We used a previously developed individual-based model (11) to assess the impact of the COVID-19 pandemic between 2020 and 2022 on TB epidemiology in California, Florida, New York, and Texas, the four states with the highest number of incident TB cases (12). The model incorporated differences in demographic composition, immigration patterns, and natural history and epidemiological complexities of TB across the four states. The model accounted for immigration from eight different regions (Mexico, Americas, China, India, Asia, Africa, Europe, Rest), with each region having specific sizes of immigrant populations that were uninfected, having latent infection or having active disease at arrival. In the model, individuals could transition through four TB states: “Uninfected”, “LTBI” for latent TB infection, “Active TB” for active TB disease, and “Post-Treatment” once individuals are successfully identified and treated for active disease (see Figure 1).

Further model details can be found in the original paper and its supplementary materials (11), and model parameters are included in the online data supplement of this paper in Table E1. The model was calibrated to reflect the baseline (no-pandemic) trend of age-specific TB incidence among US-born and non-US-born populations between 1993 to 2015 across the four states, California, Florida, New York and Texas.

We assessed three potential mechanisms of COVID-19 pandemic-related disruptions and their impact on TB epidemiology: (1) Migration-driven impact accounts for the impact of disruptions to migration patterns into the US, resulting from countries around the world closing their borders and considerably reducing entries into the US. We model this as a reduction in the importation of TB and LTBI among newly arriving non-US-born individuals as a result of the reduction in immigration rates. (2) Transmission-driven impact accounts for the impact of TB transmission from disrupted social mixing patterns, as a result of stay-at-home orders and mask wearing, and is modeled as a reduction in rates of new TB infection and reinfection. (3) Access-driven impact accounts for the potential impact of disruptions in care-seeking and TB services provision, as a result of delays in diagnosis and resources being diverted to COVID-19 testing and management. This is modeled as delays in TB diagnosis and treatment initiation. Figure 1 shows where in the model the disruptions are assumed to act, based on which mechanism is disrupted by the pandemic.

We used publicly available data to quantify each mechanism of disruption. For each mechanism of disruption, we split the data over three periods: pre-pandemic period for data up to February 2020; pandemic period for data between March 2020 and March 2021; and post-pandemic period for data after March 2021 (where available). For simplicity, we assumed that each set of data informing the mechanisms of disruption was equally informative and important, therefore we did not apply any weighting to the different sources found.

We collated data on the volume of arrivals of new non-immigrant persons (non-immigrant visas) between 2009–2021, and registration of new, lawful permanent residents between 2018–2021, in each of the four states (13,14). Since there were no state-level differences in the level of disruption across the four states, we used the same parameter for the pandemic-related disruption on migration for each state (see Figure E1 in the online data supplement for details).

To quantify the pandemic-related disruption on mobility, we analyzed Google mobility data representing the population not staying home (in millions) between 2019–2023 and US government transit data representing the change in mobility of different categories of movement compared to pre-pandemic levels, across the four states between 2020–2022 (15,16). We did not find state-level differences in the changes to the rates of travel or mobility, so assumed that the pandemic-related disruption parameter would be the same across the four states (see Figure E2 in the online supplement for details).

Finally, we were unable to identify data that described the impact of the pandemic on access to TB services. We therefore quantified the pandemic-related disruption to access by analyzing data on cancer screenings and emergency department (ED) visits, as proxies. Although not TB-specific, these data were still representative of how the pandemic may have affected care-seeking behavior, for both routine (asymptomatic) visits and emergency (symptomatic) visits. We analyzed data pre-, during and post-pandemic (respectively pre-2020, 2020–2021 and post 2021 based on data availability, see Figure E3 in the online supplement for details) (17–20). These data were not available at the state-level, so we assumed similar trends of disruptions to care-seeking across the four states.

