IMPACT OF COVID-19 ON TUBERCULOSIS CASE DETECTION IN VIET NAM: A MODELLING ANALYSIS

INTRODUCTION

Tuberculosis (TB), the disease caused by the bacterium Mycobacterium tuberculosis (M.tb), remains a critical global public health challenge [1]. Viet Nam, as a high-burden country, exemplifies the difficulties in controlling TB [1]. Its continuing high burden highlights the urgent need for improved public health strategies, and understanding its epidemic drivers may offer valuable insights for addressing TB in similar settings globally.

The COVID-19 pandemic severely disrupted global TB control, causing significant declines in TB notifications in high-burden countries like India, South Africa, Brazil, and Indonesia [2–4]. Prior research identified an 8% decrease in TB notifications in Viet Nam attributable to pandemic effects [5]. The World Health Organization (WHO) reported an 18% global decline in TB notifications in 2020, with an estimated 200,000–400,000 additional TB deaths [2, 6]. WHO’s assessments rely on national TB program reports [7], which may underestimate the true burden due to undiagnosed cases and limited surveillance capacity. Initial projections were based on notification trends rather than dynamic transmission models, which may not be the optimal approach to predicting long-term TB impact estimates. The reduction in TB notifications may be attributable to healthcare disruptions, resource diversion, reduced healthcare-seeking behavior or some combination of these factors. However, COVID-19 measures like mask-wearing and social distancing may have reduced TB transmission, which are not accounted for in WHO’s early estimates [6].

To address these challenges, we developed a deterministic compartmental model to simulate the transmission dynamics of M.tb in Viet Nam. Through this model, we aimed to quantify the long-term impacts of COVID-19 on TB control over the next decade.

METHODS

RESULTS

1. Calibration results

Candidate model comparison

All four assumptions closely matched the total population size and the prevalence of TB among adults, as shown in Figures S3 and S4. However, the fit varied when comparing the actual number of TB case notifications to the equivalent model output. Figure 2 shows this comparison under our four candidate assumptions regarding the effect of COVID-19 on M.tb dynamics. Among the assumptions, Assumption 2 provided the best fit to the observed sharp decline in TB notifications in 2021, slightly outperforming Assumption 4 based on our assessment with the EPLD-WAIC (see Table S6). In contrast, Assumption 1 and Assumption 3 (COVID-19 only reduced transmission) showed a poorer fit to the data.

• Download figure
• Open in new tab

Figure 1. Model structure

Boxes represent different compartment types: Susceptible (S), Early latent infection (E), Late latent (L), Active TB (I), Treatment (T) and Recovered (R). Subscripts indicate stratification by age (a) and pulmonary/smear status (f).

Alt text: The image depicts a compartmental model structure used in epidemiology. It consists of several boxes representing different disease states, including Susceptible (S), Early latent Infection (E), Late latent (L), Active TB (I), Treatment (T), and Recovered (R). Each box is further stratified by age (indicated by subscript ‘a’) and pulmonary/smear status (indicated by subscript ‘f’), illustrating the transitions and interactions between these different states in the model.

• Download figure
• Open in new tab

Figure 2. Model fits to the historical profile of case notifications under our four candidate assumptions regarding the effect of COVID-19 on M.tb dynamics

Solid lines represent median model estimates. Shaded areas show corresponding interquartile ranges (dark shade) and 95% credible intervals (light shade). Assumption 1: The COVID-19 pandemic had no impact on TB case detection or transmission. Assumption 2: Only case detection was reduced. Assumption 3: Only TB transmission was reduced. Assumption 4: Both case detection and TB transmission were reduced, incorporating effects from both social distancing and healthcare disruption.

Alt text: Figure 2 shows the model’s fit to historical case notifications under four different assumptions about COVID-19’s impact on tuberculosis dynamics. The model’s median estimates are shown by solid lines, while the 95% credible intervals and interquartile ranges are shown by shaded areas (dark and light, respectively). Assumption 1, neither TB case detection nor transmission was impacted by the COVID-19 pandemic; Assumption 2, only case detection was decreased; Assumption 3, only TB transmission was decreased; and Assumption 4, decreases in both case detection and TB transmission.

