The positive externalities of migrant-based TB control strategy in a Chinese urban population with internal migration: a transmission-dynamic modeling study

Shanghai has experienced major rural-to-urban migration since the 1990s (Additional file 1: Figure S1) [15]. By 2016 the Shanghai population totaled 24 million, 41% of whom were internal migrants [16]. Songjiang has a population of 1.77 million, of which 62% are internal migrants, defined as people without a Shanghai household registration status through Chinese hukou system. Most originate from western and central China. In a previous study during 2006 to 2008, migrants had higher tuberculosis rates (38.9/100,000 population) compared to local residents (27.8/100,000 population) in Songjiang (Additional file 1: Figure S2) [6]. Since the 1990s, Songjiang has participated in a comprehensive TB surveillance and reporting system involving Songjiang District Centers for Diseases Control and Prevention, a TB-designated hospital, and community-based clinics. This network provides TB services including passive case finding, diagnosis, treatment, and follow-up. Data on TB cases, including demographic, epidemiological, and clinical information, is recorded in an electronic TB Information Management System.

Currently, Songjiang applies a referral system based on TB-like symptoms to prompt TB diagnosis and notification. Although there was no specific national guideline for TB control among internal migrants, since late 2004, an enhanced program was implemented in Shanghai to extend free TB services (reimbursement for diagnosis and anti-TB drugs) to the migrant population. Despite this expansion of standard TB care policy, there was no specific routine TB screening among migrants in Songjiang. Many local manufactory industries request a general pre-employment physical examination for both migrant and resident applicants, including chest X-ray. Residents aged 65 years and above can receive a voluntary annual physical examination, including a chest X-ray examination (not specific for TB screening).

We developed an age-structured transmission-dynamic model of TB epidemiology in Songjiang District, adapted from published models (Fig. 1) [17, 18]. The model stratifies the Songjiang population into four dimensions, with a core dimension representing TB transmission, natural history, diagnosis, and treatment (Fig. 1a and Fig. S4). Additional dimensions represented differences in population demography, TB burden and susceptibility, mixing patterns, and TB care access by age, sex, and migrant status (Fig. 1b–d). Migrant populations were stratified according to the duration of stay in Shanghai (Fig. 1d), with immigration and emigration modeled explicitly. Resident individuals entering Songjiang from other districts of Shanghai were not considered “migrants” for the analysis, as their epidemiological profile is expected to be similar to those of current residents and they held the same household registration status as the Songjiang local residents. We did not simulate transmission dynamics in rural areas and calibrated TB burden in incoming migrants to match observed disease patterns in Songjiang. The full model structure results from the crossing of the four dimensions and is represented with many distinct compartments and processes and the model code was detailed in Additional file 2.

The base-case scenario assumed continuation of current TB control activities, represented by current rates of treatment-seeking and current TB care outcomes. We compared this to hypothetical scenarios describing potential expansions in TB control. Table 1 describes model scenarios and operationalization [33‐40]. In brief, scenarios one and two focus on improving TB prevention and care for elderly residents; scenarios three, four, and five target improvements in TB prevention and care for migrants. Scenario six represents the combination of migrant-focus interventions, and scenario seven represents the combination of all interventions included in other scenarios. We assumed any changes to current services would be introduced over 3 years, 2020–2022, then remain in place for the rest of the analysis period.

The model was programmed in R and C++ [41]. We used Bayesian evidence synthesis [42] to combine data and calibrate parameters to reproduce observed time-series of population and TB cases by age, sex, and migrant status. Calibration was implemented using Incremental Mixture Importance Sampling [43, 44], which produces a large number of epidemiological trajectories consistent with prior parameter distributions and calibration data (Additional file 1: Tables S4–5) [14]. We used the fitted model to simulate future outcomes under each analytic scenario. We calculated point estimates as the mean value across all simulated epidemiological trajectories and calculated 95% uncertainty intervals of the distribution of simulation results.

In sensitivity analyses, we tested the impact of alternative assumptions of a 5% annual increase and 5% annual decrease in future migrant population size, which was held flat in the main analysis. Secondly, we tested alternative assumptions of 0% and 6% annual decline in LTBI and TB prevalence among future entering migrants, as compared to the 3% decline assumed in the main analysis. Thirdly, we assessed the impact of an extended 30-year time horizon (to 2050), as compared to 15 years used for the main analysis. Table S6 in the additional file 1 describes the additional parameters used for model sensitivity analysis [33‐40].

Recent population data show rapid population growth (from 0.1 to 1.1 million between 2000 and 2014, Fig. 2a), and a much younger age distribution (Fig. 2b) for migrants in Songjiang, relative to the resident population. Similar trends were observed in the distribution of TB notifications over time (Fig. 2c) and age (Fig. 2e, f), with migrant cases clustered in the 20–35 age group while resident cases demonstrated peaks at 20–35 and 60–80 years-old. By 2016 migrants represented approximately 60% of the overall population and 74% of all reported TB cases in Songjiang, had a higher notification rate (43.5 per 100,000) compared to that among resident (24.3 per 100,000), which were consistent with previous reports [6, 14]. Additional detail on these demographic and epidemiological trends is shown in Figs. S1–3 [6, 15, 16]. The model was able to reproduce these and other recent epidemiological data, including the TB case fatality estimates by age (Fig. 2d).

