Transmission dynamics and control of tuberculosis in high-altitude regions: a modelling study in Xizang, China

Abstract

Objectives

To estimate tuberculosis (TB) incidence trends in the high-altitude Xizang, China, and to explore the key intervention strategies on achieving the WHO 2030 TB control target.

Design

We developed a susceptible–exposed–infectious–recovered transmission model using routinely reported TB surveillance data from 2004 to 2022. Scenario-based simulations were conducted to project future TB incidence under alternative intervention strategies. Model assumptions are as follows: (1) a stable population, (2) lifelong vaccine-induced immunity, (3) infectiousness of active TB cases, (4) relapse risk after recovery and (5) homogeneous mixing within the population.

Setting

Seven prefectures of Xizang Autonomous Region on the Tibetan Plateau, China.

Participants

An estimated population of approximately 3 million individuals residing in Xizang.

Interventions

We assessed the epidemiological impact of four interventions implemented independently: increasing vaccine efficacy rate, reducing transmission rates of susceptible individuals, decreasing progression rate from latent TB infection to active disease and reducing relapse rate among successfully treated patients, compared with continuation of current control measures.

Results

The estimated basic reproduction number (R₀) for TB in Xizang was 0.39 (95% CI 0.21 to 0.71) in the absence of additional interventions, which was the highest among all regions of China. Model simulations indicated that all four evaluated interventions were each likely to reduce TB incidence, but only reducing the latent-to-active TB progression had a substantial effect. A 50% reduction in the progression rate was predicted to lower TB incidence from 66.56 (62.00–70.11) to 40.54 (37.15–43.77) cases per 100 000 population, meeting the WHO 2030 TB control target.

Conclusion

Targeted management of individuals with latent TB infection should be strengthened to substantially reduce TB transmission in high-altitude areas.

STRENGTHS AND LIMITATIONS OF THIS STUDY.

• A transmission model was developed to characterise and project the high-burden tuberculosis (TB) epidemic in high-altitude regions of Xizang, China, incorporating vaccination, drug resistance and recurrence.

• This study used long-term TB surveillance data (2004–2022) from an authoritative national information system to ensure robust model calibration and validation.

• Key transmission parameters were estimated using Markov chain Monte Carlo algorithms, allowing uncertainty to be incorporated directly into the model-fitting process.

• Although the model simulated regional stratification, it assumed homogeneous population mixing and stable demographic dynamics, potentially under-representing mobility patterns and regional heterogeneity across Xizang.

• The model parameters did not sufficiently account for COVID-19-related public health interventions, which may have altered TB transmission dynamics during the validation period.

Introduction

Tuberculosis (TB), a chronic infectious disease caused by Mycobacterium tuberculosis, remains a major global public health threat.¹ For decades, it has been the leading infectious cause of mortality in adults, with an estimated 1.25 million deaths in 2023—a toll nearly twice that of HIV/AIDS.² The challenge is significantly compounded by the rise of drug-resistant TB, which is projected to account for a quarter of all TB-related deaths by 2050.³ To address this challenge, the WHO introduced the End TB Strategy, which aims for elimination (<1 case per 100 000) by 2050.⁴ China, one of the 30 high-burden countries, has aligned with these goals and aims to reduce TB incidence to <43 per 100 000 by 2030.⁵ However, a critical gap persists between ambition and reality. By 2023, global TB incidence had fallen by only 8.3%, far below the WHO 2025 milestone of a 50% reduction, indicating that current progress is insufficient to meet global and national targets.⁶

This challenge is particularly acute in high-altitude regions such as Xizang, China. Unique geographic conditions, population mobility and lifestyle factors have contributed to localised M. tuberculosis clusters.⁷ From 2005 to 2016, the average annual reported incidence in Xizang was 101.98 per 100,000—almost twice the national average and surpassed only by Xinjiang and Guizhou in China.^{1 8} High levels of drug resistance and suboptimal Bacille Calmette-Guérin vaccination coverage further exacerbate the burden. Among Tibetan populations, non-vaccination rates remain substantially higher than those reported in Han communities.⁹

