI yearly5,14. A scenario analysis on yearly TBI screening with the same 12 testing strategies was conducted for HIV-infected individuals who were at high risk for exposure to TB disease.