For these disruption parameters, we generated a relative trend of pandemic-related disruption (disruption covariate) by averaging the pandemic years’ data and dividing it by the pre-pandemic levels (up to the end of 2019), such that a value of “1” represented no disruption, and “0” represented full disruption. We extrapolated the observed trend for the remainder of the time period where data were not available (2022–2024 for migration, 2023–2024 for mobility, and 2021–2024 for access), by making optimistic and pessimistic assumptions about when trends might return to pre-pandemic levels, but assuming they’d all return to pre-pandemic levels by the end of 2024, latest (see Figure 2 and Table E2 in the online supplement for the average value of each covariate). The optimistic and pessimistic estimates were chosen to reflect estimates from simulations that assumed optimistic versus pessimistic values for the end of pandemic-related disruptions.

We used a likelihood-based approach along with monthly state-level TB treatment initiation data among US-born and non-US-born individuals between 2020 and 2023 across each of the four states (Figure 3) to infer each of the three hypothesized impacts of pandemic-related disruption: migration-driven, transmission-driven, and access-driven impacts.

We drew 100,000 combinations of the three impact parameters – transmission-driven impact (Δβ), migration-driven impact (Δm), and access-driven impact (Δω) – using Latin Hypercube sampling, allowing each of the three impact parameters to vary between [0–1], where a value of “0” for a proposed mechanism indicates the maximum impact on TB via the proposed mechanism of disruption, and “1” indicates no impact. Each impact parameter was exponentiated by their respective time-varying disruption covariate – covtranst,covmigrt,covcaret, for transmission, migration, and access, respectively – so that we could account for both the magnitude of the disruption, and the way this magnitude of disruption changed through the pre-, during and post-pandemic periods. This was represented in the updated model such that the updated, effective rates were: βeff(t)=β0·Δβ(1-covtranst) meff(t)=m0·Δm(1-covmigrt) ωeff(t)=ω0·Δω(1-covcaret) where βeff(t)meff(t) and, ωeff(t) represent the effective transmission, migration, and treatment-initiation rates, respectively, and β0,m0, and ω0 represents baseline (pre-pandemic) transmission, migration, and treatment-initiation rates, respectively.

For each combination of parameters, we compared monthly TB notification rates and model simulations using a binomial likelihood function. To identify a range of parameter values that best explained reported data, we generated likelihood profiles for each of the three parameters. For each parameter, we constructed a likelihood profile by maximizing the likelihood function over the remaining two parameters while holding the parameter being profiled constant. The 95% uncertainty interval (UI) was estimated for each parameter by finding the range of parameter values whose likelihood estimates were within 1.92 log-likelihood units of the estimated maximum, on the basis of univariate confidence limits using the χ² distribution with one degree of freedom (21).

In each of the four states, we simulated the model between 2020 and 2032 under two scenarios, no-pandemic and pandemic. In the no-pandemic scenario, we generated model projections assuming that there was no COVID-19 pandemic. In the pandemic scenario, we generated projections incorporating the three hypothesized mechanisms of impact based on the parameter ranges that were found to be consistent with the data. The projections were based on the epidemiological parameters of the baseline model (11), including the continuation of current immigration trends and the continuing decline in transmission rates as estimated in the baseline model. We conducted five replicated simulations of the model with each of the parameters consistent with the data, and estimated the mean/median and 95% UI from the simulations.

The trend of lawful non-immigrant entries saw a sharp drop in the first quarter of 2020, followed by a slow trend upwards over the following year, still quite far from pre-pandemic levels. Between April 2020 and July 2021, the volume of arrivals of new immigrants was 75–90% lower compared to the average volume in 2019 (pre-pandemic). New lawful immigrants also saw a decline at the beginning of 2020, with a 15–20% drop in 2020 and 2021, across the four states. These data, however, suggest that trends returned to pre-pandemic levels in 2022 (Figure 2A). The migration covariate, which was constructed by aggregating the two forms of data found, reached as low as 0.1 (i.e., 90% reduction, 95% uncertainty interval 0.05–0.5) during April-July of 2020, and was assumed to gradually return to 1 by the end of 2023.