Mathematical models are a crucial tool for understanding TB transmission and for informing resource allocation and policy design.¹⁰,¹² Previous studies in South Korea and the USA have evaluated control strategies using dynamic transmission models, such as the susceptible–exposed–infectious–recovered (SEIR) model with time-dependent parameters.^{13 14} However, existing models rarely account for the unique epidemiology of TB in high-altitude regions such as Xizang. Most rely on generalised assumptions regarding contact patterns, mobility and latent infection dynamics, which may not reflect the realities of nomadic and semi-nomadic populations.¹⁵,¹⁸ Moreover, no previous study has systematically evaluated multiple intervention strategies—including vaccination, transmission reduction, latent infection management and treatment optimisation—within a unified modelling framework tailored to Xizang.

In this study, we developed a customised transmission dynamics model based on the epidemic characteristics of the Xizang region. We systematically simulated the potential effects of four interventions: (1) vaccination, (2) transmission control, (3) latent infection management and (4) clinical treatment optimisation. The results aim to provide a robust scientific rationale for precision control in this high-burden setting. This research contributes essential, region-specific evidence to both China’s national TB control target and the WHO End TB targets.

Methods

Data sources

Xizang is located in southwestern China and accounts for over 50% of the total area of the Qinghai-Xizang Plateau. The average altitude of the Xizang region is over 4000 m above sea level. The highest point is Mount Everest, standing at 8848.86 m (based on the latest 2020 survey data), located on the border between China and Nepal, making it the highest peak in the world. The lowest point is near Baxika in Medog County, at the outlet of the Yarlung Zangbo Grand Canyon, with an elevation of approximately 155 m, which is the lowest area within the Xizang Autonomous Region of China. Administratively, it is divided into seven prefecture-level administrative units (Lhasa City, Chamdo City, Shannan City, Shigatse City, Nagqu City, Ngari Prefecture, Nyingchi City).

The annual TB surveillance data (1 January 2004 to 31 December 2022) were obtained from the routine surveillance system of the Xizang Centre for Disease Control and Prevention (CDC), which were collected through the National Notifiable Disease Reporting System, a standardised, web-based platform maintained by the Chinese CDC. TB was diagnosed in accordance with national guidelines, based on clinical symptoms, chest radiographic findings and laboratory tests, including sputum smear microscopy, culture and molecular assays.¹⁹ Case definitions followed the Chinese National TB Diagnostic Criteria (WS 288–2017), including clinically diagnosed cases, suspected cases and aetiologically confirmed cases. The population data were obtained from the China Statistical Yearbook, which is compiled by the National Bureau of Statistics (NBS). Key clinical and demographic data, together with regular quality checks to guarantee correctness and completeness, were recorded by qualified healthcare personnel and validated by experts. The NBS is an official agency directly under the State Council, mandated by law to organise and manage national statistical work. The Yearbook has been published annually since 1982, and its data coverage encompasses all provincial-level administrative units across the country, providing continuous, consistent and comprehensive benchmark information for research. The study was reported in accordance with the Strengthening the Reporting of Observational Studies in Epidemiology reporting checklist (online supplemental table S1).

Patient and public involvement

This study used anonymised routine TB surveillance data from the national information system and did not involve direct contact with patients. Therefore, patients and the public were not involved in the design, conduct, reporting or dissemination plans of this research.

Model construction

A detailed list of model parameters, including definitions, values, sources and ranges, is provided in table 1. The model was adapted from the TB transmission framework proposed by Blower et al¹⁰ and expanded to include vaccination coverage, vaccine efficacy and drug resistance. The total population is denoted by N, which is subdivided into the following five subpopulations: susceptible individuals (S), latent TB-infected individuals (LTBI) (E), individuals with drug-susceptible TB (I₁), individuals with multidrug-resistant TB (I₂) and recovered (R). The model was based on the following assumptions: (1) The total population remains relatively stable. (2) Successful vaccination was assumed to confer complete and lifelong immunity. (3) Individuals with active TB were assumed to be infectious. (4) Recovered individuals were at risk of relapse to an infectious state. (5) The probability of contact between any two individuals in the population is homogeneous. The model’s compartmental structure is shown in figure 1.

