Achieving a “step change” in the tuberculosis epidemic through comprehensive community-wide intervention: a model-based analysis

Background Global progress towards reducing tuberculosis (TB) incidence and mortality has consistently lagged behind the World Health Organization targets leading to a perception that large reductions in TB burden cannot be achieved. However, several recent and historical trials suggest that intervention efforts that are comprehensive and intensive can have a substantial epidemiological impact. We aimed to quantify the potential epidemiological impact of an intensive but realistic, community-wide campaign utilizing existing tools and designed to achieve a “step change” in the TB burden. Methods We developed a compartmental model that resembled TB transmission and epidemiology of a mid-sized city in India, the country with the greatest absolute TB burden worldwide. We modeled the impact of a one-time, community-wide screening campaign, with treatment for TB disease and preventive therapy for latent TB infection (LTBI). This one-time intervention was followed by the strengthening of the tuberculosis-related health system, potentially facilitated by leveraging the one-time campaign. We estimated the tuberculosis cases and deaths that could be averted over 10 years using this comprehensive approach and assessed the contributions of individual components of the intervention. Results A campaign that successfully screened 70% of the adult population for active and latent tuberculosis and subsequently reduced diagnostic and treatment delays and unsuccessful treatment outcomes by 50% was projected to avert 7800 (95% range 5450–10,200) cases and 1710 (1290–2180) tuberculosis-related deaths per 1 million population over 10 years. Of the total averted deaths, 33.5% (28.2–38.3) were attributable to the inclusion of preventive therapy and 52.9% (48.4–56.9) to health system strengthening. Conclusions A one-time, community-wide mass campaign, comprehensively designed to detect, treat, and prevent tuberculosis with currently existing tools can have a meaningful and long-lasting epidemiological impact. Successful treatment of LTBI is critical to achieving this result. Health system strengthening is essential to any effort to transform the TB response. Supplementary Information The online version contains supplementary material available at 10.1186/s12916-021-02110-5.


S-1 Model details
The model we developed evaluates the impact of a comprehensive one-time intervention and follow-up health system strengthening. It was structured to take into account natural history and transmission of TB, including buden of subclinical TB and future progression of LTBI, and aging of the population and screening of the population based on age. The model is described and schematically presented in the main text. Here, we provide the mathematical expressions of the ordinary differential equations that describe the model in the entirety. Let X {i,j,k} be the number of individuals with TB status i; where i ∈ {Uninfected, Early LTBI, Late LTBI, Asymptomatic Active TB, Symptomatic Active TB, Recovered}, living in setting j; where j ∈ {high-risk, low-risk}; and in age group k; where k ∈ {0 − 14, 15+}.
The following set of ordinary differential equations describe the model. The model parameters are described in The forces of infection that individuals are subject to in the high-and low-risk settings are given by the following equations.
Here, β is the baseline transmission rate at time t 0 (reference: symptomatic adults in the high-risk population in the year 2000), β ∆ the rate of declines in transmission rate, N h and N l are the population size of high-and lowrisk populations, respectively, σ is the mixing rate between the two populations, and T h and T l are transmission potential generated from high-and low-risk populations, respectively, as described below: high-risk: + X {i=Symptomatic Active TB,j=high-risk,k=15+} low-risk:

S-2 Model calibration
The calibration process aimed to capture key demographic and epidemiological features of TB in the urban Indian setting. We considered seven demographic and epidemiological measures for model calibration and identified a data-consistent target range for each measure, as listed in Table 1 in the main text. To calibrate the model, we first used Latin Hypercube Sampling to sample 500,000 sets of model parameter values describing TB natural history and standard of care, from ranges shown in Table S-1. For each parameter set, we simulated the model for 520 years; the first 500 years were used to bring the model to equilibrium, and the final 20 years' worth of simulations, representing the time period from 2000 to 2020, were recorded for model calibration. We assumed that transmission rate was fixed (i.e., no decline in transmission rate) for the first 500 years of simulation. For the final 20 years, we allowed transmission rate to decline (i.e., β ∆ ≥ 0), to capture decline in TB incidence. Simulations which yielded model outputs that were consistent with all calibration targets were selected, such that the calibrated model consisted of equally weighted samples of simulations in which all of the model outputs considered for calibration were within their respective calibration target ranges. Simulations of the calibrated model are provided in Fig S-