Figure 3 presents estimates for five key outcomes in 2020—effective Mtb transmissions by individuals with TB, new infections, incident TB cases, TB deaths, and years of life lost (YLL) from TB—stratified by age group and migrant status. Migrants in the 20–35-year-old age group represent a major fraction of each of these outcomes, reflecting their share of the population as well as elevated TB burden. Migrants were estimated to contribute 87.0% [95% uncertainty interval: 83.7, 89.9] of all instanced of TB transmission, and represent a smaller fraction (78.6% [74.4, 82.3]) of those newly infected, reflecting greater migrant-to-resident than resident-to-migrant transmission. Transmission from migrants was estimated to produce 91.7% [89.6, 93.9] of new infections among migrants and 69.9% [65.0, 73.6] of new infections among residents. The estimated fraction of all new infections caused by an individual of the other group (migrant to resident or resident to migrant) fitted well with the observed value (model 20.5% [18.8, 22.1], data 21.2%, supplementary materials). Residents represented an increased share of incident TB cases and 47.4% [42.3, 52.1] of TB deaths, reflecting elevated rates of reactivation TB and higher TB case fatality in the elderly resident population. Due to the younger average age, there was an estimated 35.2 [33.8, 36.7] YLL per TB death among migrants, compared to 16.5 [15.6, 17.4] YLL among residents. Migrants represented 70.2% [66.0, 74.4] of all life-years lost due to TB death.

Figure 4 reports the fraction of new TB cases in 2020 due to LTBI reactivation, recent Mtb infection, and prevalent TB among individuals migrating into Songjiang, stratified by age and migrant status. For both migrants and residents, LTBI reactivation was estimated to represent the large majority of TB cases for individuals > 50 years old and was an important source of new TB cases, representing 72.2% [67.9, 76.8] of resident TB cases and 48.0% [42.9, 54.3] of migrant TB cases. Recent infection contributed 26.9% [23.2, 32.1] of resident TB cases and 42.9% [38.6, 45.8] of migrant TB cases, and was the primary cause of TB among individuals < 30 years old. Prevalent TB at entry to Songjiang contributed only 0.8% [0.5, 1.4] and 9.1% [7.0, 11.3] of TB cases among residents and migrants respectively. Overall, recent infection was estimated to cause 38.0% [33.3, 42.3] of incident TB cases overall.

Figure 5 shows base-case projections of population and TB incidence. While the total population is projected to stabilize following rapid growth during 2000–2015 (Fig. 5a), TB cases are projected to decline. TB cases among the migrant population are projected to decline to 18.4 [17.1, 23.2] per 100,000 by 2035, a 49% [36–53] reduction from 2016. By 2035 resident TB cases are projected to decline to 11.6 [10.6, 14.4] per 100,000 (Fig. 5c), a 51% [39–55] reduction from 2016. These declines are consistent with recent TB case trends (Fig. 5, points). Migrants are projected to represent a relatively stable share of total TB cases, equivalent to 63% of total TB cases while representing 53% of the 2035 population.

Table 2 provides base-case projections for other TB outcomes. These outcomes are projected to follow similar trends to new TB cases, with TB transmission, new infections, TB deaths, and YLL due to TB all declining log-linearly over the projection period.

Figure 6 shows incremental outcomes for the targeted interventions described by scenarios 1–5 compared with the base-case, summed over 2020–2035, and stratified by age and migrant status. Among these scenarios, interventions directed at the migrant population (particularly scenarios 4 and 5) are projected to have the greatest impact on most outcomes, with benefits concentrated in younger age groups consistent with the migrant age distribution. An exception to this pattern is for TB deaths, where scenario 2 is projected to have a major impact on TB deaths in the elderly resident population. In all cases, interventions targeted at migrants are projected to produce beneficial secondary effect for residents. In contrast, interventions targeted at residents are projected to have minimal impact on migrant TB outcomes.

Table 3 summarizes incremental reductions in TB outcomes for each scenario by 2035. In general, outcomes related to TB transmission (Mtb transmission, new Mtb infections) are more sensitive to intervention than other outcomes, with the combination of all interventions (scenario 7) projected to reduce these outcomes by 20–23% by 2035 compared to the base-case, while estimated reductions in other outcomes ranged from 10 to 12%.

Sensitivity analyses representing higher future migrant populations resulted in a greater effect of migrant-focused interventions relative to resident-focused interventions, and similar results were seen for sensitivity analyses representing slower declines in TB burden among future entering migrants. Even for sensitivity analyses representing faster future declines in migration volume and migrant TB burden, interventions focused on migrants still produced the greatest reductions in TB incidence and beneficial spillovers for resident populations. Results observed over a longer time horizon were consistent with the main analysis, with ongoing log-linear declines in TB burden when outcomes were projected to 2050. Over the long-term, intervention scenarios were shown to produce a step change in TB transmission, incidence and death rates, with this percentage difference with the base case increasing slowly over the extended simulation period. Detailed results for sensitivity analyses are shown in Fig. S5 and Table S7.