Table 1. Model parameters: definitions, values and sources.

Parameter	Definition	Value (ranges)	Unit	Source
Λ	Natural population replenishment rate	15.77 (13.96, 17.94)	Per 1000 population per year	China Statistical Yearbook 2023
μ0	Natural mortality rate of the population	5.34 (4.46, 7.15)	Per 1000 population per year	China Statistical Yearbook 2023
μR	Mortality rate among cured tuberculosis patients	1.60×10⁻2 (0.90×10⁻2, 2.53×10⁻2)	Proportion (dimensionless)	Reference46
v	BCG vaccination coverage rate	0.90 (0.86, 0.99)	Proportion (dimensionless)	Reference47
p	Vaccine efficacy rate	0.54 (0.40, 0.60)	Proportion (dimensionless)	Reference48
σ1	Drug resistance rate among newly diagnosed tuberculosis patients	0.1309 (0.1203, 0.1423)	Proportion (dimensionless)	Actual data from Xizang
σ2	Drug resistance rate among relapsed tuberculosis patients	0.1916 (0.1515, 0.2392)	Proportion (dimensionless)	Actual data from Xizang
τ	Detection rate of active tuberculosis cases	0.648 (0.567, 0.839)	Proportion (dimensionless)	WHO data
γ1	The treatment success rate for drug-susceptible tuberculosis patients	0.862 (0.843, 0.875)	Proportion (dimensionless)	Reference49
γ2	The treatment success rate for drug-resistant tuberculosis patients	0.643 (0.578, 0.696)	Proportion (dimensionless)	Reference49
κ	Progression rate from latent tuberculosis infection (LTBI) to active tuberculosis disease	0.05 (0.04, 0.10)	Per year	Reference50
Parameters pending estimation
d1	Disease-induced mortality rates for drug-susceptible patients	9.93×10⁻2 (8.97×10⁻2, 10.90×10⁻2)	Per year	Data fitting based on 2004–2019 data
d2	Disease-induced mortality rates for drug-resistant patients	9.94×10⁻2 (8.95×10⁻2, 10.90×10⁻2)	Per year	Data fitting based on 2004–2019 data
ω	Relapse rate among successfully treated patients	1.55×10⁻3 (1.39×10⁻3, 1.70×10⁻3)	Per year	Data fitting based on 2004–2019 data
ρ	Progression rate from drug-susceptible to drug-resistant tuberculosis patients	1.14×10⁻3 (1.03×10⁻3, 1.25×10⁻3)	Per year	Data fitting based on 2004–2019 data
β1	Transmission rate from drug-susceptible tuberculosis patients to susceptible individuals	1.00×10–9 (9.00×10−10, 1.09×10−9)	Per person per year	Data fitting based on 2004–2019 data
β2	Transmission rate from drug-resistant tuberculosis patients to susceptible individuals	1.00×10–7 (9.06×10−8, 1.10×10−7)	Per person per year	Data fitting based on 2004–2019 data

BCG, Bacille Calmette-Guérin.

Figure 1. Compartmental diagram of the tuberculosis transmission dynamics model. The population is stratified into five compartments: susceptible individuals (S), latent tuberculosis-infected individuals (LTBI) (E), drug-susceptible tuberculosis patients (I₁), multidrug-resistant tuberculosis patients (I₂) and successfully treated individuals (R).