S-3 Modeling the effects of the one-time intervention
The one-time intervention was modeled as moving specified proportions of each latent or active TB compartment to the corresponding recovered state. The proportions were estimated as a product of proportions who would receive or successfully complete each step of the intervention, as follows: For those requiring TST, we use same reading rate and sensitivity as for adults. We also assume the same TPT uptake, completion, and efficacy as for adults. All child contacts <5y are referred for TPT. Proportion of child contacts with active TB who are diagnosed with active TB Treatment uptake and outcomes assumed same as for adults.
The proportion of children reached by the intervention was estimated by first estimating what fraction of children with a given TB status were close contacts of an adult with current active tuberculosis that could potentially be identified by the intervention, and then multiplying this by the proportion of adults with active TB whose TB was detected by the intervention (regardless of whether the adult initiated or completed treatment).
We first assumed that 20% of TB transmission to children occurs without households [38]. Then, for children with current active TB, we assumed that 50% had acquired their TB infection from an index case who still had undiagnosed active TB, and that the remainder had been infected by someone who was already treated, resolved, or deceased at the time of the adult case-finding intervention. This resulted in an estimate that 10% of children with active TB could potentially be reached by the intervention.
For children with early LTBI, we reduced the proportion with a currently-active case to 15%, based on the longer (5 year) modeled duration of early LTBI relative to the duration of active TB, such that most index cases would no longer have active disease. This resulted in an estimate that 3% of children with early LTBI could potentially be reached by the intervention. Finally, for children with late latent LTBI, we assumed that the prevalence of active TB among their adult contacts was equal to the overall prevalence of active TB among adults in that child's subpopulation (high-risk or low-risk subpopulation). We estimated that each child had close contact with an average of 2 adults, such that the probability that a child had contact with an adult case was equal to twice the prevalence of active TB among adults. This probability was estimated at the time of the intervention for each subpopulation.

S-10
S-4 Modeling the effects of medium-term health system strengthening Our compartmental transmission model included only a treatment rate parameter (ω) that applied to all symptomatic TB, and a treatment success probability (k) for those who initiated treatment.
We conceptualized ω as consisting of multiple components: an average time to care-seeking once symptomatic (t 1 ), an average time to diagnosis and initiation of treatment (t 2 ), and a probability of pretreatment loss to follow-up (p). Thus, ω = (1 − p)/(t 1 + t 2 ).
We estimated a value of 1 month for t 2 [39,40,41] and 16% for p [42]. We modeled health system strengthening as reducing each of the following by a factor m: • Time to seek care once symptomatic (t 1 ) • Time to diagnosis and treatment initiation (t 2 ) • Probability of pretreatment loss to follow up (p) • Probability of poor treatment outcomes (1 − k) Solving for t 1 in terms of ω, p, and t 2 , and applying factor m to each of t 1 , t 2 , p, and 1 − k, we modified the k and ω parameters as follows: and

S-5 Sensitivity analyses S-5.1 Sensitivity analyses of TB cases averted
We examined impact of variation in model parameters to the secondary outcome, TB cases averted over 10 years by a combined intervention of a one-time campaign plus health system strengthening. As shown in Fig. S-2, the median TB cases averted per 1 million corresponding to parameter values in the top and bottom deciles differed by more than 1,950 (25% of median) for only two of the modeled parameters: (i) reactivation rate for adults (where the difference in the outcomes corresponding to top and bottom deciles was 2,015 cases), and (ii) rate of spontaneous resolution (2,033 cases). This indicates that when (i) LTBI cases have higher expected lifetime risk of reactivating, and (ii) a higher proportion of TB cases are not diagnosed through standard TB care, the intervention is likely to be more impactful in averting potential TB cases.  Colored level-surfaces indicate additional impact on TB incidence of including health system strengthening measures with a one-time campaign, relative to the impact of the one-time campaign alone, assuming 70% coverage with one-time intervention, and a specified percentage reduction in unsuccessful treatments (x-axis) and diagnostic delays (y-axis). Red cross in panels B-D indicates the reference scenario.