Based on this preceding analysis, we selected Assumption 2 as our primary model for the following in-depth examination of M.tb dynamics in Viet Nam. Figure 3 illustrates the comparison between accepted model runs, each calibration target, and other epidemiological estimates not included in our calibration algorithm that we considered for model validation. Figure S5 presents model outputs derived from the 50 randomly accepted model runs.

• Download figure
• Open in new tab

Figure 3. Comparison between outputs and empiric observation targets other than annual case notifications for Viet Nam, under the best-fitting candidate model configuration (Assumption 2)

Model predictions are represented as median (solid line), interquartile credible interval (dark blue shade) and 95% central credible interval (light blue shade). Red points and vertical line (where available) show the point estimates and confidence interval for each calibration target, while purple points indicate estimates from literature that are presented for comparison but were not used as calibration targets. Legends indicate the data source for each target.

Alt text: Figure 3 illustrates a comparison between model outputs and empirical observation targets for Vietnam, excluding annual case notifications, under the best-fitting candidate model configuration (Assumption 2). The median is represented by a solid line, the interquartile credible interval by a dark blue line, and the 95% central credible interval (CrI) by a light blue line. Red points with vertical lines that, when available, display CrIs are used to indicate calibration targets. Estimates from the literature that are utilised for comparison but not as calibration objectives are shown by purple points. The legend provides specifics about each target’s data source.

Table S7 (Supplementary Material) details the posterior estimates and calibration metrics for the calibrated parameters of our preferred candidate model. The algorithm demonstrated highly satisfactory chain convergence, as evidenced by each parameter’s Gelman-Rubin statistic (R̂) being below 1.05 and all effective sample sizes exceeding 150. Figure S6 shows the trace plots for each parameter, illustrating their convergence and stability throughout the execution of the calibration algorithm. Figure S7 (Supplementary Material) overlays posterior distributions on the corresponding priors of the calibrated parameters. The posterior estimate for the reduction of COVID-19 on case detection was 0.33 with 94% high density interval ranging from 0.01 to 0.677. Figure S8 (Supplementary Material) shows how the posterior distributions for the natural history parameters translate into the more epidemiologically intuitive quantities of disease duration and case fatality rate for SPTB and SNTB.

Given the lack of historical data, the first 200 years of the simulation exhibited significant uncertainty, which progressively diminished in later years (Figure S10). Our calibration algorithm simulated a peak in adult pulmonary TB prevalence that began to decline from around 1986, coinciding with the establishment of the NTP. By 2017, the model predicted an adult pulmonary TB prevalence of 328 cases per 100,000 populations (95% CrI: 269–399), closely matching our prevalence survey target, which estimated bacteriologically-confirmed TB among adults at 322 (confidence interval – CI: 260– 399) per 100,000 in 2017 [12].

We estimated TB incidence in 2022 at 153 cases per 100,000 population (95% CrI: 117 - 202), which closely matches the WHO’s 2022 estimate of 176 cases per 100,000 (95% CI: 121 - 251) [11]. We estimated that approximately 18,000 people have died from TB each year since 2010, or about 18 deaths per 100,000 population. Our modelled prevalence of adult SPTB aligned closely with the results of the 2007 national prevalence survey (95 per 100,000, 95% CrI: 72 - 127, compared to 99 per 100,000, 95% CI: 78 - 125) [18], but was somewhat higher than the 2017 survey results (73 cases per 100,000, 95% CrI: 59 - 88, compared to 46 cases per 100,000, 95% CI: 32 - 68) [12]. Figure S9 (Supplementary Material) illustrates the proportion of active TB cases originating from the early latent phase, highlighting that 50% or more of these cases were due to early progression and therefore recent transmission throughout most of the recent simulation period. A minor increase in the proportional contribution of recent transmission was observed in 2021 during the COVID-19 pandemic. The modelled prevalence of LTBI in 2019 was 45% of the total population (95%CrI: 39% - 51%). This aligns closely with the estimate from Ding et al. [19], which places the LTBI prevalence at around 44% of the population, also within our CrI. Figure S109 (Supplementary Material) shows the proportions of the population by compartment based on results from the maximum likelihood run.