The model is represented by the following system of differential equations:

$${\displaystyle {\displaystyle \begin{array}{c}\frac{dS}{dt}=S\mathrm{\Lambda }-\left(1-\nu p\right)\left({\beta }_1{I}_1+{\beta }_2{I}_2\right)S-{\mu }_{0}S\\ \frac{dE}{dt}=\left(1-\nu p\right)\left({\beta }_1{I}_1+{\beta }_2{I}_2\right)S-kE-{\mu }_{0}E\\ \frac{d{I}_1}{dt}=k\left(1-{\sigma }_1\right)E+\omega \left(1-{\sigma }_2\right)R-\rho I_1-\tau {\gamma }_1{I}_1-\left({\mu }_0+d_1\right)I_1\\ \frac{d{I}_2}{dt}=k{\sigma }_{1}E+\omega {\sigma }_{2}R+\rho I_1-\tau {\gamma }_2{I}_2-\left({\mu }_0+d_2\right)I_2\\ \frac{dR}{dt}=\tau {\gamma }_1{I}_1+\tau {\gamma }_2{I}_2-\omega R-{\mu }_{R}R-{\mu }_{0}R\\ N=S+E+I_1+I_2+R\end{array}}}$$ (1)

Basic reproduction number (R₀)

The R₀ represents the average number of secondary infections generated by a single infectious individual over the course of their infectious period in a completely susceptible population. Generally, R₀=1 serves as a threshold to determine whether a disease will die out or persist. If R ₀<1, the disease-free equilibrium is locally asymptotically stable, and the infection is expected to be eliminated from the population. Conversely, if R ₀>1, an endemic equilibrium exists and is locally asymptotically stable, allowing the disease to persist within the population. In more complex dynamical systems, this conclusion may not be valid, since backward bifurcation, multistable equilibria and other non-linear behaviours can occur. However, a lower R₀ is still generally associated with improved controllability of TB transmission.^{14 20} We calculated R₀ using the next-generation matrix method.²⁰ The R₀ expression is given by

$${\displaystyle {\displaystyle \begin{array}{c}{R}_0=\left(1-vp\right)S_0\frac{{\beta }_1{T}_1+{\beta }_2{T}_2}{\left|V\right|}\end{array}}}$$ (2)

$${\displaystyle {\displaystyle \begin{array}{l}\text{where }{T}_1=(1-{\sigma }_1)\frac{1}{{\sigma }_1},T_2={\sigma }_1\frac{1}{{\sigma }_2}+(1-{\sigma }_1)\frac{\rho }{{\sigma }_{a_1{a}_2}},|V|=\frac{k+{\mu }_0}{k}.\\ a_1=\rho +{\mu }_0+d_1+\tau {\gamma }_1,a_2={\mu }_0+d_2+\tau {\gamma }_2.\end{array}}}$$

The detailed calculations are provided in the online supplemental materials.

Parameter estimation and model fitting

The parameters d₁, d₂, ω, ρ, β₁ and β₂ were estimated using a Markov chain Monte Carlo (MCMC) algorithm. Three independent MCMC chains of 50 000 iterations each were run, with the first 10 000 iterations discarded as burn-in. Priors were specified as uniform or log-uniform distributions within biologically plausible bounds (online supplemental table S2).²¹ We fitted the model to the reported annual TB incidence data in Xizang from 2004 to 2019. The model was then validated against data from 2020 to 2022 to assess its goodness of fit before being used to forecast incidence up to 2040. In this study, both model fitting and prediction targeted TB incidence. Consistent with standard infectious disease modelling practice, the model-predicted annual number of newly developed TB cases was defined as a flow-based quantity. We evaluate the fitting effect of our established model through the root mean square error (RMSE) and the mean absolute percentage error (MAPE), which are significant evaluation indicators.^{14 22} The RMSE and the MAPE are defined as follows:

$${\displaystyle {\displaystyle {\displaystyle \begin{array}{c}\text{R}\text{M}\text{S}\text{E}=\sqrt{\frac{1}{n}\sum _{i=1}^n{\left(actua{l}_i-predic{t}_i\right)}^2}\end{array}}}}$$ (3)

$${\displaystyle {\displaystyle {\displaystyle \begin{array}{c}\text{M}\text{A}\text{P}\text{E}=\frac{1}{n}\sum _{i=1}^n\left|\frac{actua{l}_i-predic{t}_i}{actua{l}_i}\right|\times 100\mathrm{\%}\end{array}}}}$$ (4)

where actual_i and predict_i represent the observed and model-predicted number of new TB cases in year i, respectively, and n is the total number of years in the validation set.