The mobility data showed moderate drops at the beginning of the pandemic, as a result of stay-at-home orders, reduced travel and movement. The data showed a 25–50% drop in volume in 2020 compared to 2019. Both forms of data showed mobility slowly returning to pre-pandemic levels in 2021. However, the volume of mobility remained far from pre-pandemic levels through 2022 and into 2023 (Figure 2B and Figure E2 in the online supplement for details). The aggregated covariate for mobility declined to the lowest level of 0.6 at the end of 2020, and was projected to return to pre-pandemic levels by mid-2024.

Finally, the data on access to care-seeking trends suggests that a modest decline occurred at the beginning of the pandemic. Data on various cancer screening showed a 10% reduction through 2020 in the proportion of individuals recently screened for breast or cervical cancer, compared to 2019; while the number of mammograms performed dropped by 100% in April-May of 2020 compared to 2019, and returned to pre-pandemic levels by the Fall of 2020. Data from EDs showed a 50% decrease in volume in April 2020 compared to 2019, then started trending upwards towards pre-pandemic levels by the end of 2020. Most forms of data saw a quick return to pre-pandemic levels by the end of 2020 (Figure 2C and Figure E3 in the Online Supplement for details). The aggregated covariate declined to 0.3 in April 2020, then was projected to return to pre-pandemic levels by 2021.

Figure 3 shows that TB treatment initiation trends, used as a proxy for TB notifications, were in a stable decline between 2009–2019, prior to the COVID-19 pandemic. In April 2020, as a result of the pandemic-related disruptions, notifications dropped by 30% in California, (from an average between the end of 2019 and March 2020 of 155 to 108 in April 2020), 20% in Florida (from an average of 38 to 30), by 55% in New York (from an average of 57 to 26), and 30% in Texas (from an average of 84 to 59). The linear trends in TB treatment initiations during the pandemic months (March 2020-March 2021) were all below the pre-pandemic levels in the four keys states: 25% in California, 30% in Florida, 20% in New York, and 30% in Texas. During the post-pandemic period, after March 2021, the linear trend of treatment initiations returned to pre-pandemic levels.

The likelihood profiles in Figure 4 allowed us to assess the strength of the evidence used to inform each mechanism disrupted by the COVID-19 pandemic. The results from the model inference suggested that the impact of all the disruptions could range from strong to no impact (the 95% confidence intervals (CI) for the likelihoods range from one, no impact, to smaller values (less than one), representing a reduction from baseline). The profiles for migration and mobility show a very homogenous distribution of the likelihood, across all values of disruption. For both of these, the 95% uncertainty intervals are [0.005–1] and therefore include both extreme values of the impact. Because of this, we were unable to statistically attribute the changes to TB notifications from the National Tuberculosis Surveillance System (NTSS) data to one specific mechanism. For the care-seeking profile, we saw a stronger signal in favor of values in the uncertainty interval [0.38–1]. However, because this interval also included values of “1”, it was still not possible to statistically attribute changes to the TB notifications to this mechanism. These results were consistent for both the aggregated profiles and the state-specific profiles, shown in Figure E5 in the online supplement.

Figure 5 shows that our model was able to capture the small dip in monthly notifications reported by the NTSS data through 2020, but both the data and the model simulations seemed to return to pre-pandemic levels in 2021.

The final part of the analyses included comparing TB projections over the next 10 years, up to 2032, with and without disruptions from the COVID-19 pandemic. Our revised model projections, which considered a range of potential impacts of the pandemic that were consistent with the monthly treatment initiation data (i.e., the impact parameters values that were within the estimated 95% UI including values of “1”, indicating no pandemic impact), were similar to our projections from the no-pandemic model. The difference in the average number of annual cases between the revised model and the no-pandemic model ranged between −70 to 70 (Figure 6).

Under the pandemic scenario, between 2020 and 2032, the TB incidence of the overall population was projected to decrease from 5.5 per 100,000 to 4.4 in California; from 3.0 per 100,000 to 2.0 in Florida; from 4.0 per 100,000 to 3.2 in New York; and from 3.0 per 100,000 to 2.5 in Texas.