Figure S11 (Supplementary Material) shows the estimated increasing profile of the rate of commencing treatment over time, with the increase beginning from around 1986, which coincides with the establishment of the Viet Nam NTP [20]. The modelled rate shows a steep inflection, accelerating most rapidly through the late 1990s, six years after the implementation of the DOTS strategy [20]. The median case detection proportion stabilized at around 0.6 from 2010 onwards, which can be interpreted as meaning that about 60% of incident cases were successfully detected and commenced treatment. Specifically, we estimated a 33% reduction in the rate of commencing treatment, during COVID-19, resulting in a 0.43-year (equivalent to 23 weeks) increase in the time to active TB onset to detection and treatment.

2. Future projection

Long-term impacts of COVID-19 on TB incidence and TB deaths

Figure 4 illustrates how the COVID-19-related health system effects may contribute to impeding future TB control efforts under Assumption 2, in comparison with a counterfactual scenario that COVID-19 had no effect on TB notifications. In 2021, we estimated 1,900 additional TB episodes (95% CrI: 200 – 5,100) and 1100 additional TB deaths associated with the effects of COVID-19 (95% CrI: 100 – 2,700). By 2035, the cumulative annual burden could rise 22,000 more TB episodes (95% CrI: 1,900 – 63,000) and 5,900 more deaths (95% CrI: 600 – 16,600).

In 2024, we estimated the average delay to diagnosing and treating an active TB case at 1.80 years, such that our scenarios are equivalent to reducing the delay time to 0.9 years (approximately 47 weeks) with a 2-fold increase, and to 0.36 years (approximately 19 weeks) with 5-fold in rate of commencing care. Our projections of the results of these scenarios are presented in Figure 5.

• Download figure
• Open in new tab

Figure 4. Impact of the COVID-19 pandemic on the cumulative number of TB episodes and TB-related deaths from 2021 to 2035 compared to a counterfactual scenario that COVID-19 had no effect on TB notifications

Projections made under Assumption 2, values are presented on a logarithmic scale. Cumulative diseased: cumulative new TB cases, Cumulative deaths: cumulative number of deaths due to TB. Lines indicate the 95% credible intervals of the estimates.

Alt text: Figure 4 displays the impact of the COVID-19 pandemic on the cumulative number of tuberculosis (TB) episodes and TB-related deaths from 2021 to 2035, compared to a counterfactual scenario where COVID-19 had no effect on TB notifications. Assumption 2 is used to make the projections, and a logarithmic scale is used to display the data. The graph labels “Cumulative deaths” and “Cumulative diseased” denote the total number of TB-related deaths and cumulative new TB cases, respectively. The graph’s lines show the estimations’ 95% confidence intervals.

• Download figure
• Open in new tab

Figure 5. Number of cumulative new TB episodes and TB-related deaths averted under two scenarios of enhancing case detection comparing to ‘status quo scenario’

Projections made under Assumption 2, values are presented on a logarithmic scale. The ‘status quo’ scenario assumes case detection remains unchanged. Scenario 1 models a 2-fold increase and Scenario 2 models a 5-fold increase in the rate at which persons with active TB commence care. Lines indicate the 95% credible intervals of the estimates.

Alt text: The number of cumulative new TB episodes and TB-related deaths prevented under two scenarios that improve case detection in comparison to a “status quo scenario” is displayed in Figure 5. The results are displayed on a logarithmic scale, and the projections are made under Assumption 2. In the ‘status quo’ scenario, case detection is assumed to stay constant. The rate at which people with active TB start receiving care is predicted to increase by two times in Scenario 1 and five times in Scenario 2. The graph’s lines show the estimations’ 95% confidence intervals.