To simulate various prevention and control strategies, we systematically varied five key parameters while holding others constant. Parameter variation ranges were determined based on practical implementability and evidence from published literature. For each scenario, the magnitude of variation and its public health meaning were specified in table 2.

Table 2. Mapping model parameters to real-world tuberculosis intervention scenarios.

Model parameters	Scenario settings	Corresponding public health measures
Vaccine efficacy rate (p)	0.6/0.7/0.8	Introduce more effective vaccines/booster shots; near-complete protection with novel vaccines or an improved BCG vaccine.
Transmission rates for drug-susceptible and multidrug-resistant tuberculosis (β₁ and β2)	Reduced by 25%/50%/75%	A 25% reduction approximates the effect of moderate improvements in protection and case management. A 75% reduction corresponds to a combination of large-scale active case finding, strict isolation and robust community-level interventions.
Progression rate from latent tuberculosis infection (LTBI) to active tuberculosis disease (k)	Reduced by 50%/60%/70%/80%/90%	Progressively scale up LTBI screening combined with preventive treatment (eg, isoniazid/rifapentine). A 50% reduction could represent the effect of a high-coverage, highly effective prevention strategy over a medium-term period.
Relapse rate among successfully treated patients (ω)	Reduced by 50%/60%/70%/80%	Implement better treatment regimens, strengthen the directly observed treatment, short-course strategy and patient follow-up and improve the rapid detection of drug resistance and timely referral.

Sensitivity analysis

To evaluate the robustness of model outcomes to uncertainty in parameter values, we performed a sensitivity analysis combining probabilistic uncertainty assessment and global sensitivity methods. Latin hypercube sampling was used to sample key model parameters from their predefined ranges or probability distributions.^{23 24} For each sampled parameter set, the model was simulated to estimate R₀. The resulting simulations were used to construct region-specific empirical density distributions of R₀ (online supplemental figure S1), which showed stable, unimodal distributions across regions with clear spatial heterogeneity in transmission potential.

In addition, partial rank correlation coefficient (PRCC) analysis was conducted to quantify the influence of selected key parameters on model outputs, including R₀ and disease incidence. PRCC provides a global sensitivity measure by assessing the strength and direction of the monotonic relationship between each parameter and the outcome while accounting for concurrent variation in other parameters.^{25 26} Both PRCC estimates and their corresponding p values were calculated to identify parameters with statistically significant effects. Detailed results of the PRCC analysis are reported in the online supplemental table S3, indicating that disease incidence was primarily driven by the progression and relapse rates.

Statistical analysis

All statistical analyses were performed using R (V.4.4.1). The MCMC parameter estimation was implemented using the deBInfer package in R.

Results

Parameter fitting results

In the model, the parameters d₁, d₂, ω, ρ, β₁ and β₂ were fitted for the Xizang and its seven prefecture-level cities through the adjustment of the model’s relevant parameters. The fitting results are summarised in table 3.

Table 3. Fitted parameter results.