S-5.2 Comparing the impact of curing LTBI versus TB disease
We compared the impact of active case finding and preventive therapy, by comparing the number of cases averted by successfully treating one case of LTBI vs. one case of TB. The number of cases averted by successfully treating one case of LTBI was estimated by dividing the total number of cases averted through preventive therapy (shown in Fig. 3, blue lines) by the total number of individuals successfully treated for LTBI as a result of preventive therapy campaign. We estimated that 30.3% of individuals with LTBI were cured through preventive therapy, as shown in Fig. 2, and the median LTBI prevalence was 39% (95% range: 31%-48%) as shown in Fig.S-1-I. Similarly, the number of cases averted by successfully treating one case of TB was estimated by dividing the number of cases averted through ACF (shown in Fig. 3, yellow lines) by the number of individuals successfully treated for TB disease as a part of ACF campaign. Approximately 40% of individuals with active TB disease was cured through ACF, as shown in Fig. 2, and the median prevalence of TB disease was 260 (95% range: 210-300) per 100,000, as shown in Fig. 1-A.
We note that the impact of treating LTBI is realized slowly over time, and is about 10-30 times smaller than that of treating TB in a per capita basis (Fig. S-3).

S-5.3 Targeting the one-time intervention
We explored the impact of targeting the intervention to the high-risk population. We considered a scenario in which individuals in the high-risk population were screened preferentially to those in the low-risk population, where the the odds ratio, i.e., ratio of the odds of screening in the high-risk population to the odds of screening in the low risk population, was 5:1. The impact of this targeted one-time intervention (without health system strengthening) was modestly larger compared to the untargeted scenario presented in the main text. The cumulative TB-related deaths averted after 10 years was 870 (655 -1,090) compared to 809 (612 -1,010); and the cumulative cases averted was 6,090 (4,310 -7,850) compared to 5,840 (4,060 -7,650) per 1 million population. (See Fig. S-4.) Figure S-4: The impact of a one-time intervention (without health system strengthening), when the intervention was targeted to the high-risk population. Shown in (A) and (B), respectively, are TB incidence rate and TB-related mortality rates, per 100,000 per year between 2000 to 2040, in model simulations without the intervention (grey), and the simulations with the intervention implemented in 2020 (red). Shown in (C) and (D) are percentage reductions in TB incidence and TB-related mortality rates, respectively. Shown in (E) and (F) are, respectively, cumulative number of TB cases and TB-related deaths averted per 1 mil. population.

Figure S-5:
The impact of a one-time intervention (without health system strengthening), with shorter duration of early LTBI. Shown in (A) and (B), respectively, are TB incidence rate and TB-related mortality rates, per 100,000 per year between 2000 to 2040, in model simulations without the intervention (grey), and the simulations with intervention implemented in 2020 (red). Shown in (C) and (D) are percentage reductions in TB incidence and TBrelated mortality rates, respectively. Shown in (E) and (F) are, respectively, cumulative number of TB cases and TB-related deaths averted per 1 mil. population.

S-19
Supplementary Materials Shrestha et al.

S-5.5 Preventive therapy targeted to recent infections
For these analyses, we assumed that as a part of the comprehensive intervention, preventive therapy was only provided to individuals in Early LTBI compartment, i.e., individuals with recent exposure. Compared to the full intervention in which preventive therapy is provided to all LTBI, this intervention resulted in lower impact: a median 35% less cases and 11% less deaths averted after 10 years of intervention. However, the number of individuals receiving preventive therapy during this targeted intervention was about one-tenth compared to the full intervention.
Figure S-6: The impact of the full intervention, when preventive therapy is limited to recent infections. Shown in (A) and (B), respectively, are TB incidence rate and TB-related mortality rates, per 100,000 per year between 2000 to 2040, in model simulations without the intervention (grey), and the simulations with intervention implemented in 2020 (red). Shown in (C) and (D) are percentage reductions in TB incidence and TB-related mortality rates, respectively. Shown in (E) and (F) are, respectively, cumulative number of TB cases and TB-related deaths averted per 1 mil. population.