Compared to a ‘status quo’ projection, enhanced case detection by increasing the rate of commencing care 5-fold may prevent approximately 433,500 (95%CrI: 249,400 – 692,200) new TB episodes, representing a 17.8% reduction (95%CrI: 13.1% - 21.9%). Additionally, this scenario may avert 109,900 (95%CrI: 68,500 – 169,900) TB-related deaths, equivalent to 34.2% (95%CrI: 31.5% - 37.0%) of all TB-related dêaths over the coming decade. Notably, enhancing TB case detection may have an even greater impact on reducing TB-related mortality.

DISCUSSION

Our results show that the decline in TB detection during COVID-19 was primarily responsible for the sharp drop in TB notifications in Viet Nam in 2021. This disruption has likely contributed to a backlog of undiagnosed cases, sustaining TB transmission and potentially leading to long-term increases in incidence and mortality. Our methodological approach, developed using open-source packages, supports data input manipulation, compartmental model construction, execution, and calibration, providing a flexible and efficient tool for developing, optimizing, and calibrating complex epidemiological models while promoting reproducibility and adaptability. [17].

he close alignment between the model-simulated indicators, including the population size, TB notifications and pulmonary TB prevalence demonstrates our model’s capacity to capture multiple key observations reflecting TB burden. We found that COVID-19 was associated with significant reductions in TB case detection, which poses a serious threat to long-term TB control efforts, which aligns with findings of other studies [20, 21]. However, Viet Nam has gained lessons from the COVID-19 pandemic to enhance TB case detection, focusing on decentralized services and proactive case-finding efforts. The “Double X” (2X) strategy, which utilizes chest radiography and GeneXpert rapid diagnostic testing, has improved early TB detection, particularly in vulnerable populations [5, 22, 23]. Continuing and expanding these programs may further strengthen TB surveillance, improve case detection rates, and mitigate long-term disruptions to TB control efforts.

The model configuration incorporating both reduced case detection and transmission reduction performed almost as well as that with reduced case detection only. However, the configuration that incorporated a reduction in case detection aligned most closely with observed trends, showing how COVID-19-related healthcare disruptions likely delayed diagnosis and treatment, leading to fewer reported cases. The detection process is obviously crucial for timely treatment commencement, and any disruption creates a backlog of undiagnosed cases, sustaining transmission within communities. Under the assumption that COVID-19 reduced case detection, our model found a small increase in the relative importance of recent transmission to driving the epidemic during this period. Over time, this accumulated backlog continues to exacerbate transmission as more cases remain undetected. Thus, while we cannot exclude that contact reduction due to COVID-19 measures like social distancing and lockdowns had a small impact on transmission, the steep decline in case detection likely played the predominant role in influencing TB notification numbers during the pandemic period. The wide credible intervals for the additional future TB cases and deaths likely reflect the uncertainty in estimating the impact of COVID-19, the compounding effects of increased transmission over time and the general uncertainty associated with our long-term projections Nevertheless, the possibility of a major adverse effect on the epidemic emphasizes the importance of restoring and strengthening TB case detection efforts to mitigate the pandemic’s lasting impact on TB control. This proactive approach not only manages the immediate effects of disruptions like those experienced during COVID-19 but also reduces long-term disease burden, saving more lives by ensuring that fewer cases go unnoticed and untreated. However, achieving this requires substantial efforts, increased investment, and strong commitment. Any major disruption now or into the future - whether due to another pandemic, other public health crises - could further exacerbate TB control challenges.