Regions	d1 (95% CI)	d2 (95% CI)	ω (95% CI)	ρ (95% CI)	β1 (95% CI)	β2 (95% CI)	R0 (95% CI)
Xizang	9.93×10–2 (8.97×10−2 to 100.90×10−2)	9.94×10–2 (8.95×10−2 to 100.90×10−2)	1.55×10–3 (1.39×10−3 to 10.70×10−3)	1.14×10–3 (1.03×10−3 to 10.25×10−3)	1.00×10–9 (9.00×10−10 to 10.09×10−9)	1.00×10–7 (9.06×10−8 to 10.10×10−7)	0.11 (0.08 to 0.15)
Lhasa	8.76×10–2 (7.89×10−2 to 90.56×10−2)	8.76×10–2 (7.92×10−2 to 90.60×10−2)	4.85×10–2 (4.35×10−2 to 50.35×10−2)	4.54×10–3 (4.11×10−3 to 40.99×10−3)	1.00×10–9 (8.98×10−10 to 10.10×10−9)	1.00×10–7 (9.03×10−8 to 10.09×10−7)	0.07 (0.03 to 0.11)
Chamdo	8.96×10–2 (8.03×10−2 to 90.82×10−2)	8.16×10–2 (7.36×10−2 to 80.92×10−2)	2.31×10–1 (2.09×10−1 to 20.54×10−1)	9.96×10–3 (9.04×10−3 to 10.10×10−2)	1.00×10–9 (9.00×10−10 to 10.10×10−9)	1.00×10–7 (9.04×10−8 to 10.10×10−7)	0.21 (0.14 to 0.30)
Shannan	7.22×10–2 (6.51×10−2 to 70.92×10−2)	5.63×10–2 (5.07×10−2 to 60.16×10−2)	1.93×10–1 (1.73×10−1 to 20.11×10−1)	1.49×10–1 (1.35×10−1 to 10.64×10−1)	1.00×10–9 (9.04×10−10 to 10.10×10−9)	1.00×10–7 (9.08×10−8 to 10.10×10−7)	0.19 (0.10 to 0.33)
Shigatse	7.08×10–2 (6.38×10−2 to 70.77×10−2)	5.76×10–2 (5.19×10−2 to 60.29×10−2)	2.01×10–1 (1.81×10−1 to 20.19×10−1)	1.38×10–1 (1.25×10−1 to 10.52×10−1)	1.00×10–9 (9.03×10−10 to 10.09×10−9)	1.00×10–7 (9.11×10−8 to 10.10×10−7)	0.39 (0.21 to 0.71)
Nagqu	4.85×10–2 (4.40×10−2 to 50.33×10−2)	5.27×10–2 (4.77×10−2 to 50.76×10−2)	1.37×10–1 (1.24×10−1 to 10.50×10−1)	8.88×10–2 (7.96×10−2 to 90.78×10−2)	1.00×10–9 (9.05×10−10 to 10.10×10−9)	1.00×10–7 (9.01×10−8 to 10.09×10−7)	0.19 (0.10 to 0.34)
Ngari	4.96×10–2 (4.44×10−2 to 50.42×10−2)	5.21×10–2 (4.72×10−2 to 50.77×10−2)	1.38×10–1 (1.24×10−1 to 10.51×10−1)	9.90×10–2 (8.88×10−2 to 100.90×10−2)	1.00×10–9 (9.03×10−10 to 10.10×10−9)	1.00×10–7 (8.99×10−8 to 10.10×10−7)	0.04 (0.02 to 0.07)
Nyingchi	7.43×10–2 (6.71×10−2 to 80.16×10−2)	6.29×10–2 (5.72×10−2 to 60.91×10−2)	1.63×10–1 (1.47×10−1 to 10.79×10−1)	6.82×10–2 (6.14×10−2 to 70.51×10−2)	1.00×10–9 (9.90×10−10 to 10.10×10−9)	1.00×10–7 (9.04×10−8 to 10.10×10−7)	0.08 (0.04 to 0.15)

Notes: CI: 95% credible interval from the posterior distribution of the MCMC estimation. All prefecture-level parameter estimates include their corresponding CIs.

Analysis of the R₀

By substituting the parameters from the literature and those fitted from empirical data into Equation 2, we calculated the R₀ for the Xizang to be 0.11, which is less than 1. This indicates a low transmission potential, meaning that each infected individual, on average, transmits the disease to fewer than one other person during their infectious period. Under current conditions, the disease is unlikely to sustain transmission and may gradually disappear. The R₀ values for the other seven prefecture-level cities are presented in table 3. Among them, Shigatse City had the highest R₀ at 0.39 (95% CI 0.21 to 0.71), followed by Chamdo, Nagqu and Shannan.

Model prediction results

Figure 2 shows the reported and model-predicted incidence of TB for the Xizang and its seven prefecture-level cities from 2004 to 2040. In the figure, the blue curve represents the model-fitted and forecasted incidence trajectory, which was calibrated against reported data from 2004 to 2019 (black dots) and validated against data from 2020 to 2022 (white dots). The model achieved a RMSE of 115.40 and a MAPE of 9.46% on the training set, with similar performance on the validation set (RMSE=119.36; MAPE=9.47%).

Figure 2. Comparison of actual and predicted tuberculosis incidence rates in the Xizang and its seven prefecture-level cities from 2004 to 2040. (A) Xizang. (B) Lhasa City. (C) Chamdo City. (D) Shannan City. (E) Shigatse City. (F) Nagqu City. (G) Ngari Prefecture. (H) Nyingchi City.

The trend in incidence shown in the figures suggests that the overall incidence rate across the entire Xizang region will exhibit a slow downward trend in the coming years. Further assessment of transmission potential in different regions revealed that Chamdo City is projected to have the highest incidence rate by 2030, while Lhasa City is expected to have a lower rate. Meanwhile, figure 2B indicates that Lhasa City was on track to meet China’s 2030 TB control target (43 per 100 000 population) by 2022.

Incidence prediction and evaluation of control strategies

Using the parameters from table 1, we predicted the TB incidence rate for Xizang from 2022 to 2040. The projected reported incidence for 2030 is 66.56 (62.00–70.11) per 100 000 population. This remains a significant gap from the targets set by the WHO 2030 End TB target (below 26.8 per 100 000) and China’s 2030 TB control target.

We simulated the effects of different TB control strategies and their varying intensities by adjusting the model parameters. Increasing the parameter p (vaccine efficacy rate) resulted in a minor reduction in TB incidence. When p was increased to 80%, the predicted reported incidence of pulmonary TB in the Xizang for 2030 was 65.93 (61.86–69.85) per 100 000 (figure 3A).

Figure 3. Changes in the incidence rate in the Xizang after adjusting five parameters. (A) Vaccine efficacy rate p. (B) Transmission rates for drug-susceptible and multidrug-resistant tuberculosis (β₁ and β₂). (C) Progression rate from latent tuberculosis infection (LTBI) to active tuberculosis disease k. (D) Relapse rate among successfully treated patients ω.

Decreasing the parameters β₁ and β₂ (transmission rates of susceptible individuals) simulated the impact of enhanced personal protection and isolation of infectious patients. The simulation showed a slight decrease in TB incidence as β₁ and β₂ were reduced. When β₁ and β₂ were reduced by 50%, the predicted incidence for 2030 was 65.87 (61.52–69.73) per 100 000 (figure 3B).

Reducing the parameter k (progression rate from LTBI to active TB) led to a significant decline in incidence. When k was reduced by 50%, the predicted incidence for 2030 dropped to 40.54 (37.15–43.77) per 100 000 (figure 3C).

Decreasing the parameter ω (relapse rate among successfully treated patients) resulted in a slight decrease in incidence. When ω was reduced by 50%, the predicted incidence for 2030 was 66.13 (61.95–70.18) per 100 000 (figure 3D).

Discussion

In this modelling study, we projected the future trajectory of TB in the Xizang region and found substantial heterogeneity in transmission potential across different prefectures. Our findings align with national-level modelling studies showing that LTBI management is critical for TB elimination in China.^{27 28} Our estimated R₀ for Xizang is 0.11, suggesting that TB transmission could eventually be interrupted under current conditions. However, this R₀ remains substantially higher than China’s average (0.06), consistent with epidemiological observations of persistently high burden in western regions.²⁹,³¹ Although elimination is theoretically possible, forecasts indicate that significant efforts are still required to achieve the WHO End TB targets.

Unlike previous studies focused on national or eastern-province averages, our work highlights the unique challenges of high-altitude settings. Our intervention simulations complement recent work by Wen et al on LTBI management but extend it by comparing multiple intervention modalities specific to resource-limited, geographically isolated settings.¹² The pronounced prefecture-level heterogeneity further underscores the necessity of subnational modelling and targeted LTBI interventions in high-burden areas to achieve China’s 2030 and the WHO End TB targets.