The modelled delay from active TB to diagnosis and treatment can be interpreted as being exponentially distributed with an average time to diagnosis of approximately 1.80 years, implying that many cases remain undetected. This delay may appear higher than prior studies which reported median diagnostic delays of 28 days in low- and lower-middle-income countries and 10 days in upper-middle-income countries [24]. In high-endemic areas, delays often exceeded 120 days, influenced by gender, rural residence, and health system inefficiencies [24, 25]. Variation in the definition of first healthcare-seeking behavior further complicates comparisons, with definitions ranging from visits to qualified providers to any contact with the NTP [24]. Additionally, asymptomatic TB constitutes a substantial proportion of cases, with 27.7 – 82.7% depending on symptom criteria [26, 27]. Similar prevalence is reported in subgroups like children and people with diabetes [28].

Our study has limitations stemming from uncertainties in TB epidemiology and natural history. While recent research provides estimates for progression rates from LTBI to active TB, LTBI reactivation remains difficult to observe [29, 30], and we relied on several estimates from low-burden settings or from pre-chemotherapy era, that may not be fully applicable to the context of Viet Nam. The difficulty in conducting longitudinal studies in high-burden areas further complicates the precision of these estimates, potentially affecting the generalizability of our findings [31]. Despite this, these estimates aligned well with observed data, offering a more accurate foundation than broad assumptions like a 5 - 10% lifetime risk [30]. Uncertainties in input parameters likely contributed to the wide range of future projections. Although we assumed that the average delay in diagnosis and treatment start would not change over the next ten years in our “status-quo” scenario, future NTP capacity expansion may reduce this delay and increase the case detection rate. Furthermore, our model implemented a synthetic social contact matrix from 2007, and an updated matrix would more accurately reflect current transmission dynamics, considering changes in social behavior, contact patterns, and demographics. While important, addressing drug-resistant (DR)-TB was beyond the scope of this study.

The major community-wide active case finding trial in Ca Mau province [32] provides opportunities for better understanding TB epidemiology in Vietnam and our modelling analysis can be considered in this context. The significant prevalence reduction observed during the trial period may suggest early progression may contribute more to TB cases than late reactivation, compared to our model estimates. However, we emphasize that our calibration process incorporated multiple epidemiological indicators, and our latency parameters - specifically early progression and late reactivation - were empirically derived, although all such estimates have their limitations. Additionally, we varied our latency parameters to some extent during calibration, and our supplementary materials indicate that early progression accounts for more than half of all cases in our model. While this discrepancy warrants further investigation, it presents a key opportunity for future research to refine our understanding of TB dynamics. Given the robustness of ACT3 as an epidemiological study, reconciling these differences could enhance the accuracy and applicability of our model to Viet Nam’s TB landscape.

Our study’s strengths include our optimization and calibration approach, which harnessed state-of-the-art publicly available software tools and successfully captured key indicators of the epidemic while also reflecting uncertainty around key input parameters. The model was informed by multiple relevant data streams, including results from the recent TB national survey. This allowed us to replicate the local TB burden and ensure that historical trends in case detection were properly reflected through TB notification data. Our pipeline, which was also applied to COVID-19 [17], can simulate a broad range of directly transmitted pathogens, and so is particularly applicable to TB. Combining domain-specific syntax with other libraries from the data science ecosystem, our framework ensures accuracy, transparency, and efficiency while supporting detailed calibration and uncertainty analysis.

CONCLUSIONS

Our study underscores the significant impact of COVID-19-related disruptions on TB case detection in Vietnam, leading to a backlog of undiagnosed cases that could sustain transmission for years. However, these impacts may be mitigated by enhancing case detection activities, which can improve early detection and reduce transmission risks. Any major disruption now or into the future - whether due to another pandemic, other public health crises - could further exacerbate TB control challenges. Addressing these risks requires sustained investment in diagnostic capacity, healthcare system resilience, and proactive policies to mitigate the long-term consequences of such disruptions on TB control.

DECLARATIONS

Corresponding author:

Viet Long Bui, School of Public Health and Preventive Medicine, Monash University, Melbourne, VIC, Australia. email viet.bui1{at}monash.edu