Overall, the Xizang region exhibits a high burden of TB incidence, with Shigatse City and Chamdo City requiring particular attention. The high incidence in this demographic may be linked to their lifestyles, lower levels of education, poor hygiene practices and limited awareness of disease prevention. Shigatse City’s high transmission potential may be related to its unique geographical location and sociocultural characteristics.³² As a major transportation hub in Xizang connecting multiple regions and neighbouring countries, its frequent population mobility increases opportunities for TB transmission.³³ Moreover, Shigatse has a complex demographic structure, with a mix of permanent residents and nomadic pastoralists whose seasonal migrations can facilitate the spread of pathogens across different areas.³⁴ Environmentally, the high altitude, cold, dry climate and limited ventilation in dwellings are also conducive to airborne transmission of TB.³⁵

In contrast, the high incidence in Chamdo City is more influenced by natural environmental and socioeconomic factors. Located in the northern section of the high-altitude Hengduan Mountains, Chamdo’s cold climate and poorly ventilated living conditions, coupled with a hypoxic environment, may weaken immune function and increase infection risk.³⁶ The region has a large population of pastoralists and farmers living in dispersed settlements, with traditional production and living habits, and their hygiene practices and self-protection awareness remain underdeveloped. The region’s lagging economic development and insufficient educational resources limit public knowledge of TB prevention and control. A 2016 survey found that the total awareness rate among all active pulmonary TB patients in Chamdo was only 28.1% and 34.5% among all respondents,³⁷ leading to delays in seeking medical care. However, since 2019, the incidence of pulmonary TB has declined significantly across all regions, a trend likely influenced by the COVID-19 pandemic and its corresponding public health and social control measures.³⁸,⁴⁰

By simulating incidence trends under different interventions, our study found that reducing the progression rate from LTBI to active TB (k) is the most effective strategy for controlling the epidemic. If k was reduced to 50% of its current level, the TB incidence in 2030 could be lowered to 40.54 per 100 000, approaching China’s 2030 TB control target. This result suggests that the large number of individuals with LTBI is a key factor contributing to the slow decline in TB incidence. Related research also indicates that the failure to promptly identify and treat the vast pool of latently infected individuals is a major reason for the persistently high burden of TB.⁴¹,⁴⁴ In 2015, the WHO End TB targets explicitly identified systematic screening and treatment of the LTBI population as a core intervention.⁴⁵ Therefore, integrating precision screening and preventive therapy into the TB control system is essential.

Although this study explored the transmission trends and effectiveness of control strategies for TB from a theoretical standpoint, it has certain limitations. First, like most compartmental models, the SEIR framework represents a simplified abstraction of TB transmission and does not explicitly account for heterogeneity in contact patterns, health-seeking behaviour or spatial mobility. With more detailed data, future studies could incorporate multiple latent compartments and use a metapopulation model with mobility data to simulate transmission pathways. Second, the model used fixed parameter estimates from historical surveillance data. In reality, these parameters may change over time due to shifts in diagnostic capacity, treatment coverage and health system performance. Major disruptions, such as the COVID-19 pandemic, can further influence these parameters. Related public health control policies may also have affected TB transmission, healthcare access and case reporting. Future studies could incorporate time-varying parameters or scenario-based analyses. Third, in evaluating control strategies, our study primarily estimated their theoretical impact on the reported incidence of TB. Practical considerations such as technical feasibility, ethical issues and economic costs were not fully explored and require further investigation. Finally, reliance on routine surveillance data may introduce misclassification and reporting biases, including underdiagnosis and delayed reporting, which could affect incidence estimates.

Conclusion

TB transmission in Xizang, while declining, exhibits significant regional heterogeneity and is largely sustained by a substantial reservoir of latent infections. Our findings indicate that prioritising the management of LTBI, through targeted screening and preventive therapy, is the most effective strategy to accelerate progress towards national and global elimination goals in this high-burden region.

Data availability statement

Data are available upon reasonable request.